SECOND ORDER LINEAR EVOLUTION

EQUATIONS WITH GENERAL DISSIPATION

FILIPPO DELL’ORO AND VITTORINO PATA

arXiv:1811.07667v1 [math.AP] 19 Nov 2018

Abstract. The contraction semigroup S(t) = etA generated by the abstract linear
dissipative evolution equation
ü + Au + f (A)u̇ = 0
is analyzed, where A is a strictly positive selfadjoint operator and f is an arbitrary non-
negative continuous function on the spectrum of A. A full description of the spectrum of
the infinitesimal generator A of S(t) is provided. Necessary and sufficient conditions for
the stability, the semiuniform stability and the exponential stability of the semigroup are
found, depending on the behavior of f and the spectral properties of its zero-set. Appli-
cations to wave, beam and plate equations with fractional damping are also discussed.

Contents
1. Introduction 2
2. Two Examples 3
2.1. Abstract wave equations with fractional damping 4
2.2. Beams and plates 4
3. The Spectral Measure of A 5
4. Functional Setting and Notation 6
5. The Linear Operator A 7
6. The Spectrum of A 8
7. The Contraction Semigroup 10
8. The Conservative Case 11
9. Stability 12
10. Exponential Stability 14
11. Semiuniform Stability 16
12. Polynomial Decay Rates 17
13. Proof of Theorem 12.1 18
14. Proof of Theorem 12.2 20
15. Applications 22
15.1. Abstract wave equations with fractional damping 22
15.2. Beams and plates without rotational inertia 24
15.3. Beams and plates with rotational inertia 25
16. Further Developments 26
Appendix: Portraits of the Spectra 27
References 30

2000 Mathematics Subject Classification. 35B35, 47D06, 35P05.

Key words and phrases. Second order equations, contraction semigroup, spectral theory, stability,
semiuniform stability, exponential stability, decay rate.
1
2 F. DELL’ORO AND V. PATA

1. Introduction
Let (H, h·, ·i, k · k) be a separable complex Hilbert space, and let
A : dom(A) ⊂ H → H
be a strictly positive selfadjoint linear operator with inverse A−1 not necessarily compact.
Let also
f : σ(A) → [0, ∞)
be a nonnegative continuous function on the spectrum σ(A) of A. Since A is strictly
positive selfadjoint, σ(A) is a nonempty closed subset of R+ = (0, ∞). Moreover, σ(A) is
compact if and only if A is a bounded operator on H.
For t > 0, we consider the abstract second order evolution equation in the unknown
variable u = u(t)
(1.1) ü + Au + f (A)u̇ = 0,
where u(0) and u̇(0) are understood to be assigned initial data and the dot stands for
derivative with respect to t. Here, f (A) is the selfadjoint operator constructed via the
functional calculus of A, namely,
Z
f (A) = f (s) dEA (s)
σ(A)

being EA the spectral measure of A (see e.g. [40]). More details on the functional calculus
will be given in Section 3.
Equation (1.1) falls within a general class of models introduced in [9] to account for
the dissipative mechanism acting in elastic systems. The operator A is usually called
elastic operator while f (A), replaced in [9] by a more general nonnegative selfadjoint
operator B, is called dissipation operator. In the last decades, these models have been the
object of intensive mathematical investigations, and nowadays the current literature on
the subject is rather vast. When the dissipation operator is comparable with the power
Aϑ for some ϑ ∈ [0, 1] of the elastic operator A (i.e. when the function f (s) controls and
is controlled by sϑ ), then the associated solution semigroup is known to be exponentially
stable and, in addition, analytic for ϑ ∈ [ 12 , 1] and of Gevrey type for ϑ ∈ (0, 12 ); see e.g.
[9, 10, 11, 12, 24, 25, 26] and the more recent contributions [23, 27, 28, 29, 30, 31, 32, 36],
among many others. At the same time, when ϑ ∈ / [0, 1], the exponential stability is
lost. In particular, for ϑ < 0, the solution semigroup is known to be semiuniformly
stable (a notion of stability weaker than the exponential one), with optimal polynomial
1
decay rate of order 2|ϑ| (see [15, 33]). The case ϑ > 1 has been analyzed in the very
recent paper [17], where well-posedness and further regularity properties of the solutions
have been discussed. The above-mentioned results are highly nontrivial, and require the
exploitation of several abstract tools from the theory of linear semigroups, combined with
quite delicate sharp computations.
On the other hand, when the dissipation operator is not comparable with Aϑ , namely,
when the function f is allowed to exhibit an arbitrary (and not necessarily polynomial)
behavior, the picture becomes even more challenging, and additional difficulties arise. In
this situation, the literature about the longterm properties of equation (1.1) is poorer and
 SECOND ORDER EQUATIONS WITH DISSIPATION 3

mainly devoted to the study of conditions under which all the solutions decay exponen-
tially to zero (see e.g. [14, Chapter VI] and the further papers [3, 18, 19, 21, 22]). Roughly
speaking, these contributions tell that exponential stability occurs whenever the following
two assumptions hold (plus possibly some extra conditions varying from paper to paper):
(i) the dissipation operator is bounded below, namely, inf s∈σ(A) f (s) > 0; and
(ii) the dissipation operator is subordinate to A, namely, sups∈σ(A) f (s)/s < ∞.
Note that within (i) the function f does not vanish on σ(A).
In light of the discussion above two natural questions arise:
⋄ What can be said on the stability of (1.1) when the dissipation operator is not
necessarily comparable with Aϑ and not necessarily bounded below, nor subordinate
to A?
⋄ In particular, what happens when the function f vanishes in some points of σ(A)?
The aim of the present work is to address these issues. After proving the existence of
the contraction semigroup S(t) of solutions for a general nonnegative continuous function
f [see Theorem 7.1], we show that S(t) is always stable, i.e. all single trajectories decay
to zero, provided that the zero-set of f

(1.2) Z = s ∈ σ(A) : f (s) = 0
has null spectral measure and is at most countable [see Theorem 9.1]. In fact, this
condition is sharp: when Z has positive spectral measure, solutions with positive constant
energy pop up. These results are attained via an explicit description of the spectrum of the
infinitesimal generator of the semigroup [see Theorems 6.1 and 6.3]. Such a description,
which seems to be new in the literature, besides having an interest by itself allows to
prove the stability of S(t) without assuming the compactness of the inverse operator
A−1 (or similar compactness conditions). On the contrary, compactness conditions are
typically used to apply the classical Sz.-Nagy-Foias theory [7, 41] or Jacobs-Glicksberg-
deLeeuw-type theorems [2, Chapter 5]. In addition, we show that conditions (i)-(ii) above
are actually necessary and sufficient in order for S(t) to be exponentially stable [see
Theorem 10.1]. In particular, we provide an elementary proof of the exponential stability
of S(t) which does not rely in any way on the linear structure of equation, and hence
can be exported to study nonlinear versions of (1.1). We also analyze an intermediate
notion of stability, the so-called semiuniform stability, proving that S(t) is semiuniformly
stable if and only if the set Z is empty and assumption (ii) is satisfied [see Theorem 11.3].
Then, we find the optimal polynomial semiuniform decay rate, again without assuming the
compactness of the inverse operator A−1 [see Theorems 12.1 and 12.2]. We finally apply
the results to some concrete physical models of waves, beams and plates with fractional
damping.
2. Two Examples
In this section, we dwell on two particular (but relevant) instances of equation (1.1). To
this end, given a bounded domain Ω ⊂ Rn with smooth boundary ∂Ω, we introduce1 the
1Itis understood that, in the real case, the results of this paper apply by considering the natural
complexifications of the involved spaces and operators.
4 F. DELL’ORO AND V. PATA

strictly positive selfadjoint Laplace-Dirichlet operator on L2 (Ω)

L = −∆ with dom(L) = H 2 (Ω) ∩ H01 (Ω),
being H 2 (Ω) and H01 (Ω) the standard Sobolev spaces on Ω. For r = 0, 1, 2, . . . , we also
define the Hilbert space (the index r will be omitted whenever zero)
r r r r
V r = dom(L 2 ), hu, viV r = hL 2 u, L 2 viL2 (Ω) , kukV r = kL 2 ukL2 (Ω) .
In particular,
V 2 = H 2 (Ω) ∩ H01 (Ω) ⊂ V 1 = H01 (Ω) ⊂ V = L2 (Ω).
2.1. Abstract wave equations with fractional damping. We consider the abstract
wave equation
(2.1) ü + Au + Aϑ u̇ = 0,
that is, equation (1.1) with a damping of the form
f (s) = sϑ , ϑ ∈ R.
In particular, the values ϑ = 0, 1 yield to the so-called weakly damped wave equation
ü + Au + u̇ = 0,
and the so-called strongly damped wave equation
ü + Au + Au̇ = 0,
respectively. A concrete realization of (2.1) is the boundary-value problem in the unknown
variable u = u(x, t) : Ω × R+ → R
(
∂tt u − ∆u + (−∆)ϑ ∂t u = 0,
(2.2)
u(x, t)|x∈∂Ω = 0,
corresponding to the choice H = V and A = L.
2.2. Beams and plates. For ϑ ∈ R and ω ≥ 0, we consider the evolution equation in
the unknown variable u = u(x, t) : Ω × R+ → R
(2.3) ∂tt u − ω∆∂tt u + ∆2 u + (−∆)ϑ ∂t u = 0,
complemented with the hinged boundary conditions
(2.4) u(x, t)|x∈∂Ω = ∆u(x, t)|x∈∂Ω = 0,
which rules the dynamics of a hinged beam (for n = 1) or plate (for n = 2) subject to
fractional dissipation. According to the formalism introduced above, (2.3)-(2.4) read
(2.5) ∂tt u + ωL∂tt u + L2 u + Lϑ ∂t u = 0.
In order to rewrite the latter equation in the abstract form (1.1), we shall treat separately
the two cases ω = 0 and ω > 0.
• If ω = 0, i.e. the rotational inertia is neglected, equation (2.5) reduces to
(2.6) ∂tt u + L2 u + Lϑ ∂t u = 0.
 SECOND ORDER EQUATIONS WITH DISSIPATION 5

The above is nothing but the particular realization of (1.1) corresponding to the choice
H = V , A = L2 with dom(A) = V 4 , and
ϑ
f (s) = s 2 .
• If ω > 0, equation (2.5) can be rewritten as
(2.7) ∂tt u + (1 + ωL)−1 L2 u + (1 + ωL)−1 Lϑ ∂t u = 0.
Endowing the space V 1 = H01 (Ω) with the equivalent Hilbert norm
1
|u|V 1 = k(1 + ωL) 2 ukL2 (Ω) ,
we now choose H = V 1 . It is then readily seen that the linear operator on H (with the
norm above)
A = (1 + ωL)−1 L2 with dom(A) = V 3
is strictly positive selfadjoint. Moreover, calling
√ ϑ−2
ωs + ω 2s2 + 4s

(2.8) f (s) = s ,
2
by means of direct calculations we find the equality
f (A) = (1 + ωL)−1 Lϑ .
In conclusion, within these choices, equation (2.7) takes the form (1.1).

3. The Spectral Measure of A

Along the paper, the functional calculus of A will be extensively used. Recall that a
spectral measure on a closed set Ω ⊂ R is a map
E : B(Ω) → P (H),
defined on the Borel σ-algebra B(Ω) of Ω with values in the space P (H) of selfadjoint
projections in H, satisfying the following properties:
• E(∅) = 0 and E(Ω) = 1.
• E(ω1 ∩ ω2 ) = E(ω1 )E(ω2 ), for all ω1 , ω2 ∈ B(Ω).
• If ω1 ∩ ω2 = ∅ then E(ω1 ∪ ω2 ) = E(ω1 ) + E(ω2 ).
• For every u, v ∈ H the set function µu,v on B(Ω) defined by
µu,v (ω) = hE(ω)u, vi
is a complex measure.
By the Spectral Theorem (see e.g. [40]), there exists a unique spectral measure EA on the
set Ω = σ(A), called the spectral measure of A, such that
Z
hAu, vi = s dµu,v
σ(A)

for all u ∈ dom(A) and v ∈ H. The integral representation above is usually written for
short as Z
A= s dEA (s).
σ(A)
6 F. DELL’ORO AND V. PATA

In addition, for every continuous complex-valued function φ on σ(A), we can define the
linear operator φ(A) by Z
φ(A) = φ(s) dEA (s)
σ(A)
with dense domain
n Z o
2
dom(φ(A)) = u ∈ H : |φ(s)| dµu,u (s) < ∞ .
σ(A)

It is well-known that φ(A) is a densely defined closed operator. Besides, φ(A) is selfadjoint
if and only if φ is real-valued. Further properties of φ(A) read as follows:
• for every u ∈ dom(φ(A)), we have the equality
Z
2
kφ(A)uk = |φ(s)|2 dµu,u (s).
σ(A)

• φ(A) is bounded if and only if φ is bounded. In which case,

kφ(A)k = sup |φ(s)|.
s∈σ(A)

• φ(A) is bounded below if and only if

inf |φ(s)| > 0.
s∈σ(A)

It is apparent to see that dom(φ(A)) endowed with the graph norm

kuk2dom(φ(A)) = kuk2 + kφ(A)uk2
is a Hilbert space. In particular, when φ(A) is bounded below, there exists c > 1 such
that
kuk2dom(φ(A)) ≤ ckφ(A)uk2.
Hence, the seminorm kφ(A)uk is actually a norm, equivalent to the graph norm.

4. Functional Setting and Notation

For r ∈ R, we define the family of Hilbert spaces (r is always omitted whenever zero)
r r r r
H r = dom(A 2 ), hu, vir = hA 2 u, A 2 vi, kukr = kA 2 uk.
If r > 0, it is understood that H −r denotes the completion of the domain, so that H −r
is the dual space of H r . Accordingly, the symbol h·, ·i will also stand for duality product
between H r and H −r . Setting
s0 = min{s : s ∈ σ(A)} > 0,
for every r1 < r2 we have the Poincaré inequality (which follows at once from the functional
calculus)
(4.1) kuk2r1 ≤ s0r1 −r2 kuk2r2 , ∀u ∈ H r2 .
In particular, the continuous and dense (but not necessarily compact) inclusion
H r2 ⊂ H r1
 SECOND ORDER EQUATIONS WITH DISSIPATION 7

holds true. Along the paper, the Poincaré inequality, as well as the Young and Hölder
inequalities, will be tacitly used several times. We conclude by defining the phase space
of our problem
H = H1 × H
endowed with the standard Hilbert product norm
k(u, v)k2H = kuk21 + kvk2 .

5. The Linear Operator A

In view of rewriting equation (1.1) as a first order ODE on H, we introduce the linear
operator A : dom(A) ⊂ H → H defined as
   
u v
A =
v −Au − f (A)v
with domain  1
 
 v ∈ H
dom(A) = (u, v) ∈ H  .
Au + f (A)v ∈ H
Remark 5.1. As customary, as we did in the definition above of the domain of A,
whenever a vector u ∈ H does not belong to H 2 we still write Au to mean the element of
the dual space H −2 acting as
.
(Au)[w] = hAw, ui, ∀w ∈ H 2 .
Analogously, whenever a vector v ∈ H does not belong to dom(f (A)), we still write f (A)v
to mean the element of the dual space dom(f (A))∗ acting as
.
(f (A)v)[w] = hf (A)w, vi, ∀w ∈ dom(f (A)).
Remark 5.2. It is readily seen that dom(A) is a dense subset of H, as it contains the
dense subspace of H
p
H 2 × dom(q(A)) where q(s) = s + [f (s)]2 .
√
Besides, it is apparent to verify that if (and only if) sups∈σ(A) f (s)/ s < ∞ then the
domain of A factorizes as
dom(A) = H 2 × H 1 .
Remark 5.3. The operator A is closed. This can either be proved directly, or deduced as
a consequence of the fact that A is the infinitesimal generator of a contraction semigroup
(as shown in the following Theorem 7.1). Moreover, it is apparent that A is injective.
If a pair (u, v) belongs to the domain of A, then the variables inherit additional regu-
larity. In particular, we have the following result.
Lemma 5.4. Let (u, v) ∈ dom(A) be arbitrarily given. Then
p
v ∈ dom( f (A)).
8 F. DELL’ORO AND V. PATA

Proof. Exploiting the conditions Au+f (A)v ∈ H and u ∈ H 1 , we find at once the relation
1 √
A− 2 f (A)v ∈ H, meaning that v ∈ dom(g(A)) where g(s) = f (s)/ s. Since v ∈ H 1 , by
the definition of dom(A), an application of the Hölder inequality yields
 12 Z  21
[f (s)]2
Z Z
f (s) dµv,v (s) ≤ s dµv,v (s) dµv,v (s)
σ(A) σ(A) σ(A) s
= kvk1 kg(A)vk < ∞.
p
The estimate above tells that v ∈ dom( f (A)), as claimed. 
Remark 5.5. Actually, from the proof of Lemma 5.4p
we infer that the variable v belongs
1
to the (more regular) space H ∩ dom(g(A)) ⊂ dom( f (A)).
We conclude the section by showing that the (densely defined) operator A is dissipative,
i.e. Re hAz, ziH ≤ 0 for all z ∈ dom(A).
Theorem 5.6. The dissipativity relation
p
(5.1) Re hAz, ziH = −k f (A)vk2 ≤ 0
holds for every z = (u, v) ∈ dom(A).
Proof. The thesis is readily obtained by direct calculations, and recalling Lemma 5.4. 
6. The Spectrum of A
In this section, we describe the spectrum σ(A) of the (closed) operator A. Besides having
some interest by itself, such a description will play a crucial role in the analysis of the
asymptotic properties of (1.1). We first state a necessary and sufficient condition for
0 6∈ σ(A).
Theorem 6.1. The operator A is bijective, i.e. 0 ∈
/ σ(A), if and only if
f (s)
(6.1) sup < ∞.
s∈σ(A) s

Proof. The (injective) operator A is bijective if and only if for any given ẑ = (û, v̂) ∈ H
the equation
(6.2) Az = ẑ
admits a (unique) solution z = (u, v) ∈ dom(A). Componentwise, this translates into
(
v = û,
Au + f (A)v = −v̂.
Substituting the first equation into the second one, we get
u = −A−1 v̂ − f (A)A−1 û.
Since A−1 v̂ ∈ H 1 , we have that u ∈ H 1 for every given û ∈ H 1 if and only if f (A)A−1 is a
bounded operator. It amounts to saying that condition (6.1) holds true. In such a case,
the couple
(−A−1 v̂ − f (A)A−1 û, û) ∈ dom(A)
 SECOND ORDER EQUATIONS WITH DISSIPATION 9

is the unique solution to (6.2). 

Remark 6.2. Observe that (6.1) is automatically satisfied when A is a bounded operator,
for σ(A) is compact and f is continuous. Hence, in this situation, it is always true that
0 belongs to the resolvent set ρ(A) of A.
We now provide a characterization of σ(A) \ {0}. To this end, for every fixed s ∈ σ(A),
we introduce the pair of complex numbers
 p
f (s) [f (s)]2 − 4s √
− ± if f (s) ≥ 2 s ,


ξs± = 2 p 2

− f (s) 4s − [f (s)]2 √
±i if f (s) < 2 s ,

2 2
which are nothing but the solutions of the second order equation
ξ 2 + f (s)ξ + s = 0.
We also consider the (possibly empty) subset of R+
n f (sn ) o
(6.3) Λ = ℓ > 0 : ∃ sn ∈ σ(A) such that sn → ∞ and lim =ℓ .
n→∞ sn

The result reads as follows.

Theorem 6.3. We have the equality
[  [ n 1o
σ(A) \ {0} = ξs± ∪ − .
ℓ ∈Λ
ℓ
s∈σ(A)

Proof. Let ξ ∈ C \ {0} and ẑ = (û, v̂) ∈ H be arbitrarily given. We look for a unique
solution z = (u, v) ∈ dom(A) to the resolvent equation
(6.4) ξz − Az = ẑ.
Written in components, we obtain the system
(
ξu − v = û,
ξv + Au + f (A)v = v̂.
Substituting the first equation into the second one, we find the expression
ξ 2 v + Av + ξf (A)v = Aŵ,
having set
ŵ = ξA−1 v̂ − û ∈ H 1 .
An exploitation of the functional calculus now yields
s
Z
v= 2
dEA (s)ŵ.
σ(A) ξ + f (s)ξ + s

Thus v ∈ H 1 for any given ŵ ∈ H 1 if and only if

 
 s 
 < ∞.
(6.5) sup  2
s∈σ(A) ξ + f (s)ξ + s 
10 F. DELL’ORO AND V. PATA

This occurs if and only if

[  1
ξ∈
/ ξs± and − ∈
/ Λ.
ξ
s∈σ(A)

Indeed, (6.5) fails to hold if and only if there is a sequence sn ∈ σ(A) for which
ξ 2 + f (sn )ξ + sn
(6.6) → 0.
sn
If sn 6→ ∞, then (up to a subsequence) sn converges to an element s ∈ σ(A), as the
spectrum is a closed set. Hence (6.6) becomes simply
ξ 2 + f (s)ξ + s = 0,
meaning that ξ = ξs+ or ξ = ξs− . If sn → ∞, then (6.6) implies that
f (sn ) 1
→− ,
sn ξ
that is, − 1ξ ∈ Λ. Once we find v, we readily get
v + û
u= ∈ H1 and Au + f (A)v = v̂ − ξv ∈ H,
ξ
meaning that (u, v) ∈ dom(A) is the unique solution to (6.4). The theorem is proved. 
The next corollary will be needed in the sequel.
Corollary 6.4. We have the equality
[ √
(σ(A) \ {0}) ∩ iR = ±i s,
s∈Z

where Z is the zero-set of f defined in (1.2).

Proof. Exploiting Theorem 6.3, we learn at once that
[ 
(σ(A) \ {0}) ∩ iR = ξs± ∩ iR.
s∈σ(A)
√
Since ξs± ∈ iR if and only if s ∈ Z, and in such a case ξs± = ±i s, we are finished. 
Remark 6.5. By means of straightforward computations it is immediate to check that,
if s ∈ σ(A) is an eigenvalue of A, then ξs± are eigenvalues of A.

7. The Contraction Semigroup

Introducing the state vector z(t) = (u(t), v(t)), we rewrite equation (1.1) as the ODE in
the phase space H
(7.1) ż(t) = Az(t).
The following holds.
Theorem 7.1. The operator A is the infinitesimal generator of a contraction semigroup
S(t) = etA : H → H.
 SECOND ORDER EQUATIONS WITH DISSIPATION 11

As a consequence, for every given initial datum z0 = (u0 , v0 ) ∈ H there exists a unique
mild solution z in the sense of Pazy [37] to equation (7.1), explicitly given by the formula
z(t) = S(t)z0 .
The associated energy reads
1
E(t) = kS(t)z0 k2H .
2
Moreover, if z0 ∈ dom(A), then z(t) ∈ dom(A) for all t ≥ 0, and the mild solution is
actually a classical one.

Proof of Theorem 7.1. In light of the Lumer-Phillips theorem (see e.g. [37]), the (densely
defined) operator A generates a contraction semigroup on H if and only if it is dissipative
and 1 − A is onto. The first fact is assured by Theorem 5.6. In order to show the second
instance, for an arbitrarily given ẑ = (û, v̂) ∈ H we look for a solution z = (u, v) ∈ dom(A)
to the equation
z − Az = ẑ.
Written in components, the latter reads
(
u − v = û,
v + Au + f (A)v = v̂.

Plugging the first equality into the second one, we find

v + Av + f (A)v = Aŵ,
where ŵ = A−1 v̂ − û ∈ H 1 . Then, owing to the functional calculus, we get
s
Z
v= dEA (s)ŵ.
σ(A) 1 + s + f (s)

Being the function f nonnegative, we have

s
sup ≤ 1,
s∈σ(A) 1 + s + f (s)

meaning that v ∈ H 1 . In turn, u = v + û ∈ H 1 as well. Finally, by comparison,

Au + f (A)v = v̂ − v ∈ H.
This completes the argument. 

8. The Conservative Case

For the sake of completeness, we preliminarily dwell on the conservative case, which is
very well known in the literature. Indeed, when the function f vanishes on the spectrum
of A, the same as saying that Z = σ(A), equation (1.1) reduces to
(8.1) ü + Au = 0.
12 F. DELL’ORO AND V. PATA

Despite its simple form, it serves as a model for several physical phenomena. With
reference to the notation of Section 2, the simplest example is the classical wave equation
with Dirichlet boundary conditions in the unknown u = u(x, t) : Ω × R+ → R
(
∂tt u − ∆u = 0,
u(x, t)|x∈∂Ω = 0,
corresponding to the choice H = V and A = L in (8.1). Another model matching the
abstract form (8.1) is the linear Klein-Gordon equation arising in Relativity Theory
∂tt u − ∆u + m2 u = 0,
in the unknown u = u(x, t) : R3 × R+ → R, where m > 0 (see e.g. [20]). It is readily seen
that, choosing H = L2 (R3 ) and
A = −∆ + m2 with dom(A) = H 2 (R3 ),
the Klein-Gordon equation takes the form (8.1). In this situation, the strictly positive
selfadjoint operator A does not have compact inverse.
The next result follows immediately from Theorems 6.1 and 6.3, observing that the set
Λ defined in (6.3) is always empty whenever f ≡ 0.
Theorem 8.1. Assume that f (s) = 0 for all s ∈ σ(A). Then, the operator A is always
bijective, and its spectrum fulfils the equality
[ √
σ(A) = ±i s.
s∈σ(A)

In particular, σ(A) is entirely contained in the imaginary axis iR.

9. Stability
In this section, we analyze the stability of S(t). Recall that S(t) is said to be stable if, for
every fixed z0 ∈ H,
lim kS(t)z0 kH = 0.
t→∞
It turns out that this property depends dramatically on the structure of the set Z defined
in (1.2).
Theorem 9.1. The following hold:
(i) If EA (Z) 6= 0 (i.e. it is not the null projection), then there exist solutions with
constant positive energy. In particular, the semigroup S(t) is not stable.
(ii) If EA (Z) = 0 and the set Z is at most countable, then the semigroup S(t) is stable.
In particular, this is always the case if Z = ∅.
In order to show the theorem, we make use of the famous Arendt-Batty-Lyubich-Vũ
stability criterion [1, 35]. Recall that, denoting by σp (A) the point spectrum of the
infinitesimal generator A, the criterion reads as follows.
Theorem 9.2 (Arendt-Batty-Lyubich-Vũ). Assume that σp (A) ∩ iR = ∅ and σ(A) ∩ iR
is at most countable. Then S(t) stable.
We now proceed with the proof of Theorem 9.1.
 SECOND ORDER EQUATIONS WITH DISSIPATION 13

Proof of Theorem 9.1. Assume first that EA (Z) is a nonnull projection. In this situation,
we can select a unit vector w ∈ EA (Z)H. In particular, the probability measure µw,w is
supported on Z. Accordingly,
Z
[f (s)]2 dµw,w (s) = 0,
σ(A)

meaning that w ∈ dom(f (A)) and f (A)w = 0. Then, by direct calculations, the solution
to (7.1) with initial datum
1
z0 = (0, A− 2 w) ∈ dom(A)
is given by
√ √ 1
z(t) = (sin(t A)A−1 w, cos(t A)A− 2 w).
Such a solution z has constant energy, for
1 1 √ √ 1
Z
1
| sin(t s)|2 + | cos(t s)|2 dµw,w (s) = kA− 2 wk2 = E(0).

E(t) =
2 σ(A) s 2
The proof of item (i) is finished. In order to show (ii), we first prove that
EA (Z) = 0 ⇒ σp (A) ∩ iR = ∅.
To this end, assume by contradiction that iλ ∈ σp (A) for some λ ∈ R. Since A is injective
(see Remark 5.3), we have λ 6= 0. Then, there is a nonnull vector z = (u, v) ∈ dom(A)
satisfying
iλz − Az = 0.
Componentwise, the equality above reads
(
iλu − v = 0,
(9.1)
iλv + Au + f (A)v = 0.
Invoking (5.1), we obtain
p Z
2
0 = Re hiλz − Az, ziH = k f (A)vk = f (s) dµv,v (s),
σ(A)

implying that
v ∈ EA (Z)H = 0.
Making use of the first equation of (9.1) and the fact that λ 6= 0, we find u = 0, reaching
the desired contradiction z = 0. At this point, Corollary 6.4 together with the assumption
that Z is either finite or countable ensure that σ(A) ∩ iR is either finite or countable. The
abstract Theorem 9.2 then allows to conclude. 
Theorem 9.1 tells in particular that S(t) is stable whenever Z = ∅, but in general
the converse is not true. Nevertheless, when the operator A−1 is compact, the sufficient
condition Z = ∅ turns out to be necessary as well.
Corollary 9.3. Assume that A−1 is a compact operator. If S(t) is stable then Z = ∅.
14 F. DELL’ORO AND V. PATA

Proof. When A−1 is compact, it is well known that the spectrum σ(A) is made of a
sequence of eigenvalues tending to infinity. In particular,
EA ({s}) 6= 0, ∀s ∈ σ(A).
This clearly yields EA (Z) 6= 0 whenever Z 6= ∅. Due to Theorem 9.1, the semigroup S(t)
is not stable. 
Remark 9.4. In the case where A−1 is not compact, the question whether or not S(t) is
stable if EA (Z) = 0 and the set Z is uncountable remains open.

10. Exponential Stability

A much stronger notion of stability is the exponential (or uniform) one. Recall that S(t)
is said to be exponentially stable if there exist κ > 0 and M ≥ 1 such that
kS(t)z0 kH ≤ Me−κt kz0 kH ,
for all z0 ∈ H. Exponential stability is equivalent to the fact that the operator norm
kS(t)kL(H) goes to zero as t → ∞. In turn, this is the same as saying that ω∗ < 0, where
ω∗ is the growth bound of S(t), defined as
ω∗ = inf ω ∈ R : kS(t)kL(H) ≤ Meωt


for some M = M(ω). From the Hille-Yosida Theorem and the boundedness of S(t), it
follows at once the relation
(10.1) σ∗ ≤ ω∗ ≤ 0,
where 
σ∗ = sup Re ξ : ξ ∈ σ(A)
is the spectral bound of A (see e.g. [37] for more details).
Theorem 10.1. The semigroup S(t) is exponentially stable if and only if
f (s)
(10.2) inf f (s) > 0 and sup < ∞.
s∈σ(A) s∈σ(A) s

We shall prove separately the necessity and the sufficiency parts.

Proof of Theorem 10.1 (Necessity). The strategy consists in showing that, if (10.2) is not
satisfied, then σ∗ ≥ 0. Due to (10.1), the latter condition implies that ω∗ = 0, i.e. S(t) is
not exponentially stable. Indeed, when
f (s)
sup = ∞,
s∈σ(A) s

we learn from Theorem 6.1 that 0 ∈ σ(A), namely, σ∗ ≥ 0. On the other hand, when
inf f (s) = 0,
s∈σ(A)

there exists a sequence sn ∈ σ(A) such that

f (sn ) → 0
 SECOND ORDER EQUATIONS WITH DISSIPATION 15

as n → ∞. Since the spectrum of A is (positive and) away from zero, it is clear that for
√
all n large we have f (sn ) < 2 sn . Exploiting Theorem 6.3, the complex numbers
p
± f (sn ) 4sn − [f (sn )]2
ξn = − ±i
2 2
±
belong to σ(A). Since Re ξn → 0, we conclude again that σ∗ ≥ 0. 
Proof of Theorem 10.1 (Sufficiency). Let z0 ∈ dom(A) be an arbitrarily fixed initial da-
tum, and let
z(t) = (u(t), u̇(t)) = S(t)z0 ∈ dom(A)
be the corresponding solution. Since dom(A) is a dense subset of H, in order to reach the
conclusion it is enough showing that the associated energy E(t) fulfills
E(t) ≤ M 2 E(0)e−2κt
for some κ > 0 and M ≥ 1 independent of z0 . Along the proof, c > 0 will denote
a generic positive constant depending only on the structural quantities of the problem
and independent of z0 . Multiplying equality (7.1) by z in H, taking the real part and
exploiting Theorem 5.6 we find the identity
d p
E = Re hAz, ziH = −k f (A)u̇k2 .
dt
Invoking the first condition of (10.2), it is apparent to see that
p Z
2
k f (A)u̇k = f (s) dµu̇,u̇ (s) ≥ cku̇k2 .
σ(A)

Hence, we get the differential inequality

d
(10.3) E + cku̇k2 ≤ 0.
dt
Next, we introduce the auxiliary functional
1 p
Φ(t) = k f (A)u(t)k2 + Re hu̇(t), u(t)i.
2
Due the Poincaré inequality (4.1) and the second condition of (10.2),
1
Z Z
(10.4) |Φ| ≤ f (s) dµu,u(s) + cE ≤ c s dµu,u (s) + cE ≤ cE.
2 σ(A) σ(A)

Moreover, by means of direct calculations,

d
(10.5) Φ + kuk21 = ku̇k2 .
dt
At this point, for all ε > 0, we define the energy-like functional
Λε (t) = E(t) + εΦ(t).
An exploitation of (10.4) yields
1
E(t) ≤ Λε (t) ≤ 2E(t)
2
16 F. DELL’ORO AND V. PATA

for every ε > 0 small enough, meaning that Λε is equivalent to E. Thus, collecting (10.3)
and (10.5), and fixing the parameter ε > 0 sufficiently small, we arrive at the differential
inequality
d
Λε + 2κΛε ≤ 0
dt
for some κ > 0. Applying the Gronwall lemma, and using once more the equivalence of
the functionals Λε and E, the conclusion follows. 

11. Semiuniform Stability

Finally, we consider an intermediate notion of stability, known as semiuniform stability.
By definition, S(t) is semiuniformly stable if there exists a nonnegative function ψ(t)
vanishing at infinity such that
(11.1) kS(t)z0 kH ≤ ψ(t)kAz0 kH , ∀z0 ∈ dom(A).
Since S(t) is a bounded semigroup (actually, a contraction), it easily follows by density
that if S(t) is semiuniformly stable then it is stable as well. Instead, if S(t) is exponentially
stable then, as a consequence of (10.1), its infinitesimal generator A is invertible with
bounded inverse, as 0 belongs to the resolvent set ρ(A) of A. This clearly implies that
S(t) is semiuniformly stable with exponential rate, i.e.
ψ(t) = M ′ e−κt with κ, M ′ > 0.
Precisely, with reference to the previous Section 10, M ′ = MkA−1 kL(H) (recall that k·kL(H)
denotes the operator norm).
The following criterion is due to Batty [4, 6].
Theorem 11.1 (Batty). The (bounded) semigroup S(t) is semiuniformly stable if and
only if σ(A) ∩ iR = ∅.
Remark 11.2. In particular, when S(t) is semiuniformly stable then 0 ∈
/ σ(A). Accord-
ingly, for every z0 ∈ H we can write
kS(t)A−1 z0 kH ≤ ψ(t)kAA−1 z0 kH = kz0 kH .
This clearly implies that any function ψ satisfying (11.1) is subject to the constraint
ψ(t) ≥ kS(t)A−1 kL(H) .
At the same time, for all z0 ∈ dom(A)
kS(t)z0 kH ≤ kS(t)A−1 kL(H) kAz0 kH ,
meaning that among the ψ complying with (11.1), the choice ψ(t) = kS(t)A−1 kL(H) is the
best possible.
We now state the necessary and sufficient condition for the semiuniform stability of S(t).
Theorem 11.3. The semigroup S(t) is semiuniformly stable if and only if
f (s)
(11.2) Z=∅ and sup < ∞.
s∈σ(A) s
 SECOND ORDER EQUATIONS WITH DISSIPATION 17

Proof. Collecting Theorem 6.1 and Corollary 6.4 we learn at once that σ(A) ∩ iR = ∅ if
and only if (11.2) holds true. Invoking Theorem 11.1, we reach the thesis. 
Remark 11.4. When the operator A is bounded, condition (11.2) is equivalent to (10.2),
being the spectrum σ(A) a compact set. Hence, in this situation, S(t) is semiuniformly
stable if and only if it is exponentially stable.

12. Polynomial Decay Rates

When S(t) is semiuniformly stable, it is of great interest to describe the decay at infinity
of ψ(t) in (11.1). In light of Remark 11.2, we concentrate on the particular function
ψ(t) = kS(t)A−1 kL(H) .
It is an easy exercise to show that ψ(t) is (Lipschitz) continuous.
Theorem 12.1. Let S(t) be semiuniformly stable. If there exists α > 0 such that
(12.1) inf sα f (s) > 0,
s∈σ(A)

1
then ψ(t) decays at least polynomially as t− 2α , namely,
1
lim sup t 2α ψ(t) < ∞.
t→∞

Theorem 12.2. Let S(t) be semiuniformly stable. If A is unbounded and there exists
β > 0 such that
(12.2) sup sβ f (s) < ∞,
s∈σ(A)

1
then ψ(t) decays at most polynomially as t− 2β , namely,
1
lim sup t 2β ψ(t) > 0.
t→∞

Remark 12.3. Actually, when A is a bounded operator, the first result above is of little
use, since in that case (cf. Remark 11.4) S(t) is semiuniformly stable if and only if it is
exponentially stable.
Remark 12.4. Note that when A is unbounded, then (12.2) is incompatible with the
first condition in (10.2). Besides, it is apparent that α ≥ β when both (12.1) and (12.2)
are satisfied (again, under the assumption A unbounded).
Theorems 12.1 and 12.2 produce an immediate corollary.
Corollary 12.5. Let S(t) be semiuniformly stable. If A is unbounded and (12.1)-(12.2)
1
simultaneously hold with α = β, then ψ(t) decays polynomially as t− 2α , and such a decay
rate is optimal.
The proofs of Theorems 12.1 and 12.2 will be given in the next two sections.
18 F. DELL’ORO AND V. PATA

13. Proof of Theorem 12.1

The strategy consists in finding a polynomial estimate from above on the growth rate of
the resolvent operator of A on the imaginary axis iR, as the latter can be linked with the
decay rate of kS(t)A−1 kL(H) making use of the following abstract result due to Borichev
and Tomilov [8].
Theorem 13.1 (Borichev-Tomilov). Assume that σ(A) ∩ iR = ∅, and let ν > 0 be fixed.
Then, we have
k(iλ − A)−1 kL(H) = O(|λ|ν ) as |λ| → ∞
if and only if
kS(t)A−1 kL(H) = O(t−1/ν ) as t → ∞.
The needed estimate is contained in the next lemma.
Lemma 13.2. Within the assumptions of Theorem 12.1, we have (σ(A) ∩ iR = ∅ and)
k(iλ − A)−1 kL(H) = O(|λ|2α ) as |λ| → ∞.
Proof. In what follows, c ≥ 0 will denote a generic constant depending only on the
structural quantities of the problem. Since S(t) is semiuniformly stable, we know from
Theorem 11.1 that σ(A) ∩ iR = ∅. Besides, due to Theorem 11.3, condition (11.2) holds
true.
For every fixed λ ∈ R and ẑ = (û, v̂) ∈ H, the resolvent equation
(13.1) iλz − Az = ẑ
has a unique solution
z = (iλ − A)−1 ẑ.
In order to prove the lemma, it is enough showing the estimate
(13.2) kzkH ≤ c|λ|2α kẑkH
for every |λ| ≥ 1. To this end, writing (13.1) componentwise, we get the system
(13.3) iλu − v = û,
(13.4) iλv + Au + f (A)v = v̂.
In light of the dissipativity property (5.1), a multiplication in H of (13.1) with z entails
the control
p
(13.5) k f (A)vk2 = Re h(iλ − A)z, ziH = Re hẑ, ziH ≤ kzkH kẑkH .
Next, multiplying in H equation (13.4) by u and exploiting (13.3), we obtain
p p
kuk21 = hv̂, ui − iλhv, ui − h f (A)v, f (A)ui
p p
= hv̂, ui + hv, ûi + kvk2 − h f (A)v, f (A)ui
p p
≤ kvk2 + k f (A)ukk f (A)vk + ckzkH kẑkH .
Invoking the second condition in (11.2),
Z  12
p
k f (A)uk ≤ c s dµu,u (s) = ckuk1 ≤ ckzkH .
σ(A)
 SECOND ORDER EQUATIONS WITH DISSIPATION 19

Hence, appealing to (13.5), we find

3/2 1/2
(13.6) kuk21 ≤ kvk2 + ckzkH kẑkH + ckzkH kẑkH .
Moreover, due to (13.3), it is also true that
(13.7) kvk1 ≤ |λ|kzkH + kẑkH .
At this point, we shall treat separately two cases.
Case α ≤ 1. An application of (12.1) and (13.5), together with the Hölder inequality,
yields
Z  21 Z  12
2 1
kvk ≤ dµv,v (s) f (s) dµv,v (s)
σ(A) f (s) σ(A)
Z  12
α
p
≤c s dµv,v (s) k f (A)vk
σ(A)
Z  α2
1/2 1/2
≤c s dµv,v (s) kvk1−α kzkH kẑkH
σ(A)

(3−2α)/2 1/2
≤ ckvkα1 kzkH kẑkH .
Estimating the term kvkα1 with the aid of (13.7), we reach the control
3/2 1/2 (3−2α)/2 (1+2α)/2
kvk2 ≤ c|λ|α kzkH kẑkH + ckzkH kẑkH .
Collecting now (13.6) and the inequality above, for every |λ| ≥ 1 we have
3/2 1/2
kzk2H ≤ 2kvk2 + ckzkH kẑkH + ckzkH kẑkH
3/2 1/2 (3−2α)/2 (1+2α)/2
≤ c|λ|αkzkH kẑkH + ckzkH kẑkH + ckzkH kẑkH ,
Finally, we estimate the first two terms on the right-hand side making use of the Young
inequality with conjugate exponents ( 34 , 4) and ( 3−2α
4 4
, 1+2α ), respectively. This leads to
1 1
kzk2H ≤ kzk2H + c|λ|4α kẑk2H + ckzkH kẑkH ≤ kzk2H + c|λ|4α kẑk2H
4 2
for all |λ| ≥ 1, which readily implies (13.2). 
Case α > 1. Exploiting (12.1), (13.5) and the Hölder inequality, we infer that
Z  12 Z  21
2 1
kvk ≤ dµv,v (s) s dµv,v (s)
σ(A) s σ(A)
Z  2α1
1
≤ α
dµv,v (s) kvk(α−1)/α kvk1
σ(A) s

(α−1)/α
p
≤ ck f (A)vk1/α kzkH kvk1
(2α−1)/2α 1/2α
≤ ckvk1 kzkH kẑkH .
20 F. DELL’ORO AND V. PATA

Recalling (13.7), we get the control

(4α−1)/2α 1/2α (2α−1)/2α (2α+1)/2α
kvk2 ≤ c|λ|kzkH kẑkH + ckzkH kẑkH .
At this point, from (13.6) and the inequality above, we obtain
3/2 1/2
kzk2H ≤ 2kvk2 + ckzkH kẑkH + ckzkH kẑkH
(4α−1)/2α 1/2α (2α−1)/2α (2α+1)/2α 3/2 1/2
≤ c|λ|kzkH kẑkH + ckzkH kẑkH + ckzkH kẑkH + ckzkH kẑkH .
Estimating the first three terms on the right-hand side making use of the Young inequality
4α 4α 4α
with conjugate exponents ( 4α−1 , 4α), ( 2α−1 , 2α+1 ) and ( 43 , 4), respectively, we end up with
1 1
kzk2H ≤ kzk2H + c|λ|4α kẑk2H + ckzkH kẑkH ≤ kzk2H + c|λ|4α kẑk2H
4 2
for all |λ| ≥ 1. The latter implies at once (13.2). 
Conclusion of the proof of Theorem 12.1. The claim merely follows by collecting the above
Lemma 13.2 and Theorem 13.1. 

14. Proof of Theorem 12.2

The key tool is the next result of Batty and Duyckaerts [6] (see also [2, Theorem 4.4.14]).
Theorem 14.1 (Batty-Duyckaerts). Assume that S(t) is semiuniformly stable. Then for
any (strictly) decreasing continuous function h : [0, ∞) → R+ vanishing at infinity and
satisfying kS(t)A−1 kL(H) ≤ h(t), there exists C > 0 such that
 1 
−1 −1
k(iλ − A) kL(H) ≤ Ch
2|λ|
for all |λ| sufficiently large.
In order to apply this abstract theorem, we need to find a (polynomial) control of the
operator norm k(iλ − A)−1 kL(H) . This is provided by the next lemma.
Lemma 14.2. Within the assumptions of Theorem 12.2, we have (σ(A) ∩ iR = ∅ and)
lim sup λ−2β k(iλ − A)−1 kL(H) > 0.
λ→∞

Proof. Take sn ∈ σ(A) with sn → ∞ (this is possible being the operator A unbounded).
Since σ(A) ∩ iR = ∅ due to Theorem 11.1, we introduce the sequence
√
ηn = k(i sn − A)−1 kL(H) .
Exploiting the functional calculus of A, and arguing exactly as in the proof of [13,
Lemma 9.2], we can find unit vectors wn ∈ H such that
.
rn1 = Awn − sn wn
.
rn2 = f (A)wn − f (sn )wn
satisfy the bounds
1
(14.1) krnı k ≤ , ı = 1, 2.
n(1 + ηn )
 SECOND ORDER EQUATIONS WITH DISSIPATION 21

Next, we define the further unit vectors

ẑn = (0, wn ) ∈ H
and we consider the resolvent equation
√
i sn zn − Azn = ẑn
which admits a unique solution
√
zn = (un , vn ) = (i sn − A)−1 ẑn .
In particular, since kẑn kH = 1, we have
√
(14.2) kzn kH = k(i sn − A)−1 ẑn kH ≤ ηn .
Writing the resolvent equation componentwise, we obtain the system
√
i sn un − vn = 0,
√
i sn vn + Aun + f (A)vn = wn .
Substituting the first equation into the second one, we find
√ √
(14.3) −sn vn + Avn + i sn f (A)vn = i sn wn .
It is readily seen that vn can be written in the form
vn = ζn wn + qn
for some ζn ∈ C and some vectors qn ⊥ wn . Due to (14.2), the controls
(14.4) |ζn | ≤ kzn kH ≤ ηn and kqn k ≤ kzn kH ≤ ηn
hold. At this point, we multiply (14.3) by wn in H. Exploiting the fact that wn and
qn are orthogonal, and after straightforward calculations, we get the equality (recall that
f (s) 6= 0 for all s ∈ σ(A) due to Theorem 11.3)
1
ζn = + ρn
f (sn )
having set
√ √
−ζn hrn1 , wn i − hqn , rn1 i − i sn ζn hrn2 , wn i − i sn hqn , rn2 i
ρn = √ .
i sn f (sn )
Note that, in light of (14.1) and (14.4), for n large enough
√
2ηn (1 + sn ) 4
|ρn | ≤ √ ≤ .
n(1 + ηn )f (sn ) sn nf (sn )
Accordingly, for n large enough
1 1
|ζn | ≥ − |ρn | ≥ .
f (sn ) 2f (sn )
Finally, making use of (14.2), the inequality above and (12.2), we arrive at
1
ηn ≥ kzn kH ≥ |ζn | ≥ ≥ csβn
2f (sn )
22 F. DELL’ORO AND V. PATA

for some structural constant c > 0. Recalling the definition of ηn , from the latter estimates
we conclude that
lim sup sn−2β k(isn − A)−1 kL(H) > 0,
n→∞
and the thesis follows. 
Conclusion of the proof of Theorem 12.2. Assume by contradiction that
1
ψ(t) = o(t− 2β ) as t → ∞,
and let h : [0, ∞) → R+ be a (strictly) decreasing continuous function satisfying
1
ψ(t) ≤ h(t) and h(t) = o(t− 2β ).
One may take for instance
h(t) = sup ψ(s) + e−t .
s≥t

Invoking now Theorem 14.1, there exists C > 0 such that

 1 
k(iλ − A)−1 kL(H) ≤ Ch−1
2|λ|
for every |λ| large enough. Since
 1 
−1
h = o(|λ|2β ) as |λ| → ∞,
2|λ|
we conclude that
k(iλ − A)−1 kL(H) = o(|λ|2β ) as |λ| → ∞,
contradicting Lemma 14.2. 

15. Applications
We now apply the results obtained so far to the examples presented in Section 2, to which
we address the reader for the notation. In what follows, we denote by
0 < λ1 < λ2 ≤ . . . ≤ λn → ∞
the sequence of eigenvalues of the Laplace-Dirichlet operator L.

15.1. Abstract wave equations with fractional damping. In view of Theorem 7.1,
equation (2.1) generates a contraction semigroup
Sϑ (t) = etAϑ : H → H.
Here, Aϑ is the particular instance of A corresponding to the choice f (s) = sϑ , namely
   
u v
Aϑ =
v −Au − Aϑ v
with domain 
v ∈ H1
 

dom(Aϑ ) = (u, v) ∈ H  .
Au + Aϑ v ∈ H
 SECOND ORDER EQUATIONS WITH DISSIPATION 23
√
Note that sups∈σ(A) f (s)/ s < ∞ if and only if ϑ ≤ 21 or A is a bounded operator. In
this situation (and only in this situation), the domain factorizes as
dom(Aϑ ) = H 2 × H 1 .
Exploiting Theorems 6.1 and 6.3, we obtain a precise description of the spectrum of Aϑ .
Theorem 15.1. The following hold:
(i) The operator Aϑ is bijective if and only if ϑ ≤ 1 or A is a bounded operator.
(ii) We have
S  ±

s∈σ(A) ξs ∪ {−1} if ϑ = 1 and A unbounded,
σ(Aϑ ) \ {0} = S  ±
s∈σ(A) ξs otherwise,


where  √
sϑ s2ϑ − 4s 1
− ± if sϑ− 2 ≥ 2,


ξs± = 2 √ 2
ϑ 2ϑ
− ± i 4s − s
s 1
if sϑ− 2 < 2.


2 2
Indeed, with reference to (6.3), the set Λ is nonempty if and only if ϑ = 1 and the
operator A is unbounded. Besides, if the latter conditions hold, then Λ = {1}.
Theorem 15.1, together with Remark 6.5, produce an immediate corollary, which pro-
vides a characterization of the spectrum of the wave equation (2.2), where A is the Laplace-
Dirichlet operator.
Corollary 15.2. Let A = L. Then, the spectrum of the corresponding operator Aϑ is
countable and is given by
S∞  ±
ξ ∪ {0} if ϑ > 1,
Sn=1  λn


∞ ±
σ(Aϑ ) = n=1 ξλn ∪ {−1} if ϑ = 1,


S∞  ±
n=1 ξλn if ϑ < 1.
Besides, the numbers ξλ±n are all eigenvalues of Aϑ .
Coming back to more general equation (2.1), the decay properties of the related semi-
group Sϑ (t) can be immediately inferred from the results of Sections 9-12, observing that
Z = ∅, due to the choice of the function f . It is readily seen that, if the operator A
is bounded, then Sϑ (t) is exponentially stable for every ϑ ∈ R (since condition (10.2) is
always satisfied). The more interesting case when A is unbounded is summarized in the
next theorem.
Theorem 15.3. Let A be unbounded. Then, the following hold:
(i) Sϑ (t) is stable for every ϑ ∈ R.
(ii) Sϑ (t) is exponentially stable if and only if ϑ ∈ [0, 1].
(iii) Sϑ (t) is semiuniformly stable if and only if ϑ ≤ 1.
1
− 2|ϑ|
(iv) If ϑ < 0, then ψ(t) = kSϑ (t)A−1 ϑ kL(H) decays polynomially as t , and such a
decay rate is optimal.
24 F. DELL’ORO AND V. PATA

15.2. Beams and plates without rotational inertia. Here we consider (2.6), that is,
equation (2.5) with ω = 0. Although in this situation the picture is formally equivalent
to the previous example with ϑ2 in place of ϑ, for completeness we provide a detailed
description of the results. Invoking Theorem 7.1, equation (2.6) generates a contraction
semigroup
0
Sϑ0 (t) = etAϑ : H0 → H0
on the space
H0 = V 2 × V.
Here, A0ϑ is the particular instance of the operator A obtained by choosing H = V , A = L2
ϑ
and f (s) = s 2 , that is    
0 u v
Aϑ =
v −L2 u − Lϑ v
with domain 
v ∈ V2
 
0 0

dom(Aϑ ) = (u, v) ∈ H  2 .
L u + Lϑ v ∈ V
√
Being the operator A = L2 unbounded, we have sups∈σ(A) f (s)/ s < ∞ if and only if
ϑ ≤ 1. In this situation (and only in this situation), the domain takes the form
dom(A0ϑ ) = V 4 × V 2 .
Moreover, since the spectrum of the operator L2 is entirely made by eigenvalues and reads
[∞
2
σ(L ) = {λ2n },
n=1
an exploitation of Theorems 6.1 and 6.3, together with Remark 6.5, yields a complete
description of σ(A0ϑ ).
Theorem 15.4. The spectrum of A0ϑ is countable and is given by
S
∞  ±
ξ ∪ {0} if ϑ > 2,
 n=1 λ2n


S∞ ±
σ(A0ϑ ) =

n=1 ξλ2n ∪ {−1} if ϑ = 2,


 ∞ ξ±
S 
if ϑ < 2.
n=1 λ2n

Explicitly,  p
λϑn λ2ϑ
n − 4λn
2
− ± if λϑ−1 ≥ 2,


n
ξλ±2n = 2 p 2
ϑ 2 2ϑ
− n ± i 4λn − λn
λ

if λϑ−1 < 2.

n
2 2
Besides, the numbers ξλ±2n are all eigenvalues of A0ϑ .
Here, we made use of the fact that the set Λ defined in (6.3) is nonempty if and only
if ϑ = 2, and in this case Λ = {1}.
Finally, the stability properties of the semigroup Sϑ0 (t) can be inferred from the results
of Sections 9-12, noting that Z = ∅.
Theorem 15.5. The following hold:
 SECOND ORDER EQUATIONS WITH DISSIPATION 25

(i) Sϑ0 (t) is stable for every ϑ ∈ R.

(ii) Sϑ0 (t) is exponentially stable if and only if ϑ ∈ [0, 2].
(iii) Sϑ0 (t) is semiuniformly stable if and only if ϑ ≤ 2.
1
(iv) If ϑ < 0, then ψ(t) = kSϑ0 (t)(A0ϑ )−1 kL(H0 ) decays polynomially as t− |ϑ| , and such a
decay rate is optimal.
15.3. Beams and plates with rotational inertia. The last example concerns equa-
tion (2.5) with ω > 0, which can be rewritten in the form (2.7). Analogously to the
previous examples, appealing to Theorem 7.1 we infer that equation (2.7) generates a
contraction semigroup
ω
Sϑω (t) = etAϑ : Hω → Hω
on the space
Hω = V 2 × V 1 .
This time, Aωϑ is the particular instance of A corresponding to H = V 1 , A = (1+ωL)−1 L2 ,
with f given by (2.8). Explicitly,
   
ω u v
Aϑ =
v −(1 + ωL)−1 L2 u − (1 + ωL)−1 Lϑ v
with domain
  2

v ∈ V
dom(Aωϑ ) ω

= (u, v) ∈ H  .
(1 + ωL)−1 L2 u + (1 + ωL)−1 Lϑ v ∈ V 1
√
Since A is an unbounded operator, a closer look to (2.8) tells that sups∈σ(A) f (s)/ s < ∞
if and only if ϑ ≤ 23 . In this situation (and only in this situation), the domain factorizes
as
dom(Aωϑ ) = V 3 × V 2 .
Similarly to the case ω = 0, the spectrum of A is entirely made by eigenvalues and is
given by
∞ 
λ2n
[ 
σ(A) = .
n=1
1 + ωλn
Accordingly, making use of (2.8), together with Theorems 6.1 and 6.3 and Remark 6.5,
we readily get a complete description of the spectrum of Aωϑ .
Theorem 15.6. The spectrum of Aωϑ is countable and is given by
 S∞  ±
ξ ∪ {0} if ϑ > 2,
Sn=1  νn


∞
σ(Aωϑ ) = ±
n=1 ξνn ∪ {−1} if ϑ = 2,


 S∞  ±
n=1 ξνn if ϑ < 2,
having set
λ2n
νn = .
1 + ωλn
26 F. DELL’ORO AND V. PATA

Explicitly,  p
f (νn ) [f (νn )]2 − 4νn √
− ± if f (νn ) ≥ 2 νn ,


ξν±n = 2 p 2
2
− f (νn ) ± i 4νn − [f (νn )] √

if f (νn ) < 2 νn ,

2 2
where f is given by (2.8). Besides, the numbers ξν±n are all eigenvalues of Aωϑ .
Exactly as in the previous example, we have exploited the fact that the set Λ defined
in (6.3) is nonempty if and only if ϑ = 2, and in this case Λ = {1}.
We conclude by summarizing the stability properties of Sϑω (t) which, again, can be
readily inferred from the results of Sections 9-12, observing that Z = ∅ and
f (s) ∼ ω ϑ−2 sϑ−1 as s → ∞.
Theorem 15.7. The following hold:
(i) Sϑω (t) is stable for every ϑ ∈ R.
(ii) Sϑω (t) is exponentially stable if and only if ϑ ∈ [1, 2].
(iii) Sϑω (t) is semiuniformly stable if and only if ϑ ≤ 2.
1
(iv) If ϑ < 1, then ψ(t) = kSϑω (t)(Aωϑ )−1 kL(Hω ) decays polynomially as t− 2(1−ϑ) , and
such a decay rate is optimal.
Remark 15.8. Note that the presence of the rotational inertia ω > 0 changes the expo-
nential stability interval of ϑ from [0, 2] to [1, 2].

16. Further Developments

We finally discuss some possible developments, which might be deepened in future works.
I. As pointed out in Remark 9.4, it would be interesting to investigate the stability of S(t)
where the operator A−1 is not compact and Z is an uncountable set with null spectral
measure. As shown in the proof of Theorem 9.1, if EA (Z) = 0 then it is always true
that no eigenvalues of A lie on the imaginary axis. Nevertheless, Corollary 6.4 tells that
the set σ(A) ∩ iR is uncountable. Hence, in this situation, the Arendt-Batty-Lyubich-Vũ
stability criterion cannot be applied.
II. Another problem concerns the behavior at infinity of the resolvent operator (iλ − A)−1
on the imaginary axis when the exponential stability of S(t) occurs. Due to the Gearhart-
Prüss theorem [16, 38], such a resolvent operator is (defined and) bounded on the whole
imaginary axis. If for instance
 
−1 1
k(iλ − A) kL(H) = O as |λ| → ∞
log |λ|
then S(t) is eventually differentiable (see e.g. [37, Theorem 4.9]), while if
 
−1 1
k(iλ − A) kL(H) = O as |λ| → ∞
|λ|
then S(t) is analytic (see e.g. [34, Theorem 1.3.3]). Among other reasons, such regularity
properties have a certain relevance since eventually differentiable semigroups or analytic
 SECOND ORDER EQUATIONS WITH DISSIPATION 27

semigroups are know to fulfill the spectrum determined growth (SDG) condition (see e.g.
[14, Corollary 3.12]). In the notation of Section 10, this means that the growth bound
ω∗ of S(t) equals the spectral bound σ∗ of its infinitesimal generator A. Note that, in
the proof of the necessity part of Theorem 10.1, we have already shown that the SDG
condition is satisfied with ω∗ = σ∗ = 0 whenever condition (10.2) fails.
III. An intriguing and possibly challenging task would be to investigate semiuniform (or
semiuniform-like) decay rates of S(t) which are not necessarily of polynomial type, making
use of recent abstract results obtained in [5] (see also [39]) dealing with fine decay scales
of strongly continuous semigroups.

Appendix: Portraits of the Spectra

We illustrate some particular instances of the spectra of the operators Aϑ , A0ϑ and Aωϑ
discussed in Section 15.
Portraits of σ(Aϑ ). Choosing H = V = L2 (0, π) and A = L, the eigenvalues λn are
equal to
λn = n2 , n = 1, 2, 3, . . .
Accordingly, the eigenvalues ξλ±n of Aϑ take the form
 √
n2ϑ n4ϑ − 4n2
− ± if n2ϑ−1 ≥ 2,


±
ξλn = 2 √ 2
2ϑ
n 4n2 − n4ϑ
if n2ϑ−1 < 2.

−

±i
2 2
Making use of Corollary 15.2 and the software Mathematica R , we have the following

pictures of σ(Aϑ ), corresponding to the cases ϑ = −1, 0, 1.

Figure 1. The case ϑ = −1 [“subdamped” wave equation].

28 F. DELL’ORO AND V. PATA

Figure 2. The case ϑ = 0 [weakly damped wave equation].

Figure 3. The case ϑ = 1 [strongly damped wave equation]. Behavior around

zero (left) and global behavior (right).

Portraits of σ(A0ϑ ). Choosing H = V = L2 (0, π), the eigenvalues λ2n of the operator
A = L2 are equal to
λ2n = n4 , n = 1, 2, 3, . . .
Therefore, the eigenvalues ξλ±2n of A0ϑ are given by
 √
n2ϑ n4ϑ − 4n4
− ± if n2(ϑ−1) ≥ 2,


±
ξλ2n = 2 √ 2
2ϑ
 n 4n4 − n4ϑ
− ±i if n2(ϑ−1) < 2.

2 2
Making use of Theorem 15.4 and the software Mathematica R , we get the following pictures

of σ(A0ϑ ), corresponding to the choices ϑ = −1, 0, 1.

 SECOND ORDER EQUATIONS WITH DISSIPATION 29

Figure 4. The case ϑ = −1 [“subdamped” beam equation].

Figure 5. The case ϑ = 0 [beam equation with frictional damping].

Figure 6. The case ϑ = 1 [beam equation with Kelvin-Voigt damping]. Be-

havior around zero (left) and global behavior (right).
1 1
Portraits of σ(Aωϑ ). Choosing H = V = H0 (0, π), the eigenvalues νn of the operator
−1 2
A = (1 + ωL) L are equal to
n4
νn = .
1 + ωn2
30 F. DELL’ORO AND V. PATA

Hence, the eigenvalues ξν±n of Aωϑ read

 1h
q
2 i
n4 n4 4n4 n4 2n2


− f 1+ωn
 2 ∓ f 1+ωn2
− 1+ωn2
if f 1+ωn 2 ≥ √1+ωn 2,
± 2
ξ νn =
 − 1 f n4
∓ i
h q
2 i
4n4 n4 n4 2n2


− f if f 1+ωn < √1+ωn 2,

1+ωn 2 1+ωn 2 1+ωn 2 2
2
where f is given by (2.8). Making use of Theorem 15.6 and the software Mathematica R,
ω
we obtain the following pictures of σ(Aϑ ), corresponding to the choices ϑ = 0, 1.

Figure 7. The case ϑ = 0 and ω = 1 [beam equation with rotational inertia

and frictional damping].

Figure 8. The case ϑ = 1 with ω = 1 (left) and ω = 1/200 (right) [beam equa-
tion with rotational inertia and Kelvin-Voigt damping]. Note that the spectrum
becomes close to two straight lines as ω → 0. Compare with Figure 6.

References
[1] W. Arendt and C.J.K. Batty, Tauberian theorems and stability of one-parameter semigroups, Trans.
Amer. Math. Soc. 306 (1988), 837–852.
[2] W. Arendt, C.J.K. Batty, M. Hieber and F. Neubrander, Vector-valued Laplace transforms and
Cauchy problems, Birkhäuser, Basel, 2011.
 SECOND ORDER EQUATIONS WITH DISSIPATION 31

[3] A. Bátkai and K.-J. Engel, Exponential decay of 2 × 2 operator matrix semigroups, J. Comput. Anal.
Appl. 6 (2004), 153–163.
[4] C.J.K. Batty, Asymptotic behaviour of semigroups of operators, in “Functional analysis and operator
theory”, vol. 30, Banach Center Publ. Polish Acad. Sci., Warsaw, 1994.
[5] C.J.K. Batty, R. Chill and Y. Tomilov, Fine scales of decay of operator semigroups, J. Eur. Math.
Soc. (JEMS) 18 (2016), 853–929.
[6] C.J.K. Batty and T. Duyckaerts, Non-uniform stability for bounded semi-groups on Banach spaces,
J. Evol. Equ. 8 (2008), 765–780.
[7] C.D. Benchimol, A note on weak stabilizability of contraction semigroups, SIAM J. Control Opti-
mization 16 (1978), 373–379.
[8] A. Borichev and Y. Tomilov, Optimal polynomial decay of functions and operator semigroups, Math.
Ann. 347 (2010), 455–478.
[9] G. Chen and D.L. Russell, A mathematical model for linear elastic systems with structural damping,
Quart. Appl. Math. 39 (1981/82), 433–454.
[10] S. Chen and R. Triggiani, Proof of extensions of two conjectures on structural damping for elastic
systems, Pacific J. Math. 136 (1989), 15–55.
[11] S. Chen and R. Triggiani, Gevrey class semigroups arising from elastic systems with gentle dissipa-
tion: the case 0 < α < 12 , Proc. Amer. Math. Soc. 110 (1990), 401–415.
[12] S. Chen and R. Triggiani, Characterization of domains of fractional powers of certain operators
arising in elastic systems, and applications, J. Differential Equations 88 (1990), 279–293.
[13] V. Danese, F. Dell’Oro and V. Pata, Stability analysis of abstract systems of Timoshenko type, J.
Evol. Equ. 16 (2016), 587–615.
[14] K.-J. Engel and R. Nagel, One-parameter semigroups for linear evolution equations, Springer-Verlag,
New York, 2000.
[15] L.H. Fatori, M.Z. Garay and J.E. Muñoz Rivera, Differentiability, analyticity and optimal rates of
decay for damped wave equations, Electron. J. Differential Equations 48 (2012), 13 pp.
[16] L. Gearhart, Spectral theory for contraction semigroups on Hilbert space, Trans. Amer. Math. Soc.
236 (1978), 385–394.
[17] M. Ghisi, M. Gobbino and A. Haraux, Local and global smoothing effects for some linear hyperbolic
equations with a strong dissipation, Trans. Amer. Math. Soc. 368 (2016), 2039–2079.
[18] G.R. Goldstein, J.A. Goldstein and G. Perla Menzala, On the overdamping phenomenon: a general
result and applications, Quart. Appl. Math. 71 (2013), 183–199.
[19] G.R. Goldstein, J.A. Goldstein and G. Reyes, Overdamping and energy decay for abstract wave
equations with strong damping, Asymptot. Anal. 88 (2014), 217–232.
[20] W. Greiner, Relativistic quantum mechanics. Wave equations. Third edition, Springer-Verlag, Berlin,
2000.
[21] R.O. Griniv and A.A. Shkalikov, Exponential stability of semigroups associated with some operator
models in mechanics. (Russian), translation in Math. Notes 73 (2003), 618–624.
[22] R.O. Griniv and A.A. Shkalikov, Exponential energy decay of solutions of equations corresponding to
some operator models in mechanics. (Russian), translation in Funct. Anal. Appl. 38 (2004), 163–172.
[23] A. Haraux and M. Ôtani, Analyticity and regularity for a class of second order evolution equations,
Evol. Equ. Control Theory 2 (2013), 101–117.
[24] F. Huang, On the holomorphic property of the semigroup associated with linear elastic systems with
structural damping, Acta Math. Sci. (English Ed.) 5 (1985), 271–277.
[25] F. Huang, On the mathematical model for linear elastic systems with analytic damping, SIAM J.
Control Optim. 26 (1988), 714–724.
[26] F. Huang and K. Liu, Holomorphic property and exponential stability of the semigroup associated
with linear elastic systems with damping, Ann. Differential Equations 4 (1988), 411–424.
[27] B. Jacob and C. Trunk, Location of the spectrum of operator matrices which are associated to second
order equations, Oper. Matrices 1 (2007), 45–60.
[28] B. Jacob and C. Trunk, Spectrum and analyticity of semigroups arising in elasticity theory and
hydromechanics, Semigroup Forum 79 (2009), 79–100.
32 F. DELL’ORO AND V. PATA

[29] I. Lasiecka and R. Triggiani, Control Theory for Partial Differential Equations: Continuous and
Approximation Theories, Cambridge University Press, Cambridge, 2000.
[30] I. Lasiecka and R. Triggiani, Domains of fractional powers of matrix-valued operators: a general
approach, Oper. Theory Adv. Appl., 250, Birkhäuser/Springer, Cham, 2015.
[31] K. Liu and Z. Liu, Analyticity and differentiability of semigroups associated with elastic systems with
damping and gyroscopic forces, J. Differential Equations 141 (1997), 340–355.
[32] Z. Liu and J. Yong, Qualitative properties of certain C0 semigroups arising in elastic systems with
various dampings, Adv. Differential Equations 3 (1998), 643–686.
[33] Z. Liu and Q. Zhang, A note on the polynomial stability of a weakly damped elastic abstract system,
Z. Angew. Math. Phys. 66 (2015), 1799–1804.
[34] Z. Liu and S. Zheng, Semigroups associated with dissipative systems, Chapman & Hall/CRC, Boca
Raton, 1999.
[35] Y.I. Lyubich and Q.P. Vũ, Asymptotic stability of linear differential equations in Banach spaces,
Studia Math. 88 (1988), 37–42.
[36] D. Mugnolo, A variational approach to strongly damped wave equations, Functional analysis and
evolution equations, 503–514, Birkhäuser, Basel, 2008.
[37] A. Pazy, Semigroups of linear operators and applications to partial differential equations, Springer-
Verlag, New York, 1983.
[38] J. Prüss, On the spectrum of C0 -semigroups, Trans. Amer. Math. Soc. 284 (1984), 847-857.
[39] J. Rozendaal, D. Seifert and R. Stahn, Optimal rates of decay for operator semigroups on Hilbert
spaces, arXiv: 1709.08895.
[40] W. Rudin, Functional analysis, McGraw-Hill, New York-Düsseldorf-Johannesburg, 1973.
[41] B. Sz-Nagy and C. Foias, Harmonic analysis of operators on Hilbert space, North-Holland Publishing
Company, Amsterdam-London, 1970.

Politecnico di Milano - Dipartimento di Matematica

Via Bonardi 9, 20133 Milano, Italy
E-mail address: [email protected]
E-mail address: [email protected]