Second Order Linear Evolution Equations With General Dissipation
Second Order Linear Evolution Equations With General Dissipation
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
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
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).
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)
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 .
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
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
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
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.
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)
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.
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 learn from Theorem 6.1 that 0 ∈ σ(A), namely, σ∗ ≥ 0. On the other hand, when
inf f (s) = 0,
s∈σ(A)
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)
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.
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.
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
(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
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
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
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
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].
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.
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
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.