Local null controllability of a model system for strong

interaction between internal solitary waves

Jon Asier Bárcena-Petisco, Sergio Guerrero, Ademir Pazoto

To cite this version:

Jon Asier Bárcena-Petisco, Sergio Guerrero, Ademir Pazoto. Local null controllability of a model
system for strong interaction between internal solitary waves. 2020. �hal-02376059v2�

HAL Id: hal-02376059

https://hal.archives-ouvertes.fr/hal-02376059v2
Preprint submitted on 21 Oct 2020

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Local null controllability of a model system for strong interaction
between internal solitary waves

Jon Asier Bárcena-Petisco*„ … Sergio Guerrero§ Ademir F. Pazoto¶

October 19, 2020

Abstract

In this paper we prove the local null controllability property for a nonlinear coupled
system of two Korteweg-de Vries equations posed on a bounded interval and with a source
term decaying exponentially on t = T . The system was introduced by Gear and Grimshaw
to model the interactions of two dimensional, long, internal gravity waves propagation in a
stratified fluid. We address the controllability problem by means of a control supported on
an interior open subset of the domain and acting on one equation only. The proof consists
mainly on proving the controllability of the linearized system, which is done by getting
a Carleman estimate for the adjoint system. While doing the Carleman we improve the
techniques for dealing with the fact that the solutions of dispersive and parabolic equations
with a source term in L2 have a limited regularity. A local inversion theorem is applied to
get the result for the nonlinear system.

Key words: Carleman inequalities, internal control, Korteweg-de Vries systems, null controllability
AMS subject classification: 35Q53, 93B05, 93C20
Abbreviated title: Local null controllability of a system for solitary waves
Acknowledgements: The first author has been supported by grants from Région Ile-de-France.
The first and second authors have been supported by the Agence Nationale de la Recherche (France)
research project IFSMACS(ANR-15-CE40-0010). The third author was supported by Conselho
Nacional de Desenvolvimento Cientı́fico e Tecnológico (Brazil) and Coordenação de Aperfeiçoamento
de Pessoal de Nı́vel Superior through Program Print-Capes (Brazil).

*
Sorbonne Université, Laboratoire Jacques-Louis Lions, F-75005, Paris, France
„
Present address: [1] Departamento de Matemáticas, Universidad Autónoma de Madrid, 28049 Madrid, Spain.
[2] Chair of Computational Mathematics, Fundación Deusto, Avenida de las Universidades 24, 48007 Bilbao,
Basque Country, Spain; [email protected]; Orcid: 0000-0002-6583-866X
…
Corresponding author
§
Sorbonne Université, Laboratoire Jacques-Louis Lions, F-75005, Paris, France; [email protected]
¶
Instituto de Matemática, Universidade Federal do Rio de Janeiro, C.P. 68530 - CEP 21941-909, Rio de Janeiro
(RJ), Brazil; [email protected]

1
1 Introduction

Nonlinear dispersive partial differential equations naturally appear as models describing wave
phenomena in various branches of physics and engineering, such as quantum mechanics, non-
linear optics, plasma physics, water waves, and atmospheric sciences. They have received wide
attention from the applied science community due to their importance in applications and have
also been studied extensively from the theoretical point of view, providing a framework for
development of analytical ideas and tools.

In this context, the interaction of internal gravity waves propagating in one horizontal di-
rection has been considered by Gear and Grimshaw [26]. More precisely, the authors derived
a model to describe strong interactions of two long internal gravity waves in a stratified fluid,
where the two waves are assumed to correspond to different modes of the linearized equations
of motion. It can be written as

ut + uux + uxxx + avxxx + k1 vvx + k2 (uv)x = 0,

(1.1)
cvt + rvx + vvx + abuxxx + vxxx + k2 buux + k1 b(uv)x = 0,

where a, b, c, r, k1 , k2 are real parameters, the unknowns u and v are real valued functions of
the variables x and t and subscripts indicate partial differentiation. Observe that (1.1) has
the structure of a pair of Korteweg-de Vries (KdV) equations with both linear and nonlinear
coupling terms.

We also refer to [6] for an extensive discussion on the physical relevance of the system in its
full structure.

1.1 Setting of the problem

In this paper, we are concerned with the study of the controllability properties of the system
(1.1), posed on a bounded interval (0, L), with homogeneous boundary conditions. The main
purpose is to steer the solutions of the system to the rest by means of a control supported on
an interior open subset ω of (0, L), acting on one equation only. As usual, we denote

Q ∶= (0, T ) × (0, L), Qω ∶= (0, T ) × ω, for any open subset ω of (0, L),
(1.2)
1ω , the characteristic function on ω.

With the notation above, our attention is given to the following distributed control system

2
with initial data (u0 , v 0 ), source term (h1 , h2 ) and control f :
⎧
⎪
⎪
⎪
⎪ut + uux + uxxx + avxxx + k1 vvx + k2 (uv)x = h1 , in Q,
⎪
⎪
⎪
⎪
⎪
⎪cvt + rvx + vvx + abuxxx + vxxx + k2 buux + k1 b(uv)x = h2 + f 1ω , in Q,
⎪
⎪
⎪
⎪
⎨u(⋅, 0) = u(⋅, L) = ux (⋅, L) = 0, on (0, T ), (1.3)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪v(⋅, 0) = v(⋅, L) = vx (⋅, L) = 0, on (0, T ),
⎪
⎪
⎪
⎪ 0 0
⎩u(0, ⋅) = u , v(0, ⋅) = v , on (0, L).
⎪
⎪

Under these conditions, considerations will be given to the following null controllability
problem:

Given T > 0, an initial state (u0 , v 0 ) in a certain space, and a source term (h1 , h2 ) in a
certain space, can one find an appropriate control f , so that the system (1.3) admits a solution
(u, v) which satisfies (u(T, ⋅), v(T, ⋅)) = (0, 0)?

Our main result gives a positive answer to this question. The proof combines the analysis
of the linearized system and an inverse function theorem for the full system. In order to ana-
lyze the linearized system, we follow the classical duality approach [21, 32], which reduces the
null controllability property to the proof of an observability inequality for the solutions of the
corresponding adjoint system. Here, this is done by using Carleman estimates.

In order to overcome some technical difficulties that will become clear later (see fourth
comment in Section 4.3), we assume that the coefficients a, b and c satisfy:

a ≠ 0, b, c > 0 and a2 b < 1. (1.4)

To our knowledge, this problem has not been addressed in the literature and the existing
developments do not allow to give an immediate answer to it. Indeed, the controllability problem
for system (1.1) was first solved when the control acts through the boundary conditions [8, 9, 17,
35]. The results are obtained combining the analysis of the linearized system and the Banach’s
fixed-point theorem. In order to study the linear problem, the authors follow closely the ideas
introduced by Rosier [36] while studying the boundary controllability of the KdV equation (based
on compactness arguments, a spectral analysis of the state operator and multipliers). But, being
different from other systems, the length L of the spatial domain may play a crucial role in
determining the controllability of the system, specially when some configurations of the controls
input are allowed to be used. This phenomenon, the so-called critical length phenomenon, was
observed for the first time by Rosier in [36]. Roughly speaking, it was proved the existence of a
finite dimensional subspace M of L2 (0, L), which is not reachable by the KdV equation, when
starting from the origin, if L belongs to a countable set of lengths.

3
 More recently, by using the approach used in the present paper, the internal null control-
lability for the Hirota-Satsuma system was addressed in [16]. However, due to the complexity
of the system (1.1), which has coupling constituted by third order terms, the control problem
presents new difficulties we have to deal with. Then, we develop techniques applicable to this
more complicated situation, which allow to slightly improve the result present in [16] and, in
addition, can also be applied to other dispersive and parabolic systems.

To conclude, we can mention the work [10] where the authors consider the problem of con-
trolling pointwise, by means of a time dependent Dirac measure supported at a given point, the
linear system associated with (1.1) on the unit circle. The results are obtained by means of
spectral analysis and Fourier expansion of the solutions.

1.2 Main results

Before stating our main result, we introduce the space

′ ′
H p,p (Q) ∶= H p (0, T ; L2 (0, L)) ∩ L2 (0, T ; H p (0, L)), for p, p′ ≥ 0,

endowed with its natural norm, where Q was defined in (1.2). In addition, to shorten the
notation we denote

ˆ U ∶= (u, v), U 0 ∶= (u0 , v 0 ),

′
ˆ V = V 2 , for V any Banach space (like L2 , H01 or H p,p ),

ˆ Φ ∶= (ϕ, φ) the adjoint variable,

ˆ C (resp. q) a generic positive constant (resp. polynomial of positive coefficients) that may
be different each time.

The main result that we prove in this paper concerns the local null controllability of (1.3):

Theorem 1.1. Let L > 0, T > 0 and assume that (1.4) is satisfied. Then, there is C > 0 such
that for all h1 , h2 satisfying
C
e (T −t)12 (h1 , h2 ) ∈ L2 (0, T ; H−1 (0, L)),

there is δ > 0 such that for all ∥U 0 ∥ < δ there is f ∈ L2 (Qω ) and U ∈ C([0, T ]; L2 (0, L)) satisfying
that (U, f ) is a solution of (1.3) and U (T, ⋅) = 0.

4
 As usual, we obtain the null controllability of (1.3) by proving the null controllability of the
linearized system around 0. That system is given by
⎧
⎪
⎪
⎪
⎪LU = (h1 , h2 + f 1ω ), in Q,
⎪
⎪
⎪
⎨U (⋅, 0) = U (⋅, L) = Ux (⋅, L) = 0, on (0, T ), (1.5)
⎪
⎪
⎪
⎪
⎪
⎪
⎪U (0, ⋅) = U 0 , on (0, L),
⎩
for h1 , h2 ∈ L2 (0, T ; H−1 (0, L)) decaying exponentially at t = T as specified in Section 4.1,
U 0 ∈ L2 (0, L) and
LU ∶= (ut + uxxx + avxxx , cvt + rvx + abuxxx + vxxx ). (1.6)

First, in Section 2 we study the main properties of the associated Cauchy problem, which is
given by
⎧
⎪
⎪
⎪
⎪LU = G, in Q,
⎪
⎪
⎪
⎨U (⋅, 0) = U (⋅, L) = Ux (⋅, L) = 0, on (0, T ), (1.7)
⎪
⎪
⎪
⎪
⎪
⎪
⎪U (0, ⋅) = U 0 , on (0, L),
⎩
for U 0 ∈ L2 (0, L) and G ∈ L2 (0, T ; H−1 (0, L)). Afterwards, in order to show the null control-
lability of (1.5) we follow the classical duality approach [21, 32], which reduces the problem to
prove an observability inequality for the solutions of the corresponding adjoint system, which is
given by
⎧
⎪
⎪
⎪
⎪L∗ Φ = G, in Q,
⎪
⎪
⎪
⎨Φ(⋅, 0) = Φ(⋅, L) = Φx (⋅, 0) = 0, on (0, T ), (1.8)
⎪
⎪
⎪
⎪
⎪
⎪
⎪Φ(T, ⋅) = ΦT , on (0, L),
⎩
for ΦT ∈ L2 (0, L), G ∈ L2 (Q) and

L∗ Φ ∶= (−ϕt − ϕxxx − abφxxx , −cφt − rφx − aϕxxx − φxxx ). (1.9)

In fact, we study the main properties of (1.8) in Section 2, and then we prove a Carleman
estimate in Section 3.

One of the main contributions of this paper is to make it clear the difficulty of getting Carleman
estimates for any well-posed dispersive or parabolic system with a force G ∈ L2 (Q) or G = 0.
Indeed, to overcome the fact that solution Φ is not regular enough (because G ∈ L2 (Q)) the
proof of the Carleman estimate presents the novelty that we decompose Φ as the sum of the
solution of three Cauchy problems. In the literature, the adjoint variable has been decomposed
as the sum of the solution of two Cauchy problems (see, for instance, [13]), but for (1.8) we

5
clearly need more. This implies some changes in the proof with respect to the existing results,
which are explained in Section 3.2 and can be applied for many other dispersive and parabolic
systems.

1.3 Historical background

There is a huge literature on controllability results with controls having a reduced number
of components concerning parabolic equations (see the survey [4] and the references therein).
For instance, there are a lot of results about Navier-Stokes-like systems, such as [7, 13, 14, 18,
19, 24, 30, 33]. It has also been studied, for example, the controllability in cascade-like systems
[29, 15], the null controllability in the context of linear thermoelasticity [31], the controllability to
trajectories in phase-field models [3], the existence of insensitizing controls for the heat equation
[20], and the controllability in reaction-diffusion systems [2]. However, as for KdV-like systems,
it is a very recent research topic which has been studied only in [10] and [16].

The main difference of the problem we consider in this paper with respect to the above cited
papers, except [10], is the coupling. Indeed, in the papers cited above (and in the literature, as
far as we know) the coupling is constituted by a zero, first or second order term. However, in
this paper the coupling is constituted by a third order term. In this sense, the main contribution
of this paper is in the technicalities of the Carleman, which provides a technique to move from
the controllability of the dipersive linearized system with null force to the controllability of the
linearized system with a force decaying in T (usually a necessary step to prove the local null
controllability of some nonlinear systems).

In [10], the authors consider the problem of controlling pointwise, by means of a time de-
pendent Dirac measure supported by a given point, the linearized system on the unit circle.
More precisely, by means of spectral analysis and Fourier expansion they prove, under general
assumptions on the physical parameters of the system, a pointwise observability inequality which
leads to the pointwise controllability by using one control function. It is also possible to find
the sharp time of the controllability.

By contrast, the study of the boundary controllability properties is considerably more devel-
oped. The first result was obtained in [34], when the model is posed on a periodic domain and
r = 0. In this case, a diagonalization of the main terms allows to decouple the corresponding
linear system into two scalar KdV equations and to use the previous results available in the lit-
erature. Later on, in [35], the authors proved the local exact boundary controllability property
for the nonlinear system (1.1) posed on a bounded interval. Their result was improved in [9],
[17] and in [8], where the same boundary control problem was addressed with a different set
of boundary conditions. In all those works, except in [17], the results were proved by applying

6
the duality approach and some ideas introduced by Rosier in [36]. Then, it was discovered that
whether the system is exactly controllable depends on the length of the spatial domain, i.e.,
if L does not lie in the so-called countable set of critical lengths. More precisely, the observ-
ability inequality is proved with the aid of a compactness argument that leads the issue to a
nonstandard unique continuation principle for the eigenvalues of the state operator associated
with the model. Then, the problem is solved by using the Paley-Wiener method introduced
by Rosier, which allows the authors determine explicitly all the critical lengths for which the
exact controllability fails for the linearized system. The same problem was addressed in [17]
when only two controls act on the Neumann boundary conditions. In this case, the analysis of
the linearized system is much more complicated. Therefore, the authors used a direct approach
based on the multiplier technique that gives the observability inequality for small values of the
length L and large time of control T .

Before closing this section, we mention the work [12], in which the boundary controllability
problem was addressed for a Boussinesq system of KdV type.

The rest of the paper is organized as follows: in Section 2 we study the regularity estimates
of some elliptic and evolution problems, in Section 3 we prove a Carleman inequality for (1.8),
and finally in Section 4 we end the proof of Theorem 1.1.

2 Some elliptic and evolution problems

In this section we first prove a Poincaré inequality and secondly we prove some existence,
uniqueness and regularity results for (1.7) and (1.8).

2.1 A Poincaré inequality

In this subsection we prove the following Poincaré inequality:

Proposition 2.1. Let L > 0 and i ∈ N. Then, there is C > 0 such that for all w ∈ H 2+i (0, L)
satisfying
w(0) = w(L) = w′ (0) = 0, (2.1)

we have the estimate

∥w∥H 2+i (0,L) ≤ C∥wxxx ∥H i−1 (0,L) . (2.2)

Remark 2.2. By symmetry, for all i ∈ N we also have (2.2) for all w ∈ H 2+i (0, L) satisfying

w(0) = w(L) = w′ (L) = 0.

7
 Proposition 2.1, for i ≥ 1, is well-known and has been used in the literature (see for instance
[27]). However, as far as we know, there is no reference for i = 0.

In order to prove Proposition 2.1, we use the following interpolation result:

Lemma 2.3. Let L > 0. Then, there is C > 0 such that for all z ∈ H 1 (0, L) we have that

∥zx ∥L2 (0,L) ≤ C(∥z∥L2 (0,L) + ∥zxx ∥H −1 (0,L) ). (2.3)

L
Proof of Lemma 2.3. We fix χ ∈ D(0, L) such that ∫0 χ = 1. Then, for any ψ ∈ D(0, L) we have
the estimate below
L L x L L L
∣∫ zx ψ∣ = ∣∫ zx (∫ (ψ(s) − χ(s) ∫ ψ(z)dz) ds) dx + ∫ zx χ ∫ ψ(s)dsdx∣
0 0 0 0 x 0 0
x L L L
= ∣⟨−zxx , ∫ (ψ(s) − χ(s) ∫ ψ(z)dz) ds⟩ −∫ zχx ∫ ψ(s)ds∣
0 0 H −1 (0,L)×H01 (0,L) 0 0

≤ C(∥zxx ∥H −1 (0,L) + ∥z∥L2 (0,L) )∥ψ∥L2 (0,L) . (2.4)

Consequently, (2.3) is obtained from Riesz theorem and the density of D(0, L) in L2 (0, L).

Proof of Proposition 2.1. It suffices to prove Proposition 2.1 for i = 0. Indeed, once that case is
proved, we have the following estimate

∥w∥H 2+i (0,L) ≤ C(∥w∥H 2 (0,L) + ∥wxxx ∥H i−1 (0,L) )

≤ C(∥wxxx ∥H −1 (0,L) + ∥wxxx ∥H i−1 (0,L) ) ≤ C∥wxxx ∥H i−1 (0,L) .

We prove the case i = 0 by contradiction.

If Proposition 2.1 is false for i = 0, there is a sequence wj ∈ H 2 (0, L) satisfying (2.1) such that

∥wj ∥H 2 (0,L) = 1, (2.5)

and
j
∥wxxx ∥H −1 (0,L) → 0. (2.6)
By compactness, we can suppose that wj is convergent in H 7/4 (0, L) to some function w (see, for
instance, [1]). The convergence implies that w satisfies (2.1) and that wxxx = 0 as a distribution.
Hence, we can prove easily that w = 0. Consequently,

∥wj ∥H 1 (0,L) → 0, (2.7)

which together with Lemma 2.3 for z = wx and (2.6) implies that
j
∥wxx ∥L2 (0,L) → 0. (2.8)

Thus, we find from (2.7) and (2.8) that ∥wj ∥H 2 (0,L) → 0, contradicting (2.5).

8
2.2 Existence, uniqueness and regularity properties of the solutions of some
dissipative evolution problems

We have the following result for (1.7), which is a consequence of the results presented in [35,
Section 2.1] and classical semigroup theory:

Lemma 2.4. Let L > 0, T > 0 and assume that (1.4) is satisfied. Then, for any U 0 ∈ L2 (0, L)
and G ∈ L1 (0, T ; L2 (0, L)) system (1.7) admits a unique solution

U ∈ C([0, T ]; L2 (0, L)) ∩ L2 (0, T ; H1 (0, L)).

Moreover, there is C > 0 independent of U 0 and G such that

∥U ∥C([0,T ];L2 (0,L))∩L2 (0,T ;H1 (0,L)) ≤ C(∥U 0 ∥L2 (0,L) + ∥G∥L1 (0,T ;L2 (0,L)) ).

In addition, if U 0 ∈ D(0, L) and G ∈ D(Q) we have that U ∈ C∞ (Q).

In a similar way, we have an analogous result for the adjoint system:

Lemma 2.5. Let L > 0, T > 0 and assume that (1.4) is satisfied. Then, for any ΦT ∈ L2 (0, L)
and G ∈ L1 (0, T ; L2 (0, L)) system (1.8) admits a unique solution

Φ ∈ C([0, T ]; L2 (0, L)) ∩ L2 (0, T ; H1 (0, L)).

Moreover, there is C > 0 independent of ΦT and G such that

∥Φ∥C([0,T ];L2 (0,L)))∩L2 (0,T ;H1 (0,L)) ≤ C(∥ΦT ∥L2 (0,L) + ∥G∥L1 (0,T ;L2 (0,L)) ).

In addition, if ΦT ∈ D(0, L) and G ∈ D(Q) we have that Φ ∈ C∞ (Q).

To continue with, we prove more accurate results about the regularity of the solutions of
(1.8):

Proposition 2.6. Let L > 0 and assume that (1.4) is satisfied. Then:

1. For all T > 0, ΦT ∈ L2 (0, L) and G ∈ L2 (0, T ; H−1 (0, L)) system (1.8) admits a unique
solution
Φ ∈ C([0, T ]; L2 (0, L)) ∩ L2 (0, T ; H1 (0, L)).

Moreover, there is C > 0 independent of T , ΦT and G such that

∥Φ∥C([0,T ];L2 (0,L)))∩L2 (0,T ;H1 (0,L)) ≤ C(1 + T )(∥ΦT ∥L2 (0,L) + ∥G∥L2 (0,T ;H−1 (0,L)) ). (2.9)

9
 2. There is C > 0 such that for all T > 0 and G ∈ L2 (Q) we have

Φ ∈ L2 (0, T ; H2 (0, L)) ∩ H 1 (0, T ; H−1 (0, L))

satisfying the estimate

∥Φ∥L2 (0,T ;H2 (0,L))∩H 1 (0,T ;H−1 (0,L)) ≤ C(1 + T )∥G∥L2 (Q) , (2.10)

for Φ the energy solution of (1.8) with initial value ΦT = 0.

3. There is C > 0 such that for all T > 0 and G ∈ L2 (0, T ; H10 (0, L)) we have Φ ∈ H1,3 (Q)
satisfying the estimate

∥Φ∥H1,3 (Q) ≤ C(1 + T 3 )∥G∥L2 (0,T ;H1 (0,L)) , (2.11)

0

for Φ the energy solution of (1.8) with initial value ΦT = 0.

For the proof of Proposition 2.6 we need the following result, which is proved in [27]:

Lemma 2.7. Let L > 0 and k > 0. Then, there is C > 0 such that for all T > 0 and g ∈ L2 (Q)
we have w ∈ L2 (0, T ; H 2 (0, L)) satisfying the estimate

∥w∥L2 (0,T ;H 2 (0,L)) ≤ C∥g∥L2 (Q) , (2.12)

for w the energy solution of

⎧
⎪
⎪
⎪
⎪−wt − kwxxx = g, in Q,
⎪
⎪
⎪
⎨w(⋅, 0) = w(⋅, L) = wx (⋅, 0) = 0, on (0, T ),
⎪
⎪
⎪
⎪
⎪
⎪
⎪w(T, ⋅) = 0 on (0, L).
⎩

Proof of Proposition 2.6. Step 1: some first facts and presentation of the proof. The uniqueness
result of statement 1 is a consequence of the linearity of the problem and of Lemma 2.5 applied
to ΦT = 0 and G = 0. Thus, we just have to focus on the existence results and the regularity
estimates (2.9)-(2.11). For that purpose, thanks to linearity and density results, it actually
suffices to prove (2.9)-(2.11) when ΦT ∈ D(0, L) (or ΦT = 0) and G ∈ D(Q). We remind the
reader that under those hypothesis thanks to Lemma 2.5 the energy solution of (1.8) belongs
to C∞ (Q). Hence, the rest of the proof is structured as follows: we first diagonalize (1.8) and
secondly we prove the estimates (2.9)-(2.11).

10
 Step 2: diagonalization of (1.8). We first remark that (1.8) is given by
⎧ ⎡ ⎤ ⎡ ⎤
⎪
⎪ ⎢0 0 ⎥ ⎢ g1 ⎥
⎪ ∗ ∗ ⎢ ⎥ ∗ ⎢ ⎥
⎪
⎪
⎪
⎪−Φ t − M Φ xxx − ⎢ ⎥ Φ x = ⎢ ⎥, in Q,
⎪ ⎢ r ⎥ ⎢ g2 ⎥
⎪
⎪ ⎢0 c ⎥ ⎢c⎥
⎨ ⎣ ⎦ ⎣ ⎦ (2.13)
⎪
⎪
⎪Φ ∗
(⋅, 0) = Φ ∗
(⋅, L) = Φ ∗
(⋅, 0) = 0, on (0, T ),
⎪
⎪ x
⎪
⎪
⎪
⎪ ∗ T ∗
⎩Φ (T, ⋅) = (Φ ) ,
⎪ on (0, L),
for
⎡ ⎤ ⎡ ⎤ ⎡ T⎤
⎢ 1 ab⎥ ⎢ϕ⎥ ⎢ϕ ⎥
M ∶= ⎢⎢ a 1 ⎥⎥ , Φ∗ ∶= ⎢⎢ ⎥⎥ , (ΦT )∗ ∶= ⎢⎢ T ⎥⎥ .
⎢c c ⎥ ⎢φ⎥ ⎢φ ⎥
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
We have that the eigenvalues of M are
√ √
1 1 2 1−a2 b 2 2
1 + c ± (1 + c ) − 4 c 1 + c ± (1 − 1c ) + 4 ac b
1
λ± ∶= = , (2.14)
2 2
which are strictly positive and different by (1.4). Thereby, there is a matrix P such that
⎡ ⎤
⎢λ+ 0 ⎥
P M P −1 = ⎢⎢ ⎥.
⎥
⎢ 0 λ− ⎥
⎣ ⎦
Then, if
Φ̃ ∶= P Φ∗ , (2.15)
we have that Φ̃ satisfies the system
⎧ ⎡ ⎤
⎪
⎪ ⎢λ+ 0 ⎥
⎪ ⎢ ⎥
⎪
⎪
⎪
⎪−Φ̃t − ⎢ ⎥ Φ̃xxx − M̃ Φ̃x = G̃, in Q,
⎪ ⎢ ⎥
⎪
⎪ ⎢ 0 λ− ⎥
⎨ ⎣ ⎦ (2.16)
⎪
⎪
⎪Φ̃(⋅, 0) = Φ̃(⋅, L) = Φ̃x (⋅, 0) = 0, on (0, T ),
⎪
⎪
⎪
⎪
⎪
⎪ T ∗
⎩Φ̃(T, ⋅) = P (Φ ) ,
⎪ on (0, L),
for
⎡ ⎤ ⎡ ⎤
⎢0 0 ⎥ −1 ⎢g1 ⎥
⎢
M̃ ∶= P ⎢ ⎥ P , G̃ ∶= P ⎢ ⎥. (2.17)
⎥ ⎢ g2 ⎥
⎢0 rc ⎥ ⎢c⎥
⎣ ⎦ ⎣ ⎦
Clearly, it suffices to get the estimates for system (2.16), which is easier because the coupling
appears only on the first-order spatial term.

Step 3: proof of (2.9). If we multiply (2.16) by Φ̃, after integrating by parts and applying
Cauchy-Schwarz inequality we obtain
δ
∥Φ̃∥2C([0,T ];L2 (0,L)) ≤ Cδ T ∥G̃∥2L2 (0,T ;H−1 (0,L)) + ∥Φ̃∥2L2 (0,T ;H1 (0,L)) . (2.18)
T 0

Thus, if we multiply (2.16) by (L + 1 − x)Φ̃, (2.9) is obtained by using classical estimates,

integration by parts, (2.18) to deal with the first order term and (2.15).

11
 Step 4: proof of (2.10). We recall that from now on

Φ̃T = P (ΦT )∗ = 0. (2.19)

Thanks to (2.16) we know that system (1.8) behaves in the higher order terms as two independent
KdV equations. In particular, by considering (2.19), (2.12) (g is taken as the components of
G̃ + M̃ Φ̃x ) and (2.9) we find that

∥Φ̃∥L2 (0,T ;H2 (0,L)) ≤ C(1 + T )∥G̃∥L2 (Q) . (2.20)

Hence, we obtain from (2.15), (2.17) and (2.20) that

∥Φ∥L2 (0,T ;H2 (0,L)) ≤ C(1 + T )∥G∥L2 (Q) . (2.21)

We can easily get (2.10) from (2.21) and (1.8).

Step 5: proof of (2.11). Considering (2.19), we remark that Ψ ∶= Φ̃xxx is the solution of the
system
⎧
⎪ ⎡ ⎤
⎪
⎪ ⎢λ+ 0 ⎥
⎪ ⎢ ⎥
⎪
⎪
⎪Ψt + ⎢ ⎥ Ψxxx + M̃ Ψx = G̃xxx , in Q,
⎪ ⎢ ⎥
⎪
⎪
⎪ ⎢ 0 λ− ⎥
⎪
⎪ ⎣ ⎦
⎪
⎪
⎪
⎪Ψ(⋅, 0) = −M̂ M̃ Φ̃x (⋅, 0) = 0,
⎪ on (0, T ),
⎨
⎪
⎪
⎪Ψ(⋅, L) = −M̂ M̃ Φ̃x (⋅, L), on (0, T ),
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪Ψx (⋅, 0) = −M̂ M̃ Φ̃xx (⋅, 0), on (0, T ),
⎪
⎪
⎪
⎪
⎩Ψ(T, ⋅) = 0, on (0, L),
⎪
⎪
for
⎡ −1 ⎤
⎢λ+ 0 ⎥⎥
⎢
M̂ ∶= ⎢ ⎥.
⎢ 0 λ−1− ⎥
⎣ ⎦
2 3
Using that G̃ ∈ L (0, T ; H0 (0, L)) and considering that the solution can be defined by transpo-
sition as in [27, Proposition 1] we find the following estimate:

∥Ψ∥L2 (Q) ≤ C(1 + T )(∥G̃∥L2 (0,T ;H1 (0,L)) + ∥Φ̃x (⋅, L)∥L2 (0,T ) + ∥Φ̃xx (⋅, 0)∥L2 (0,T ) ). (2.22)
0

Recalling that Ψ = Φ̃xxx and (2.20) we obtain from (2.22) that

∥Φ̃∥L2 (0,T ;H3 (0,L)) ≤ C((1 + T 2 )∥G̃∥L2 (0,T ;H1 (0,L)) + (1 + T )∥Φ̃xx (⋅, 0)∥L2 (0,T ) ). (2.23)
0

Consequently, using the following trace estimate

δ
∥Φ̃xx (⋅, 0)∥L2 (0,T ) ≤∥Φ̃∥L2 (0,T ;H3 (0,L)) + Cδ (1 + T )∥Φ̃∥L2 (0,T ;H2 (0,L)) ,
1+T
we find from (2.15), (2.17) and (2.23) that

∥Φ∥L2 (0,T ;H3 (0,L)) ≤ C(1 + T 3 )∥G∥L2 (0,T ;H1 (0,L)) . (2.24)
0

In addition, the L2 -norm of Φt is estimated from (1.8) and (2.24), and thus we obtain (2.11).

12
 Two consequences of Proposition 2.6 are the following:

Corollary 2.8. Let L > 0, p ∈ N∗ and assume that (1.4) is satisfied. Then, there is a polynomial
q such that for all T > 0 and

G ∈ H p−1 (0, T ; H10 (0, L)) ∩ L2 (0, T ; H3p−3 (0, L))

satisfying
∂ti G(T, ⋅) = 0, ∀i ∈ {0, 1, . . . , p − 2},

we have that Φ ∈ Hp,3p (Q) and the following estimate holds

∥Φ∥Hp,3p (Q) ≤ q(T )(∥G∥H p−1 (0,T ;H1 (0,L))∩L2 (0,T ;H3p−3 (0,L)) ), (2.25)
0

for Φ the energy solution of (1.8) with initial value ΦT = 0, and q a polynomial of degree 3p.

Corollary 2.9. Let L > 0, p ∈ N∗ and assume that (1.4) is satisfied. Then, there a polynomial
q such that for all T > 0 and

G ∈ H p (0, T ; H−1 (0, L)) ∩ L2 (0, T ; H3p−2 (0, L))

satisfying
∂ti G(T, ⋅) = 0, ∀i ∈ {0, 1, . . . , p − 1},

we have that Φ ∈ H p (0, T ; H1 (0, L)) ∩ L2 (0, T ; H3p+1 (0, L)) and the following estimate holds

∥Φ∥H p (0,T ;H1 (0,L))∩L2 (0,T ;H3p+1 (0,L)) ≤ q(T )∥G∥H p (0,T ;H−1 (0,L))∩L2 (0,T ;H3p−2 (0,L)) , (2.26)

for Φ the energy solution of (1.8) with initial value ΦT = 0, and q a polynomial of degree 3p + 1.

Corollary 2.8 is proved by induction, using the inductive hypothesis on Φt and diagonalising L∗
as in (2.16). The time estimate of Corollary 2.9 is proved by using (2.9) on ∂tp Φ. The space
estimate of Corollary 2.9 is proved by diagonalising L∗ as in (2.16) and induction: first by using
(2.9) on Φt , and then by applying the inductive hypothesis on Φt . Indeed, we have the following
continuous inclusions:

H p (0, T ; H−1 (0, L)) ∩ L2 (0, T ; H3p−2 (0, L))

3(p−1)−2 p−1 − p1
⊂ H 1 (0, T ; H p (0, L)) ⊂ H 1 (0, T ; H3(p−1)−2 (0, L)) .

Remark 2.10. We have for (1.7) regularity results analogous to those in Proposition 2.6, Corollary
2.8 and Corollary 2.9.

13
3 A Carleman inequality for the solutions of (1.8)

In order to shorten the notation, we define for a function g ∈ C([0, L]) and any subset X ⊂
[0, L] the integral:
∫ g ∶= ∑ g(x).
X x∈X

The structure of this section is the following: first we present the weights, and secondly we
state and prove the Carleman inequality.

3.1 Definition of the weights and an auxiliary result

Let L > 0 and ω ⊂ (0, L) be an open subdomain. We consider η ∈ C 3 ([0, L]) a function
satisfying:
∥η∥∞ = 1, inf ∣η ′ ∣ > 0, sup η ′′ < 0, η(0) = η(L) = 0. (3.1)
(0,L)∖ω (0,L)

With that auxiliary function we define the following classical weights:

eλ(2m+2) − eλ(2m+η(x)) eλ(2m+η(x))

α(t, x) ∶= , ξ(t, x) ∶= ,
tm (T − t)m tm (T − t)m
(3.2)
∗ ∗
α (t) ∶= sup α(t, x), ξ (t) ∶= inf ξ(t, x),
x∈(0,L) x∈(0,L)

for m ∈ R+ and λ > 0 to be fixed later. Similar weights were first defined in [25], but this version
is taken from [23]. We remark that (3.1)3 and (3.1)4 imply the equalities:

α∗ (t) = α(t, 0) = α(t, L), ξ ∗ (t) = ξ(t, 0) = ξ(t, L). (3.3)

We recall that for all i ∈ N∗ there is C > 0 such that if λ ≥ C we have the estimate:

∣∂ti ξ∣ + ∣∂ti α∣ ≤ CT i (ξ ∗ )1+i/m . (3.4)

Next, we define the weights:

eλ(2m+2) − eλ(2m+η(x))
β(t, x) ∶= , β ∗ (t) ∶= sup β(t, x), (3.5)
(T 2 l(t/T ))m x∈(0,L)

for l(t) ∈ C ∞ ([0, 1]) a positive function satisfying:

l(t) = t(1 − t), in [3/4, 1], l(t) > t(1 − t), in [0, 3/4), l′ (t) < 0, in [0, 1]. (3.6)

An easy consequence of (3.6) is that

β ∗ ≤ α∗ in [0, T ]. (3.7)

14
As for these weights, we have the following bound for λ ≥ C:

eλ(2m+2) − e2mλ
∣βt∗ ∣ = ∣mT l′ (t/T ) ∣ ≤ CT (β ∗ )1+1/m . (3.8)
(T 2 l(t/T ))m+1

In general, for all i ∈ N∗ there is C > 0 such that if λ ≥ C we have the estimates:

∣∂ti β ∗ ∣ ≤ CT i (β ∗ )1+i/m , ∣∂ti β ∗ ∣ ≤ CT i (ξ ∗ )1+i/m . (3.9)

Finally, for all δ > 0 and k ∈ R+ there is C > 0 such that for all s ∈ R+ :
∗
(sβ ∗ )k ≤ Ceδsβ . (3.10)

Indeed, xk e−δx ∈ L∞ (R+ , dx).

In order to prove a Carleman inequality for (1.8) we need the following technical result:

Proposition 3.1. Let L > 0, i ≥ 0, j, k1 , k2 ∈ R, k3 ∈ R ∖ {0}, m ≥ 1/2 and ω ⊂ (0, L). There is
C > 0 such that for all λ ≥ C, s ≥ C(T m + T 2m ), and w ∈ H 2,3 (Q) we have the estimate

s1+i λ2+j ∬ e−2sα ξ 1+i ∣wxx ∣2 + s3+i λ4+j ∬ e−2sα ξ 3+i ∣wx ∣2 + s5+i λ6+j ∬ e−2sα ξ 5+i w2
Q Q Q

≤ C(si λj ∬ e−2sα ξ i g̃ 2 + s1+i λ2+j ∬ e−2sα ξ 1+i ∣wxx ∣2 + s5+i λ6+j ∬ e−2sα ξ 5+i w2
Q Qω Qω
T
+ s−1+i λ−1+j ∫ ∫ e−2sα ξ −1+i ∣wt ∣2 ), (3.11)
0 {0,L}

for
g̃ ∶= wt + k1 wx + k2 wxx + k3 wxxx . (3.12)

Proposition 3.1 can be proved for i = j = k1 = k2 = 0 by following, step by step, with some
small changes the proof presented in [11, Proposition 3.1] and in [5, Theorem 3.1] (which are
based on the proof presented in [25] for the heat equation). In addition, we have for any i ≥ 0
that si/2 e−sα ξ i/2 w ∈ H 2,3 (Q), so we just have to replace w by si/2 ξ i/2 w in (3.11) and do the usual
absorptions. As for j, it suffices to multiply (3.11) by λj at both sides of the inequality. Finally,
we can prove it for any k1 , k2 ∈ R by considering that wt + k3 wxxx = g̃ − k1 wx − k2 wxx , and then
absorbing the first and second order term with the left-hand side of (3.11).

3.2 The Carleman estimate

In this section we prove the following Carleman estimate:

15
Proposition 3.2. Let L > 0, ω ⊂ (0, L) be an open subset, m ≥ 12 and assume that (1.4) is
satisfied. There is C > 0 such that for all λ ≥ C, s ≥ C(T m + T 3m ), ΦT ∈ L2 (0, L) and G ∈ L2 (Q)
we have the estimate
T ∗ −10sβ ∗ ∗ −10sβ ∗ ∗
s6 λ6 ∫ e−2sα (ξ ∗ )6 ∥Φ∥2H2 (0,L) ≤ C (∬ e−sα φ2 + ∬ e−8sβ ∣G∣2 ) , (3.13)
0 Qω Q

for Φ the energy solution of (1.8).

Remark 3.3. We have that s ≥ C(T m + T 3m ) implies that s ≥ CT 2m , which implies that sξ ∗ ≥ 1.

The structure of the proof is the following: first we reduce it to the case in which the solution
is regular, second we decompose Φ as the sum of the solutions of three Cauchy problems, third
we get regularity estimates for the first two, fourth we deduce a PDE that just contains the
second component of the third auxiliary variable, fifth we get a Carleman estimate for the third
auxiliary variable, sixth we estimate the trace terms, seventh we estimate the local terms, and
finally we deduce a Carleman estimate just in terms of a local term of φ.
Remark 3.4. As explained in the introduction, the proof of Proposition 3.2 presents the novelty
that we decompose Φ as the sum of the solutions of three Cauchy problems to overcome the
fact that the adjoint variable is not regular enough. The main difference in considering three
Cauchy problems instead of two (as it has been done in the literature) is that in many parts of
the proof we need to use the weights (3.5) instead of (3.2) as a way to ensure that the solutions
of those Cauchy problems belong to L2 (Q). In addition, there is a second novelty which is to use
a factorization to simplify the sixth order differential equation that we get when we uncouple.

Proof of Proposition 3.2. Step 1: reducing to regular solutions. By linearity arguments, density
arguments and Lemma 2.5 it suffices to prove (3.13) for G ∈ D(Q) and ΦT ∈ D(0, L). In those
cases we have by Lemma 2.5 that Φ ∈ C∞ (Q). Thus, from now we suppose that G ∈ D(Q) and
ΦT ∈ D(0, L).

Step 2: Decomposition of Φ. We consider that

∗
e−5sβ Φ = Φ̂ + Φ + Φ̃, (3.14)

for Φ̂ ∶= (ϕ̂, φ̂) the solution of the system

⎧ ∗
⎪
⎪
⎪L∗ Φ̂ = e−5sβ G, in Q,
⎪
⎪
⎪
⎪
⎨Φ̂(⋅, 0) = Φ̂(⋅, L) = Φ̂x (⋅, 0) = 0, on (0, T ), (3.15)
⎪
⎪
⎪
⎪
⎪
⎪
⎪Φ̂(T, ⋅) = 0, on (0, L),
⎩

16
Φ ∶= (ϕ, φ) the solution of the system
⎧
⎪
⎪
⎪
⎪L∗ Φ = 5sβt∗ Φ̂, in Q,
⎪
⎪
⎪
⎨Φ(⋅, 0) = Φ(⋅, L) = Φx (⋅, 0) = 0, on (0, T ), (3.16)
⎪
⎪
⎪
⎪
⎪
⎪
⎪Φ(T, ⋅) = 0, on (0, L),
⎩
and Φ̃ ∶= (ϕ̃, φ̃) the solution of the system
⎧ ∗
⎪
⎪
⎪L∗ Φ̃ = 5sβt∗ (e−5sβ Φ − Φ̂), in Q,
⎪
⎪
⎪
⎪
⎨Φ̃(⋅, 0) = Φ̃(⋅, L) = Φ̃x (⋅, 0) = 0, on (0, T ), (3.17)
⎪
⎪
⎪
⎪
⎪
⎪
⎪Φ̃(T, ⋅) = 0, on (0, L).
⎩
We remark that because of (3.14), Φ̃ satisfies:
⎧
⎪
⎪
⎪
⎪L∗ Φ̃ = 5sβt∗ (Φ + Φ̃), in Q,
⎪
⎪
⎪
⎨Φ̃(⋅, 0) = Φ̃(⋅, L) = Φ̃x (⋅, 0) = 0, on (0, T ), (3.18)
⎪
⎪
⎪
⎪
⎪
⎪
⎪Φ̃(T, ⋅) = 0, on (0, L).
⎩
Moreover, from Proposition 2.6, Corollary 2.8 and Corollary 2.9 we deduce that

Φ̂, Φ, Φ̃ ∈ C∞ (Q).

Step 3: estimates of the solutions of (3.15) and (3.16). First of all, we obtain from (2.10)
that:
∗
∥Φ̂∥L2 (0,T ;H2 (0,L)) ≤ q(T )∥e−5sβ G∥L2 (Q) . (3.19)

We recall that q stands for a generic polynomial of positive coefficient that might be different
each time it appears.

Moreover, using (2.9) for system (3.15) and (3.6)3 , we have that for all δ > 0 there is a
polynomial q of positive coefficients such that, for all t ∈ [0, T ],
∗ ∗ (t) ∗
∥Φ̂(t, ⋅)∥L2 (0,L) ≤ q(T − t)∥e−5sβ G∥L2 ((t,T )×(0,L)) ≤ q(T )e−δsβ ∥e−(5−δ)sβ G∥L2 (Q) . (3.20)
∗
∗
This allows us to estimate ∥sβtt Φ̂∥L2 (Q) and ∥(sβt∗ )2 Φ̂∥L2 (Q) in terms of ∥e−(5−δ)sβ G∥L2 (Q) .
Consequently, looking at the system satisfied by βt∗ Φ̂, from (2.10), (3.9)1 , (3.10) and (3.20) we
get
∗
∥sβt∗ Φ̂∥L2 (0,T ;H2 (0,L))∩H 1 (0,T ;H−1 (0,L)) ≤ q(T )∥e−(5−δ)sβ G∥L2 (Q) . (3.21)

Thus, using (2.26) for p = 1, from (3.21) if follows that

∗
∥Φ∥H 1 (0,T ;H1 (0,L))∩L2 (0,T ;H4 (0,L)) ≤ q(T )∥e−(5−δ)sβ G∥L2 (Q) . (3.22)

17
 Finally, since λ ≥ C and s ≥ C(T m + T 3m ), from (3.10) and (3.6)3 we obtain the estimate
s1/m ∗ ∗
T ≤C2
≤ C(sβ ∗ (0))1/m ≤ Cesδβ (0) ≤ Cesδβ (t) , ∀t ∈ [0, T ]. (3.23)
T
Consequently, taking δ > 0 small enough and keeping the interesting terms, for λ ≥ C and
s ≥ C(T m + T 3m ) we have
∗
∥Φ̂∥L2 (0,T ;H2 (0,L)) + ∥Φ∥H 1 (0,T ;H1 (0,L))∩L2 (0,T ;H4 (0,L)) ≤ C∥e−4sβ G∥L2 (Q) . (3.24)

Step 4: uncoupling (3.18). We now get an equation containing just φ̃. By applying the
operator −∂t − ∂x3 to the second equation of (3.18)1 and using the first equation of (3.18)1 , we
obtain the PDE:

cφ̃tt + cφ̃txxx + rφ̃tx + r∂x4 φ̃ − a2 b∂x6 φ̃ + φ̃txxx + ∂x6 φ̃

= 5asβt∗ ∂x3 (ϕ + ϕ̃) − 5s(∂t + ∂x3 )(βt∗ (φ + φ̃)). (3.25)

With the objective of simplifying the left-hand side of (3.25), we show the existence of
κ1 , κ2 , κ̃1 , κ̃2 ∈ R such that the following equality is satisfied:

c(∂t + κ̃1 ∂x + κ1 ∂x3 )(∂t + κ̃2 ∂x + κ2 ∂x3 )φ̃

= κ̃1 κ̃2 ∂x2 φ̃ − 5sβt∗ ∂x3 (ϕ + ϕ̃) − 5s(∂t + ∂x3 )(βt∗ (φ + φ̃)). (3.26)

The higher order terms form the system:

⎧
⎪ 1
⎪κ1 + κ2 = 1 + c ,
⎪
⎨ (3.27)
⎪ 1−a2 b
⎪
⎪
⎩κ 1 κ 2 = c .

Thus, we search for κ1 such that:

c+1 1 − a2 b
κ21 − κ1 + = 0. (3.28)
c c
The discriminant of (3.28) is given by:
(c + 1)2 1 − a2 b c2 + 2c + 1 − 4c + 4a2 bc (c − 1)2 + 4a2 bc
− 4 = = > 0;
c2 c c2 c2
hence, equation (3.28) and system (3.27) have two real solutions. As for the coefficients of the
terms of order 4, we find the system
⎧
⎪ r
⎪κ̃1 + κ̃2 = c ,
⎪
⎨
⎪ r
⎩κ2 κ̃1 + κ1 κ̃2 = c ,
⎪
⎪
which clearly has at least a real solution (κ̃1 , κ̃2 ). Consequently, the decomposition (3.26) is
possible. In particular, we define the auxiliary variable

θ ∶= (∂t + κ̃2 ∂x + κ2 ∂x3 )φ̃. (3.29)

18
Step 5: a Carleman estimate for Φ̃. We have the following Carleman estimates:

ˆ If we use Proposition 3.1 on φ̃ for (k1 , k2 , k3 ) = (κ̃2 , 0, κ2 ) and consider the homogeneous
Dirichlet boundary conditions, we obtain that:

s6 λ7 ∬ e−2sα ξ 6 ∣φ̃xx ∣2 + s8 λ9 ∬ e−2sα ξ 8 ∣φ̃x ∣2 + s10 λ11 ∬ e−2sα ξ 10 φ̃2

Q Q Q

≤ C (s5 λ5 ∬ e−2sα ξ 5 θ2 + s6 λ7 ∬ e−2sα ξ 6 ∣φ̃xx ∣2 + s10 λ11 ∬ e−2sα ξ 10 φ̃2 ) . (3.30)

Q Qω Qω

ˆ If we use Proposition 3.1 on φ̃x and recall the boundary conditions of Φ̃, we get:

s4 λ5 ∬ e−2sα ξ 4 ∣φ̃xxx ∣2 + s6 λ7 ∬ e−2sα ξ 6 ∣φ̃xx ∣2 + s8 λ9 ∬ e−2sα ξ 8 ∣φ̃x ∣2

Q Q Q

≤ C(s3 λ3 ∬ e−2sα ξ 3 ∣θx ∣2 + s4 λ5 ∬ e−2sα ξ 2 ∣φ̃xxx ∣2 + s8 λ9 ∬ e−2sα ξ 8 ∣φ̃x ∣2

Q Qω Qω
T
+ s2 λ2 ∫ ∫ e−2sα ξ 2 ∣φ̃tx ∣2 ). (3.31)
0 {L}

ˆ If we use Proposition 3.1 on θ for (k1 , k2 , k3 ) = (κ̃1 , 0, κ1 ), we get from c > 0, (3.29) and
(3.26) the following estimate:

s2 λ3 ∬ e−2sα ξ 2 ∣θxx ∣2 + s4 λ5 ∬ e−2sα ξ 4 ∣θx ∣2 + s6 λ7 ∬ e−2sα ξ 6 θ2

Q Q Q
2
≤ C(sλ ∬ e−2sα ξ [κ̃1 κ̃2 φ̃xx − 5sβt∗ (ϕxxx + ϕ̃xxx ) − 5s(∂t + ∂x3 ) (βt∗ (φ + φ̃))]
Q
T
+ s2 λ3 ∬ e−2sα ξ 2 ∣θxx ∣2 + s6 λ7 ∬ e−2sα ξ 6 θ2 + ∫ ∫ e−2sα ∣θt ∣2 ). (3.32)
Qω Qω 0 {0,L}

ˆ From (3.29) we have:

s4 λ5 ∬ e−2sα ξ 4 ∣φ̃t ∣2 ≤ Cs4 λ5 ∬ e−2sα ξ 4 (∣φ̃xxx ∣2 + ∣φ̃x ∣2 + θ2 ). (3.33)

Q Q

ˆ If we consider the second equation of (3.18)1 , it follows that:

2
s4 λ5 ∬ e−2sα ξ 4 ∣ϕ̃xxx ∣2 ≤ Cs4 λ5 ∬ e−2sα ξ 4 (∣φ̃xxx ∣2 + ∣φ̃x ∣2 + ∣φ̃t ∣2 + s2 (βt∗ )2 φ + s2 (βt∗ )2 φ̃2 ).
Q Q
(3.34)

ˆ Thanks to the second equation of (3.18)1 , (3.29) and Poincaré inequality (Proposition 2.1
for i = 0) we obtain:
T ∗
s6 λ6 ∫ e−2sα (ξ ∗ )6 ∥ϕ̃∥2H 2 (0,L)
0
2
≤ Cs6 λ6 ∬ e−2sα (ξ ∗ )6 (∣φ̃xx ∣2 + ∣φ̃x ∣2 + φ̃2 + θ2 + s2 (βt∗ )2 φ + s2 (βt∗ )2 φ̃2 ). (3.35)
Q

19
 ˆ Considering the first equation of (3.18)1 the following holds:

∗
s7/2 λ4 ∬ e−2sα (ξ ∗ )7/2 ∣ϕ̃t ∣2
Q
∗
≤ Cs7/2 λ4 ∬ e−2sα ((ξ ∗ )7/2 (∣ϕ̃xxx ∣2 + ∣φ̃xxx ∣2 + (5sβt∗ )2 (ϕ̃2 + ϕ2 )). (3.36)
Q

We recall that because of (3.9)2 we have for λ ≥ C and s ≥ C(T m + T 3m ) the estimate

(sβt∗ )2 ≤ Cs2+2/m (ξ ∗ )2+2/m .

Thus, the zero-order term of ϕ̃ in the right-hand side of (3.36) can be absorbed by the
left-hand side of (3.35) if m ≥ 4, λ ≥ C and s ≥ C(T m + T 3m ).

So, considering (3.9)2 , (3.24) and (3.30)-(3.36), after some easy absorptions and majorations
we obtain for m ≥ 4, λ ≥ C and s ≥ C(T m + T 3m ) that:

3
s4 λ5 ∬ e−2sα ξ 4 ∣φ̃t ∣2 + ∑ s10−2i λ11−2i ∬ e−2sα ξ 10−2i ∣∂xi φ̃∣2
Q i=0 Q
∗ T ∗
+ s7/2 λ4 ∬ e−2sα (ξ ∗ )7/2 ∣ϕ̃t ∣2 + s4 λ5 ∬ e−2sα ξ 4 ∣ϕ̃xxx ∣2 + s6 λ6 ∫ e−2sα (ξ ∗ )6 ∥ϕ̃∥2H 2 (0,L)
Q Q 0
2
+ ∑ s6−2i λ7−2i ∬ e−2sα ξ 6−2i ∣∂xi θ∣2 + ∥Φ̂∥2L2 (0,T ;H2 (0,L)) + ∥Φ∥2H 1 (0,T ;H1 (0,L))∩L2 (0,T ;H4 (0,L))
i=0 Q
3
≤ C( ∑ s10−2i λ11−2i ∬ e−2sα ξ 10−2i ∣∂xi φ̃∣2 + ∑ s6−2i λ7−2i ∬ e−2sα ξ 6−2i ∣∂xi θ∣2
i=0 Qω i=0,2 Qω

T T ∗
+∫ ∫ e−2sα ∣θt ∣2 + s2 λ2 ∫ ∫ e−2sα ξ 2 ∣φ̃tx ∣2 + ∬ e−8sβ ∣G∣2 ). (3.37)
0 {0,L} 0 {L} Q

Step 6: absorption of the trace terms. Let us first show how to absorb the trace term of φ.
We recall (3.3). From trace interpolation estimates and Young inequality we get:

T T
3/2 1/2
s2 λ2 ∫ ∫ e−2sα ξ 2 ∣φ̃tx ∣2 ≤ Cs2 λ2 ∫ e−2sα∗ (ξ ∗ )2 ∥φ̃t ∥H 2 (0,L) ∥φ̃t ∥L2 (0,L)
0 {L} 0
∗ T ∗
≤ C (s4 λ4 ∬ e−2sα (ξ ∗ )4 ∣φ̃t ∣2 + s4/3 λ4/3 ∫ e−2sα (ξ ∗ )4/3 ∥φ̃t ∥2H 2 (0,L) ) . (3.38)
Q 0

It is clear that the first term in the right-hand side of (3.38) can be absorbed by the left-hand
side of (3.37) for λ ≥ C. As for the second term, we deduce that:

T ∗
s4/3 λ4/3 ∫ e−2sα (ξ ∗ )4/3 ∥φ̃t ∥2H 2 (0,L)
0
T ∗ ∗
≤C∫ ∣(s2/3 λ2/3 e−sα (ξ ∗ )2/3 )t ∣2 ∥φ̃∥2H 2 (0,L) + ∥s2/3 λ2/3 e−sα (ξ ∗ )2/3 φ̃∥2H 1 (0,T ;H 2 (0,L)) . (3.39)
0

20
It can be proved easily that the first term in the right-hand side of (3.39) can be absorbed by
the left-hand side of (3.37) using that m ≥ 4. As for the second one, using (2.10) on Φt we have:
∗ ∗
∥s2/3 λ2/3 e−sα (ξ ∗ )2/3 φ̃∥2H 1 (0,T ;H 2 (0,L)) ≤ (1 + T 2 )(∥(s2/3 λ2/3 e−sα (ξ ∗ )2/3 )t Φ̃∥2H1,0 (Q)
∗ ∗
+ ∥s2/3 λ2/3 e−sα (ξ ∗ )2/3 sβt∗ Φ̃∥2H1,0 (Q) + ∥s2/3 λ2/3 e−sα (ξ ∗ )2/3 sβt∗ Φ∥2H1,0 (Q) ). (3.40)

The third term in the right-hand side of (3.40) can be estimated by (3.24). Moreover, the first
two terms in the right-hand side of (3.40) can be absorbed by the left-hand side of (3.37) if
m ≥ 12. Indeed, the most problematic term is
∗
s10/3 λ4/3 ∬ e−2sα (∣αt∗ ∣2 + ∣βt∗ ∣2 )(ξ ∗ )4/3 ∣Φ̃t ∣2 ,
Q

which can be absorbed by the left-hand side of (3.37) by considering (3.4), (3.9)2 and that
CT 2+2 = CT 4 ≤ s2/m . The other terms can be estimated similarly.

Let us now estimate the trace term of θ. Because of (3.29) and the Dirichlet boundary
conditions, we have that:

θt = κ̃2 φ̃tx + κ2 φ̃txxx , on {0, L} × (0, T ).

In addition, considering that the source term in (3.17) satisfies Dirichlet boundary conditions
and that L∗ can be diagonalize as in (2.16), we obtain from (3.18) that there is κ3 ∈ R such that

θt (⋅, 0) = 0, on (0, T ), θt (⋅, L) = κ3 φ̃tx (⋅, L), on (0, T ).

Thus, by Remark 3.3 we obtain

T T
∫ ∫ e−2sα ∣θt ∣2 ≤ Cs2 λ2 ∫ ∫ e−2sα ξ 2 ∣φ̃tx ∣2 ,
0 {0,L} 0 {L}

a term which has already been treated.

Summing up, after the operations on this step and keeping the interesting terms in (3.37) we
get if m ≥ 12, λ ≥ C and s ≥ C(T m + T 3m ) the following estimate:
3
s4 λ5 ∬ e−2sα ξ 4 ∣φ̃t ∣2 + ∑ s10−2i λ11−2i ∬ e−2sα ξ 10−2i ∣∂xi φ̃∣2
Q i=0 Q
∗ T ∗
+ s7/2 λ4 ∬ e−2sα (ξ ∗ )7/2 ∣ϕ̃t ∣2 + s4 λ5 ∬ e−2sα ξ 4 ∣ϕ̃xxx ∣2 + s6 λ6 ∫ e−2sα (ξ ∗ )6 ∥ϕ̃∥2H 2 (0,L)
Q Q 0

+ ∥Φ̂∥2L2 (0,T ;H2 (0,L)) + ∥Φ∥2H 1 (0,T ;H1 (0,L))∩L2 (0,T ;H4 (0,L))
3
≤ C( ∑ s10−2i λ11−2i ∬ e−2sα ξ 10−2i ∣∂xi φ̃∣2 + ∑ s6−2i λ7−2i ∬ e−2sα ξ 6−2i ∣∂xi θ∣2
i=0 Qω i=0,2 Qω

∗
+ ∬ e−8sβ ∣G∣2 ). (3.41)
Q

21
 Step 7: absorption of the local terms. Using space interpolation and Young inequality, we
obtain that:
3 T 15 ∗
10−2i 11−2i
∑s λ ∬ e−2sα ξ 10−2i ∣∂xi φ̃∣2 ≤ C ∫ e− 8 sα ∥φ̃∥2H 3 (ω)
i=0 Qω 0

sα∗ ∗ ∗
T
− 15 1/2 3/2 13
≤C∫ e 8 ∥φ̃∥L2 (ω) ∥φ̃∥H 4 (ω) ≤ C (∬ e−sα φ̃2 + ∥e− 12 sα φ̃∥2L2 (0,T ;H 4 (0,L)) ) . (3.42)
0 Qω

The second term on the right hand-side of (3.42) can be bounded with (2.26) for p = 1:
13 ∗ 13 ∗
∥e− 12 sα φ̃∥2L2 (0,T ;H 4 (0,L)) ≤ C(∥(e− 12 sα )t Φ̃∥2H 1 (0,T ;H−1 (0,L))∩L2 (0,T ;H1 (0,L))
13 ∗
+ ∥sβt∗ e− 12 sα Φ̃∥2H 1 (0,T ;H−1 (0,L))∩L2 (0,T ;H1 (0,L))
13 ∗
+ ∥sβt∗ e− 12 sα Φ∥2H 1 (0,T ;H−1 (0,L))∩L2 (0,T ;H1 (0,L)) ). (3.43)

The third term in the right-hand side of (3.43) can be estimated with (3.24). As for the other
two terms, they can be absorbed by the left-hand side of (3.41).

In order to study the local term of θ, we substitute θ by the right-hand side of (3.29). The
spatial terms can be treated similarly as before. As for φ̃t we obtain for λ ≥ C and s ≥ C(T m +
T 3m ) the estimate

6−2i 7−2i 63 ∗ 2/3 4/3

∑ s λ ∬ e−2sα ξ 6−2i (∂xi ∂t φ̃)2 ≤ C ∬ e− 32 sα ∥φ̃t ∥L2 (ω) ∥φ̃t ∥H 3 (ω)
i=0,2 Qω Q

15 ∗ T 129 ∗
≤ C (∬ e− 8 sα φ̃2t + ∫ e− 64 sα ∥φ̃t ∥2H 3 (0,L) ) . (3.44)
Qω 0

We can treat the last term at the left hand-side of (3.44) as before thanks to interpolation
results, Corollary 2.8 (for p = 2), Corollary 2.9 (for p = 1), (3.23) and (3.7). As for the first term
in the right-hand side of (3.44), we can simply integrate by parts:
15 ∗ 1 − 15 sα∗ 15 ∗
∬ e− 8 sα ∣φ̃t ∣2 = ∬ (e 8 )tt φ̃2 − ∬ e− 8 sα φ̃tt φ̃
Qω 2 Qω Qω
∗ 11 ∗
≤ C (∬ e−sα φ̃2 + ∬ e− 4 sα ∣φ̃tt ∣2 ) . (3.45)
Qω Q

The last term in the right-hand side of (3.45) can be estimated as before with the help of
Corollary 2.8 (for p = 2), Corollary 2.9 (for p = 1), (3.23) and (3.7).

So, summing up and keeping the interesting terms, from (3.42) we get
T ∗
s6 λ6 ∫ e−2sα (ξ ∗ )6 ∥Φ̃∥2H2 (0,L) + ∥Φ̂∥2L2 (0,T ;H2 (0,L)) + ∥Φ∥2L2 (0,T ;H2 (0,L))
0
∗ ∗
≤ C( ∬ e−sα φ̃2 + ∬ e−8sβ ∣G∣2 ). (3.46)
Qω Q

22
 Step 8: returning to the variable Φ. Considering (3.14) and (3.24), from (3.46) we deduce
that
T ∗
s6 λ6 ∫ e−2sα (ξ ∗ )6 ∥Φ̃∥2H2 (0,L) + ∥Φ̂∥2L2 (0,T ;H2 (0,L)) + ∥Φ∥2L2 (0,T ;H2 (0,L))
0
∗ −10sβ ∗ ∗
≤ C (∬ e−sα φ2 + ∬ e−8sβ ∣G∣2 ) . (3.47)
Qω Q

Therefore, from (3.47) and (3.14) we obtain (3.13).

Remark 3.5. The careful reader can remark that in Steps 2 and 3 of the proof of Proposition 3.2
we have just used that the dynamics of (1.8) are dispersive, that we could replicate those steps
if the dynamics were diffusive and that it is in steps 4 and 5 where we have used the properties
∗
of the specific equation. In addition, the technique in step 2 for decomposing e−5sβ Φ as the
∗
solution of three Cauchy problem can be generalized to decomposing e−5sβ Φ as the solution of N
Cauchy problems, which could be useful in situations in which we need to have higher regularity
for the solutions of analogous Cauchy problems. Thus, the effort of getting a Carleman estimate
in dispersive or diffusive system with source term G = 0 or G ∈ L2 (Q) is very similar.

4 The nonlinear control problem

In the following, we denote

Lr (ρ(t)(0, T ); V ) ∶= {w ∶ ρ(t)w ∈ Lr (0, T ; V )},

for V a Banach space, ρ a measurable function and r ∈ [1, +∞]. Moreover, we define

∥w∥Lr (ρ(t)(0,T );V ) ∶= ∥ρ(t)w∥Lr (0,T ;V ) .

Similarly, we define
C(ρ(t)[0, T ]; V ) ∶= {w ∶ ρ(t)w ∈ C([0, T ]; V )}.

In addition, in this section we fix (m, λ, s) so that Proposition 3.2 is satisfied and
1 ∗
∣sβt∗ ∣ ≤ Ce 8 sβ . (4.1)

In that sense, in the following the generic constant C may also depend on s, λ and T .

The structure of this section is the following: first we study the controllability of the linearized
problem, second we end the proof of Theorem 1.1, and thirdly we make some remarks about
similar control problems which can be studied by using the technique presented in this paper.

23
4.1 Null controllability of (1.5)

Let E be the couples (U, f ) such that:

∗
ˆ U ∈ L2 (e4sβ (0, T ); L2 (0, L)),
7 ∗ 7 ∗
ˆ U ∈ L2 (e 2 sβ (0, T ); H10 (0, L)) ∩ C(e 2 sβ [0, T ]; L2 (0, L)),
∗
ˆ f 1ω ∈ L2 (e5sβ (0, T ); L2 (0, L)),
∗
ˆ LU − (0, f 1ω ) ∈ L2 (e7sβ (0, T ); H−1 (0, L)).

Then, we have the following controllability result:

∗
Proposition 4.1. Let U 0 ∈ L2 (0, L), h1 , h2 ∈ L2 (e7sβ (0, T ); H −1 (0, L)) and assume that (1.4)
is satisfied. Then, there is a solution of (1.5) satisfying (U, f ) ∈ E.

Remark 4.2. If (U, f ) ∈ E, we have that U (T, ⋅) = 0.

For the proof of Proposition 4.1 we follow the approach introduced in [25] which is based on
duality. In particular, we first prove the following auxiliary result:

Lemma 4.3. Let L > 0, ω ⊂ (0, L) be an open subset and assume that (1.4) is satisfied. There
is C > 0 such that
T ∗ ∗ ∗ ∗
∫ e−14sβ ∥Φ∥2H1 (0,L) + ∥e−7sβ Φ∥2C([0,T ];L2 (0,L)) ≤ C (∬ e−10sβ φ2 + ∬ e−8sβ ∣G∣2 ) , (4.2)
0 0 Qω Q

for Φ the energy solution of (1.8) and (m, λ, s) fixed in the beginning of this section.

Proof of Lemma 4.3. It is a straight consequence of (3.5), (3.6)1 , Proposition 2.6 applied to
∗
e−7sβ Φ in [3T /4, T ] and (3.13) that the following estimate holds:
T ∗ ∗ ∗ ∗
∫ e−14sβ ∥Φ∥2H1 (0,L) + ∥e−7sβ Φ∥2C([3T /4,T ];L2 (0,L)) ≤ C (∬ e−10sβ φ2 + ∬ e−8sβ ∣G∣2 ) .
3T /4 0 Qω Q
(4.3)
Let χ be a positive function supported in [0, 7/8] such that χ = 1 in [0, 3/4]. By considering
(3.5), the problem satisfied by χ(t/T )Φ, (2.9) and (4.3) we easily obtain (4.2).

Proof of Proposition 4.1. Let P0 be the subspace of C3 ([0, T ] × [0, L]) of the functions Φ such
that
Φ(⋅, 0) = Φ(⋅, L) = Φx (⋅, 0) = 0, on [0, T ].

Let a ∶ P0 × P0 → R be the bilinear form

∗ ∗
a(Φ̂, Φ) = ∬ e−10sβ φ̂φ + ∬ e−8sβ (L∗ Φ̂) ⋅ (L∗ Φ) (4.4)
Qω Q

24
and ` ∶ P0 → R the linear form
L T
`(Φ) = ∫ U 0 ⋅ Φ(0, x)dx + ∫ ⟨(h1 , h2 ), Φ⟩H−1 (0,L)×H1 (0,L) . (4.5)
0 0 0

Notice that the observability inequality (4.2) holds for every Φ ∈ P0 taking G = L∗ Φ. Con-
sequently, a is a scalar product in P0 and there is C > 0 such that for all Φ ∈ P0 we have
that
∗ √
∣`(Φ)∣ ≤ C(∥e7sβ (h1 , h2 )∥L2 (0,T ;H−1 (0,L)) + ∥(u0 , v0 )∥L2 (0,L) ) a(Φ, Φ).

In the following we denote P the completion of P0 with respect to a (and keep denoting a and
` to their only continuous extensions). By Riesz theorem there is Φ̂ ∈ P such that

a(Φ̂, Φ) = `(Φ), ∀Φ ∈ P.

We define (U, f ) by
∗ ∗
U ∶= e−8sβ L∗ Φ̂, f ∶= −e−10sβ φ̂1ω . (4.6)

It can be proved easily that U is the unique solution by transposition of

⎧
⎪
⎪
⎪
⎪LU = (h1 , h2 + f 1ω ), in Q,
⎪
⎪
⎪
⎨U (⋅, 0) = U (⋅, L) = Ux (⋅, L) = 0, on (0, T ), (4.7)
⎪
⎪
⎪
⎪
⎪
⎪
⎪U (0, ⋅) = U 0 , on (0, L).
⎩
Indeed, let Φ be the solution of (1.8) for G ∈ D(Q) and for ΦT = 0. Then, we have by Lemma
2.5 that Φ ∈ C∞ (Q) ∩ P0 , which implies that
T ∗ ∗
∫ ⟨U, G⟩H−1 (0,L),H1 (0,L) = ∬ e−8sβ (L∗ Φ̂) ⋅ G = ∬ e−8sβ (L∗ Φ̂) ⋅ (L∗ Φ)
0 0 Q Q
∗ L T
= a(Φ̂, Φ) − ∬ e−10sβ φ̂φ = ∫ U 0 ⋅ Φ(0, x)dx + ∫ ⟨(h1 , h2 + f 1ω ), Φ⟩H−1 (0,L)×H1 (0,L) .
Qω 0 0 0

So we just have to use the density of D(Q) in L2 (0, T ; H10 (0, L)). Consequently, since f, h1 , h2 ∈
L2 (0, T ; H −1 (0, L)), we obtain from Remark 2.10 that

U ∈ L2 (0, T ; H10 (0, L)) ∩ C([0, T ]; L2 (0, L)).

As for the estimates, from (4.6) we get

∗ ∗
∬ e10sβ ∣f ∣2 + ∬ e8sβ ∣U ∣2 < +∞. (4.8)
Qω Q

Moreover, since U is the solution by transposition of (4.7) we have that

∗
LU − (0, f 1ω ) = (h1 , h2 ) ∈ L2 (e7sβ (0, T ); H−1 (0, L)). (4.9)

25
 7 ∗
Finally, we consider that Ũ = e 2 sβ U satisfies:
⎧ 7 ∗ 7 ∗
⎪
⎪
⎪
⎪ LŨ = 72 sβt∗ e 2 sβ U + e 2 sβ (h1 , h2 + f 1ω ), in Q,
⎪
⎪
⎪
⎨Ũ (⋅, 0) = Ũ (⋅, L) = Ũx (⋅, L) = 0, on (0, T ),
⎪
⎪
⎪
⎪ 7 ∗
⎪
⎪
⎪ Ũ (0, ⋅) = e 2 sβ (0) U 0 , in (0, L).
⎩
∗
Moreover, from U ∈ L2 (e4sβ (0, T ); H−1 (0, L)) and (4.1) we obtain
7 ∗ 7 sβ ∗ 7 ∗
sβt e 2 U + e 2 sβ (h1 , h2 + f 1ω ) ∈ L2 (0, T ; H−1 (0, L)).
2
Consequently, from Remark 2.10 we find
7 ∗ 7 ∗
U ∈ L2 (e 2 sβ (0, T ); H10 (0, L)) ∩ C(e 2 sβ [0, T ]; L2 (0, L)). (4.10)

Thereby, we obtain from (4.8)-(4.10) that (U, f ) ∈ E, which along with being the solution of
(4.7) completes the proof.

4.2 Proof of Theorem 1.1

We now prove the local null controllability of system (1.3) by a local inversion argument.
We recall that the components of U are denoted by u and v. We consider h1 , h2 as in the
∗
statement of Theorem 1.1, with C large enough so that h1 , h2 ∈ L2 (e7sβ (0, T ); H−1 (0, L)). Let
∗
G ∶ E → L2 (e7sβ (0, T ); H−1 (0, L)) × L2 (0, L) be an operator defined by:

G(U, f ) ∶= (ut + uux + uxxx + avxxx + k1 vvx + k2 (uv)x − h1 ,

cvt + rvx + vvx + abuxxx + vxxx + k2 buux + k1 b(uv)x − h2 − f 1ω , u(0, ⋅), v(0, ⋅)). (4.11)

We recall that if G is C 1 and if DG[0] is surjective, there exists δ > 0 such that if ∥U 0 ∥ < δ there
exists (U, f ) ∈ E such that
G(U, f ) = (0R2 , U 0 ).
Hence, by Remark 4.2 we just have to check those two hypotheses to conclude the proof of
Theorem 1.1:

ˆ In order to see that G is C 1 , since G is the sum of an affine form and some bilinear forms,
it suffices to see that it is a bounded operator. The fact that the affine form is continuous
is a straight consequence of the definition of E. As for the bilinear terms of (4.11), we
7 ∗
can prove that they are bounded by considering that for all w1 ∈ L∞ (e 2 sβ (0, T ); L2 (0, L))
7 ∗
and w2 ∈ L2 (e 2 sβ (0, T ); H01 (0, L)) we have the estimate

∥w1 ∂x w2 ∥L2 (e7sβ∗ (0,T );H −1 (0,L)) ≤ C∥w1 ∂x w2 ∥L2 (e7sβ∗ (0,T );L1 (0,L))
≤ C∥w1 ∥ 7 ∗ ∥w2 ∥ 7 ∗ .
L∞ (e 2 sβ (0,T );L2 (0,L)) L2 (e 2 sβ (0,T );H01 (0,L))

26
 ˆ As for the surjectivity of DG[0], we remark that

DG[0](U, f ) = (LU − (0, f 1ω ), U (0, ⋅));

thus, it is a consequence of Proposition 4.1.

4.3 Some additional remarks

The technique presented in this paper can be used to solve other controllability problems:

ˆ We can prove the local null controllability of the following control system
⎧
⎪
⎪
⎪
⎪ut + uux + uxxx + avxxx + k1 vvx + k2 (uv)x = h1 + f 1ω , in Q,
⎪
⎪
⎪
⎪
⎪
⎪cvt + rvx + vvx + abuxxx + vxxx + k2 buux + k1 b(uv)x = h2 , in Q,
⎪
⎪
⎪
⎪
⎨u(⋅, 0) = u(⋅, L) = ux (⋅, L) = 0, on (0, T ), (4.12)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪v(⋅, 0) = v(⋅, L) = vx (⋅, L) = 0, on (0, T ),
⎪
⎪
⎪
⎪ 0 0
⎩u(0, ⋅) = u , v(0, ⋅) = v , on (0, L),
⎪
⎪
assuming (1.4). Indeed, we can do that by replicating the proof of Theorem 1.1 step by
step.

ˆ We can slightly improve the result presented in [16] and show that the general Hirota-
Satsuma system can be controlled with two forces supported in any two control domains
(in [16] the authors need that one of the control domains touches the boundary). The proof
is very similar to the one of Theorem 1.1, as we only have to do some minor adjustments.

ˆ An interesting question that remains open for the control system (1.3) is the possibility
of getting exact controllability results similar to those presented in [11] but with a control
acting in one component; i.e. to determine if for L̃ ∈ (0, L) and u∗ , v ∗ ∈ L2 (0, L̃) we can
obtain (u, v)(T, ⋅)∣(0,L̃) = (u∗ , v ∗ ) when ω is compactly included in (L̃, L).

ˆ The second and third hypothesis in (1.4) are needed for the well-posedness of system (1.3)
and its linearized system. Indeed, we use them in the proof of Proposition 2.6 to diagonalize
higher order terms. In addition, these conditions have a physical meaning as explained
in [26, Section 4]. Moreover, the first hypothesis in (1.4) is only necessary to prove the
controllability of the linearized system (as otherwise we have two independent equations).
In fact, determining the null controllability of (1.3) when a = 0 remains an interesting
open problem (in that case the other two hypotheses of (1.3) are still necessary to ensure
the well-posedness of the system). Finally, there is no extra hypothesis for the coefficients
which just multiply lower order terms (i.e. for r, k1 and k2 ) because the important terms
for the well-posedness and controllability properties are the higher order ones.

27
References

[1] R. A. Adams and J. J.F. Fournier. Sobolev spaces. Elsevier, 2003.

[2] F. Ammar-Khodja, A. Benabdallah, and C. Dupaix. Null-controllability of some reaction–

diffusion systems with one control force. J. Math. Anal. App., 320(2):928–943, 2006.

[3] F. Ammar-Khodja, A. Benabdallah, C. Dupaix, and I. Kostin. Controllability to the trajec-

tories of phase-field models by one control force. SIAM J. Control Optim., 42(5):1661–1680,
2003.

[4] 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.
F., 1(3):267–306, 2011.

[5] F. D. Araruna, E. Cerpa, A. Mercado, and M. C. Santos. Internal null controllability of a

linear Schrödinger–KdV system on a bounded interval. J. Differ. Equations, 260(1):653–
687, 2016.

[6] J. L. Bona, G. Ponce, J.-C. Saut, and M. M. Tom. A model system for strong interaction
between internal solitary waves. Comm. Math. Phys., 143(2):287–313, 1992.

[7] J. A. Bárcena-Petisco. Null controllability of a penalized Stokes problem in dimension two

with one scalar control. Asymptotic Anal., 117(3-4):161–198, 2020.

[8] R. A. Capistrano-Filho, F. A. Gallego, and A. F. Pazoto. Neumann boundary controllability

of the Gear-Grimshaw system with critical size restrictions on the spatial domain. Z. Angew.
Math. Phys., 67(5):Art. 109, 36, 2016.

[9] R. A. Capistrano-Filho, F. A. Gallego, and A. F. Pazoto. Boundary controllability of a

nonlinear coupled system of two Korteweg–de Vries equations with critical size restrictions
on the spatial domain. Math. Control Signals Syst., 29(1):Art. 6, 37, 2017.

[10] R. A. Capistrano-Filho, V. Komornik, and A. F. Pazoto. Pointwise control of the linearized

Gear–Grimshaw system. Evol. Equ. Control The., 9(3):693–719, 2020.

[11] R. A. Capistrano-Filho, A. F. Pazoto, and L. Rosier. Internal controllability of the

Korteweg–de Vries equation on a bounded domain. ESAIM: COCV, 21(4):1076–1107, 2015.

[12] R. A. Capistrano-Filho, A. F. Pazoto, and L. Rosier. Control of a Boussinesq system of

KdV-KdV type on a bounded interval. ESAIM: COCV, 25:58, 2019.

28
[13] N. Carreño and S. Guerrero. Local null controllability of the N -dimensional Navier–Stokes
system with N − 1 scalar controls in an arbitrary control domain. J. Math. Fluid Mech.,
pages 1–15, 2013.

[14] N. Carreño, S. Guerrero, and M. Gueye. Insensitizing controls with two vanishing compo-
nents for the three-dimensional Boussinesq system. ESAIM: COCV, 21(1):73–100, 2015.

[15] N. Carreño, E. Cerpa, and A. Mercado. Boundary controllability of a cascade system

coupling fourth-and second-order parabolic equations. Syst. & Control Lett., 133:104542,
2019.

[16] N. Carreño, E. Cerpa, and E. Crépeau. Internal null controllability of the generalized
Hirota-Satsuma system. ESAIM:COCV, 26, 2020.

[17] E. Cerpa and A. F. Pazoto. A note on the paper “On the controllability of a coupled system
of two Korteweg-de Vries equations”. Commun. Contemp. Math., 13(1):183–189, 2011.

[18] J.-M. Coron and S. Guerrero. Null controllability of the N -dimensional Stokes system with
N − 1 scalar controls. J. Differ. Equations, 246(7):2908–2921, 2009.

[19] 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.

[20] L. De Teresa and E. Zuazua. Identification of the class of initial data for the insensitizing
control of the heat equation. Commun. Pure Appl. Anal., 8(1):457–471, 2009.

[21] S. Dolecki and D. L. Russell. A general theory of observation and control. SIAM J. Control
Optimization, 15(2):185–220, 1977.

[22] 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: COCV,
12(3):442–465, 2006.

[23] E. Fernández-Cara and S. Guerrero. Global carleman inequalities for parabolic systems and
applications to controllability. SIAM J. Control. and Optim., 45(4):1395–1446, 2006.

[24] E. Fernández-Cara, S. Guerrero, O. Yu. 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.

[25] A. V. Fursikov and O. Yu. Imanuvilov. Controllability of evolution equations. Number 34.
Seoul National University, 1996.

29
[26] J. A. Gear and R. Grimshaw. Weak and strong interactions between internal solitary waves.
Stud. Appl. Math., 70(3):235–258, 1984.

[27] O. Glass and S. Guerrero. Some exact controllability results for the linear KdV equation
and uniform controllability in the zero-dispersion limit. Asymptotic Anal., 60(1-2):61–100,
2008.

[28] O. Glass and S. Guerrero. Uniform controllability of a transport equation in zero diffusion–
dispersion limit. Math. Mod. Meth. Appl. S., 19(09):1567–1601, 2009.

[29] 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.

[30] S. Guerrero. Controllability of systems of Stokes equations with one control force: existence
of insensitizing controls. Ann. I. H. Poincaré, 24:1029–1054, 2007.

[31] G. Lebeau and E. Zuazua. Null-Controllability of a System of Linear Thermoelasticity.

Arch. Ration. Mech. An., 141(4):297–329, 1998.

[32] J.-L. Lions. Contrôlabilité exacte, perturbations et stabilisation de systèmes distribués.

Tome 1, volume 8 of Recherches en Mathématiques Appliquées [Research in Applied Mathe-
matics]. Masson, Paris, 1988. Contrôlabilité exacte. [Exact controllability], With appendices
by E. Zuazua, C. Bardos, G. Lebeau and J. Rauch.

[33] J.-L. Lions and E. Zuazua. A generique uniqueness result for the Stokes system and its
control theoretical consequences. Partial differential equations and applications: Collected
Papers in Honor of Carlo Pucci, 177:221–235, 1996.

[34] S. Micu and J. H. Ortega. On the controllability of a linear coupled system of Korteweg-de
Vries equations. In Mathematical and numerical aspects of wave propagation (Santiago de
Compostela, 2000), pages 1020–1024. SIAM, Philadelphia, PA, 2000.

[35] S. Micu, J. H. Ortega, and A. F. Pazoto. On the controllability of a coupled system of two
Korteweg–de Vries equations. Commun. Contemp. Math., 11(05):799–827, 2009.

[36] L. Rosier. Exact boundary controllability for the Korteweg-de Vries equation on a bounded
domain. ESAIM: COCV, 2:33–55, 1997.

30