Controllability results for cascade systems of m coupled N -dimensional

Stokes and Navier-Stokes systems by N − 1 scalar controls

arXiv:2108.09856v1 [math.AP] 22 Aug 2021

∗1 †2
Takéo Takahashi , Luz de Teresa , and Ying Wu-Zhang‡2
1
Université de Lorraine, CNRS, Inria, IECL, F-54000 Nancy, France
2
Instituto de Matematicas, Universidad Nacional Autónoma de México, Circuito Exterior, C.U.,
C. P. 04510 Ciudad de México, México

August 24, 2021

Contents
1 Introduction 2

2 Preliminaries 4

3 The Carleman estimate for the adjoint system 6

4 Proof of the main results 15

4.1 Final state observability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Proof of Theorem 1.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 Proof of Theorem 1.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Abstract
In this paper we deal with the controllability properties of a system of m coupled Stokes systems or m coupled
Navier-Stokes systems. We show the null-controllability of such systems in the case where the coupling is in
a cascade form and when the control acts only on one of the systems. Moreover, we impose that this control
has a vanishing component so that we control a m × N state (corresponding to the velocities of the fluids) by
N − 1 distributed scalar controls. The proof of the controllability of the coupled Stokes system is based on a
Carleman estimate for the adjoint system. The local null-controllability of the coupled Navier-Stokes systems is
then obtained by means of the source term method and a Banach fixed point.

Keywords: Null controllability, Navier-Stokes systems, Carleman estimates

2010 Mathematics Subject Classification. 76D05, 35Q30, 93B05, 93B07, 93C10
∗ [email protected]
† [email protected]. Partially supported by Conacyt, project A1-S-17475
‡ [email protected]

1
1 Introduction
Controllability issues related to a single parabolic equation or to a single Stokes or Navier-Stokes system have been
intensively studied in the last fifty years giving rise to interesting techniques, new challenges and open problems.
See some seminal results [12,16,25] for the heat equation and [8,14,21] for the Navier-Stokes system. The literature
is vast and it is difficult to mention all the intensive studies about this subject. However, it is only in the last fifteen
years that the challenging issue of controlling coupled parabolic systems has attracted the interest of the control
community. This kind of systems appears mathematically in optimal control theory as a characterization of the
optimal control (with one equation coupled to its adjoint) but also appears, for example, in the study of chemical
reactions (see e.g. [11], [7]), and in a wide variety of mathematical biology and physical situations (see e.g. [20] ). In
the case of scalar (heat) coupled equations an important number of challenging problems has been solved (see [1] for
a survey of results until 2011) and sometimes the results have been surprising [2–4]. In the case of coupled Stokes
or Navier-Stokes systems, to our knowledge, only some cases of two coupled systems have been treated [5, 6, 18, 29].
Here our aim is to generalize results for a m scalar cascade system [17] to a m N -dimensional Stokes or Navier-Stokes
cascade system but including an extra deal: to eliminate one component on the N -dimensional control.
Let us be more specific: we consider a bounded domain Ω of RN (N = 2, 3) whose boundary ∂Ω is regular enough.
Let T > 0 and let ω ⊂ Ω be a (arbitrary small) nonempty open subset which will usually be referred as the control
domain. We will use the notation Q = Ω × (0, T ) and Σ = ∂Ω × (0, T ).
In this article, we are interested in the null controllability of a coupled system of m Stokes or Navier-Stokes
systems, with m > 2:

Pm

 ∂t y (i) − ∆y (i) + ∇p(i) + ε y (i) · ∇ y (i) = Ai,j y (j) + Di v1ω in Q, (1 6 i 6 m)


 j=1
∇ · y (i) = 0 in Q, (1 6 i 6 m)

 (i)

 y = 0 on Σ, (1 6 i 6 m)
 (i) (i)
y (·, 0) = y0 in Ω, (1 6 i 6 m)

with ε = 0 for Stokes systems and ε = 1 for Navier-Stokes systems, where Ai,j ∈ MN (R) and where Di ∈ MN,r (R)
for some r ∈ N∗ . We have denoted by 1ω the characteristic function of ω. Let us notice that we take in this work
the viscosities of the fluids constant and equal to 1 to simplify.
We can write the above systems in a more compact way as


 ∂t y − ∆y + ∇p = Ay + Dv1ω in Q,

∇ · y = 0 in Q,
(1.1)

 y=0 on Σ,


y|t=0 = y0 in Ω,
or 
∂t y − ∆y + (y · ∇)y + ∇p = Ay + Dv1ω


in Q,
∇ · y = 0 in Q,
(1.2)
y = 0
 on Σ,


y|t=0 = y0 in Ω,
where we set

 
y = y (1) , . . . , y (m) , p = p(1) , . . . , p(m) , y0 = y0(1) , . . . , y0(m) ,



∆y = ∆y (1) , . . . , ∆y (m) , ∇p = ∇p(1) , . . . , ∇p(m) ,





(y · ∇) y = y (1) · ∇ y (1) , . . . , y (m) · ∇ y (m) , ∇ · y = ∇ · y (1) , . . . , ∇ · y (m)
and  
P
m P
m

Ay = A1,j y (j) , . . . , Am,j y (j) Dv = D1 v, . . . , Dm v .
j=1 j=1

2
 In this work, we will focus in the particular case where the partitioned matrix A has the form
 
A1,1 A1,2 A1,3 ... A1,m
A2,1 A2,2 A2,3 ... A2,m 
 
 0 A3,2 A3,3 ... A3,m 
A=  (1.3)
 .. .. .. .. .. 
 . . . . . 
0 0 . . . Am,m−1 Am,m
with all the blocks under the diagonal non zero and such that all its blocks are scalar matrices. More precisely, our
hypotheses on A are
Ai,j = ai,j IN , ai,i−1 6= 0 (2 6 i 6 m), ai,j = 0 if i > j + 2. (1.4)
We also control (1.1) or (1.2) by acting only on one of the Stokes or of the Navier-Stokes systems, for instance the
first one, and with N − 1 scalar controls on this system. Thus, without loss of generality, we assume
 
  1 0
1
r = N − 1, Dj = 0 (j > 2), D1 = (if N = 2) or D1 = 0 1 (if N = 3). (1.5)
0
0 0
The above choice of the matrix A corresponds to a particular coupling considered in the context of the null-
controllability of systems of m linear heat equations, see [17] and our aim is to extend this result in the case of
coupled of Stokes or Navier-Stokes systems.
In order to state our main results, we recall some standard functional spaces associated with the Stokes system:

H = y ∈ L2 (Ω)N : ∇ · y = 0 in Ω, y · n = 0 on ∂Ω (1.6)
and 
V = y ∈ H01 (Ω)N : ∇ · y = 0 in Ω . (1.7)
Our main result is the following theorem.
m
Theorem 1.1. Assume (1.4) and (1.5). Then, for any T > 0 and for any
2 2 N −1 (1) (m)

y0 ∈ H , there exists a control
v ∈ L (0, T ; L (ω) ) such that the corresponding solution y = y , ..., y to (1.1) satisfies
y(·, T ) = 0 in Ω.
Remark 1.2. As a consequence, we deduce that we can control the system (1.1) of N × m scalar equations with
N − 1 scalar controls.
From Theorem 1.1 and a general method to deal with the controllability of nonlinear parabolic systems, we
deduce the local null controllability of the system (1.2):
Theorem 1.3. Assume (1.4) and (1.5). Then, for any T > 0, there exists δ > 0 such that, for any y0 ∈ V m
satisfying
ky0 kV m 6 δ,
there exists v ∈ L2 (0, T ; L2(ω)N −1 ) such that the corresponding solution y to (1.2) satisfies
y(·, T ) = 0 in Ω.
In order to prove Theorem 1.1, we introduce the adjoint system:

 −∂t ϕ(i) − ∆ϕ(i) + ∇π (i) = P Aj,i ϕ(j)
 i+1

 in Q, (1 6 i 6 m − 1)

 j=1


 −∂ ϕ(m) − ∆ϕ(m) + ∇π (m) = P A ϕ(j)
 m
t j,m in Q,
j=1 (1.8)



 ∇ · ϕ(i) = 0 in Q, (1 6 i 6 m)


 ϕ(i) = 0
 on Σ, (1 6 i 6 m)

 (i) (i)
ϕ (·, T ) = ϕT in Ω, (1 6 i 6 m)

3
 (i) (i)
with ϕT ∈ H (1 6 i 6 m) and we denote by ϕj , j = 1, . . . , N the coordinates of ϕ(i) . Note that by setting

ϕ(m+1) ≡ 0, am+1,i = 0 (1 6 i 6 m),

we can write the first above equations as

i+1
X
− ∂t ϕ(i) − ∆ϕ(i) + ∇π (i) = aj,i ϕ(j) in Q, (1 6 i 6 m). (1.9)
j=1

Following a standard duality argument (see, for instance, [34, Theorem 11.2.1, p.357]), Theorem 1.1 will be obtained
as a consequence of the following observability inequality:
Z X
m  2 ZZ N
X −1  
 (i)   (1) 2
ϕ (x, 0) dx 6 C(T ) ϕj  dxdt. (1.10)
Ω i=1 ω×(0,T )
j=1

for some C depending on T , Ω and ω.

The proof of (1.10) is based on (global) Carleman inequalities introduced in [16] for the controllability of parabolic
equations. Such inequalities have been used by many authors to deal with Stokes of Navier-Stokes systems (for
instance, [22] or [14]). The case of controls with some vanishing components was considered in [15], [9] and [6]. We
follow here the method introduced in [9]. As a first step, one can get rid of the pressure by applying a differential
operator on (1.9) (or on components of (1.9)) such as curl or ∆. This leads to a system of coupled heat equations but
without prescribed boundary conditions. Using results such as [13] or [24], one can obtain a Carleman estimate with
boundary terms that can be absorbed by some standard arguments. Let us note that to recover the observability
on the components that are not observed, one has to use the divergence-free condition on ϕ(i) and the Dirichlet
boundary conditions. Unhappily in this process, one loses part of the weights on these components. The last part
(i)
of the proof of (1.10) consists in estimating the local terms associated with ϕj , i > 1 and this is done by using
(1.4). An important work related to this subject is [10] where the authors obtain the local null controllability of the
Navier-Stokes system in dimension 3 with a control having two vanishing components. In that case, the method is
based on a different linearization and on a different approach based on results of Gromov.
This paper is organized as follows. In Section 2, we introduce the weights for our Carleman estimates and we
recall some results, in particular some Carleman estimates for other systems. Section 3 corresponds to the statement
and to the proof of the Carleman estimate for (1.8). Finally, in Section 4, we use this estimate to prove Theorem 1.1
and Theorem 1.3.

2 Preliminaries
In the whole article, we use the notation C for a generic positive constant that depends on Ω, ω. We can assume to
simplify that T ∈ (0, 1) and that will allow us to avoid some dependence on T for some constants. Finally we only
write the proof in the case N = 2, the case N = 3 can be done similarly.
To write our Carleman inequalities, we introduce standard weights and functions. First, we consider a nonempty
domain ω0 such that ω0 ⊂ ω. Then, using [16] (see also [34, Theorem 9.4.3, p.299]), there exists η 0 ∈ C 2 (Ω)
satisfying
η 0 > 0 in Ω, η 0 = 0 on ∂Ω, max η 0 = 1, ∇η 0 6= 0 in Ω \ ω0 . (2.1)
Ω

Then, we define the following functions:

exp {λ(2ℓ + 2)} − exp{λ(2ℓ + η 0 (x))} exp{λ(2ℓ + η 0 (x))}

α(x, t) = , ξ(x, t) = , (2.2)
tℓ (T − t)ℓ tℓ (T − t)ℓ

exp {λ(2ℓ + 2)} − exp{2λℓ} exp{2λℓ}

α∗ (t) = max α(x, t) = , ξ ∗ (t) = min ξ(x, t) = , (2.3)
x∈Ω tℓ (T − t)ℓ x∈Ω tℓ (T − t)ℓ

4
where ℓ > 14, λ > 1. In literature ℓ > 3 is enough. However, to prove our result we need to use ℓ > 14 (see Proof
of Proposition 3.2).
Note that we have the following useful relations: there exists C > 0 depending on Ω such that
|∂t α| + |∂t ξ| 6 CT ξ 1+1/ℓ , (2.4)
 ∗ ′  ∗ ′    
(α )  + (ξ )  6 CT (ξ ∗ )1+1/ℓ , (α∗ )′′  + (ξ ∗ )′′  6 CT 2 (ξ ∗ )1+2/ℓ , (2.5)
C
ξ∗ >
, (2.6)
T 2ℓ
|∇α| = |∇ξ| 6 Cλξ, |∆α| = |∆ξ| 6 Cλ2 ξ. (2.7)
Weights of the kind (2.2) were first considered in [16]. In its present form, these weights have already been used
in [19] in order to obtain Carleman estimates for the controllability of strongly coupled parabolic equations and
later in [18] for the existence of insensitizing controls for Stokes systems.
Now, we recall some standard results. The first one is a Carleman estimate for the gradient operator, it is stated
and proved in [9]:
Lemma 2.1. Let r ∈ R. There exists C > 0 depending only on r, Ω and ω0 such that, for every T > 0, s > C and
every u ∈ L2 (0, T ; H 1(Ω)),
 
ZZ ZZ ZZ
 
e−2sα sr+2 ξ r+2 |u|2 dxdt 6 C  e−2sα sr ξ r |∇u|2 dxdt + e−2sα sr+2 ξ r+2 |u|2 dxdt. .
Q Q ω0 ×(0,T )

The second result is a Carleman estimate for the heat equation with non-homogeneous Dirichlet boundary con-
ditions. It is proved in [24].
Lemma 2.2. There exists a constant C > 0 such that for any λ > C, s > C, f0 , f1 , ..., fN ∈ L2 (Q) and



u ∈ L2 0, T ; H 1(Ω) ∩ H 1 0, T ; H −1(Ω)
satisfying
N
X
∂t u − ∆u = f0 + ∂j fj in Q,
j=1

we have
ZZ ZZ
2 2
2
e−2sα s−1 ξ −1 |∇u| + sξ |u| dxdt 6 C e−2sα sξ |u| dxdt
ω0 ×(0,T )
Q
2 2
∗ ∗
+ e−sα s−1/4 (ξ ∗ )−1/4+1/ℓ u 2 + e−sα s−1/4 (ξ ∗ )−1/4 u 1 1
L (Σ) H 4 , 2 (Σ)

ZZ N ZZ
X
2 2
+ e−2sα s−2 ξ −2 |f0 | dxdt + e−2sα |fj | dxdt . (2.8)
Q j=1 Q

Recall that  1/2

2 2
kuk 1,1 = kukH 1/4 (0,T ;L2 (∂Ω)) + kukL2 (0,T ;H 1/2 (∂Ω)) .
H 4 2 (Σ)
The next result concerns the regularity of the solutions of the Stokes system


 ∂t u − ∆u + ∇p = f in Q,

∇ · u = 0 in Q,
(2.9)

 u = 0 on Σ,


u(·, 0) = 0 in Ω.

5
This result can be found in [33].
Lemma 2.3. Assume T > 0. For any f ∈ L2 (Q)N , there exists a unique solution

u ∈ L2 (0, T ; H 2(Ω)N ) ∩ H 1 (0, T ; H), p ∈ L2 (0, T ; H 1 (Ω)/R)

to the Stokes system (2.9) and there exists a constant C > 0 depending only on Ω such that

kuk2L2 (0,T ;H 2 (Ω)N ) + kuk2H 1 (0,T ;L2 (Ω)N ) 6 C kf k2L2 (0,T ;L2 (Ω)N ) . (2.10)

Assume moreover that

f ∈ L2 (0, T ; V )
then
u ∈ L2 (0, T ; H 3 (Ω)N ) ∩ H 1 (0, T ; V )
and there exists a constant C > 0 depending only on Ω such that
2 2 2
kukL2 (0,T ;H 3 (Ω)N ) + kukH 1 (0,T ;V ) 6 C kf kL2 (0,T ;V ) . (2.11)

Assume moreover that

f ∈ L2 0, T ; H 3(Ω)N ∩ H 1 (0, T ; V ) , f (·, 0) = 0 in Ω.
2 5 N 1 3 N 2
Then, u ∈ L (0, T ; H (Ω) ) ∩ H (0, T ; H (Ω) ) ∩ H (0, T ; V ) and there exists a constant C > 0 depending only
on Ω such that
 
2 2 2
kukL2 (0,T ;H 5 (Ω)N ) + kukH 1 (0,T ;H 3 (Ω)N ) + kukH 2 (0,T ;V ) 6 C kf k2L2(0,T ;H 3 (Ω)N ) + k∂t f k2L2 (0,T ;V ) .

3 The Carleman estimate for the adjoint system

As before, we denote by C various positive constants which depends only on Ω and ω (they depend also in general
on the choice of η 0 and ω0 but one can consider that η 0 as well as ω0 depend Ω and ω). Without any lack of
generality, we treat the case of dimension N = 2 and j = 2. The same proof can be performed in the general case.
Our aim is to estimate the following quantity associated with the solutions of the system (1.8):
ZZ   2  2  2   
 (i)   (i)   (i)   (i) 2
I1 (s, ϕ(i) ) := e−2sα s−1 ξ −1 ∇3 ∆ϕ1  + sξ ∇2 ∆ϕ1  + s3 ξ 3 ∇∆ϕ1  + s5 ξ 5 ∆ϕ1  dxdt
Q
ZZ  2
∗  
+ e−2sα (sξ ∗ )5 ϕ(i)  dxdt. (3.1)
Q

Thus, our main result states as follows:

Theorem 3.1. There exists C > 0 depending on the geometry such that for any s > C and for any ϕT ∈ H m , the
solution (ϕ(1) , ..., ϕ(m) ) of (1.8) satisfies
m
X ZZ  
m+3
−6 2m+3 −6  (1) 2
I1 (s, ϕ(i) ) 6 C e−2sα s2 ξ ϕ1  dxdt. (3.2)
i=1
ω×(0,T )

In order to prove the above proposition, we first start by estimating each I1 (s, ϕ(i) ) (i = 1, . . . , m) independently.

6
Proposition 3.2. Let ωb ⊂ Ω be a nonempty open set such that ω0 ⋐ ω b . Then, there exists a constant C such that
for any s > C and for any ϕT ∈ H m , the solution ϕ of (1.8) satisfies

ZZ  
(i)  (i) 2
I1 (s, ϕ ) 6 C e−2sα s5 ξ 5 ∆ϕ1  dxdt
ω
b ×(0,T )

m ZZ 
X  2  2 
   (j) 
dxdt
∗
+ e−2sα s9/2 (ξ ∗ )5 ϕ(j)  + e−2sα s−2 ξ −2 ∇2 ∆ϕ1  (1 6 i 6 m). (3.3)
j=1 Q

Proof of Proposition 3.2. First taking the divergence of (1.9), we remark that

∆π (i) = 0 in Q (1 6 i 6 m).

Thus, using a standard method for the Stokes system (see, for instance [9]), we apply the operator ∇2 ∆ to the first
components of (1.9), and we deduce
i+1
X
(i) (i) (j)
− ∂t ∇2 ∆ϕ1 − ∆∇2 ∆ϕ1 = aj,i ∇2 ∆ϕ1 in Q, (1 6 i 6 m). (3.4)
j=1

b1 and for s > sb1 ,

Applying Lemma 2.2 to the above equations, we deduce that for λ > λ
ZZ   2  2 
−2sα −1 −1  3 (i)   2 (i) 
e s ξ ∇ ∆ϕ1  + sξ ∇ ∆ϕ1  dxdt
Q
ZZ  2 2
 (i)  ∗ (i)
6C e−2sα sξ ∇2 ∆ϕ1  dxdt + e−sα s−1/4 (ξ ∗ )−1/4+1/ℓ ∇2 ∆ϕ1 2
ω0 ×(0,T ) L (Σ)4

2 i+1 Z Z
X  2
(i)  (j) 
e−2sα s−2 ξ −2 ∇2 ∆ϕ1  dxdt . (3.5)
∗
+ e−sα (sξ ∗ )−1/4 ∇2 ∆ϕ1 1 1 +
H 4 , 2 (Σ)4
j=1 Q

The rest of the proof is divided into several steps:

• In Step 1, we complete the left-hand side of (3.5) with weighted integrals of ϕ(i) in Q, and adding some local
terms in the right-hand side.
• In Step 2, we obtain an upper bound of the boundary terms.
• Finally, in Step 3, we estimate the local terms that do not appear in (3.3).
(i)
Step 1. We apply Lemma 2.1 with u = ∇∆ϕ1 and r = 1: for any s > C and λ > C,
 
ZZ  2 ZZ  2 ZZ  2
 (i)    (i)   (i)  
e−2sα s3 ξ 3 ∇∆ϕ1  dxdt 6 C  e−2sα sξ ∇2 ∆ϕ1  dxdt + e−2sα s3 ξ 3 ∇∆ϕ1  dxdt . (3.6)
Q Q ω0 ×(0,T )

(i)
Then, we apply Lemma 2.1 with u = ∆ϕ1 and r = 3: for any s > C and λ > C,
 
ZZ  2 ZZ  2 ZZ  2
 (i)    (i)   (i)  
e−2sα s5 ξ 5 ∆ϕ1  dxdt 6 C  e−2sα s3 ξ 3 ∇∆ϕ1  dxdt + e−2sα s5 ξ 5 ∆ϕ1  dxdt . (3.7)
Q Q ω0 ×(0,T )

7
 Now, using the divergence condition of ϕ(i) , we have
     
 (i)   (i)   (i) 
∂2 ϕ2  = ∂1 ϕ1  6 ∇ϕ1  .

Then using the Poincaré inequality and the ellipticity of the Laplace operator with Dirichlet boundary conditions,
we deduce the existence of a constant C depending on Ω such that
Z   Z  
 (i) 2  (i) 2
ϕ  dx 6 C ∆ϕ1  dx.
Ω Ω

Combining the above relation with (3.5), (3.6) and (3.7), we deduce that I1 (s, ϕ(i) ) defined by (3.1) satisfies
ZZ   2  2   
 (i)   (i)   (i) 2
I1 (s, ϕ(i) ) 6 C e−2sα sξ ∇2 ∆ϕ1  + s3 ξ 3 ∇∆ϕ1  + s5 ξ 5 ∆ϕ1  dxdt
ω0 ×(0,T )
2 2
∗ (i) ∗ (i)
+ e−sα s−1/4 (ξ ∗ )−1/4+1/ℓ ∇2 ∆ϕ1 2 + e−sα (sξ ∗ )−1/4 ∇2 ∆ϕ1 1 1
L (Σ)4 H 4 , 2 (Σ)4

i+1 Z Z
X  2
 (j) 
+ e−2sα s−2 ξ −2 ∇2 ∆ϕ1  dxdt . (3.8)
j=1 Q

Step 2. In this step, we get rid of the boundary terms in the right-hand side of (3.8). To estimate the first term,
we notice that since ℓ > 4, (ξ ∗ )−1/4+1/ℓ is bounded in (0, T ) (see (2.6)). Thus
2 2
−sα∗ −1/4 ∗ −1/4+1/ℓ 2 (i) ∗ (i)
e s (ξ ) ∇ ∆ϕ1 2 4 6 Cs−1/2 e−sα ∇2 ∆ϕ1 2 4
L (Σ) L (Σ)
 
2
∗ (i) ∗ (i) ∗ (i)
6 Cs−1/2 e−sα ∇2 ∆ϕ1 2 4 + s1/2 e−sα (ξ ∗ )1/2 ∇2 ∆ϕ1 2 4 s−1/2 e−sα (ξ ∗ )−1/2 ∇3 ∆ϕ1 2 8
L (Q) L (Q) L (Q)
ZZ      
2 2
∗  (i)   (i) 
6 Cs−1/2 e−2sα sξ ∇2 ∆ϕ1  + s−1 ξ −1 ∇3 ∆ϕ1  dxdt 6 Cs−1/2 I1 (s, ϕ(i) ). (3.9)
Q

For the second boundary term, we will use a trace inequality (see, for instance, [27, Theorem 2.1, p.9]) and by
an interpolation argument (see, for instance, [26, Corollary 9.2, p.46]), we have
2 2
∗ −1/4 −sα∗ 2 (i) ∗ (i)
(ξ ) e ∇ ∆ϕ1 1,1 6 C (ξ ∗ )−1/4 e−sα ∇2 ∆ϕ1
H 4 2 (Σ)4 L2 (0,T ;H 1 (Ω)4 )∩H 1/2 (0,T ;L2 (Ω)4 )
2
∗ (i)
6 C (ξ ∗ )−1/4 e−sα ∇2 ∆ϕ1 2 . (3.10)
L (0,T ;H (Ω) )∩H (0,T ;H −1 (Ω)4 )
1 4 1

Our goal is to estimate the last term. First we consider the function
∗
θ1 (t) := s3/2 (ξ ∗ )3/2−1/ℓ e−sα .
From (2.5)-(2.6), for s > C, ∗
|θ1′ | 6 Cs5/2 (ξ ∗ )5/2 e−sα . (3.11)
(i) (i)
Then, from (1.8), (θ1 ϕ , θ1 π ) is the solution of

 (i)

(i)

(i)

i+1
P


 −∂ t θ 1 ϕ − ∆ θ 1 ϕ + ∇ θ 1 π = aj,i θ1 ϕ(j) − θ1′ ϕ(i) in Q,

 j=1


∇ · θ1 ϕ(i) = 0 in Q, (3.12)





 θ1 ϕ(i) = 0 on Σ,


θ1 ϕ(i) |t=T = 0 in Ω.

8
Applying Lemma 2.3 to the above system and using (3.11), we obtain
2

θ1 ϕ(i)
L2 (0,T ;H 2 (Ω)2 )∩H 1 (0,T ;L2 (Ω)2 )
 
i+1
X 2
 3/2 ∗ 3/2 −sα∗ (j) 2 ∗ .
6C s (ξ ) e ϕ + s5/2 (ξ ∗ )5/2 e−sα ϕ(i)
L2 (Q)2 L2 (Q)2
j=1

Now, combining the above estimate with an interpolation inequality, we deduce

2 ∗ 2−1/(2ℓ) −sα∗ (i) 2 5/2 ∗ 5/2 −sα∗ (i)
s (ξ ) e ϕ 6 C θ1 ϕ(i) s (ξ ) e ϕ
L2 (0,T ;H 1 (Ω)2 ) L2 (0,T ;H 2 (Ω)2 ) L2 (Q)2
 
i+1
X 2
3/2 ∗ 3/2 −sα∗ (j) 2
6C  . (3.13)
∗
s (ξ ) e ϕ 2 + s5/2 (ξ ∗ )5/2 e−sα ϕ(i) 2
L (Q)2 L (Q)2
j=1

Second, we introduce ∗
θ2 = s(ξ ∗ )1−3/(2ℓ) e−sα .
From (2.5)-(2.6), for s > C, ∗
|θ2′ | 6 Cs2 (ξ ∗ )2−1/(2ℓ) e−sα . (3.14)
Then, from (1.8), (θ2 ϕ(i) , θ2 π (i) ) is the solution of




i+1
P


 −∂t θ2 ϕ(i) − ∆ θ2 ϕ(i) + ∇ θ2 π (i) = aj,i θ2 ϕ(j) − θ2′ ϕ(i) in Q,

 j=1


∇ · θ2 ϕ(i) = 0 in Q, (3.15)




 θ2 ϕ(i) = 0 on Σ,



θ2 ϕ(i) |t=T = 0 in Ω.

Using that the right-hand side of the first equation is divergence free and with null trace on ∂Ω, we can apply
Lemma 2.3 to the above system and we deduce
!
2 i+1
X 2
(i) (j) ′ (i) 2
θ2 ϕ 6C θ2 ϕ + θ2 ϕ .
L2 (0,T ;H 3 (Ω)2 )∩H 1 (0,T ;V ) L2 (0,T ;V ) L2 (0,T ;V )
j=1
∗
Using θ2 6 Cs2 (ξ ∗ )2−1/(2ℓ) e−sα and combining the above result with (3.14) and (3.13)
2 m
X
5/2 ∗ 5/2 −sα∗ (j) 2
θ2 ϕ(i) 2 6C s (ξ ) e ϕ . (3.16)
L (0,T ;H 3 (Ω)2 )∩H 1 (0,T ;V ) 2 L (Q)2
j=1

Finally, we introduce ∗
θ3 = (ξ ∗ )−5/(2ℓ) e−sα .
From (2.5)-(2.6), for s > C, ∗
|θ3′ | 6 Cθ2 , |θ3′′ | 6 Cs2 (ξ ∗ )2−1/(2ℓ) e−sα . (3.17)
(i) (i)
Then, from (1.8), (θ3 ϕ , θ3 π ) is the solution of

 (i)

(i)

(i)

i+1
P


 −∂ t θ 3 ϕ − ∆ θ 3 ϕ + ∇ θ 3 π = Aj,i θ3 ϕ(j) − θ3′ ϕ(i) in Q,

 j=1


∇ · θ3 ϕ(i) = 0 in Q, (3.18)





 θ3 ϕ(i) = 0 on Σ,


θ3 ϕ(i) |t=T = 0 in Ω.

9
We can again apply Lemma 2.3 to the above estimate to deduce
2

θ3 ϕ(i)
L2 (0,T ;H 5 (Ω)2 )∩H 1 (0,T ;H 3 (Ω)2 )∩H 2 (0,T ;V )

Xi+1 2   2

6C θ3 ϕ(j) 2 + ∂t θ3 ϕ(j) 2
L (0,T ;H 3 (Ω)2 ) L (0,T ;H 1 (Ω)2 )
j=1
2   2 

+ θ3′ ϕ(i) 2 + ∂t θ3′ ϕ(i) 2 . (3.19)
L (0,T ;H 3 (Ω)2 ) L (0,T ;H 1 (Ω)2 )

Using θ3 6 Cθ2 and applying (3.17), (3.16), and (3.13), we deduce from (3.19)
 
2 Xm
 5/2 ∗ 5/2 −sα∗ (j) 2 .
θ3 ϕ(i) 2 5 2 1 3 2 2
6 C s (ξ ) e ϕ 2 (3.20)
L (0,T ;H (Ω) )∩H (0,T ;H (Ω) )∩H (0,T ;V ) L (Q)2
j=1

Using the trace inequality (3.10) and that ℓ > 14, we have
2 2
∗ −1/4 −sα∗ 2 (i) ∗ (i)
(ξ ) e ∇ ∆ϕ1 1 , 1 6 C (ξ ∗ )−1/4 e−sα ∇2 ∆ϕ1
H 4 2 (Σ)4 L2 (0,T ;H 1 (Ω)4 )∩H 1 (0,T ;H −1 (Ω)4 )
2 2
∗ (i)
6 C (ξ ∗ )−1/4 e−sα ϕ1 6 C θ3 ϕ(i) . (3.21)
L2 (0,T ;H 5 (Ω))∩H 1 (0,T ;H 3 (Ω)) L2 (0,T ;H 5 (Ω)2 )∩H 1 (0,T ;H 3 (Ω)2 )

Combining (3.20) and (3.21), we deduce that

2 m
X
−sα∗ ∗ −1/4 2 (i) −1/2 5/2 ∗ 5/2 −sα∗ (j) 2
e (sξ ) ∇ ∆ϕ1 1 1 6 Cs s (ξ ) e ϕ 2 .
H 4 , 2 (Σ)4 L (Q)2
j=1

Putting together (3.8), the above relation and (3.9), we deduce at this step the following inequality:

ZZ   2  2   
 (i)   (i)   (i) 2
I1 (s, ϕ(i) ) 6 C e−2sα sξ ∇2 ∆ϕ1  + s3 ξ 3 ∇∆ϕ1  + s5 ξ 5 ∆ϕ1  dxdt
ω0 ×(0,T )

m ZZ 
X   2   2

   (j) 
dxdt . (3.22)
∗
+ e−2sα s9/2 (ξ ∗ )5 ϕ(j)  + e−2sα s−2 ξ −2 ∇2 ∆ϕ1 
j=1 Q

Step 3. To estimate the local terms, we proceed in a standard way: we consider ω1 an open subset satisfying
ω0 ⋐ ω1 ⋐ ω
b and
η1 ∈ Cc2 (ω1 ), η1 ≡ 1 in ω0 , η1 > 0.
Then, an integration by parts gives
ZZ  2 ZZ  2
−2sα ∂2 (i) −2sα ∂2 (i)
e sξ ∆ϕ1 dxdt 6 η1 e sξ ∆ϕ1 dxdt
ω0 ×(0,T ) ∂xk ∂xq ω1 ×(0,T ) ∂xk ∂xq
ZZ
∂
∂2 (i) ∂ (i)
=− η1 e−2sα sξ ∆ϕ1 ∆ϕ1 dxdt
ω1 ×(0,T ) ∂xk ∂xk ∂xq ∂xq
ZZ
∂3 (i) ∂ (i)
− η1 e−2sα sξ 2 ∆ϕ1 ∆ϕ1 dxdt.
ω1 ×(0,T ) ∂xk ∂xq ∂x q

10
Using (2.7) and Young’s inequality, we deduce from the above relation that there exists C > 0 such that for all
ε > 0,
ZZ  2 ZZ   2  2 
−2sα  2 (i)  −2sα −1 −1  3 (i)   2 (i) 
e sξ ∇ ∆ϕ1  dxdt 6 ε e s ξ ∇ ∆ϕ1  + sξ ∇ ∆ϕ1  dxdt
ω0 ×(0,T )
Q
ZZ  2
C  (i) 
+ e−2sα s3 ξ 3 ∇∆ϕ1  dxdt. (3.23)
ε ω1 ×(0,T )

(i)
Now we estimate, in an analogous way, the local term associated with ∇∆ϕ1 : we consider

η2 ∈ Cc2 (b
ω ), η2 ≡ 1 in ω1 , η2 > 0.

Then, integrating by parts, we obtain

ZZ  2 ZZ  2
−2sα 3 3 ∂ (i) −2sα 3 3 ∂ (i)
e s ξ ∆ϕ1 dxdt 6 η2 e s ξ ∆ϕ1 dxdt
ω1 ×(0,T ) ∂xk ω
b ×(0,T ) ∂xk
ZZ
∂
∂ (i) (i)
=− η2 e−2sα s3 ξ 3 ∆ϕ1 ∆ϕ1 dxdt
ω
b ×(0,T ) ∂x k ∂xk
ZZ
∂2 (i) (i)
− η2 e−2sα s3 ξ 3 2 ∆ϕ1 ∆ϕ1 dxdt.
ω
b ×(0,T ) ∂xk

Using (2.7) and the Young’s inequality, we deduce from the above relation that there exists C > 0 such that for
all ε > 0,
ZZ  2 ZZ   2  2 
 (i)   (i)   (i) 
e−2sα s3 ξ 3 ∇∆ϕ1  dxdt 6 ε e−2sα s3 ξ 3 ∇∆ϕ1  + sξ ∇2 ∆ϕ1  dxdt
ω1 ×(0,T )
Q
ZZ  
C  (i) 2
+ e−2sα s5 ξ 5 ∆ϕ1  dxdt. (3.24)
ε ω
b ×(0,T )

The above estimate, together with (3.22) and (3.23) implies (3.3). This concludes the proof of Proposition 3.2.
We are now in a position to prove Theorem 3.1.
Proof of Theorem 3.1. Summing (3.3) for i = 1, . . . , m, and taking s > C for a constant C large enough, we deduce
that
m
X Xm ZZ  
(i)  (i) 2
I1 (s, ϕ ) 6 C e−2sα s5 ξ 5 ∆ϕ1  dxdt. (3.25)
i=1 i=1 ω
b ×(0,T )

In order to get rid of the local terms in the right-hand side (except the term corresponding to i = 1), we introduce
a sequence of open sets Oi , (0 6 i 6 m) such that

b =: O0 ⋐ O1 ⋐ ... ⋐ Oi ⋐ ... ⋐ Om ⋐ ω
ω

and functions
ζi ∈ Cc2 (Oi ) such that ζi ≡ 1 in Oi−1 , ζi > 0 (1 6 i 6 m).
Then, we consider the equation m−1 of (1.9), we apply the Laplace operator on the first component of this equation
and we multiply it by ζ1 :
m−1
X
(m) (m−1) (m−1) (j)
ζ1 am,m−1 ∆ϕ1 = −ζ1 ∂t ∆ϕ1 − ζ1 ∆2 ϕ1 − ζ1 aj,m−1 ∆ϕ1 . (3.26)
j=1

11
 Then, using the above equation combined with (1.4) and integrating by parts, we deduce
ZZ   ZZ  
 (m) 2  (m) 2
e−2sα s5 ξ 5 ∆ϕ1  dxdt 6 ζ1 e−2sα s5 ξ 5 ∆ϕ1  dxdt
ω
b ×(0,T ) O1 ×(0,T )
 
ZZ   m−1
X
−1 (m)  (m−1) (m−1) (j)
= ζ1 e−2sα s5 ξ 5 ∆ϕ1 ∂t ∆ϕ1 + ∆2 ϕ1 + aj,m−1 ∆ϕ1  dxdt. (3.27)
am,m−1 O1 ×(0,T ) j=1

Integrating by parts and applying ∆ to the first component of the equation (1.9) with i = m, we obtain
ZZ ZZ
−2sα 5 5 (m) (m−1)
(m) (m−1)
ζ1 e s ξ ∆ϕ1 ∂t ∆ϕ1 dxdt = − ζ1 ∂t e−2sα s5 ξ 5 ∆ϕ1 ∆ϕ1 dxdt
O1 ×(0,T ) O1 ×(0,T )
ZZ
(m) (m−1)
+ ζ1 e−2sα s5 ξ 5 ∆2 ϕ1 ∆ϕ1 dxdt
O1 ×(0,T )
m ZZ
X (j) (m−1)
+ ζ1 e−2sα s5 ξ 5 aj,m ∆ϕ1 ∆ϕ1 dxdt. (3.28)
j=1 O1 ×(0,T )

By integrating by parts, we also find

ZZ    ZZ
(m) (m−1)
(m) (m−1)
ζ1 e−2sα s5 ξ 5 ∆ϕ1 ∆2 ϕ1 dxdt = ∆ ζ1 e−2sα s5 ξ 5 ∆ϕ1 ∆ϕ1 dxdt
O1 ×(0,T ) O1 ×(0,T )
ZZ
(m) (m−1)
+ ζ1 e−2sα s5 ξ 5 ∆2 ϕ1 ∆ϕ1 dxdt
O1 ×(0,T )
ZZ

(m) (m−1)
+ 2∇ ζ1 e−2sα s5 ξ 5 · ∇∆ϕ1 ∆ϕ1 dxdt. (3.29)
O1 ×(0,T )

Let us now estimate the right-hand side of (3.28) and (3.29) and the last term in (3.27).
Using (2.4), we have 

∂t e−2sα s5 ξ 5  6 Ce−2sα s6 ξ 6+1/ℓ .
Combining this inequality and Young’s inequality, we deduce the existence of C such that for any s > C and for
any ε > 0,
Z Z 


 −2sα 5 5 (m) (m−1) 
 ζ1 ∂t e s ξ ∆ϕ1 ∆ϕ1 dxdt
 O1 ×(0,T ) 
ZZ   ZZ  
 (m) 2 C  (m−1) 2
6ε e−2sα s5 ξ 5 ∆ϕ1  dxdt + e−2sα s7 ξ 7+2/ℓ ∆ϕ1  dxdt. (3.30)
ε
Q O1 ×(0,T )

Using (2.7), we have  

∆(ζ1 e−2sα s5 ξ 5 ) 6 Ce−2sα s7 ξ 7 .
Combining this inequality and Young’s inequality, we deduce the existence of C such that for any s > C and for
any ε > 0,
Z Z 


 (m) (m−1) 
 ∆ ζ1 e−2sα s5 ξ 5 ∆ϕ1 ∆ϕ1 dxdt
 O1 ×(0,T ) 
ZZ   ZZ  
 (m) 2 C  (m−1) 2
6ε e−2sα s5 ξ 5 ∆ϕ1  dxdt + e−2sα s9 ξ 9 ∆ϕ1  dxdt. (3.31)
ε
Q O1 ×(0,T )

12
 Using (2.7), we have  
∇(ζ1 e−2sα s5 ξ 5 ) 6 Ce−2sα s6 ξ 6 .

Combining this inequality and Young’s inequality, we deduce the existence of C such that for any s > C and for
any ε > 0,
Z Z 


 −2sα 5 5 (m) (m−1) 
 2∇ ζ1 e s ξ · ∇∆ϕ1 ∆ϕ1 dxdt
 O1 ×(0,T ) 
ZZ  2 ZZ  
 (m)  C  (m−1) 2
6ε e−2sα s3 ξ 3 ∇∆ϕ1  dxdt + e−2sα s9 ξ 9 ∆ϕ1  dxdt. (3.32)
ε
Q O1 ×(0,T )

Finally, the other terms can be estimated by using Young’s inequality:

Z Z 
 
 −2sα 5 5 2 (m) (m−1) 
 ζ1 e s ξ ∆ ϕ ∆ϕ1 dxdt
 O1 ×(0,T ) 
ZZ  2 ZZ  
 (m)  C  (m−1) 2
6ε e−2sα sξ ∆2 ϕ1  dxdt + e−2sα s9 ξ 9 ∆ϕ1  dxdt, (3.33)
ε
Q O1 ×(0,T )

 
 m ZZ 
X (m−1) 
 ζ1 e −2sα 5 5 (j)
s ξ aj,m ∆ϕ ∆ϕ1 dxdt

j=1 O1 ×(0,T ) 
Xm ZZ   ZZ  
 (j) 2 C  (m−1) 2
6ε e−2sα s5 ξ 5 ∆ϕ1  dxdt + e−2sα s5 ξ 5 ∆ϕ1  dxdt, (3.34)
j=1
ε
Q O1 ×(0,T )

 
 ZZ m−1
X 
 −1 (m) (j) 
 ζ1 e −2sα 5 5
s ξ ∆ϕ1 aj,m−1 ∆ϕ1 dxdt
 am,m−1
 O1 ×(0,T ) j=1 
ZZ   m−1 Z Z  
 (m) 2 C X  (j) 2
6ε e−2sα s5 ξ 5 ∆ϕ1  dxdt + e−2sα s5 ξ 5 ∆ϕ1  dxdt. (3.35)
ε j=1
Q O1 ×(0,T )

The combination of (3.27) with (3.30), (3.31), (3.32), (3.33), (3.34), and (3.35) yields the existence of a constant
C such that for any s > C and for any ε > 0,
ZZ   Xm ZZ  
 (m) 2 C  (m−1) 2
e−2sα s5 ξ 5 ∆ϕ1  dxdt 6 ε I1 (s, ϕ(j) ) + e−2sα s9 ξ 9 ∆ϕ1  dxdt
ω
b ×(0,T ) j=1
ε
O1 ×(0,T )
m−2 Z Z  
C X  (j) 2
+ e−2sα s5 ξ 5 ∆ϕ1  dxdt. (3.36)
ε j=1
O1 ×(0,T )

(m−1)
We can repeat the same analysis for ∆ϕ1 and we obtain the existence of a constant C such that for any s > C

13
and for any ε > 0,
ZZ   Xm ZZ  
−2sα 9 9 (m−1) 2 C  (m−2) 2
e s ξ ∆ϕ1  dxdt 6 ε I1 (s, ϕ(j) ) + e−2sα s17 ξ 17 ∆ϕ1  dxdt.
O1 ×(0,T ) j=1
ε
O2 ×(0,T )
m−3 Z Z  
C X  (j) 2
+ e−2sα s13 ξ 13 ∆ϕ1  dxdt. (3.37)
ε j=1
O2 ×(0,T )

Proceeding by induction, we can estimate all the local terms and we deduce from (3.25) that
m
X ZZ  2
(m+1)
+1 2(m+1) +1  (1) 
I1 (s, ϕ(i) ) 6 C e−2sα s2 ξ ∆ϕ1  dxdt. (3.38)
i=1 Om ×(0,T )

(1)
Finally, we estimate the above local term in terms of ϕ1 . In order to do this, we consider ω
e an open subset
satisfying Om ⋐ ωe ⋐ ω and
ζe ∈ Cc2 (e
ω ) such that ζe ≡ 1 in Om , ζe > 0.
Then by integrating by parts, we obtain
ZZ  2 ZZ  2
(m+1)
+1 2(m+1) +1  (1)  e −2sα s2(m+1) +1 ξ 2(m+1) +1 ∆ϕ(1)  dxdt
e−2sα s2 ξ ∆ϕ1  dxdt 6 ζe 1
Om ×(0,T ) ω
e ×(0,T )
ZZ  
=− e −2sα s2(m+1) +1 ξ 2(m+1) +1 ∆ϕ(1) · ∇ϕ(1) dxdt
∇ ζe 1 1
ω
e ×(0,T )
ZZ
− e −2sα s2(m+1) +1 ξ 2(m+1) +1 ∆∇ϕ(1) · ∇ϕ(1) dxdt.
ζe (3.39)
1 1
ω
e ×(0,T )

Using (2.7), we have  

 e −2sα 2(m+1) +1 2(m+1) +1  (m+1)
+2 2(m+1) +2
∇(ζe s ξ ) 6 Ce−2sα s2 ξ .
Combining this inequality and Young’s inequality, we deduce the existence of C such that for any s > C and for
any ε > 0,
ZZ  2 ZZ   2   
(m+1)
+1 2(m+1) +1  (1)   (1)   (1) 2
e−2sα s2 ξ ∆ϕ1  dxdt 6 ε e−2sα s3 ξ 3 ∇∆ϕ1  + s5 ξ 5 ∆ϕ1  dxdt
Om ×(0,T )
Q
ZZ  
C (m+2)
−1 2(m+2) −1  (1) 2
+ e−2sα s2 ξ ∇ϕ1  dxdt. (3.40)
ε
ω
e ×(0,T )

Then, we consider
ζ ∈ Cc2 (ω) such that ζ ≡ 1 in ω
e, ζ >0
and we integrate by parts:
ZZ  2 ZZ  
(m+2)
−1 2(m+2) −1  (1)  (m+2)
−1 2(m+2) −1  (1) 2
e−2sα s2 ξ ∇ϕ1  dxdt 6 ζe−2sα s2 ξ ∇ϕ1  dxdt
ω
e ×(0,T ) ω×(0,T )
ZZ  
(m+2)
−1 2(m+2) −1 (1) (1)
=− ∇ ζe−2sα s2 ξ · ∇ϕ1 ϕ1 dxdt
ω×(0,T )
ZZ
(m+2)
−1 2(m+2) −1 (1) (1)
− ζe−2sα s2 ξ ∆ϕ1 ϕ1 dxdt.
ω×(0,T )

14
 Using (2.7), we have  
 (m+2)
−1 2(m+2) −1  (m+2) (m+2)
∇(ζe−2sα s2 ξ ) 6 Ce−2sα s2 ξ2 .

Combining this inequality and Young’s inequality, we deduce the existence of C such that for any s > C and for
any ε > 0,
ZZ  
(m+2)
−1 2(m+2) −1  (1) 2
e−2sα s2 ξ ∇ϕ1  dxdt
ω
e ×(0,T )
 
ZZ ZZ
(1) (1)
6 ε e−2sα s6 ξ 6 |∇ϕ1 |2 dxdt + e−2sα s5 ξ 5 |∆ϕ1 |2 dxdt
Q Q
ZZ  
C (m+3)
−6 2(m+3) −6  (1) 2
+ e−2sα s2 ξ ϕ1  dxdt. (3.41)
ε
ω×(0,T )

Using Lemma 4 in [9] (see [23] for the proof) and the definition (3.1) of I1 , we deduce that
ZZ   ZZ   ZZ  
 (i) 2  (i) 2  (i) 2
I1 (s, ϕ(i) ) + e−2sα s8 ξ 8 ϕ1  dxdt > e−2sα s5 ξ 5 ∆ϕ1  dxdt + e−2sα s8 ξ 8 ϕ1  dxdt
ω×(0,T ) Q ω×(0,T )
 
ZZ  2 ZZ  2
 (i)   (i) 
>C e−2sα s8 ξ 8 ϕ1  dxdt + e−2sα s6 ξ 6 ∇ϕ1  dxdt .
Q Q

This estimate, combined with (3.38), (3.40) and (3.41) yields the conclusion of Theorem 3.1.

4 Proof of the main results

4.1 Final state observability
In this section, we use Theorem 3.1 in order to prove the final state observability of the adjoint system (1.8).
Lemma 4.1. Assume T ∈ (0, 1) and ω is non empty open set of Ω. Then, there exists C > 0 and ℓ > 14 such that
for any ϕ0 ∈ H m , the solution ϕ of (1.8) satisfies
Z X
m  2 ZZ  
 (i)  C  (1) 2
ϕ (x, 0) dx 6 Ce T 2ℓ ϕ1  dxdt. (4.1)
Ω i=1 ω×(0,T )

Proof. First, we consider an energy estimate of the adjoint system (1.8). Multiplying each equation (1.9) by ϕ(i)
and integrating by parts, we deduce
Z X
m   Z X
m   Z X
m
1 d  (i) 2  (i) 2
− ϕ  dx + ∇ϕ  dx = Aj,i ϕ(i) · ϕ(j) dx. (4.2)
2 dt
Ω i=1 Ω i=1 Ω i,j=1

Thus, using the Grönwall lemma, there exists C > 0 such that
Z X
m  2
 (i) 
t 7→ eCt ϕ (x, t) dx
Ω i=1

15
is nondecreasing. In particular, for some constant C > 0,

Z X 3T
Z /4Z X
m  2 m  2
 (i)  2 CT  (i) 
ϕ (x, 0) dx 6 e ϕ (x, t) dx dt. (4.3)
i=1
T i=1
Ω T /4 Ω

On the other hand, from (3.2) and (3.1), we deduce that

ZZ m 
X  ZZ  
∗  (i) 2 2m+3 −6  (1) 2
e−2sα (sξ ∗ )5 ϕ  dxdt 6 C e−2sα (sξ) ϕ1  dxdt. (4.4)
Q i=1
ω×(0,T )

Using that for t ∈ [T /4, 3T /4],

3T 2 T2
6 t(T − t) 6 ,
16 4
we deduce, from (2.3), the existence of two constants C1 , C2 > 0 such that for t ∈ [T /4, 3T /4],
C1 C2
α∗ (t) 6 , ξ ∗ (t) >
T 2ℓ T 2ℓ
and consequently, for some constant C3 > 0,
C3
e−2sα (sξ ∗ )5 > e− T 2ℓ .
∗
(4.5)

Similarly, from (2.2), there exist two constants c1 , c2 > 0 such that for (x, t) ∈ Ω × [0, T ],
c1 c2
α(x, t) > , ξ(x, t) 6
tℓ (T − t)ℓ tℓ (T − t)ℓ
and consequently, for some constant c3 > 0,
 2m+3 −6
−2sα 2m+3 −6 −2s
c1
tℓ (T −t)ℓ
c2
e (sξ) 6e s 6 c3 . (4.6)
t (T − t)ℓ
ℓ

Combining (4.3), (4.4), (4.5) and (4.6), we deduce that for some constant C,
Z Xm  2 ZZ  
 (i)  C C T+ 1  (1) 2
ϕ (x, 0) dx 6 e ( T 2ℓ ) ϕ1  dxdt.
i=1
T
Ω ω×(0,T )

This implies that (4.1).

4.2 Proof of Theorem 1.1

We recall that H and V are defined by (1.6) and (1.7). We also set

H := H m , V := V m . (4.7)

We define P0 : L2 (Ω)N → H the Leray projector and the projection P defined as:
 m


P : L2 (Ω)N → H, y = y (1) , . . . , y (m) 7→ P0 y (1) , . . . , P0 y (m) .

We also consider the unbounded operator in H defined by

 m
D(A) := H 2 (Ω)N ∩ V , Ay := P∆y + Ay. (4.8)

16
Finally, we define the control operator B ∈ L(U, H) by





U := L2 (ω)N −1 , Bv = B v1 , . . . , vN −1 := P0 v1 , . . . , vN −1 , 0 1ω , 0, . . . , 0 . (4.9)

With the above definition, we can write (1.1) as

(
y ′ = Ay + Bv,
(4.10)
y(0) = y0 .

As it is well-known (see, for instance, [34, p.357]), system (4.10) is null-controllable in time T > 0 if and only if
there exists K(T ) > 0 such that
Z T
T A∗ 2 ∗ tA∗ 2
e ϕ0 6 K(T )2 B e ϕ0 dt (ϕ0 ∈ H). (4.11)
H 0 H

One can check that  m

D(A∗ ) = H 2 (Ω)N ∩ V , A∗ y = P∆y + A∗ y,
where A∗ is the transpose of the matrix A and that
    
(1) (1)
B∗ϕ = ϕ1 , . . . , ϕN −1 .
|ω |ω

Thus, we deduce Theorem 1.1 from Lemma 4.1.

4.3 Proof of Theorem 1.3

In order to prove Theorem 1.3, we recall a method introduced in [28,32] to deal with the controllability of nonlinear
parabolic systems. We
consider H and U two Hilbert spaces, A : D(A) → H the infinitesimal generator of an
analytic semigroup etA t>0 and B ∈ L(U, H) a control operator. We also assume the final state observability for
all T > 0 Z T

T A∗ 2 ∗ tA∗ 2
e ϕ0 6 K(T )2 B e ϕ0 dt (ϕ0 ∈ H), (4.12)
H 0 H

with K : (0, ∞) → [0, ∞) continuous and non increasing. Let us consider T > 0 and suppose there exist ρ0 , ρ1 , ρ ∈
C 0 ([0, T ], R+ ), non increasing, positive in [0, T ) such that ρ0 (T ) = ρ1 (T ) = ρ(T ) = 0 and such that, for some
constant q > 1,     
1
ρ0 (t) := ρ1 (q 2 (t − T ) + T )K((q − 1)(T − t)) t ∈ T 1 − 2 ,T , (4.13)
q
ρ0 6 Cρ, ρ1 6 Cρ, |ρ′ |ρ0 6 Cρ2 (t ∈ [0, T ]) . (4.14)
for some constant C > 0. We denote by L2ρ1 (0, T ; H) the space
 
f
L2ρ1 (0, T ; H) := f ∈ L2 (0, T ; H) ; ∈ L2 (0, T ; H)
ρ1

and we define similarly L2ρ0 (0, T ; U).

Then we can consider the control problem
(
y ′ = Ay + Bv + f,
(4.15)
y(0) = y0 .

We have the following result (see [28]):

17
Theorem 4.2. With the above assumptions, there exists a bounded operator
   
1/2
ET ∈ L D (−A) × L2ρ1 (0, T ; H), L2ρ0 (0, T ; U)
 
such that for any y0 ∈ D (−A)1/2 and for any f ∈ L2ρ1 (0, T ; H), the solution y of (4.15) with u = ET (y0 , f )
satisfies   
y 1/2
∈ L2 (0, T ; D(A)) ∩ C 0 [0, T ]; D (−A) ∩ H 1 (0, T ; H).
ρ
Moreover there exists a constant C such that
 
y
6 C ky k 1/2 + kf kL2 (0,T ;H) .
ρ 2 0 D ((−A) ) ρ1
L (0,T ;D(A))∩C 0 ([0,T ];D ((−A)1/2 ))∩H 1 (0,T ;H)

Remark 4.3. Note that in [28], A is assumed to be self-adjoint negative but the result can be extended to the
case where A is the generator of an analytic semigroup. Indeed, the hypothesis used in the proof is the maximal
regularity of (4.15) for v = 0.
Remark 4.4. Since ρ(T ) = 0, the above result implies in particular that y(T ) = 0, that is the null-controllability
of (4.15).
In the previous section, we have defined for our problem the spaces H, U, A and B, see (4.7), (4.8) and (4.9).
To show that A is the generator of an analytic semigroup, we first note that P∆ is self-adjoint negative (see, for
instance, [31, Theorem 2.1.1, p.128]) and thus is the generator of an analytic semigroup. Then using a perturbation
argument (see, for instance, [30, Theorem 2.1, p.80]), we deduce that A is the generator of an analytic semigroup
in H.
Finally, applying Lemma 4.1, we deduce that (4.12) holds with
CK
K(T ) = CK e T 2ℓ ,

for some constant CK > 0.

Let us consider  
1
q ∈ 1, 2 4ℓ .
and let us set C0 C1 C⋆
− − −
ρ0 (t) := CK e (T −t)2ℓ , ρ1 (t) := e (T −t)2ℓ , ρ(t) := e (T −t)2ℓ

with C0 , C1 , C⋆ some positive constants such that

C1 CK C1 C1
C0 := − > , < C⋆ < C0 < C1 .
q 4ℓ (q − 1)2ℓ 2 2

Then we can check that (4.13) and (4.14) hold and we have moreover that

ρ2 6 ρ1 . (4.16)

Consequently, we deduce from Theorem 4.2 a controllability result on the system

 Pm

 ∂t y (1) − ∆y (1) + ∇p(1) = A1,j y (j) + ve1 1ω + f (1) in Q,



 j=1

 ∂t y (i) − ∆y (i) + ∇p(i) = P Ai,j y (j) + f (i)
 m
in Q, (2 6 i 6 m)
j=i−1 (4.17)



 ∇ · y (i) = 0 in Q, (1 6 i 6 m)



 y (i) = 0 on Σ, (1 6 i 6 m)
 (i) (i)
y (·, 0) = y0 in Ω. (1 6 i 6 m)

18
More precisely, there exists
  m 
ET ∈ L V × L2ρ1 (0, T ; L2 (Ω)N ), L2ρ0 (0, T ; L2 (ω))

 m
such that for any y0 ∈ V and for any f = f (1) , . . . , f (m) ∈ L2ρ1 (0, T ; L2 (Ω)N ), the solution y of (4.17) with the
control v = ET (y0 , f ) satisfies
y  m   m 
∈ L2 (0, T ; H 2 (Ω)N ) ∩ C 0 [0, T ]; H 1 (Ω)N ∩ H 1 (0, T ; H). (4.18)
ρ
Moreover we have the following estimate
 
y
C ky k + kf k
ρ 2 6 0 V Lρ1 (0,T ;[L (Ω) ] ) .
2 2 N m (4.19)
L (0,T ;[H 2 (Ω)N ]m )∩C 0 ([0,T ];[H 1 (Ω)N ]m )∩H 1 (0,T ;H)

We are now in a position to prove Theorem 1.3:

Proof of Theorem 1.3. First we notice that y is solution of (1.2) if it is a solution of (4.17) with
 
f = −(y (1) · ∇)y (1) , ..., −(y (m) · ∇)y (m) .

Thus, we consider the mapping

NT : f ∈ BR 7→ (−(y (1) · ∇)y (1) , ..., −(y (m) · ∇)y (m) ),
where ( )
 2 
N m
f
BR := f∈ L2ρ1 (0, T ; L (Ω)
); 6R
ρ1 L2 (0,T ;[L2 (Ω)N ]m )
where R > 0 is such that
ky0 kV 6 R.
We are going to show that for R small enough (and thus ky0 kV small enough), NT (BR ) ⊂ BR and that (NT )|BR
is a strict contraction. Using the Banach fixed point theorem we deduce the existence of a fixed point of NT . The
corresponding solution y of (4.17) is a solution of (1.2) and from (4.18), we deduce that y(·, T ) = 0.
It thus remains to prove that for R small enough, NT (BR ) ⊂ BR and that (NT )|BR is a strict contraction. In
order to do this, we first note that, using (4.16), Sobolev’s embeddings and Hölder’s inequalities, we have
Z Z   Z T Z    2 2
T  v · ∇w 2  v w  v ∇w 2
  dxdt 6  · ∇ dxdt 6 C
 ρ1   ρ ρ  ρ ∞ ρ 2
0 Ω 0 Ω L (0,T ;L6 (Ω)N ) L (0,T ;L6 (Ω)N )
2 2
v w
6 C
ρ ∞

. (4.20)
L (0,T ;H 1 (Ω)N ) ρ L2 (0,T ;H 2 (Ω)N )

Using this relation and (4.19), we deduce that

 2
NT (f )
6 C ky k + kf k 6 4CR2 6 R,
ρ1 2 0 V 2 2 N m
Lρ1 (0,T ;[L (Ω) ] )
L (0,T ;[L2 (Ω)N ]m )

for R small enough. For such R, we have NT (BR ) ⊂ BR .

Now, let us consider fe, fb ∈ BR and let us write f = fe − fb. We consider the solution ye (resp. yb) the solution of
(4.17) associated with the control ve = ET (y0 , fe) (resp. vb = ET (y0 , fb)). Then, y := ye − yb is the solution of (4.17)
associated with the control v := ET (0, f ) and thus

y
6 C kf kL2ρ (0,T ;[L2 (Ω)N ]m ) .
ρ 2 1
L (0,T ;[H 2 (Ω)N ]m )∩C 0 ([0,T ];[H 1 (Ω)N ]m )∩H 1 (0,T ;H)

19
Using this and (4.20), we obtain
       
N (fe) N (fb) ye y y yb
T T
− 6 ·∇ + ·∇
ρ1 ρ1 ρ ρ L2 (0,T ;[L2 (Ω)N ]m ) ρ ρ L2 (0,T ;[L2 (Ω)N ]m )
L2 (0,T ;[L2 (Ω)N ]m )

6 CR kf kL2 2 (Ω)N ]m ) .
ρ1 (0,T ;[L

Thus for R small enough, (NT )|BR is a strict contraction and this ends the proof of Theorem 1.3

References
[1] F. Ammar-Khodja, A. Benabdallah, M. González-Burgos, and L. de Teresa. Recent results on the controllability
of linear coupled parabolic problems: a survey. Math. Control Relat. Fields, 1(3):267–306, 2011.
[2] F. Ammar Khodja, A. Benabdallah, M. González-Burgos, and L. de Teresa. Minimal time for the null con-
trollability of parabolic systems: the effect of the condensation index of complex sequences. J. Funct. Anal.,
267(7):2077–2151, 2014.
[3] F. Ammar Khodja, A. Benabdallah, M. González-Burgos, and L. de Teresa. New phenomena for the null con-
trollability of parabolic systems: minimal time and geometrical dependence. J. Math. Anal. Appl., 444(2):1071–
1113, 2016.
[4] A. Benabdallah, F. Boyer, and M. Morancey. A block moment method to handle spectral condensation
phenomenon in parabolic control problems. Ann. H. Lebesgue, 3:717–793, 2020.
[5] N. Carreño, S. Guerrero, and M. Gueye. Insensitizing controls with two vanishing components for the three-
dimensional Boussinesq system. ESAIM Control Optim. Calc. Var., 21(1):73–100, 2015.
[6] N. Carreño and M. Gueye. Insensitizing controls with one vanishing component for the Navier-Stokes system.
J. Math. Pures Appl. (9), 101(1):27–53, 2014.
[7] F. Conforto, L. Desvillettes, and R. Monaco. Some asymptotic limits of reaction-diffusion systems appearing
in reversible chemistry. Ric. Mat., 66(1):99–111, 2017.
[8] J.-M. Coron and A. V. Fursikov. Global exact controllability of the 2D Navier-Stokes equations on a manifold
without boundary. Russian J. Math. Phys., 4(4):429–448, 1996.
[9] J.-M. Coron and S. Guerrero. Null controllability of the N -dimensional Stokes system with N − 1 scalar
controls. J. Differential Equations, 246(7):2908–2921, 2009.
[10] J.-M. Coron and P. Lissy. Local null controllability of the three-dimensional Navier-Stokes system with a
distributed control having two vanishing components. Invent. Math., 198(3):833–880, 2014.
[11] P. Érdi and J. Tóth. Mathematical models of chemical reactions. Nonlinear Science: Theory and Applications.
Princeton University Press, Princeton, NJ, 1989. Theory and applications of deterministic and stochastic
models.
[12] H. O. Fattorini and D. L. Russell. Uniform bounds on biorthogonal functions for real exponentials with an
application to the control theory of parabolic equations. Quart. Appl. Math., 32:45–69, 1974/75.
[13] E. Fernández-Cara, M. González-Burgos, S. Guerrero, and J.-P. Puel. Null controllability of the heat equation
with boundary Fourier conditions: the linear case. ESAIM Control Optim. Calc. Var., 12(3):442–465, 2006.
[14] E. Fernández-Cara, S. Guerrero, O. Y. Imanuvilov, and J.-P. Puel. Local exact controllability of the Navier-
Stokes system. J. Math. Pures Appl. (9), 83(12):1501–1542, 2004.

20
[15] E. Fernández-Cara, S. Guerrero, O. Y. Imanuvilov, and J.-P. Puel. Some controllability results for the N -
dimensional Navier-Stokes and Boussinesq systems with N − 1 scalar controls. SIAM J. Control Optim.,
45(1):146–173, 2006.
[16] A. V. Fursikov and O. Y. Imanuvilov. Controllability of evolution equations, volume 34 of Lecture Notes Series.
Seoul National University, Research Institute of Mathematics, Global Analysis Research Center, Seoul, 1996.
[17] M. González-Burgos and L. de Teresa. Controllability results for cascade systems of m coupled parabolic PDEs
by one control force. Port. Math., 67(1):91–113, 2010.
[18] S. Guerrero. Controllability of systems of Stokes equations with one control force: existence of insensitizing
controls. Ann. Inst. H. Poincaré Anal. Non Linéaire, 24(6):1029–1054, 2007.
[19] S. Guerrero. Null controllability of some systems of two parabolic equations with one control force. SIAM J.
Control Optim., 46(2):379–394, 2007.
[20] M. Iida, H. Monobe, H. Murakawa, and H. Ninomiya. Vanishing, moving and immovable interfaces in fast
reaction limits. J. Differential Equations, 263(5):2715–2735, 2017.
[21] O. Y. Imanuvilov. On exact controllability for the Navier-Stokes equations. ESAIM Control Optim. Calc. Var.,
3:97–131, 1998.
[22] O. Y. Imanuvilov. Remarks on exact controllability for the Navier-Stokes equations. ESAIM Control Optim.
Calc. Var., 6:39–72, 2001.
[23] O. Y. Imanuvilov and J.-P. Puel. Global Carleman estimates for weak solutions of elliptic nonhomogeneous
Dirichlet problems. Int. Math. Res. Not., (16):883–913, 2003.
[24] O. Y. Imanuvilov, J.-P. Puel, and M. Yamamoto. Carleman estimates for parabolic equations with nonhomo-
geneous boundary conditions. Chin. Ann. Math. Ser. B, 30(4):333–378, 2009.
[25] G. Lebeau and L. Robbiano. Contrôle exacte de l’équation de la chaleur. In Séminaire sur les Équations aux
Dérivées Partielles, 1994–1995, pages Exp. No. VII, 13. École Polytech., Palaiseau, 1995.
[26] J.-L. Lions and E. Magenes. Non-homogeneous boundary value problems and applications. Vol. I. Die
Grundlehren der mathematischen Wissenschaften, Band 181. Springer-Verlag, New York-Heidelberg, 1972.
Translated from the French by P. Kenneth.
[27] J.-L. Lions and E. Magenes. Non-homogeneous boundary value problems and applications. Vol. II. Die
Grundlehren der mathematischen Wissenschaften, Band 182. Springer-Verlag, New York-Heidelberg, 1972.
Translated from the French by P. Kenneth.
[28] Y. Liu, T. Takahashi, and M. Tucsnak. Single input controllability of a simplified fluid-structure interaction
model. ESAIM Control Optim. Calc. Var., 19(1):20–42, 2013.
[29] C. Montoya and L. de Teresa. Robust Stackelberg controllability for the Navier-Stokes equations. NoDEA
Nonlinear Differential Equations Appl., 25(5):Paper No. 46, 33, 2018.
[30] A. Pazy. Semigroups of linear operators and applications to partial differential equations, volume 44 of Applied
Mathematical Sciences. Springer-Verlag, New York, 1983.
[31] H. Sohr. The Navier-Stokes equations. Modern Birkhäuser Classics. Birkhäuser/Springer Basel AG, Basel,
2001. An elementary functional analytic approach, [2013 reprint of the 2001 original] [MR1928881].
[32] T. Takahashi. Boundary local null-controllability of the Kuramoto-Sivashinsky equation. Math. Control Signals
Systems, 29(1):Art. 2, 21, 2017.

21
[33] R. Temam. Behaviour at time t = 0 of the solutions of semilinear evolution equations. J. Differential Equations,
43(1):73–92, 1982.
[34] M. Tucsnak and G. Weiss. Observation and control for operator semigroups. Birkhäuser Advanced Texts:
Basler Lehrbücher. [Birkhäuser Advanced Texts: Basel Textbooks]. Birkhäuser Verlag, Basel, 2009.

22