J. Math. Anal. Appl.

495 (2021) 124693

Contents lists available at ScienceDirect

Journal of Mathematical Analysis and Applications

[Link]/locate/jmaa

Rational energy decay rate for the wave equation with delay term
on the dynamical control
Gilbert Bayili a , Serge Nicaise b,∗ , Roland Silga a
a
Université Joseph KI-ZERBO, Laboratoire de Mathématiques et Informatique (LAMI), UFR, Sciences
Exactes et appliquées, Département de Mathématiques, 03 B.P.7021 Ouagadougou 03, Burkina Faso
b
Université Polytechnique Hauts-de-France, LAMAV - Laboratoire de Mathématiques et leurs
Applications de Valenciennes, FR CNRS 2956, F-59313 Valenciennes, France

a r t i c l e i n f o a b s t r a c t

Article history: We consider the wave equation with a delay term in the dynamical control. If the
Received 24 June 2020 delay term is small enough, we establish the strong stability of this system, but
Available online 14 October 2020 show that it is not exponentially stable. We then prove that the system with delay
Submitted by D. Donatelli
has the same rational decay rate than the one without delay.
Keywords: © 2020 Elsevier Inc. All rights reserved.
Time delay
Dynamical control
Strong stability
Lack of uniform stability
Rational decay

1. Introduction

Many evolution equations coming from applications (in biology, electrical engineering and mechanics, ...)
present a time-delay in the feedback law modeling mechanical time lag for instance [15,32,2]. In many cases,
arbitrarily small delays in the feedback may destabilize the system, see for instance [10–12,16,24–26,31,36].
Therefore the stability issue of systems with delay is of theoretical and practical importance.
Different methods have been recently developed to prove some exponential or polynomial decay of the
energy of wave type equations with delay, we refer to [25–27,3,28,4] and the references citing them. In this
paper, we study a wave equation set in a bounded domain (in any space dimension) with a dynamical control
and prove that if the delay term is small enough, then the system with delay has the same (polynomial)
decay rate than the one without delay. The main (and simple) idea is to use a duality argument already
used in [1, §2] in a fully different context and in [7] for a system without dynamical control. Hence our
purpose is to generalize these results to a system with delayed dynamical control.

* Corresponding author.
E-mail addresses: bgilbert8@[Link] (G. Bayili), [Link]@[Link] (S. Nicaise), silgaroland@[Link] (R. Silga).

[Link]
0022-247X/© 2020 Elsevier Inc. All rights reserved.
2 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693

Let us shortly describe the distributed parameter systems that we will analyse. Let Ω ⊂ Rn be an open
bounded domain with a lipschitz boundary Γ. We assume that Γ is divided into two open parts ΓD and ΓN ,
i.e. Γ = ΓD ∪ ΓN with meas (ΓD ) = 0 and meas (ΓN ) = 0.
In this domain Ω, we consider the following wave equation with delay term on the dynamical control
⎧
⎪
⎪ utt (x, t) − Δu(x, t) = 0 in Ω × (0, +∞),
⎪
⎪
⎪
⎪ u(x, t) = 0 on ΓD × (0, +∞),
⎪
⎪
⎪
⎪ ∂u
⎪
⎨ (x, t) + η(x, t) = 0 on ΓN × (0, +∞),
∂ν
ηt (x, t) − ut (x, t) + β1 η(x, t) + β2 η(x, t − τ ) = 0 on ΓN × (0, +∞), (1.1)
⎪
⎪
⎪ u(x, 0) = u , u (x, 0) = u in Ω,
⎪
⎪
⎪ 0 t 1
⎪
⎪
⎪
⎪ η(x, 0) = η on Γ ,
⎪ 0 N
⎩ η(x, t − τ ) = f (x, t − τ ) on Γ × (0, τ ),
0 N

∂u
where ν(x) denotes the outer unit normal vector to the point x ∈ Γ and is the normal derivative of u.
∂ν
Besides η denotes the dynamical control, τ > 0 is the time delay, β1 and β2 are positive constants and the
initial data (u0 , u1 , f0 ) belong to a suitable space (precisely described below). The damping of the system
is made via the indirect damping mechanism on ΓN .
Our first goal is to show that this system is well-posed and is strongly stable under the assumption (that
we suppose to hold in the whole paper)

β2 < β1 . (1.2)

Afterwards, we show that this system is not exponentially stable. Finally we show that if the system (1.1)
without delay (i.e., with β2 = 0) is polynomially stable, then system (1.1) with delay inherits the same
polynomial decay rate.
The paper is organized as follows: section 2 is devoted to the well-posedness and strong stability of
problem (1.1). In section 3 we establish the non uniform stability, and finally in section 4 we prove the
rational stability.

2. Well posedness and strong stability

2.1. Well posedness

In this section we will give the well posedness for the problem (1.1) using the semigroup theory, and then
establish a strong stability result. Let us set

z(x, ρ, t) = η(x, t − τ ρ), x ∈ ΓN , ρ ∈ (0, 1) , t ∈ (0, +∞) .

The problem (1.1) is now equivalent to

⎧
⎪
⎪ utt (x, t) − Δu(x, t) = 0 in Ω × (0, +∞),
⎪
⎪
⎪
⎪ τ zt (x, ρ, t) + zρ (x, ρ, t) = 0 in ΓN × (0, 1) × (0, +∞) ,
⎪
⎪
⎪
⎪ u(x, t) = 0 on ΓD × (0, +∞),
⎪
⎪
⎪
⎪
⎨ ∂ν u(x, t) + η(x, t) = 0 on ΓN × (0, +∞),
ηt (x, t) − ut (x, t) + β1 η(x, t) + β2 z(x, 1, t) = 0 on ΓN × (0, +∞), (2.1)
⎪
⎪
⎪
⎪ z(x, 0, t) = η(x, t) on ΓN × (0, +∞) ,
⎪
⎪
⎪
⎪ u(x, 0) = u0 , ut (x, 0) = u1 in Ω,
⎪
⎪
⎪
⎪ η(x, 0) = η0 on ΓN ,
⎪
⎪
⎩
z(x, ρ, 0) = f0 (x, −ρτ ) on ΓN × (0, τ ).
 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693 3

We set

T
U = (u, ut , η, z) .

Then we have
 T
Ut = (ut , utt , ηt , zt )T = ut , Δu, (ut − β1 η − β2 z(., 1))|ΓN , −τ −1 zρ .

Therefore problem (2.1) can be rewritten in an abstract framework:


Ut = AU,
T (2.2)
U(0) = (u0 , u1 , η0 , f0 (·, − · τ )) ,

where the operator A is defined by

T
A (u, v, η, z) = v, Δu, γ (v − β1 η − β2 z(·, 1)) , −τ −1 zρ
T
,

with domain

T ∂ν u = −η
D(A) = (u, v, η, z) ∈ D(Δ) × HΓ1D (Ω) × L2 (ΓN ) × L2 ΓN × H 1 (0, 1) on ΓN ,
z (·, 0) = η

where

HΓ1D (Ω) = u ∈ H 1 (Ω) , u = 0 on ΓD ,

γ is the trace mapping from H 1 (Ω) into L2 (ΓN ), and D(Δ) is the “maximal” domain of the Laplace operator

D(Δ) = u ∈ HΓ1D (Ω) | Δu ∈ L2 (Ω), and ∂ν u ∈ L2 (ΓN ) .

Let us now introduce the Hilbert space

H = HΓ1D (Ω) × L2 (Ω) × L2 (ΓN ) × L2 (ΓN × (0, 1))

with the norm

 2
 T 2
(u, v, η, z)  = ∇u2L2 (Ω)n + v2L2 (Ω) + ηL2 (ΓN ) + ζz2L2 (ΓN ×(0,1))
H

and the natural associated inner product

⎛ ⎞ ⎛ ⎞
 u u∗ 
⎜ v ⎟ ⎜ v∗ ⎟    1
⎜ ⎟ ⎜ ⎟ ∗ ∗ ∗
⎜ ⎟,⎜ ∗ ⎟ = (∇u · ∇ū + vv̄ ) dx + η η̄ dΓ + ζ z(x, ρ)z̄ ∗ (x, ρ) dρ,
⎝η⎠ ⎝η ⎠
ΓN 0
z∗
Ω ΓN
z H

where, here and below, ζ is a fixed constant satisfying

τ β1 < ζ < τ (2β1 − β2 ), (2.3)

which always exists due to the assumption (1.2).

4 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693

Proposition 2.1. The operator A defined above is m-dissipative.

Proof. By Green’s formula and Cauchy-Schwarz’s inequality, we see that

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
 u u   v u 
⎜v⎟ ⎜v⎟ ⎜ ⎟ ⎜v⎟
⎜ ⎟ ⎜ ⎟ ⎜ Δu ⎟ ⎜ ⎟
A⎜ ⎟,⎜ ⎟ = ⎜ ⎟,⎜ ⎟
⎝η⎠ ⎝η⎠ ⎝ γ (v − β1 η − β2 z(·, 1)) ⎠ ⎝ η ⎠
z z H
−τ −1 zρ z H
⎛
  
= ⎝ ∇v · ∇ū dx + Δu v̄ dx + (v − β1 η − β2 z(·, 1)) η̄ dΓ
Ω Ω ΓN
⎞
 1
−ζτ −1 zρ (x, ρ)z̄(x, ρ) dρdΓ⎠
ΓN 0
 
=− ηv̄ dΓ + (v − β1 η − β2 z(·, 1)) η̄ dΓ
ΓN ΓN
  1
−1 1
−ζτ |z(x, ρ)|2 dΓ
2 0
ΓN
  
ζτ −1 2
= −β1 |η|2 dΓ − β2 z(x, 1)η̄ dΓ − |z(x, 1)| dΓ
2
ΓN ΓN ΓN
−1 
ζτ 2
+ |η| dΓ
2
ΓN
  
2 ζτ −1
≤ −β1 |η| dΓ + β2 |z(·, 1)η̄| dΓ − |z(x, 1)|2 dΓ
2
ΓN ΓN ΓN
−1 
ζτ
+ |η|2 dΓ
2
ΓN
  
β2 β2
≤ −β1 |η|2 dΓ + |z(x, 1)|2 dΓ + |η|2 dΓ
2 2
ΓN ΓN ΓN
−1  −1 
ζτ ζτ
− |z(x, 1)|2 dΓ + |η|2 dΓ.
2 2
ΓN ΓN

This proves that

⎛ ⎞ ⎛ ⎞
 u u 
⎜v ⎟ ⎜v ⎟ 2
A⎝ ⎠,⎝ ⎠ ≤ Λ1 ηL2 (ΓN ) + Λ2 z (·, 1) 2L2 (ΓN ) , (2.4)
η η
z z H

with
β2 ζ β2 ζ
Λ1 = −β1 + + τ −1 , Λ2 = − τ −1 (2.5)
2 2 2 2
that are negative constants due to (2.3). We therefore deduce that A is dissipative.
 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693 5

Now, we will show that the operator λI − A is surjective for all λ ≥ 0.

Given F = (f, g, h, k)T ∈ H, we look for U = (u, v, η, z)T ∈ D(A) solution of

(λI − A) U = F

that is
⎧
⎪
⎪ λu − v = f
⎪
⎨ λv − Δu = g
(2.6)
⎪
⎪ λη − (v − β1 η − β2 z(·, 1)) = h
⎪
⎩ λz + τ −1 z = k.
ρ

Assuming for the moment that we have found η and z with appropriate regularity such that

z (·, 0) = η. (2.7)

Then using (2.7) we deduce from the fourth equation of (2.6) that

ρ
−λτ ρ −λτ ρ
z (·, ρ) = ηe + τe k (·, σ) eλτ σ dσ ∈ L2 (ΓN × (0, 1)) ; (2.8)
0

consequently

1
−λτ −λτ
z (·, 1) = ηe + τe k (·, σ) eλτ σ dσ. (2.9)
0

Eliminating v from the first identity in (2.6), namely

v = λu − f, (2.10)

and using (2.9) in the third identity of (2.6), we find

η = λCλ u − f˜λ on ΓN , (2.11)

where for shortness we set

⎧ 1
⎪
⎪
⎪ Cλ = λ + β1 + β2 e−λτ ,
⎨ ⎛ ⎞
1 (2.12)
⎪
⎪ ˜ ⎝ −λτ
k (·, σ) eλτ σ dσ − h⎠ .
⎪
⎩ fλ = Cλ f + β2 τ e
0

It follows that u verifies

⎧
⎪
⎨ −Δu + λ u = g + λf in Ω
2

u = 0 on ΓD (2.13)
⎪
⎩ ∂ u + λC u = f˜ on Γ .
ν λ λ N

Multiplying the first equation of (2.13) by ϕ ∈ HΓ1D (Ω) and integrating formally by parts we get
6 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693

aλ (u, ϕ) = Lλ (ϕ), ∀ ϕ ∈ HΓ1D (Ω) , (2.14)

where the bilinear and continuous form aλ is given by

  
aλ (u, ϕ) = ∇u∇ϕ dx + λ 2
uϕ dx + Cλ λ uϕ dΓ
Ω Ω ΓN

while the linear form Lλ is

 
Lλ (ϕ) = (g + λf ) ϕ dx + f˜λ ϕ dΓ.
Ω ΓN

Since aλ is clearly strongly coercive on HΓ1D (Ω) × HΓ1D (Ω) and Lλ is continuous on HΓ1D (Ω), the Lax-
Milgram theorem ensures that the problem (2.14) admits a unique weak solution u ∈ HΓ1D (Ω).
Moreover, considering test function φ ∈ D (Ω), we recover the first identity of (2.13) in D (Ω) that is in
distributions sense; consequently Δu ∈ L2 (Ω) since f and g belong to L2 (Ω). Coming back to (2.14), and
again applying Green’s formula, we find that

∂ν u + λCλ u = f˜λ on ΓN .

Further since f˜λ and u belong to L2 (ΓN ), we deduce that u belongs to D(Δ). Furthermore, from (2.6) and
(2.13) we define v and η by (2.10) and (2.11) respectively, which gives the regularity v = λu −f ∈ HΓ1D (Ω) ⊂
L2 (Ω), η ∈ L2 (ΓN ) and the boundary condition ∂ν u = −η on ΓN . Finally we have found U ∈ D(A) which
verifies (2.6). This shows that the operator A is m-dissipative on H and then generates a C0 -semigroup of
contractions in H. Thanks to Lumer-Phillips’ theorem, problem (1.1) is well posed. 

2.2. Strong stability

The main result of this subsection is the following.

Theorem 2.2. The C0 -semigroup etA t≥0

is strongly stable on the energy space H, that is for any U0 ∈ H,

 
lim etA U0 H = 0.
t−→0

Since it is easy to see that the resolvent of A is not compact, then the classical methods such as Lasalle’s
invariance principle [18] or the spectral decomposition theory of Sz-Nagy-Foias, Foguel and Benchimol
[13,8,33] are not applicable in this case. Then we will study the strong stability of system (1.1) by using
a general criterion of Arendt-Batty [5]. Following this criterion, the C0 -semigroup of contractions etA is
strongly stable if σ(A) ∩ iR is countable and no eigenvalue of A lies on the imaginary axis, where σ(A)
denotes the spectrum of A.
First of all, let us look for the spectrum on the imaginary axis, and next analyse the behaviour of the
resolvent on the imaginary axis.

Lemma 2.3. For all λ ∈ R we have

ker (iλI − A) = {0} ;

more precisely, there is no eigenvalue of A on the imaginary axis.

 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693 7

Proof. Let us set U = (u, v, η, z)T . Given λ ∈ R, let U ∈ ker (iλI − A). It follows that

iλU − AU = 0; (2.15)

which is equivalent to
⎧
⎪
⎪ iλu − v = 0 in Ω
⎪
⎨ iλv − Δu = 0 in Ω
(2.16)
⎪
⎪ iλη − v + β1 η + β2 z (·, 1) = 0 on ΓN
⎪
⎩ iλz + τ −1 z = 0 on Γ × (0, 1) .
ρ N

Recalling the dissipativity property (2.4) of A, we get

⎛ ⎞ ⎛ ⎞
 u u 
⎜v⎟ ⎜v⎟ 2
0= A⎝ ⎠,⎝ ⎠ ≤ Λ1 ηL2 (ΓN ) + Λ2 z (·, 1) 2L2 (ΓN ) ≤ 0, (2.17)
η η
z z H

and as Λ1 and Λ2 from (2.5) are negative constants, we deduce that


η = 0 on ΓN
(2.18)
z = 0 on ΓN × (0, 1) .

Combining (2.18) and the third equation of (2.16) gives

v = 0 on ΓN . (2.19)

Using now (2.19) in the first equation of (2.16) it follows that

u = 0 on Γ. (2.20)

Furthermore U ∈ D(A) implies that

∂ν u = −η = 0 on ΓN . (2.21)

Combining the first and the second equation of (2.16) and using (2.20) and (2.21) we get
⎧
⎪ 2
⎨ Δu + λ u = 0 in Ω
u = 0 on Γ (2.22)
⎪
⎩ ∂ u = 0 on Γ .
ν N

Thanks to Holmgren’s theorem (see [21]), the unique solution of system (2.22) is u = 0. Finally we conclude
that U = 0, the proof of Lemma 2.3 is thus completed. 

Lemma 2.4. Equipped with the norm

uD(Δ) = ∇uL2 (Ω)n + ΔuL2 (Ω) + ∂ν uL2 (ΓN ) ,

D(Δ) is compactly embedded into HΓ1D (Ω).

8 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693

Proof. Let (uk )k∈N be a bounded sequence in D(Δ), or equivalently

∇uk L2 (Ω)n + Δuk L2 (Ω) + ∂ν uk L2 (ΓN ) ≤ C, ∀k ∈ N, (2.23)

for a positive constant C. Then by Kondrachov embedding theorem and a trace theorem, there exist u ∈
HΓ1D (Ω) and a subsequence, still denoted by (uk )k∈N for shortness such that

uk → u strongly in L2 (Ω) as k → ∞, (2.24)

uk → u strongly in L2 (ΓN ) as k → ∞. (2.25)

Now by setting fk := Δuk in Ω and gk := ∂ν uk on ΓN , we notice that

  
|∇(uk − u )| dx = −
2
(fk − f )(uk − u )dx + (gk − g )(uk − u )dx
Ω Ω ΓN

and by Cauchy-Schwarz’s inequality and (2.23) we deduce that


|∇(uk − u )|2 dx ≤ 2C uk − u L2 (Ω) + uk − u L2 (ΓN ) .
Ω

By (2.24) and (2.25), this right-hand side tends to zero as k goes to infinity and we deduce that (uk )k∈N is
a Cauchy sequence in HΓ1D (Ω), hence the conclusion. 

Lemma 2.5. For all λ ∈ R we have

R (iλI − A) = H.

Proof. Since we have already shown that A is surjective, it suffices to treat the case λ = 0. Hence let λ ∈ R∗
and F = (f, g, h, k)T ∈ H. Then we look for U = (u, v, η, z)T ∈ D (A) such that

(iλI − A) U = F. (2.26)

Equivalently, we have
⎧
⎪
⎪ iλu − v = f in Ω
⎪
⎨ iλv − Δu = g in Ω
(2.27)
⎪ iλη − v + β1 η + β2 z (·, 1) = h on ΓN
⎪
⎪
⎩ iλz + τ −1 z = k on Γ × (0, 1) .
ρ N

Proceeding as in the proof of Proposition 2.1 we arrive to problem (2.13) with λ replaced by iλ, its variational
formulation being (2.14) with λ replaced by iλ. Since aiλ is no more coercive in HΓ1D (Ω), we use a compact
perturbation argument. Namely we introduce the (unbounded) operator Aiλ from D(A) ⊂ H to H by

Aiλ (u, v, η, z)T = (−v, −Δu, iλη − v + β1 η + β2 z(·, 1), iλz + τ −1 zρ )T , ∀(u, v, η, z)T ∈ D(A).

We first show that Aiλ is an isomorphism from D(A) to H. Indeed given F = (f, g, h, k)T ∈ H, we look for
U = (u, v, η, z)T ∈ D(A) solution of

Aiλ U = F,
 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693 9

or equivalently,
⎧
⎪
⎪ −v = f in Ω
⎪
⎨ −Δu = g in Ω
(2.28)
⎪
⎪ iλη − v + β1 η + β2 z (·, 1) = h on ΓN
⎪
⎩ iλz + τ −1 z = k on Γ × (0, 1) .
ρ N

This directly yields v = −f and as in the proof of Proposition 2.1, we find

ρ
−iλτ ρ −iλτ ρ
z (·, ρ) = ηe + τe k (·, σ) eiλτ σ dσ.
0

Eliminating v and z (·, 1) in the third identity of (2.28), we find

η = −f˜iλ on ΓN , (2.29)

where f˜iλ was defined in (2.12). It follows that u verifies

⎧
⎪
⎨ −Δu = g in Ω
u = 0 on ΓD (2.30)
⎪
⎩ ∂ u = f˜ on Γ .
ν iλ N

Since g ∈ L2 (Ω) and f˜iλ belongs to L2 (ΓN ), it is well-known that a unique solution u exists in HΓ1D (Ω),
with the regularity

Δu ∈ L2 (Ω), and ∂ν u ∈ L2 (ΓN ).

This furnishes a unique solution (u, v, η, z)T ∈ D(A) of (2.28).

Now we notice that for (u, v, η, z)T ∈ D(A), we have
T T
((iλI − A) − Aiλ ) (u, v, η, z) = (iλu, iλv, 0, 0) .

By Lemma 2.4 and again the Kondrachov embedding theorem, we deduce that (iλI − A) − Aiλ is a compact
operator from D(A) (equipped with the graph norm) into H, because (using the inequality (a + b + c)2 ≤
4(a2 + b2 + c2 ) valid for all real numbers a, b, c)

uD(Δ) + ∇vL2 (Ω)n = ∇uL2 (Ω)n + ΔuL2 (Ω) + ηL2 (ΓN ) + ∇vL2 (Ω)n
≤ 2(u, v, η, z)T H + A(u, v, η, z)T H .

As Aiλ is an isomorphism, we deduce that iλI − A is a Fredholm operator of index zero. As Lemma 2.3
guarantees its injectivity, we deduce that it is also surjective. 

Proof of Theorem 2.2. Lemma 2.3 ensures that there is no eigenvalue of A on the imaginary axis, while
Lemma 2.5 guarantees that σ(A) ∩ iR = ∅. That achieves the proof of Theorem 2.2. 

3. Lack of exponential stability

In this section, we will show that the system (1.1) is not exponential stable. Our argument is based on a
frequency domain approach for exponential stability (see Huang [17] and Prüss [29]), more precisely on the
next result.
10 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693

Lemma 3.1. A C0 -semigroup etA t≥0

of contractions on a Hilbert space H is exponentially stable, namely
satisfies
 tA 
e U0  ≤ Ce−ωt U0 H ∀ U0 ∈ H, ∀ t ≥ 0, (3.1)
H

for some positive constants C and ω if and only if

ρ(A) ⊃ {iβ, β ∈ R} ≡ iR (3.2)

and
 
 −1 
sup (iβ − A)  <∞ (3.3)
β∈R L(H)

where ρ(A) denotes the resolvent set of the operator A.

The main result of current section is the following.

Theorem 3.2. The system (1.1) is not exponentially stable in the energy space H.

Proof. As the condition (3.2) is guaranteed by Lemma 2.3, it suffices now to build a contradiction following
some techniques used in [22,34,30]. More precisely, we prove that the condition (3.3) is not satisfied in the
sense that there exist a positive real number C and some sequences of λn ∈ iR, Un = (un , vn , ηn , zn )T ∈ D(A)
and Fn = (fn , gn , hn , kn )T ∈ H, with n ∈ N such that

(λn − A)Un = Fn , ∀n ∈ N, (3.4)

Un H ≥ C, ∀n ∈ N, (3.5)

and

lim Fn H = 0. (3.6)

n→+∞

Following [30, Theorem 3.1], let (μ2n )n∈N be the sequence of eigenvalues of the Laplacian with Dirichlet
boundary condition on ΓD and Robin boundary condition on ΓN (repeated according to their mulitplicity)
and let ϕn be its associated normalized eigenfunction, more precisely, solution of
⎧
⎪
⎨ −Δϕn = μn ϕn in Ω,
2

ϕn = 0 on ΓD , (3.7)
⎪
⎩ ∂ ϕ + ϕ = 0 on Γ ,
ν n n N

with

ϕn L2 (Ω) = 1. (3.8)

Without loss of generality, we can assume that all μn are positive. Then for all n ∈ N, we chose
ϕn ϕn
λn = iμn , un = , vn = ϕn , zn (·, ρ) = ηn e−iμn τ ρ and ηn = .
λn λn

With this choice, we easily check that (un , vn , ηn , zn )T belongs to D(A) since by the Robin condition on ΓN
in (3.7) and the definition of zn , we have
 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693 11

zn (·, 0) = ηn = −∂ν un = 0 on ΓN .

Further, we see that (un , vn , ηn , zn )T is solution of

⎧
⎪
⎪ λn un − vn = 0
⎪
⎨ λ v − Δu = 0
n n n
(3.9)
⎪
⎪ λ η − (vn − β1 ηn − β2 zn (·, 1)) = hn ,
⎪ n n
⎩ λ z + τ −1 z (·, ρ) = 0,
n n n,ρ

where

β1 + β2 e−iμn τ
hn = ϕn .
iμn

Equivalently this means that (un , vn , ηn , zn )T is solution of (3.4) with fn = gn = kn = 0 and hn defined
above. Now we notice that

2
Un H = ∇un 2L2 (Ω)n + vn 2L2 (Ω) + ηn 2L2 (ΓN ) + ζzn 2L2 (ΓN ×(0,1)) ≥ vn 2L2 (Ω) = ϕn 2L2 (Ω) = 1,

which proves (3.5) with C = 1. Further straightforward computations yield

2 (|β1 | + |β2 |)2

Fn H = hn 2L2 (ΓN ) ≤ ϕn 2L2 (ΓN ) .
|μn |2

But using the trace estimate of interpolation type (see [14, Theorem [Link]] or [23, Theorem 1.4.4]), there
exists a positive constant Ctr (independent of n) such that

1
|ϕn (x)|2 dσ(x) ≤ Ctr ϕn H 1 (Ω) ϕn L2 (Ω) ≤ Ctr (1 + μ2n ) 2 .
ΓN

Inserting this estimate in the previous one, we find

1
2 Ctr (|β1 | + |β2 |)2 (1 + μ2n ) 2
Fn H ≤ −→ 0 as n −→ ∞. (3.10)
|μn |2

Hence (3.6) is also satisfied. This shows that the resolvent of A is not uniformly bounded on the imaginary
axis. 

4. Rational stabilization result

In this section we will prove the rational stability of problem (1.1) using again a frequency domain
approach, the rational stability of problem (1.1) without delay, and a duality argument. First of all we recall
the following result due to Borichev and Tomilov [9]:

Theorem 4.1. Let A be the generator of a C0 -semigroup of bounded operators on a Hilbert space H. Assume
that

iR ⊂ ρ (A) . (4.1)

Let l be a positive real number. Then we have the polynomial decay

12 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693

 tA 
e U0  ≤ C U0 
D(A) , t > 0, U0 ∈ D(A), (4.2)
t1/l

if and only if

1  

−1 
lim sup  (iλ − A)  < ∞. (4.3)
|λ|→+∞ |λ|
l

Problem (1.1) without delay is the following one

⎧
⎪
⎪ utt (x, t) − Δu(x, t) = 0 in Ω × (0, +∞),
⎪
⎪
⎪
⎪ u(x, t) = 0 on ΓD × (0, +∞),
⎪
⎪
⎨ ∂u
(x, t) + η(x, t) = 0 on ΓN × (0, +∞),
∂ν (4.4)
⎪
⎪ ηt (x, t) − ut (x, t) + β1 η(x, t) = 0 on ΓN × (0, +∞),
⎪
⎪
⎪
⎪
⎪
⎪ u(x, 0) = u0 , ut (x, 0) = u1 in Ω,
⎩
η(x, 0) = η0 on ΓN .

This problem is well-posed in

H0 := HΓ1D (Ω) × L2 (Ω) × L2 (ΓN ) (4.5)

endowed with the norm

 2
 T 2
(u, v, η)  := ∇u2(L2 (Ω))n + v2L2 (Ω) + ηL2 (ΓN ) . (4.6)
H0

The generator of its semigroup is A0 defined by

T T
A0 (u, v, η) := (v, Δu, γ (v − β1 η)) (4.7)

with domain
 
T
D(A0 ) := (u, v, η) ∈ H 2 (Ω) ∩ HΓ1D (Ω) × HΓ1D (Ω) × L2 (ΓN ) : ∂ν u + η = 0 on ΓN . (4.8)

The main result of this section is the following:

Theorem 4.2. Assume that the semigroup generated by A0 in H0 decays polynomially, namely

 tA 
e  U0  ≤ C U0 
D(A ) , ∀ U0 ∈ D(A0 ), ∀ t > 0, (4.9)
t1/l

for some positive real number l. Then the semigroup of system (1.1) inherits the same polynomial decay,
namely

 tA 
e U0  ≤ C U0 
D(A) , ∀ U0 ∈ D(A), ∀ t > 0. (4.10)
t1/l

Proof. As the condition (4.1) is already checked in Lemma 2.3, we only need to check the condition (4.3).
T
So according to the above theorem, we will establish that for any λ ∈ R and F = (f, g, h, k) ∈ H, the
T
solution U = (u, v, η, z) ∈ D(A) of

(iλI − A) U = F (4.11)
 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693 13

satisfies

U H ≤ C(1 + |λ|l )F H ; (4.12)

where C is positive constant (independent of λ and F ). Now we consider the solution U ∗ = (u∗ , v ∗ , η ∗ ) of
T

   
u∗ u
(iλI − A0 ) v∗ = v (4.13)
η∗ η

that due to our assumption satisfies

   
 ∗ ∗ ∗ T  T
(u , v , η )  ≤ C∗ |λ|l (u, v, η)  (4.14)
H0 H0

where C∗ is a positive constant.

Note further that the dissipativeness of A0 directly yields

β1 |η ∗ |2 dΓ ≤ ((iλI − A0 )U ∗ , U ∗ )H0 ≤ (u, v, η) H0 U ∗ H0 . (4.15)
ΓN

On the other hand the system (4.13) can be equivalently written as

⎧
⎪ ∗ ∗
⎨ iλu − v = u on Ω
iλv − Δu∗ = v on Ω
∗ (4.16)
⎪
⎩ iλη ∗ − v ∗ + β η ∗ = η on Γ .
1 N

Integrating by parts and using (4.16) we get

⎛ ⎞ ⎛ ∗⎞ ⎛ ⎞ ⎛ ∗⎞
 u u   iλu − v u 
∗
⎜v⎟ ⎜v ⎟ ⎜ iλv − Δu ⎟ ⎜ v∗ ⎟
(iλI − A) ⎝ ⎠ , ⎝ ∗ ⎠ = ⎝ iλη − v + β η + β z (·, 1) ⎠ , ⎝ ∗ ⎠
η η 1 2 η
z 0 −1 0
H iλz + τ zρ H
 
= ∇ (iλu − v) · ∇u∗ dx + (iλv − Δu) v ∗ dx
Ω Ω

+ (iλη − v + β1 η + β2 z (·, 1)) η ∗ dΓ
ΓN
  
= iλ ∇u.∇u∗ dx − ∇v.∇u∗ dx + iλ vv ∗ dx
Ω Ω Ω
 
− Δu v ∗ dx + (iλη − v + β1 η + β2 z (·, 1)) η ∗ dΓ
Ω ΓN
  
= iλ ∇u.∇u∗ dx + Δu∗ v dx + η ∗ v dΓ
Ω Ω ΓN
  
+iλ vv ∗ dx + ∇u · ∇v ∗ dx + ηv ∗ dΓ
Ω Ω ΓN

+ (iλη − v + β1 η + β2 z (·, 1)) η ∗ dΓ
ΓN
14 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693

 
= ∇u · ∇(−iλu∗ + v ∗ )dx + v(−iλv ∗ + Δu∗ )dx
Ω Ω
 
+ η(−iλη ∗ + v ∗ −β1 η ∗ )dΓ+ (2β1 η + β2 z (·, 1))η ∗ dΓ
ΓN ΓN
  
=− |∇u|2 dx − |v|2 dx − |η|2 dΓ
Ω Ω ΓN

+ (2β1 η + β2 z (·, 1))η ∗ dΓ.
ΓN

Recalling (4.11) and using (4.6), the above relation can be rewritten as
⎞ ⎛
 2
 u∗  
 T ⎜ v∗ ⎟
(u, v, η)  = − F, ⎝ ∗ ⎠ + (2β1 η + β2 z (·, 1))η ∗ dΓ. (4.17)
H0 η
0 H
ΓN

Applying Cauchy-Schwarz’s and Young’s inequalities, we get

 2   4β 2 β2
 T
≤ F H (u∗ , v ∗ , η ∗ )T H + 1 ηL2 (ΓN ) + 2 z (·, 1)L2 (ΓN ) + ε η ∗ L2 (ΓN ) ,
2 2 2
(u, v, η)  (4.18)
H0 0 ε ε

for all ε > 0. Furthermore, using the dissipativity property (2.4) of A, we have
⎛ ⎞ ⎛ ⎞
 u u 
⎜v ⎟ ⎜v ⎟ 2
(iλI − A) ⎝ ⎠ , ⎝ ⎠ = − AU, U H ≥ −Λ1 ηL2 (ΓN ) − Λ2 z (·, 1) 2L2 (ΓN ) ;
η η
z z H

where we recall that Λ1 and Λ2 are negative constants. Then Cauchy-Schwarz’s inequality guarantees that

2 −1
ηL2 (ΓN ) + z (·, 1) 2L2 (ΓN ) ≤ (min{−Λ1 , −Λ2 }) F H U H . (4.19)

This estimate in (4.18) yields

 2  
 T C
≤ F H (u∗ , v ∗ , η ∗ )T H + F H U H + ε η ∗ L2 (ΓN ) ,
2
(u, v, η) 
H0 0 ε

for all ε > 0, and a positive constant C independent of ε. At this stage we make use of (4.14) and (4.15) to
find
 2   C∗ |λ|l ε
 T  T C
(u, v, η)  ≤ C∗ F H |λ|l (u, v, η)  + F H U H + (u, v, η) 2H0 ,
H0 H0 ε β1

for all ε > 0, and a positive constant C independent of ε and λ. Using again Young’s inequality in the first
term of this right-hand side, we arrive at

 2 C∗2  2 C∗ |λ|l ε
 T  T C
(u, v, η)  ≤ F 2H + δ|λ|2l (u, v, η)  + F H U H + (u, v, η) 2H0 ,
H0 4δ H0 ε β1

C∗ |λ|l ε 1
for all ε, δ > 0. Now by fixing ε and δ so that β1 = 4 and δ|λ|2l = 14 , we get
 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693 15

 2
 T
(u, v, η)  ≤ C1 |λ|2l F 2H + |λ|l F H U H , (4.20)
H0

with a positive constant C1 independent of λ.

Now by its definition, we have
 2
2  T
U H = (u, v, η)  + ζz2L2 (ΓN ×(0,1)) , (4.21)
H0

therefore it remains to estimate the L2 -norm of z. But it is easy to check that (compare with (2.8))

ρ
−iλτ ρ −iλτ ρ
z (·, ρ) = ηe + τe k (·, σ) eiλτ σ dσ,
0

hence we have

zL2 (ΓN ×(0,1)) ≤ ηL2 (ΓN ) + τ kL2 (ΓN ×(0,1)) .

By (4.19), we obtain

−1
z2L2 (ΓN ×(0,1)) ≤ 2 (min{−Λ1 , −Λ2 }) F H U H + 2τ 2 F 2H .

Inserting this estimate in (4.21) and using (4.20), we obtain

2
U H ≤ C2 1 + |λ|2l F 2H + 1 + |λ|l F H U H ,

with a positive constant C2 independent of λ. Applying Young’s inequality, we obtain (4.12). The proof of
Theorem 4.2 is thus completed. 

To the best of our knowledge the polynomial decay (4.9) for system (4.4) is not true in a general setting.
Since it holds under some geometrical assumptions (described below), our previous result allows to get the
same decay rate for our system (1.1).

Corollary 4.3. In addition to the assumptions from section 1, assume that there exists x0 ∈ Rn such that

(x − x0 ) · ν(x) > 0, ∀x ∈ ΓN ,
(x − x0 ) · ν(x) ≤ 0, ∀x ∈ ΓD .

Then (4.10) holds with l = 2.

Proof. With our additional assumptions, the results from Theorem 3.2 of [30] (see also [34] and Theorem
2.1 of [35] if n = 1) guarantee that (4.9) holds with l = 2, hence the conclusion directly follows from
Theorem 4.2. 

Corollary 4.4. In addition to the assumptions from section 1, suppose that one of the following assumptions
holds:
1. the boundary of Ω is of class C ∞ and ΓN satisfies the Geometric Control Condition (GCC). Recall
[6] that the GCC can be formulated as follows: ΓN satisfies the Geometric Control Condition if there exists
T > 0 such that every geodesic traveling at speed one issued from Ω at time t = 0 intersects ΓN before time
T.
16 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693

2. the boundary of Ω is of class C ∞ and there exists a vector field h ∈ C 2 (Ω, Rn ) such that

h(x) · ν(x) ≤ 0, ∀x ∈ ΓD ,

and the symetrized Jacobian matrix Jh + Jh is strictly positive definite in Ω.

3. the boundary of Ω is of class C 2 and there exists a vector field h ∈ C 2 (Ω, Rn ) such that

h(x) · ν(x) > 0, ∀x ∈ ΓN ,

h(x) · ν(x) ≤ 0, ∀x ∈ ΓD ,

and the symetrized Jacobian matrix Jh + Jh is strictly positive definite in Ω.

Then (4.9) holds with l = 3.

Proof. First we notice that by [6, §5] in case 1., [20, Theorem 1.2] in case 2., and [19, Theorem 1] in case
3., the wave equation with standard damping on ΓN :
⎧
⎪
⎪ u (x, t) − Δu(x, t) = 0 in Ω × (0, +∞),
⎪ tt
⎪
⎨ u(x, t) = 0 on ΓD × (0, +∞),
∂u (4.22)
⎪
⎪ (x, t) + ut (x, t) = 0 on ΓN × (0, +∞),
⎪
⎪
⎩ ∂ν
u(x, 0) = u0 , ut (x, 0) = u1 in Ω,

is exponentially stable. Hence we apply [1, Proposition 2.2] with B = −β1 and C = −1 to deduce that (4.9)
holds with l = 3. Again the conclusion directly follows from Theorem 4.2. 

Corollary 4.5. In addition to the assumptions from section 1, assume that Ω = (0, 1)2 is the unit square of
R2 and ΓN = {1} × (0, 1). Then (4.10) holds with l = 5.

Proof. Due to [1, Remark 2.6] and the arguments stated in the proof of the previous Corollary, (4.9) holds
with l = 5, hence the conclusion directly follows from Theorem 4.2. 

References

[1] Z. Abbas, S. Nicaise, The multidimensional wave equation with generalized acoustic boundary conditions II: polynomial
stability, SIAM J. Control Optim. 53 (4) (2015) 2582–2607.
[2] C. Abdallah, P. Dorato, J. Benitez-Read, R. Byrne, Delayed positive feedback can stabilize oscillatory systems, in: ACC’
1993, San Francisco, 1993, pp. 3106–3107.
[3] E.M. Ait Benhassi, K. Ammari, S. Boulite, L. Maniar, Feedback stabilization of a class of evolution equations with delay,
J. Evol. Equ. 9 (1) (2009) 103–121.
[4] K. Ammari, S. Nicaise, Stabilization of Elastic Systems by Collocated Feedback, Lecture Notes in Mathematics, vol. 2124,
Springer, Cham, 2015.
[5] W. Arendt, C.J.K. Batty, Tauberian theorems and stability of one-parameter semigroups, Trans. Am. Math. Soc. 306 (2)
(1988) 837–852.
[6] C. Bardos, G. Lebeau, J. Rauch, Sharp sufficient conditions for the observation, control, and stabilization of waves from
the boundary, SIAM J. Control Optim. 30 (5) (1992) 1024–1065.
[7] G. Bayili, A. Ben Aissa, S. Nicaise, Same decay rate of second order evolution equations with or without delay, Syst.
Control Lett. 141 (2020) 104700.
[8] C.D. Benchimol, A note on weak stabilization of contraction semigroups, SIAM J. Control Optim. 16 (1978) 373–379.
[9] A. Borichev, Y. Tomilov, Optimal polynomial decay of functions and operator semigroups, Math. Ann. 347 (2010) 455–478.
[10] R. Datko, Not all feedback stabilized hyperbolic systems are robust with respect to small time delays in their feedbacks,
SIAM J. Control Optim. 26 (3) (1988) 697–713.
[11] R. Datko, Two examples of ill-posedness with respect to time delays revisited, IEEE Trans. Autom. Control 42 (4) (1997)
511–515.
[12] R. Datko, J. Lagnese, M.P. Polis, An example on the effect of time delays in boundary feedback stabilization of wave
equations, SIAM J. Control Optim. 24 (1) (1986) 152–156.
[13] S.R. Foguel, Powers of contraction in Hilbert space, Pac. J. Math. 13 (1) (1963) 551–561.
 G. Bayili et al. / J. Math. Anal. Appl. 495 (2021) 124693 17

[14] P. Grisvard, Elliptic Problems in Nonsmooth Domains, Monographs and Studies in Mathematics, vol. 24, Pitman, Boston–
London–Melbourne, 1985.
[15] K.P. Hadeler, Delay equations in biology, in: Functional Differential Equations and Approximation of Fixed Points,
Proc. Summer School and Conf., Univ. Bonn, Bonn, 1978, in: Lecture Notes in Math., vol. 730, Springer, Berlin, 1979,
pp. 136–156.
[16] J. Hale, S. Verduyn Lunel, Introduction to Functional Differential Equations, Applied Mathematical Sciences, vol. 99,
Springer-Verlag, New York, 1993.
[17] F.L. Huang, Strong asymptotic stability of linear dynamical systems in Banach spaces, J. Berlin 35 (1985) 585–603.
[18] V. Komornic, Exact Controllability and Stabilization. The Multiplier Method, Research in Applied Mathematics, vol. 36,
1994.
[19] J. Lagnese, Decay of solutions of wave equations in a bounded region with boundary dissipation, J. Differ. Equ. 50 (2)
(1983) 163–182.
[20] I. Lasiecka, R. Triggiani, Uniform stabilization of the wave equation with Dirichlet or Neumann feedback control without
geometrical conditions, Appl. Math. Optim. 25 (2) (1992) 189–224.
[21] J.-L. Lions, Contrôlabilité exacte, perturbations et stabilisation de systèmes distribués. Tome 1. Contrôlabilité exacte
(Exact controllability), Recherches en Mathématiques Appliquées (Research in Applied Mathematics), vol. 8, Masson,
Paris, 1988. With appendices by E. Zuazua, C. Bardos, G. Lebeau and J. Rauch.
[22] Z. Liu, B. Rao, Frequency domain approach for the polynomial stability of a system of partially damped wave equations,
J. Math. Anal. Appl. 335 (2) (2007) 860–881.
[23] Z. Liu, S. Zheng, Semigroups Associated with Dissipative Systems, Chapman & Hall/CRC Research Notes in Mathematics,
vol. 398, Chapman & Hall/CRC, Boca Raton, FL, 1999.
[24] H. Logemann, R. Rebarber, G. Weiss, Conditions for robustness and nonrobustness of the stability of feedback systems
with respect to small delays in the feedback loop, SIAM J. Control Optim. 34 (2) (1996) 572–600.
[25] S. Nicaise, C. Pignotti, Stability and instability results of the wave equation with a delay term in the boundary or internal
feedbacks, SIAM J. Control Optim. 45 (5) (2006) 1561–1585 (electronic).
[26] S. Nicaise, J. Valein, Stabilization of the wave equation on 1-D networks with a delay term in the nodal feedbacks, Netw.
Heterog. Media 2 (3) (2007) 425–479 (electronic).
[27] S. Nicaise, J. Valein, Stabilization of second order evolution equations with unbounded feedback with delay, ESAIM Control
Optim. Calc. Var. 16 (2) (2010) 420–456.
[28] C. Pignotti, A note on stabilization of locally damped wave equations with time delay, Syst. Control Lett. 61 (1) (2012)
92–97.
[29] J. Pruss, On the spectrum of c0 -semigroups, Trans. Am. Math. Soc. 284 (1984) 847–857.
[30] B. Rao, L. Toufayli, A. Wehbe, Stability and controllability of a wave equation with dynamical boundary control, Math.
Control Relat. Fields 5 (2) (2015) 305–320.
[31] R. Rebarber, S. Townley, Robustness with respect to delays for exponential stability of distributed parameter systems,
SIAM J. Control Optim. 37 (1) (1999) 230–244.
[32] I.H. Suh, Z. Bien, Use of time delay action in the controller design, IEEE Trans. Autom. Control 25 (1980) 600–603.
[33] B. Sz-Nagy, C. Foias, Analyse Harmonique des opérateurs de l’espace de Hilbert, Masson, Paris, 1967.
[34] L. Toufayli, Stabilisation polynomiale et contrôlabilité exacte des équations des ondes par des contrôles indirects et
dynamiques, PhD thesis, Université de Strasbourg, 2013.
[35] A. Wehbe, Rational energy decay rate for a wave equation with dynamical control, Appl. Math. Lett. 16 (3) (2003) 357–364.
[36] G.Q. Xu, S.P. Yung, L.K. Li, Stabilization of wave systems with input delay in the boundary control, ESAIM Control
Optim. Calc. Var. 12 (4) (2006) 770–785 (electronic).