See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/342546970

The Cauchy problem for the Moore-Gibson-Thompson equation in the

dissipative case

Preprint · June 2020

CITATIONS READS

0 81

2 authors:

Wenhui Chen Ryo Ikehata

Shanghai Jiao Tong University Hiroshima University
22 PUBLICATIONS   73 CITATIONS    114 PUBLICATIONS   1,958 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

To make a theorem which has come to mind. View project

All content following this page was uploaded by Ryo Ikehata on 30 June 2020.

The user has requested enhancement of the downloaded file.

 The Cauchy problem for the Moore-Gibson-Thompson equation in the
dissipative case
Wenhui Chen∗1 and Ryo Ikehata2
arXiv:2006.00758v2 [math.AP] 26 Jun 2020

1
School of Mathematical Sciences, Shanghai Jiao Tong University, 200240 Shanghai, China
2
Department of Mathematics, Division of Educational Sciences, Graduate School of Humanities
and Social Sciences, Hiroshima University, 739-8524 Higashi-Hiroshima, Japan

Abstract
In this paper, we study the Cauchy problem for the linear and semilinear Moore-Gibson-Thompson (MGT)
equation in the dissipative case. Concerning the linear MGT model, by utilizing WKB analysis associated with
Fourier analysis, we derive some L2 estimates of solutions, which improve those in the previous research [48].
Furthermore, asymptotic profiles of the solution and an approximate relation in a framework of the weighted L1
space are derived. Next, with the aid of the classical energy method and Hardy’s inequality, we get singular limit
results for an energy and the solution itself. Concerning the semilinear MGT model, basing on the obtained sharp
L2 estimates and constructing time-weighted Sobolev spaces, we investigate global (in time) existence of Sobolev
solutions with different regularities. Finally, under a sign assumption on initial data, nonexistence of global (in
time) weak solutions is proved by applying a test function method.

Keywords: Moore-Gibson-Thompson equation, third-order hyperbolic equation, Fourier analysis, asymptotic

profiles, singular limit, global existence.

AMS Classification (2010) Primary: 35L30, 35L76; 35B40; Secondary: 74D05, 34K26, 35A01, 35B44.

1 Introduction
In the last two decades, researches of the Moore-Gibson-Thompson (MGT) equation, which is lin-
earized by a model for the wave propagation in viscous thermally relaxing fluids and is widely applied
in medical as well as industrial uses of high-intensity ultrasound e.g. lithotripsy, thermotherapy or
ultrasound cleaning, have caught a lot of attention. The MGT model is considered through the
third-order (in time) strictly hyperbolic partial differential equation as follows:

τ uttt + utt − c2 ∆u − b∆ut = 0, (1)

where the scalar unknown u = u(t, x) ∈ R denotes an acoustic velocity. The MGT model (1) exhibits
a variety of dynamical behaviors for solutions, which heavily depend on the physical parameters
in the equation. To be specific, concerning the model (1), c stands for the speed of sound and τ
denotes the thermal relaxation in the view of the physical context of acoustic waves. Moreover, the
parameter b = βc2 concerns the diffusivity of the sound carrying τ ∈ (0, β].
Actually, one may distinguish behaviors of solutions to the model (1) according to the dissipative
case when τ ∈ (0, β) and the conservative case when τ = β. Precisely, in the case of bounded
∗ Corresponding author: Wenhui Chen ([email protected])
 2

domains for the linear MGT model, there exists a transition from the case τ ∈ (0, β) with an energy
being exponentially stable to the limit case τ = β with an energy being conserved. Concerning some
studies for the linear or nonlinear MGT equations, we refer interested readers to the related works
[42, 54, 34, 20, 36, 35, 41, 33, 40, 49, 7, 16, 39, 17, 38, 48, 5, 1, 15, 6, 51, 47, 50, 11, 12, 4, 43, 44]
and references therein.
It is well-known that to study qualitative properties of solutions to the linear problem is not
only significant for us to understand some underlying physical phenomena, it is also the crucial
point for proving existence results of solutions to its corresponding nonlinear models. Let us come
to the Cauchy problem for the linear MGT equation which has been firstly studied by the recent
paper [48]. By reducing the third-order (in time) equation to the first-order (in time) coupled
system, the authors of [48] employed energy methods in the Fourier space combined with suitable
Lyapunov functionals to derive some energy estimates, and eigenvalues expansions to investigate
some estimates for the solution itself. However, the obtained estimates for solutions in [48] seem
not sharp, especially, in some low-dimensional cases. In this paper, we will improve their results
and derive some optimal estimates. What’s more, in the view of the limit case τ = 0, the linear
MGT equation formally turns out to be the viscoelastic damped wave equation. For this reason,
one may conjecture that there exist some relations between them. We will answer this conjecture
from two points of view which are singular limits and approximate relation in the sense of diffusion
phenomena, respectively.
Our first aim in this paper is to investigate qualitative properties of solutions to the following
linear MGT equation in the dissipative case:

τ u + utt − ∆u − β∆ut = 0, x ∈ Rn , t > 0,
ttt
(2)
u(0, x) = 0, u (0, x) = 0, u (0, x) = u (x), x ∈ Rn ,
t tt 2

where τ ∈ (0, β) and n > 1. Without loss of generality, we set the speed of the sound by c2 = 1 in
the last equation. To be specific, in Section 2 by preparing representation of solutions in the Fourier
space and using asymptotic expansions of eigenvalues as well as WKB analysis, we deduce some
L2 estimates of solutions to the Cauchy problem (2) for initial data taken from L2 space with or
without additional Lm regularity carrying m ∈ [1, 2). By a different treatment of some singularities,
our results of L2 estimates improve those in [48], especially, the estimates of solutions in one and
two spatial dimensions. Moreover, the regular assumption on initial data is relaxed. Later in
Section 3 we obtain asymptotic profiles of the solution to the Cauchy problem (2) in a framework
of weighted L1 data, where we provide sharp estimates for lower bounds and upper bounds of the
solution itself in the L2 norm. Namely, in the consideration of L2 data with additional weighted
L1 regularity, the derived estimates are optimal for any n > 1. In Subsection 3.2, in the frame of
L2 space, we describe an approximate relation (strongly related to diffusion phenomena) between
the linear MGT equation and the linear viscoelastic damped wave equation, where gained decay
rates are obtained for one- and two-dimensional cases. Next, in Section 4 we consider the singular
limit problem, in which we find the solution of the linear MGT equation converges to the solution
of the linear viscoelastic damped wave equation as the thermal relaxation tending to 0, i.e. τ → 0+ .
Particularly, under different assumptions for initial data, we observe different rates of such tendency
with respect to τ .
Our next purpose is to consider the Cauchy problem for the semilinear MGT equation in the
 3

dissipative case with the nonlinearity of power type, namely,


τ u + utt − ∆u − β∆ut = |u|p , x ∈ Rn , t > 0,
ttt
(3)
u(0, x) = 0, ut (0, x) = 0, utt (0, x) = u2 (x), x ∈ Rn ,

where τ ∈ (0, β), n > 1 and p > 1. Recently, the blow-up results of the Cauchy problem for the
semilinear MGT equation in the conservative case, i.e. the limit case τ = β, with the nonlinearity
of power type |u|p in [11], or of derivative type |ut |p in [12] have been obtained by applying iteration
methods with suitable slicing procedure for unbounded multipliers. These works interpret the
semilinear MGT equation in the conservative case as the semilinear wave equation with power
source nonlinearities. Nevertheless, this statement does not hold anymore for the MGT equation in
the dissipative case due to the damping effect that we derived in the corresponding linear problem.
For this reason, it seems interesting to study existence as well as nonexistence of global (in time)
solutions to the semilinear MGT models in the dissipative case.
Let us now turn to the Cauchy problem (3). To the best of authors’ knowledge, not only global
(in time) existence but also blow-up results for (3) are still open. We will answer these questions
in the present paper. By making use of the improved L2 − L2 estimates with an additional L1
regularity and employing Banach’s fixed point theory, we prove global (in time) existence of small
data Sobolev solutions to the Cauchy problem (3) in Section 5. Particularly, we analyze the interplay
effect between dimension n, regularity s and power p on the existence of global (in time) Sobolev
solution such that

u ∈ C([0, ∞), H s(Rn )),

with some positive parameters s. Soon afterward in Section 6, we apply a test function method to
prove nonexistence of global (in time) weak solutions to the semilinear Cauchy problem (3) if the
power p fulfills some conditions. We should underline that the result in the one-dimensional case is
optimal due to the blow-up result holding for any 1 < p < ∞.
Lastly, throughout Sections 2, 3, 5 and 6, we will consider the MGT equations with vanishing
first and second data. Indeed, non-vanishing third data will exert some dominant influences on the
total estimates and existence results of solutions. We expect that one may derive the corresponding
results with non-vanishing data by following the same approaches as we did later without any
additional difficulties. Clearly, additional regularities for initial data would be necessary.
Notation: We give some notations to be used in this paper. Later, c and C denote some positive
constants, which may be changed from line to line. We denote that f . g if there exists a positive
constant C such that f 6 Cg and, analogously, for f & g. We denote ⌈r⌉ := min{y ∈ N : 0 < r 6 y}
as the positive ceiling function. BR stands for the ball around the origin with radius R in Rn .
Moreover, Ḣqs (Rn ) with s > 0 and 1 6 q < ∞, denote the Riesz potential spaces based on the
Lebesgue spaces Lq (Rn ). Finally, |D|s with s > 0 stands for the pseudo-differential operator with
the symbol |ξ|s.
 4

2 Estimates of solutions to the linear MGT equation in the dissipative

case
2.1 Pointwise estimates in the Fourier space
At first, we apply the partial Fourier transform with respect to spatial variables to the Cauchy
problem (2). Then, it yields the following initial value problem for the third-order |ξ|-dependent
ordinary differential equation:

τ û + ûtt + β|ξ|2ût + |ξ|2û = 0, ξ ∈ Rn , t > 0,
ttt
(4)
û(0, ξ) = 0, ût (0, ξ) = 0, ûtt (0, ξ) = û2 (ξ), ξ ∈ Rn ,

whose solution can be given by

 
X exp(λj (|ξ|)t)
c ξ)û (ξ) := 
û(t, ξ) = K(t, Q  û2 (ξ), (5)
2
j=1,2,3 k=1,2,3, k6=j (λj (|ξ|) − λk (|ξ|))

where λj = λj (|ξ|) with j = 1, 2, 3, are three pairwise distinct roots to the cubic equation

τ λ3 + λ2 + β|ξ|2λ + |ξ|2 = 0. (6)

Here, the case for multiple roots can be regarded as a zero measure set with respect to |ξ|, and
precisely, the discriminant of (6) is zero, that is
   
△Cub = |ξ|2 −4β 3 τ |ξ|4 + 18βτ + β 2 − 27τ 2 |ξ|2 − 4 = 0,

if and only if
q
2 2
18βτ + β 2 − 27τ 2 ± (18βτ + β 2 − 27τ 2 )2 − 64β 3 τ
|ξ| = 0 or |ξ| = . (7)
8β 3τ
Under these preparations, we just need to discuss the case when the cubic equation (6) does not
have any roots of double multiply. Estimates of solutions in a zero measure set (7) do not give any
influence on total estimates. Indeed, the pointwise estimates of solutions in the zero measure set
were shown in [48].
Remark 2.1. The principal symbol of the equation in (2) is given by τ η 3 − βη|ζ|2q . Thus, the
2 2
characteristic equation η(τ η − β|ζ| ) = 0 has pairwise distinct real roots η = 0, η = β/τ |ζ| and
q
η = − β/τ |ζ|. In other words, the linear MGT equation in the dissipative case is strictly hyperbolic.
Therefore, it is clear that the Cauchy problem (2) is well-posedness, e.g. there exists a unique Sobolev
solution u ∈ C([0, ∞), H s(Rn )) for s ∈ [0, 2] if u2 ∈ H s−2 (Rn ) ⊂ L2 (Rn ). Furthermore, the theory
in the strictly hyperbolic equation (see, for example, Section 3.4 in [53]) shows that finite propagation
speed property holds.
Before deriving some L2 estimates of solutions in the next subsection, we will prepare pointwise
estimates of solutions in the Fourier space by investigating asymptotic behaviors of the kernel
c ξ). It is well-known that the explicit formula of the cubic equation (6) can be uniquely
function K(t,
 5

given by Cardano’s formula. Nevertheless, this would be a complex way to analyze behaviors of the
kernel. To overcome the difficulty, we will employ asymptotic expansions of eigenvalues in small and
large frequency zones, and demonstrate an exponential stability of solutions in bounded frequency
zone. We define these zones in Fourier space by
Zint (ε) := {ξ ∈ Rn : |ξ| < ε ≪ 1} ,
Zmid(ε, N) := {ξ ∈ Rn : ε 6 |ξ| 6 N} ,
Zext (N) := {ξ ∈ Rn : |ξ| > N ≫ 1} .
Let us set the cut-off functions χint (ξ), χmid(ξ), χext (ξ) ∈ C∞ (Rn ) owning their supports in Zint (ε),
Zmid (ε/2, 2N) and Zext (N), respectively. Furthermore, they fulfill χmid (ξ) = 1 − χint (ξ) − χext (ξ) for
all ξ ∈ Rn .
Proposition 2.1. Let τ ∈ (0, β). Then, the solution û = û(t, ξ) to the initial value problem (4)
fulfills the following estimates:
! !
| sin(|ξ|t)| − β−τ |ξ|2t 1
χint (ξ)|û(t, ξ)| . χint (ξ) | cos(|ξ|t)| + e 2 + e− τ t |û2(ξ)|, (8)
|ξ|
χmid(ξ)|û(t, ξ)| . χmid (ξ)e−ct|û2 (ξ)|, (9)
1 β−τ 1
χext (ξ)|û(t, ξ)| . χext (ξ) 2 e− min{ 2βτ , β }t |û2 (ξ)|, (10)
|ξ|
for some constants c > 0.
Proof. Let us begin with estimating the solution û(t, ξ) for small frequencies. Motivated by the
recent research [48], we deduce that the eigenvalues λj (|ξ|) with j = 1, 2, 3, have the asymptotic
expansions for |ξ| → 0 such that
(0) (1) (2)
λj (|ξ|) = λj + λj |ξ| + λj |ξ|2 + · · · , (11)
(k)
where the coefficients λj ∈ C for all k ∈ N0 . What we need now is the dominant part of pairwise
distinct eigenvalues. So, by plugging (11) into (6) and processing lengthy but straightforward
computations, until different characteristic roots appear, the eigenvalues behave asymptotically for
|ξ| → 0 as
β−τ 2
λ1,2 (|ξ|) = ±i|ξ| − |ξ| + O(|ξ|3),
2
1
λ3 (|ξ|) = − + (β − τ )|ξ|2 + O(|ξ|3).
τ
Let us denote a |ξ|-dependent function by
1 3
T0 (|ξ|) := − (β − τ )|ξ|2 = O(1) as |ξ| → 0. (12)
τ 2
According to the representation of the kernel given in (5), the Fourier transform of the kernel
localized in small frequency zone can be estimated by
β−τ 2 ! 1 2
c ξ)| e− 2 |ξ| t | sin(|ξ|t)| e− τ t+(β−τ )|ξ| t
χint (ξ)|K(t, . χint (ξ) 2 | cos(|ξ|t)| + T0 (|ξ|) + χ int (ξ)
T0 (|ξ|) + |ξ|2 |ξ| T02 (|ξ|) + |ξ|2
! !
| sin(|ξ|t)| − β−τ |ξ|2 t 1 2
. χint (ξ) | cos(|ξ|t)| + e 2 + e− τ t+(β−τ )|ξ| t ,
|ξ|
 6

which immediately implies the desired estimate (8).

Next, let us turn to the case for large frequencies. The eigenvalues to the cubic equation (6) for
|ξ| → ∞ have the asymptotic expansions such that
(0) (1) (2) (3)
λj (|ξ|) = λ̄j |ξ|2 + λ̄j |ξ| + λ̄j + λ̄j |ξ|−1 + · · · , (13)
(k)
where the coefficients λ̄j ∈ C for all k ∈ N0 . By substituting (13) into (6), it yields that the
eigenvalues have asymptotic behaviors for |ξ| → ∞ as follows:
1
λ1 (|ξ|) = − + O(|ξ|−1),
β
√
β β−τ
λ2,3 (|ξ|) = ±i √ |ξ| − + O(|ξ|−1).
τ 2βτ
Hence, the next chain inequalities hold:
   √   √   
   
− β1 t sin √ β |ξ|t  cos √ β |ξ|t 
 e   τ   τ   − β−τ t 
c ξ)| . χ (ξ)  
χext (ξ)|K(t, 2 +   +  2  e 2βτ 
ext   2  
β−3τ
2βτ
+ βτ |ξ|2 |ξ| β−3τ
2βτ
+ βτ |ξ|2
β−3τ
2βτ
+ βτ |ξ|2
 
1 β−τ 1
. χext (ξ) 2 e− 2βτ t + e− β t .
|ξ|
The previous estimate combined with the formula of solution in (5) proves our desired assertion
(10).
Finally, let us prove an exponential decay estimate of solutions localized in bounded frequency
zone. With the aim of deriving an exponential stability of solutions, we now follow the idea of
Subsection 2.3 in [9]. Let us assume that there exists an eigenvalue λ = id with d ∈ R\{0}. In
other words, according to (6), the non-zero real number d should fulfill the equalities
 
id τ d2 − β|ξ|2 = 0 and d2 − |ξ|2 = 0.

Due to the settings that d 6= 0 and τ ∈ (0, β), it immediately finds a contradiction. Namely, there
does not exists any pure imaginary roots to the cubic equation (6) for ξ ∈ Zmid(ε, N). Viewing
the expansions of eigenvalues, we know Re λj (|ξ|) < 0 for any j = 1, 2, 3 as ξ ∈ Zint (ε) ∪ Zext (N).
Therefore, by applying the compactness of bounded frequency zone Zmid(ε, N) and the continuity
of the eigenvalues, the derivation of the exponential decay estimates (9) and the proof of this
proposition are complete.

2.2 L2 estimates of solutions

Basing on the pointwise estimates shown in Proposition 2.1, we next investigate L2 − L2 estimates
with or without additional Lm regularity with m ∈ [1, 2), respectively. These estimates will play
an essential role in the forthcoming part to consider global (in time) existence of solutions to the
semilinear MGT model.
 7

Theorem 2.1. Let τ ∈ (0, β). Then, the solution u = u(t, x) to the Cauchy problem (2) fulfills the
following estimates:

(1 + t)1− 2s ku
2 kH max{s−2,0} (Rn ) if s ∈ [0, 1),
k |D|su(t, ·)kL2(Rn ) . 1
−s
(1 + t) 2 2 ku2 kH max{s−2,0} (Rn ) if s ∈ [1, ∞),
for any t > 0.
Proof. By applying Proposition 2.1 and the Parseval equality, we arrive at
! !
| sin(|ξ|t)| − β−τ |ξ|2t
s s 1
k |D| u(t, ·)kL2 (Rn ) . χint (ξ)|ξ|

| cos(|ξ|t)| + e 2 + e− τ t

ku2 kL2 (Rn )
|ξ| L∞ (Rn )


+ e−ct ku2 kL2 (Rn ) + e−ct χext (ξ)|ξ|s−2û2 (ξ) , (14)
L2 (Rn )

with the aid of the norm inequality k · kL2 (Rn ) 6 k · kL∞ (Rn ) k · kL2 (Rn ) .
Let us estimate the first L∞ norm on the right-hand side of (14). Obviously, by using | cos(|ξ|t)| 6
1, then for any t > 0 we get

s − β−τ |ξ|2 t β−τ
|ξ|2 t s
χint (ξ)|ξ| | cos(|ξ|t)|e 2 . χint (ξ)|ξ|se− 2 ∞ n . (1 + t)− 2 .
L∞ (Rn ) L (R )

We will divide the remaindering estimate into two parts.

Concerning the case for small time, i.e. t ∈ [0, 1], one may directly obtain

| sin(|ξ|t)|
β−τ 2 β−τ 2
χint (ξ)|ξ|s−1| sin(|ξ|t)|e− 2 |ξ| t . t χint (ξ)|ξ|s e− 2 |ξ| t . 1,
L∞ (Rn ) |ξ|t L∞ (Rn )

which immediately shows bounded estimates for small time.

For another, concerning the case for large time, i.e. t ∈ (1, ∞), one applies | sin(|ξ|t)| 6 1 to have
  s−1
β−τ 2 − s−1
− β−τ

|ξ|2 t 1 s
χint (ξ)|ξ|s−1| sin(|ξ|t)|e− 2 |ξ| t .t 2 χint (ξ) |ξ|2 t 2
e 2
. t2−2
L∞ (Rn ) L∞ (Rn )

for s ∈ [1, ∞). In the case s ∈ [0, 1), we do by another way that

 s | sin(|ξ|t)| − β−τ |ξ|2t
s−1 β−τ 2 1− 2s 1− 2s
χint (ξ)|ξ| | sin(|ξ|t)|e− 2 |ξ| t .t
2 2
χint (ξ) |ξ| t e 2
∞ n
. t .
L∞ (Rn ) |ξ|t L (R )

Thus, it completes that


(1 + t)1− 2s if s ∈ [0, 1),
β−τ 2
χint (ξ)|ξ|s−1| sin(|ξ|t)|e− 2 |ξ| t . 1
L∞ (Rn ) (1 + t) −s
2 2 if s ∈ [1, ∞),

for any t > 0. Particularly, decay estimates hold for any s ∈ (1, ∞).
On the other hand, we know


χext (ξ)|ξ|s−2û2 (ξ) . ku2 kH max{s−2,0} (Rn ) ,
L2 (Rn )

where we used χext (ξ)|ξ|s−2 . 1 if s ∈ [0, 2] and χext (ξ)|ξ|s−2 . (1 + |ξ|2)(s−2)/2 if s ∈ (2, ∞).
Summarizing the derived estimates, the proof is now complete.
 8

Theorem 2.2. Let τ ∈ (0, β). Then, the solution u = u(t, x) to the Cauchy problem (2) fulfills the
following estimates:

 F(t)ku
2 kH max{s−2,0} (Rn )∩Lm (Rn ) if 2sm + (2 − m)n < 2 + m,
k |D|s u(t, ·)kL2(Rn ) .
(1 + t) 12 − 2s − n(2−m) ku
4m
2 kH max{s−2,0} (Rn )∩Lm (Rn ) if 2sm + (2 − m)n > 2 + m,
for any t > 0, where s ∈ [0, ∞) and m ∈ [1, 2). In the above case 2sm + (2 − m)n < 2 + m, the
time-dependent coefficient is denoted by
 n(2−m)


(1 + t)1−s− 2m if 2sm + (2 − m)n < 2m,
 1 s n(2−m) 2−m
−2−
F(t) := (1 + t) 2 4m (ln(e + t)) 2m if 2sm + (2 − m)n = 2m,

 n(2−m) 2+m−2sm−(2−m)n
(1 + t) 12 − 2s − 4m + 2(2+m) if 2sm + (2 − m)n > 2m.
Remark 2.2. Let us consider the special case m = 1. The estimates stated in Theorem 2.2 improve
those results of Theorem 5.1 and Theorem 5.3 in [48]. For example, concerning the estimate of the
solution itself, i.e. s = 0, according to Theorem 2.2, we arrive at
 1

 (1 + t) 2 ku2 kL2 (R)∩L1 (R) if n = 1,

 1
ku(t, ·)kL2(Rn ) . (ln(e + t)) 2 ku2kL2 (R2 )∩L1 (R2 ) if n = 2,



(1 + t) 21 − n4 ku
2 kL2 (Rn )∩L1 (Rn ) if n > 3,
where the derived estimates in the low-dimensional cases n = 1 and n = 2 are sharper than those
in [48]. General speaking, we replace the restriction s + n > 3 in Theorem 5.3 shown in [48] by
2s + n > 3, which allows us to get shaper estimates in a larger admissible range of dimensions, e.g.
n = 2 with s = 1/2. For another, the requirement of the regularity for initial data is relaxed from
H s to H max{s−2,0} .
Proof. We may start by discussing the case for small frequencies. Employing Hölder’s inequality
and the Hausdorff-Young inequality, one has
kχint (D)|D|su(t, ·)kL2 (Rn )
! !
| sin(|ξ|t)| − β−τ |ξ|2 t
s − 1
t
. χint (ξ)|ξ| | cos(|ξ|t)| + e 2 + e τ 2m ku2 kLm (Rn )
|ξ| 2−m n
L (R )
  β−τ 2
1
. χint (ξ) |ξ|s| cos(|ξ|t)| + |ξ|s−1 | sin(|ξ|t)| e− 2 |ξ| t 2m n ku2 kLm (Rn ) + e− τ t ku2 kLm (Rn )
L 2−m (R )
Z ε  2−m
2(s−1)m 2m (β−τ )m 2 2m s n(2−m)
+n−1
. r 2−m | sin(rt)| 2−m e− 2−m
r t
dr ku2 kLm (Rn ) + (1 + t)− 2 − 4m ku2kLm (Rn ) ,
0

where we applied
Z ε  2−m
(β−τ )m 2
β−τ 2 2sm
+n−1 2m
− r t 2m
χint (ξ)|ξ|s| cos(|ξ|t)|e− 2 |ξ| t 2m . r 2−m | cos(rt)| 2−m e 2−m dr
L 2−m (Rn ) 0
s n(2−m)
. (1 + t)− 2 − 4m .
Let us now estimate the term including the sine function which is denoted by
Z ε  2−m
2(s−1)m 2m (β−τ )m 2 2m
+n−1
G(t) := r 2−m | sin(rt)| 2−m e− 2−m
r t
dr .
0
 9

 
Due to the interplay between the diffusive part from exp − (β−τ
2−m
)m 2
r t and the oscillating part from
| sin(rt)|/r, one should analyze a delicate equilibrium as well as the singularity as r → 0+ in the
case for negative power of r. This treatment is the difference from those in [48]. For one thing, as
usual approach by considering t ∈ [0, 1], we find
 ! 2m  2−m
Z 2m
ε 2ms
+n−1 | sin(rt)| 2−m (β−τ )m
− 2−m r 2 t
G(t) = t  r 2−m e dr  . 1,
0 rt
where we used 2ms/(2 − m) + n − 1 > 0. For another, we consider t ∈ (1, ∞) to derive
Z ε 2(s−1)m+(n−1)(2−m)
 2−m
2(s−1)m+n(2−m) (β−τ )m 2 1 2m 1 s n(2−m)
− 2 − r t 2
G(t) = t 4m (r t) 2(2−m) e 2−m d(r t) 2 . t2−2− 4m ,
0

where we restricted 2sm + (2 − m)n > 2 + m to guarantee the nonnegativity of the power for r 2 t
in the integral term, otherwise, a singularity will come as r → 0+ .
Let us use another approach to get the result when 2sm + (2 − m)n < 2 + m for t ∈ (1, ∞). Setting
1
a new variable ω = rt 2 , it holds that
s−1 n(2−m) 2−m
G(t) . t− 2
− 4m
( I(t)) 2m , (15)
where the time-dependent function on the right-hand side is defined by
Z Z !
t−1/α ∞ 2sm+(2−m)n−(2+m) 2m (β−τ )m 2
(1)
I(t) := I (t) + I (t) := (2)
+ ω 2−m | sin(t1/2 ω)| 2−m e− 2−m
ω
dω.
0 t−1/α

Here, we used WKB analysis to separate the integral over (0, ∞) to (0, t−1/α ) and [t−1/α , ∞) carrying
a suitable positive constant α to be determined later. The choice of the parameter α is helpful for
us to understand sharper estimates.
To estimate I(1) (t), by the boundedness of | sin(y)/y|, we obtain
Z  2m
 sin(t1/2 ω)  2−m
t−1/α 2sm+(2−m)n (β−τ )m 2
m   −1
(1)
I (t) . t 2−m   ω 2−m e− 2−m
ω
dω
0  t ω 
1/2

mα−2sm−(2−m)n+2−m
Z t−1/α mα−2sm−(2−m)n
.t (2−m)α dω . t (2−m)α , (16)
0

where we observed 2sm + (2 − m)n > (2 − m) for any s ∈ [0, ∞) and m ∈ [1, 2).
To investigate the estimate for I(2) (t), we divide the discussion into three cases. If 2sm+(2−m)n <
2m, then we may directly apply integration by parts to find
Z ∞ 2sm+(2−m)n−(2+m) (β−τ )m 2
I(2) (t) . ω 2−m e− 2−m
ω
dω
t−1/α
 
2−m 2sm+(2−m)n−2m (β−τ )m 2 ω=∞
. ω 2−m e− 2−m ω 
2sm + (2 − m)n − 2m ω=t−1/α
Z ∞
2(β − τ )m 2sm+(2−m)n−2m (β−τ )m
+1 − 2−m ω 2
+ ω 2−m e dω
2sm + (2 − m)n − 2m t−1/α
2m−2sm−(2−m)n (β−τ )m −2/α
Z ∞ 2sm+(2−m)n+2−3m (β−τ )m 2
− t
.t (2−m)α e 2−m − ω 2−m e− 2−m
ω
dω
t−1/α
2m−2sm−(2−m)n
.t (2−m)α . (17)
 10

By considering (16) and (17), in the case 2sm + (2 − m)n < 2m, we may obtain the sharp estimates
mα−2sm−(2−m)n 2m−2sm−(2−m)n 2m−2sm−(2−m)n
I(t) . t (2−m)α +t (2−m)α .t 2(2−m) ,
providing that mα − 2sm − (2 − m)n = 2m − 2sm − (2 − m)n if and only if α = 2.
Let us turn to the case 2sm + (2 − m)n = 2m. Therefore, by the similar procedure to the above,
we estimate
Z ∞ (β−τ )m 2
I(2) (t) . ω −1e− 2−m
ω
dω
t−1/α
  ω=∞ Z
−
(β−τ )m 2
ω  2(β − τ )m ∞ (β−τ )m 2
. (ln ω)e 2−m 
 + ω| ln ω|e− 2−m
ω
dω
ω=t−1/α 2−m t−1/α
Z ∞
1 (β−τ )m −2/α (β−τ )m 2
. (ln t)e− 2−m t + ω| ln ω|e− 2−m ω dω . ln t. (18)
α 0

Then, by choosing α = 2 again, it follows from (16) and (18) that

mα−2sm−(2−m)n
I(t) . t (2−m)α + ln t . ln t,
when 2sm + (2 − m)n = 2m.
In the remaindering case 2sm + (2 − m)n > 2m, we found that
Z ∞ 2sm+(2−m)n−(2+m) (β−τ )m 2
(2)
I (t) . ω 2−m e− 2−m
ω
dω
t−1/α
2+m−2sm−(2−m)n
Z ∞ (β−τ )m 2 2+m−2sm−(2−m)n
.t (2−m)α e− 2−m
ω
dω . t (2−m)α . (19)
0

For the moment, we would like to remark that since

v
Z u
∞ (β−τ )m
− 2−m ω 2
u π(2 − m)
e dω = t ,
0 2(β − τ )m
the restriction on the dissipative case, i.e. τ ∈ (0, β), acts a pivotal part in the way that
Z ∞ (β−τ )m 2
lim− e− 2−m
ω
dω = ∞,
τ →β 0

which somehow shows the limit case τ = β having singularities.

Combining (16) and (19), it yields
mα−2sm−(2−m)n 2+m−2sm−(2−m)n m(2+m−2sm−(2−m)n)
I(t) . t (2−m)α +t (2−m)α .t (2+m)(2−m) ,
where we chose α = (2 + m)/m to guarantee the optimality of the last competition.
All in all, from (15), for t ∈ (1, ∞) we assert that
 n(2−m)
 1−s− 2m
t


if 2sm + (2 − m)n < 2m,
1 n(2−m) 2−m
− s − 4m
G(t) . t 2 2 (ln t) 2m if 2sm + (2 − m)n = 2m,

 1 n(2−m) 2+m−2sm−(2−m)n
 − s − 4m +
t 2 2 2(2+m) if 2sm + (2 − m)n > 2m,
in the case 2sm + (2 − m)n < 2 + m.
The estimates of solutions for bounded frequencies and large frequencies are exactly the same as
those in Theorem 2.1. Thus, the summary of the derived estimates completes the proof.
 11

3 Asymptotic profiles in a framework of weighted L1 space

3.1 Optimal estimates with weighted L1 data
In this subsection, we will derive asymptotic profiles for the linear MGT equation in the dissipative
case in a framework of L1,1 space, where
 Z 
L1,1 (Rn ) := f ∈ L1 (Rn ) : kf kL1,1 (Rn ) := (1 + |x|)|f (x)|dx < ∞ .
Rn

As a preparation, we now define a time-dependent function

 1

 t2 if n = 1,

 1
Dn (t) := (ln t) 2 if n = 2,


t− n−2
4 if n > 3.

In order to derive asymptotic profiles of solutions, we will estimate upper bounds and lower bounds
of the solution itself with u2 ∈ L2 (Rn )∩L1,1 (Rn ). Before processing these estimates, let us introduce
R
the notation for the integral of f (x) by Pf := Rn f (x)dx, and recall Lemma 2.1 from [26].
Lemma 3.1. Let us assume f ∈ L1,1 (Rn ). Then, the following estimate holds:

|fˆ(ξ)| 6 C1 |ξ| kf kL1,1(Rn ) + |Pf |,

with a positive constant C1 > 0.

Moreover, due to the support condition for χint (ξ), by minor modifications of some derived
lemmas in [27, 30], one may show the validity of Lemma 3.2. Or one may use the inequality
 
ˆ |ξ|)kL2(Rn ) & kf(t,
kχint (ξ)f(t, ˆ |ξ|)kL2(Rn ) − kχmid (ξ)f(t,
ˆ |ξ|)kL2(Rn ) + kχext (ξ)f(t,
ˆ |ξ|)kL2(Rn )

for t ≫ 1, where fˆ(t, |ξ|) = | sin(|ξ|t)|e−c|ξ| t /|ξ| or fˆ(t, |ξ|) = | cos(|ξ|t)|e−c|ξ| t and estimates
2 2

ˆ |ξ|)kL2(Rn ) + kχext (ξ)f(t,

kχmid (ξ)f(t, ˆ |ξ|)kL2(Rn ) . e−c0 t ,

with a suitable constant c0 > 0, to prove the next lemma.

Lemma 3.2. Let n > 1. The following estimates hold:

| sin(|ξ|t)| −c|ξ|2t

Dn (t) . χint (ξ) e
2 n
. Dn (t),
|ξ| L (R )

n −c|ξ|2t n
t− 4 . χint (ξ)| cos(|ξ|t)|e 2 n . t− 4 ,
L (R )

with c > 0, for t ≫ 1.

Let us state our result on asymptotic profiles of the solution. Particularly, in one and two-
dimensional cases, we can easily observe the glow-up properties of the solution u(t, ·) in the L2
norm for the linear MGT equation in the dissipative case with initial data belonging to L2 ∩ L1,1 .
 12

Theorem 3.1. Let τ ∈ (0, β). Let us assume u2 ∈ L2 (Rn ) ∩ L1,1 (Rn ) and |Pu2 | =
6 0. Then, the
solution u = u(t, x) to the Cauchy problem (2) fulfills the following estimates:
Dn (t)|Pu2 | . ku(t, ·)kL2(Rn ) . Dn (t)ku2 kL2 (Rn )∩L1,1 (Rn )
for any t ≫ 1.
Remark 3.1. According to Theorem 3.1 and concerning t ≫ 1, we may observe that the decay
rate for the estimates of ku(t, ·)kL2(Rn ) from the above and the below are the same for any n > 1.
Moreover, u2 ∈ L1,1 (Rn ) implies |Pu2 | < ∞ for n > 1. Namely, the decay estimates stated in
Theorem 3.1 are optimal for all spatial dimensions in a framework of weighted L1 space.
Proof. Initially, let us estimate upper bounds of solutions by modifying the estimate for small
frequencies. The philosophy of derivative is essentially the same as those in Theorem 2.2. By
applying Lemma 3.1 and Proposition 2.1, we arrive at
 β−τ 1

|ξ|2 t
χint (ξ)|û(t, ξ)| . χint (ξ) (|ξ| | cos(|ξ|t)| + | sin(|ξ|t)|) e− 2 + |ξ|e− τ t ku2kL1,1 (Rn )
! !
| sin(|ξ|t)| − β−τ |ξ|2t 1
+ χint (ξ) | cos(|ξ|t)| + e 2 + e− τ t |Pu2 |.
|ξ|
Clearly, for the sake of the polar co-ordinate transform, we may deduce
 

+ |ξ|e− τ t
β−τ 2 1
χint (ξ) (|ξ| | cos(|ξ|t)| + | sin(|ξ|t)|) e− 2 |ξ| t
L2 (Rn )
Z ε  1 Z ε 1
2 2 1
n+1 2 −(β−τ )r 2 t n−1 2 −(β−τ )r 2 t
. r | cos(rt)| e dr + r | sin(rt)| e dr + e− τ t
0 0
− n+2 −n − τ1 t −n
.t 4 +t 4 +e .t4

for t ≫ 1. Repeating the same procedure as those in Theorem 2.2, we have

n
kχint (D)u(t, ·)kL2(Rn ) . t− 4 ku2kL1,1 (Rn ) + Dn (t)|Pu2 | (20)
for t ≫ 1. In the case for bounded and large frequencies, we just need to use the same estimates
as those in Theorem 2.1. Finally, by using the fact that |Pu2 | 6 ku2kL1,1 (Rn ) , we are able to prove
upper bound estimates for the solution itself in the L2 norm.
Let us now turn to lower bound estimates. According to the study in Section 2, we may represent
the solution for small frequencies by
χint (ξ)û(t, ξ) = χint (ξ)I(t, |ξ|)û2(ξ)
= χint (ξ)(I1 (t, |ξ|) + I2 (t, |ξ|) + I3 (t, |ξ|))û2(ξ),
c ξ) which was shown in
where for the sake of convenience in the proof, we denoted I(t, |ξ|) := K(t,
(5), furthermore, we introduced
  
β−τ
exp ±i|ξ| − 2
|ξ|2 + O(|ξ|3) t
I1,2 (t, |ξ|) := 
1
,
(±2i|ξ| + O(|ξ|3)) τ
± i|ξ| − 23 (β − τ )|ξ|2 + O(|ξ|3)
  
exp − τ1 + (β − τ )|ξ|2 + O(|ξ|3) t
I3 (t, |ξ|) :=  1  .
τ
+ i|ξ| − 32 (β − τ )|ξ|2 + O(|ξ|3) 1
τ
− i|ξ| − 23 (β − τ )|ξ|2 + O(|ξ|3)
 13

By omitting the terms containing O(|ξ|3), we may regard next three functions as the leading term
of Ij (t, |ξ|) for small frequencies:
  
β−τ
exp ±i|ξ| − 2
|ξ|2 t
J1,2 (t, |ξ|) := 
1
,
±2i|ξ| τ
± i|ξ| − 32 (β − τ )|ξ|2
  
exp − τ1 + (β − τ )|ξ|2 t
J3 (t, |ξ|) :=  1  ,
τ
+ i|ξ| − 32 (β − τ )|ξ|2 1
τ
− i|ξ| − 23 (β − τ )|ξ|2
respectively, whose sum can be shown by
β−τ 2 !
X e− 2 |ξ| t sin(|ξ|t) 1 3 2
J(t, |ξ|) := Jk (t, |ξ|) = 2 T0 (|ξ|) + e− τ t+ 2 (β−τ )|ξ| t − cos(|ξ|t) . (21)
k=1,2,3 T0 (|ξ|) + |ξ|2 |ξ|
where we recalled (12). Now, we should be carefully analyze that the error estimates between the
leading term Jj (t, |ξ|) the formulas Ij (t, |ξ|) for j = 1, 2, 3, individually. It proves the additional
decay estimates. Concerning the case for J1 (t, |ξ|), denoting
1 3
g1 (|ξ|) := + i|ξ| − (β − τ )|ξ|2 = O(1) as |ξ| → 0,
τ 2
we may handle
 3 
 eO(|ξ| )t 1 
− β−τ|ξ|2 t 
χint (ξ)|I1(t, |ξ|) − J1 (t, |ξ|)| . χint (ξ)e 2 
 2i|ξ|g1 (|ξ|) + O(|ξ|3 )
− 
2i|ξ|g1(|ξ|) 
   
 2i|ξ|g (|ξ|) eO(|ξ|3 )t − 1 + O(|ξ|3 ) 
β−τ 2  1 
. χint (ξ)e− 2 |ξ| t  

 4|ξ|2(g1 (|ξ|))2 + O(|ξ|4) 
 Z 1 
β−τ
|ξ|2 t 1 3 )ts
. χint (ξ)e− 2 O(|ξ|4)t eO(|ξ| ds − O(|ξ|3)
|ξ|2 0
 2 2t

. χint (ξ) O(|ξ|2)te−c|ξ| t + O(|ξ|)e−c|ξ| ,
since there exists a constant c > 0 such that
Z 1
− β−τ |ξ|2 t 3 )ts β−τ
|ξ|2 t 1 2 2
χint (ξ)e 2 eO(|ξ| ds . χint (ξ)e− 4 e− 4 ((β−τ )−O(|ξ|))|ξ| t . χint (ξ)e−c|ξ| t .
0
Next, by repeating the same way as the previous one, we get
 2 2t

χint (ξ)|I2(t, |ξ|) − J2 (t, |ξ|)| . χint (ξ) O(|ξ|2)te−c|ξ| t + O(|ξ|)e−c|ξ| .
Considering the last term, by defining
 2
1 3
g2 (|ξ|) := − (β − τ )|ξ|2 + |ξ|2 = O(1) as |ξ| → 0,
τ 2
one has
 3 
 eO(|ξ| )t 1 

− τ1 t+(β−τ )|ξ|2 t
χint (ξ)|I3 (t, |ξ|) − J3 (t, |ξ|)| . χint (ξ)e 
 g2 (|ξ|) + O(|ξ|3 )
− 
g2 (|ξ|) 
 
 g (|ξ|)O(|ξ|3)t R 1 eO(|ξ|3 )ts ds − O(|ξ|3 ) 
1 2  2 
. χint (ξ)e− τ t+(β−τ )|ξ| t  0
2 3


(g2 (|ξ|)) + O(|ξ| )
. χint (ξ)e−ct O(|ξ|3)t . χint (ξ)e−ct
 14

for any t ≫ 1, which still provides us an exponential decay.

In conclusion, we can claim
  2
χint (ξ)|(I(t, |ξ|) − J(t, |ξ|))û2(ξ)| . χint (ξ) O(|ξ|2)t + O(|ξ|) e−c|ξ| t |û2 (ξ)|. (22)

Let us decompose initial data by

û2 (ξ) = Pu2 + A(ξ) − iB(ξ),

where
Z Z
A(ξ) := u2 (x)(1 − cos(x · ξ))dx and B(ξ) := u2 (x) sin(x · ξ)dx.
Rn Rn

In the view of Lemma 2.2 in [27], these ξ-dependent functions can be controlled by

|A(ξ)| + |B(ξ)| . |ξ| ku2kL1,1 (Rn ) .

As a consequence, we may represent the solution in the Fourier space by

û(t, ξ) = I(t, |ξ|)Pu2 + (A(ξ) − iB(ξ))I(t, |ξ|).

From the derived estimate (22), it yields

−1
χint (D)u(t, ·) − χint (D) Fξ→x (J(t, |ξ|))P u2
L2 (Rn )

6 kχint (ξ)(I(t, |ξ|) − J(t, |ξ|))kL2 (Rn ) |Pu2 | + kχint (ξ)(A(ξ) − iB(ξ))I(t, |ξ|)kL2 (Rn )
 
. χint (ξ) |ξ|2t + |ξ| e−c|ξ| t
2
|Pu2 | + kχint (ξ)|ξ|I(t, |ξ|)kL2(Rn ) ku2 kL1,1 (Rn )
L2 (Rn )
−n
.t 4 ku2kL1,1 (Rn ) (23)

for t ≫ 1, where we used

β−τ
|ξ|2 t n
kχint (ξ)|ξ|I(t, |ξ|)kL2(Rn ) . χint (ξ)(|ξ| + 1)e− 2 2 n . t− 4 .
L (R )

Additionally, let us recall the function J(t, |ξ|) in (21). By employing Lemma 3.2 and the Parseval
equality, it is valid that

−1
χint (D) Fξ→x (J(t, |ξ|)) 2 n
L (R )
!
sin(|ξ|t)
1 3 2
− t+ 2 (β−τ )|ξ| t β−τ 2
& χint (ξ) T0 (|ξ|) + e τ − cos(|ξ|t) e− 2 |ξ| t
|ξ| 2 n
L (R )
 
 − 2 |ξ| t
β−τ 2 − 2 |ξ| t
β−τ 2 

&  χint (ξ) H(t, |ξ|)e 2 n − χint (ξ) cos(|ξ|t)e 2 n 
L (R ) L (R )
−n
& | Dn (t) − t 4 | & Dn (t)

for t ≫ 1, where we denoted

sin(|ξ|t) 1 3 2
H(t, |ξ|) := T0 (|ξ|) + e− τ t+ 2 (β−τ )|ξ| t ,
|ξ|
 15

moreover, we used upper bound estimates as follows:

β−τ 2
χint (ξ) H(t, |ξ|)e− 2 |ξ| t . Dn (t)
L2 (Rn )

and estimates from the below such that

β−τ 2
χint (ξ) H(t, |ξ|)e− 2 |ξ| t 2 n
L (R )
 
 
 1 2 | sin(|ξ|t)| β−τ 2 
&  χint (ξ)e− τ t+(β−τ )|ξ| t 2 n − χint (ξ) e− 2 |ξ| t 
 L (R ) |ξ| 2 n 
L (R )

& |e−ct − Dn (t)| & Dn (t)

for any n > 1 and t ≫ 1.
Finally, by using the Minkowski inequality, we conclude

−1
kχint (D)u(t, ·)kL2(Rn ) > χint (D) Fξ→x (J(t, |ξ|)) |Pu2 |
L2 (Rn )

−1
− χint (D)u(t, ·) − χint (D) Fξ→x (J(t, |ξ|))Pu2
L2 (Rn )
−n
& Dn (t)|Pu2 | − t 4 ku2 kL1,1 (Rn )
for t ≫ 1. Actually, in the above by taking t ≫ 1, the time-dependent coefficients of |Pu2 | play
dominant influence for all n > 1. Thus, with the help of the fact that
ku(t, ·)kL2(Rn ) > kχint (D)u(t, ·)kL2(Rn ) & Dn (t)|Pu2 |,
the proof is complete.

3.2 Approximate relation in one- and two-dimensional cases

Our purpose in this part is to give an approximate relation between the linear MGT equation
and the linear viscoelastic damped wave equation (or the strongly damped wave equation). This
approximate relation is strongly related to the so-called diffusion phenomenon (see, for example,
[45]), which bridges a connection for the damped wave equation and the heat equation such that
 
τ udw dw
+ udw x ∈ Rn , t > 0, =⇒ −∆v h + vth = 0, x ∈ Rn , t > 0,
tt − ∆u t = 0,
udw (0, x) = 0, udw (0, x) = udw (x), x ∈ Rn , τ = 0 v h (0, x) = udw n
t 1 1 (x), x ∈ R .

It is well-known that the decay rates of udw (t, ·) and v h (t, ·) in the L2 norm are the same. Further-
more, the decay estimates of the difference

dw
u (t, ·) − v h (t, ·)
L2 (Rn )

is faster than the decay estimates for each of them in the L2 norm. The gained decay rate is
(1 + t)−1 . Namely, diffusion phenomena bridge the connection between second-order (in time)
evolution equations and first-order (in time) evolution equations.
Before giving our result, let us recall some derived estimates of solutions to the following linear
Cauchy problem:

ũ − ∆ũ − β∆ũt = 0, x ∈ Rn , t > 0,
tt
(24)
ũ(0, x) = 0, ũ (0, x) = ũ (x), x ∈ Rn ,
t 1
 16

where β > 0. The Cauchy problem for the viscoelastic damped wave equation has been deeply
studied in [52, 32, 14, 27, 28, 30, 3, 2] and references therein. Particularly, in the paper [27], the
author proved estimates of solutions to (24) as follows:
kũ(t, ·)kL2 (Rn ) . Dn (t)kũ1 kL2 (Rn )∩L1,1 (Rn ) (25)
for t ≫ 1, providing that |Pũ1 | =
6 0. Concerning the Cauchy problem, we found that the estimates
for the linear MGT equation (2) in Theorem 3.1, and for the viscoelastic damped wave equation
(24) in (25), are exactly the same. Therefore, we conjecture that behaviors of solutions for the linear
MGT equation are similar to those for the linear viscoelastic damped wave equation, especially the
decay property. Furthermore, it becomes interesting to derive the approximate relation between
them with suitable initial data, and to find a gained decay rate.
From the previous study, we know the decay rates of (L2 ∩ L1,1 ) − L2 estimates are determined
by the behavior of the eigenvalues for small frequencies only. In the case for bounded and large
frequencies, the behaviors of the eigenvalues together with the suitable regularity for initial data
show immediately some exponential decays. For this reason, the next approximate relation is
explained by the behavior of solutions localized in small frequency zone, which is the most interesting
one.
Theorem 3.2. Let τ ∈ (0, β). Let us assume u2 ∈ L1,1 (Rn ) and |Pu2 | = 6 0. Then, the solution
u = u(t, x) to the Cauchy problem (2) and the solution ũ = ũ(t, x) to the Cauchy problem (24) with
ũ1 (x) = u2(x) fulfill the following estimates:
1 n
kχint (D) (u(t, ·) − τ ũ(t, ·))kL2 (Rn ) . t 2 − 4 ku2 kL1,1 (Rn )
for any n > 1 and t ≫ 1.
Remark 3.2. By subtracting τ ũ(t, ·) in the L2 norm, we find the derived estimates for u(t, ·) in
1 1
Theorem 3.1 can be improved t− 4 if n = 1 and (ln t)− 2 if n = 2 for t ≫ 1. It is still open that the
gained decay rate for n > 3.
Remark 3.3. Indeed, from Theorems 3.1 and 3.2, one may derive
ku(t, ·) − τ ũ(t, ·)kL2(Rn ) = kχint (D)(u(t, ·) − τ ũ(t, ·))kL2 (Rn ) + k(1 − χint (D))(u(t, ·) − τ ũ(t, ·))kL2 (Rn )
1 n 1 n
. t 2 − 4 ku2kL1,1 (Rn ) + e−ct ku2 kL2 (Rn ) . t 2 − 4 ku2kL2 (Rn )∩L1,1 (Rn )
for t > t0 ≫ 1. Moreover, concerning 0 6 t 6 t0 , it is trivial that
ku(t, ·) − τ ũ(t, ·)kL2 (Rn ) . ku(t, ·)kL2(Rn ) + τ kũ(t, ·)kL2(Rn ) . ku2 kL2 (Rn )∩L1,1 (Rn ) .
Therefore, the approximate relation holds for all t > 0 and the whole spaces such that
1 n
ku(t, ·) − τ ũ(t, ·)kL2 (Rn ) . (1 + t) 2 − 4 ku2kL2 (Rn )∩L1,1 (Rn ) ,
where we assumed u2 ∈ L2 (Rn ) ∩ L1,1 (Rn ). Namely, the solution for the linear MGT equation
approximate to that for the linear viscoelastic damped wave equation at least for n = 1, 2.
Proof. By applying the partial Fourier transform ũˆ(t, ξ) = Fx→ξ (ũ(t, x)), let us recall the derived
inequality stated in Lemma 2.1 in [27] such that
!

ˆ ξ) − sin(|ξ|t) e− 2 |ξ|2t Pu2
β n
χint (ξ)

ũ(t,

. t− 4 ku2kL1,1 (Rn ) (26)
|ξ| L2 (Rn )
 17

for t ≫ 1. Again, ũ(t, x) is the solution to the viscoelastic damped wave equation (24) with initial
data choosing by ũ1 (x) = u2 (x).
We notice that the difference of the solutions can be decomposed by three components as follows:
!
ˆ ξ) = (û(t, ξ) − J(t, |ξ|)Pu2 ) + J(t, |ξ|) − τ sin(|ξ|t) e− 2 |ξ|2t Pu2
β
û(t, ξ) − τ ũ(t,
|ξ|
!
sin(|ξ|t) − β |ξ|2 t
+ τ e 2 Pu2 − τ ũˆ(t, ξ) .
|ξ|
Therefore, employing the Parseval equality and the norm inequality, we arrive at
 
ˆ ξ)
kχint (D) (u(t, ·) − τ ũ(t, ·))kL2 (Rn ) = χint (ξ) û(t, ξ) − τ ũ(t,
L2 (Rn )
!
sin(|ξ|t) − β |ξ|2t

. kχint (ξ) (û(t, ξ) − J(t, |ξ|)Pu2 )kL2 (Rn ) + χint (ξ) J(t, |ξ|) − τ e 2 |Pu2 |
|ξ|
L2 (Rn )
!
sin(|ξ|t)
ˆ ξ) − − β
|ξ| 2t
+ τ χint (ξ) ũ(t, e 2 Pu2
|ξ| 2 n
L (R )
−n
.t 4 ku2 kL1,1 (Rn ) + J(t)|Pu2 |,

where we used (23) and (26). In the above inequality, we denoted

!
sin(|ξ|t) − β |ξ|2t

J(t) := χint (ξ)

J(t, |ξ|) − τ e 2

.
|ξ| L2 (Rn )

In other words, we just need to estimate J(t) in the remaindering part of the proof.
Recalling (12), from the definition of J(t) in the last subsection, we may estimate

− β−τ |ξ|2 t  
e 2 1 3 2
J(t) . χint (ξ) 2 2
− cos(|ξ|t) + e− τ t+ 2 (β−τ )|ξ| t
T0 (|ξ|) + |ξ| 2 n
L (R )
τ 2t !
sin(|ξ|t) − β |ξ|2t T0 (|ξ|)e 2 |ξ|

+ χint (ξ) e 2 −τ
|ξ| T02 (|ξ|) + |ξ|2 2 n
L (R )
(1) (2)
=: J (t) + J (t).

For one thing, it is clear that

J(1) (t) . χint (ξ)| cos(|ξ|t)|e−c|ξ|
2t n
+ e−ct . t− 4
L2 (Rn )

for t ≫ 1. Before estimating J(2) (t), the explicit computation shows the identity as follows:
   
τ
|ξ|2 t τ
|ξ|2 t 3
T0 (|ξ|)e 2 −τ T02 (|ξ|) + |ξ| 2
= T0 (|ξ|) e 2 − 1 + τ (β − τ )|ξ|2 − τ |ξ|2
2
Z 1  
τ 2 τ
|ξ| 2 ts 3
= |ξ| t T0 (|ξ|) e2 ds + τ T0(|ξ|) (β − τ ) − 1 |ξ|2.
2 0 2
 18

Thus, we compute
Z 1
β 2 τ 2
J(2) (t) . t χint (ξ)| sin(|ξ|t)|e− 2 |ξ| t |ξ| | T0 (|ξ|)| e 2 |ξ| ts ds
0 L2 (Rn )
 

−β |ξ| 2t 

3 

+ χint (ξ)| sin(|ξ|t)|e 2  T0 (|ξ|) (β − τ ) − 1 |ξ| 2 n
2 L (R )

−c|ξ|2t 1 n
−
. t χint (ξ)|ξ| | sin(|ξ|t)|e 2 n . t2 4
L (R )

for t ≫ 1. Summarizing the derived estimates, one has

n 1 n
kχint (D) (u(t, ·) − τ ũ(t, ·))kL2 (Rn ) . t− 4 ku2 kL1,1 (Rn ) + t 2 − 4 |Pu2 |,

and the proof is immediately complete.

Remark 3.4. Although we can observe a similar decay property between the linear MGT equation
(2) and the linear viscoelastic damped wave equation (24), there is a great difference between them.
Recalling that the property of finite propagation speed (FPS) is valid for the linear MGT equation
in the conservative case, we refer to Section 2 in [11]. Actually, the property of FPS holds for the
linear MGT equation even in the dissipative case (see Remark 2.1). To estimate the propagation
speed, we construct an energy for (2), namely,
Z  2 Z  2
 1   1 
   
EFPS [u](t) := β ∇ut (t, x) +

∇u(t, x) dx + τ utt (t, x) +

ut (t, x) dx
Λβ,τ β Λβ,τ β
!Z
1 τ
+ 1− |ut (t, x)|2 dx,
β β Λβ,τ

where the domain is defined by

 q 
Λβ,τ := (t, x) : t ∈ [0, T ], |x − x0 | 6 β/τ (T − t) .

So, taking the derivative with respect to t, we arrive at

Z ! !
d 1 1
EFPS[u](t) = 2β ∇ ut (t, x) + u(t, x) · ∇ utt (t, x) + ut (t, x) dx
dt Λβ,τ β β
Z ! !
1 1
+ 2τ utt (t, x) + ut (t, x) uttt (t, x) + utt (t, x) dx
Λβ,τ β β
!Z
2 τ
+ 1− ut (t, x)utt (t, x)dx
β β Λβ,τ
s  2 s  2
Z   Z  
β  1  β  1 
−β ∇ut (t, x) +

∇u(t, x) dS − τ utt (t, x) +

ut (t, x) dS
τ ∂Λβ,τ β τ ∂Λβ,τ β
s !Z
1 β τ
+ 1− |ut (t, x)|2 dS.
β τ β ∂Λβ,τ
 19

Integration by parts and the Cauchy-Schwarz inequality yield

Z ! !
1 1
2β ∇ ut (t, x) + u(t, x) · ∇ utt (t, x) + ut (t, x) dx
Λβ,τ β β
Z ! !
1 1
6 −2β ∆ut (t, x) + ∆u(t, x) utt (t, x) + ut (t, x) dx
Λβ,τ β β
s  2 s Z  2
βZ 
 1 
 τ 
 1 

+β ∇ut (t, x) + ∇u(t, x) dS + β utt (t, x) + ut (t, x) dS.
τ ∂Λβ,τ  β  β ∂Λβ,τ  β 

Finally, summarizing them, one has

!Z Z
d τ 1
EFPS [u](t) 6 −2 1 − |utt (t, x)|2 dx − √ |ut (t, x)|2 dS 6 0.
dt β Λβ,τ βτ ∂Λβ,τ

In other words, for all t ∈ [0,nT ], it is valid

q that E
o FPS
[u](t) 6 EFPS[u](0) = 0 if u(0, x) = ut (0, x) =
0
utt (0, x) = 0 in a set Λβ,τ := |x − x0 | 6 β/τ T . According to the definition of energy EFPS[u](t),
it follows immediately that ut (t, x) = 0 and ∇u(t, x) = 0 in Λβ,τ q
. This implies u(t, x) = 0 in
Λβ,τ due to u(0, x) = 0 in Λ0β,τ . The propagation speed reads as β/τ . In the limit case τ = β,
the propagation speed is equal to 1, which corresponds to the statement of Section 2 in [11]. In
particular, formally taking τ = 0, i.e. the viscoelastic damped wave equation, the propagation speed
is infinite. However, the property of FPS does not hold anymore in the linear viscoelastic damped
wave equation, and the solution to (24) has some smoothing effects.
Remark 3.5. Actually, there is another aspect to analyze the linear MGT equation. Let us consider
the conservative case (τ = β) in the Cauchy problem (2), whose local (in spaces) energy with R > 0
can be defined by
1 1
EMGT,R [u](t) := k∂t (βut (t, ·) + u(t, ·))kL2(BR ) + k∇(βut (t, ·) + u(t, ·))kL2(BR ) .
2 2
Motivated by [11], we may rewrite the Cauchy problem (2) with τ = β by the next way:

(βu + u)tt − ∆(βut + u) = 0, x ∈ Rn , t > 0,
t
(βu + u)(0, x) = 0, (βu + u) (0, x) = βu (x), x ∈ Rn .
t t t 2

Then, by applying Theorem 1.2 with c(x) ≡ 1, n > 3 and L = 0 in the recent paper [8], we can get

EMGT,R [u](t) = O(t−1 )

for each R > 0 and t ≫ 1, provided that u2 ∈ L1 (Rn ) and

Z
(1 + |x|)|u2 (x)|2 dx < ∞.
Rn

In other words, local (in spaces) energy for the linear MGT equation in the conservative case τ = β
decays with an algebraic decay order.
 20

4 Singular limit problem

In this section, we focus on the following Cauchy problem for the singular limit problem of the form:

τ u + uτ,tt − ∆uτ − β∆uτ,t = 0, x ∈ Rn , t > 0,
τ,ttt
(27)
u (0, x) = u (x), u (0, x) = u (x), u (0, x) = u (x), x ∈ Rn ,
τ 0 τ,t 1 τ,tt 2

where τ ∈ (0, β) with β > 0. Moreover, the time-derivative for the unknown uτ = uτ (t, x) is denoted
by uτ,t := ∂t uτ , and similarly for uτ,tt as well as uτ,ttt . Particularly, we consider τ to be a small
parameter such that 0 < τ ≪ β. In other words, our main purpose in the section is to understand
the asymptotic profiles of the solution uτ = uτ (t, x) as τ → 0+ . This property has been studied
between damped wave equations and heat equations. We refer readers to [37, 25, 29, 13, 23, 19, 31]
and references therein. Nevertheless, concerning the study of the Cauchy problem for the linear
MGT equation, it seems new from the knowledge of authors.
Let us introduce the Cauchy problem for the viscoelastic damped wave equation, namely,

v − ∆v − β∆vt = 0, x ∈ Rn , t > 0,
tt
(28)
v(0, x) = u (x), v (0, x) = u (x), x ∈ Rn ,
0 t 1

where β > 0. As mentioned in the last section, the Cauchy problem for the viscoelastic damped
wave equation has been widely studied. For instance, considering Theorem 14.3.2 and Corollary
14.3.1 in the book [18], we know solutions to the Cauchy problem (28) fulfill
2  

|D|k v(t, ·) 2 n 6 C (1 + t)−k ku0 k2H k (Rn ) + (1 + t)−(k−1) ku1k2H k−1 (Rn ) for k > 1,
L (R )
2  
k
|D| vt (t, ·) 2 n 6 C (1 + t)−(k+1) ku0k2H k (Rn ) + (1 + t)−k ku1k2H k (Rn ) for k > 1.
L (R )

Therefore, it is easy to observe that


X
j k
2

C(1 + t)−1 k(u 2
0 , u1 )kH 4 (Rn )×H 4 (Rn ) if u1 6= 0,
∇ ∂t v(t, ·) 2 6 (29)
j,k=1,2
L (Rn ) C(1 + t)−2 ku0k2H 4 (Rn ) if u1 = 0,

X
1+j k
2

Ck(u 2
0 , u1 )kH 2 (Rn )×H 2 (Rn ) if u1 6= 0,
∇ ∂t v(t, ·) 2 6 (30)
L (Rn ) C(1 + t)−1 ku0k2H 2 (Rn ) if u1 = 0,
j,k=0,1

where we employed k∇j f (t, ·)kL2 (Rn ) ≈ k |D|j f (t, ·)kL2 (Rn ) .
Finally, let us define w = w(t, x) such that

w(t, x) := uτ (t, x) − v(t, x), (31)

where uτ = uτ (t, x) is the solution to the Cauchy problem (27) and v = v(t, x) is the solution to
the Cauchy problem (28).

4.1 Singular limit for an energy

Theorem 4.1. Let us assume (u0, u1 , u2 ) ∈ H 4 (Rn ) × H 4 (Rn ) × L2 (Rn ), where u0 and u1 are not
zero simultaneously. Then, the difference w = w(t, x) defined in (31) fulfills the following estimates
 21

for small τ such that 0 < τ ≪ β:

Z t Z t
E[w](t) + (2 − ε1 − 2τ k) kwηη (η, ·)k2L2 (Rn ) dη + (2βk − ε1 − 2) k∇wη (η, ·)k2L2 (Rn ) dη
0  0
Cτ 2 ln(e + t)k(u 2
0 , u1 )kH 4 (Rn )×H 4 (Rn ) if u1 6= 0,
6 τ ku2 − ∆u0 − β∆u1 k2L2 (Rn ) +
Cτ 2 ku 2
0 kH 4 (Rn ) if u1 = 0,
h i  i
where C is a positive constant independent of τ , and k ∈ 2+ε1 2−ε1
2β
, 2τ carrying ε1 ∈ 0, 2β−2τ
β+τ
.
Moreover, the energy E[w](t) is defined by
2
1

E[w](t) := β ∇wt (t, ·) + ∇w(t, ·) + τ kwtt (t, ·) + kwt (t, ·)k2L2 (Rn )
β
L2 (Rn )
!
1
+ k(1 − τ k)kwt (t, ·)k2L2 (Rn ) + k − k∇w(t, ·)k2L2 (Rn ) .
β
Remark 4.1. The assumption that u0 and u1 are not zero simultaneously, is natural to guarantee
nontrivial solution for the viscoelastic damped wave equation (28).
Remark 4.2. Actually, the choice of parameters k and ε1 can be independent of τ for a small
value of τ > 0. For example, by taking a small τ such that 0 < τ 6 39β/41, we can fix ε1 = 1/20
and k = 41/(40β). Particularly, by considering τ → 0+ , we can immediately enlarge the choice of
parameters k and ε1 .
Remark 4.3. Let us consider t ∈ (0, T ). In Theorem 4.1 with T < ∞, we may observe
• if u2 6= ∆u0 + β∆u1 , it holds E[w](t) = O(τ ) as τ → 0+ ;
• if u2 = ∆u0 + β∆u1 , it holds E[w](t) = O(τ 2 ) as τ → 0+ .
So, the speeds of convergence are different under different choices of initial data. However, concern-
ing T = ∞, the property for singular limit holds if and only if u1 = 0. Otherwise, we found that
E[w](t) = O(ln(e + t)) as t → ∞. In conclusion, the choice for initial data is extremely important
in the consideration of singular limit property.
Proof. Let us act τ ∂t on the equation in (28) and then add itself to arrive at

τ vttt + vtt − ∆v − β∆vt = τ (∆vt + β∆vtt ). (32)

Let us recall w = w(t, x) as a difference such that w(t, x) = uτ (t, x) − v(t, x). Then, by subtracting
the equation in (27) with (32), we have

τ wttt + wtt − ∆w − β∆wt = −τ (∆vt + β∆vtt ). (33)

To achieve our aim, we next will apply the classical energy method for the Cauchy problem. For
one thing, we construct an energy as follows:
Z
E1[w](t) := τ kwtt (t, ·)k2L2 (Rn ) + βk∇wt(t, ·)k2L2 (Rn ) − 2 ∆w(t, x)wt (t, x)dx.
Rn
 22

It shows that
Z Z
d
E1[w](t) = 2τ wttt (t, x)wtt (t, x)dx + 2β ∇wtt (t, x) · ∇wt (t, x)dx
dt ZR
n
ZR
n

−2 ∆wt (t, x)wt (t, x)dx − 2 ∆w(t, x)wtt (t, x)dx

Rn Rn Z
= −2kwtt (t, ·)k2L2 (Rn ) + 2k∇wt (t, ·)k2L2 (Rn ) − 2τ (∆vt (t, x) + β∆vtt (t, x))wtt (t, x)dx,
Rn

where we considered (33).

For another, let us introduce the other auxiliary energy
Z
E2 [w](t) := k∇w(t, ·)k2L2(Rn ) + kwt (t, ·)k2L2(Rn ) + 2τ wtt (t, x)wt (t, x)dx.
Rn

Taking the derivative with respect to time variable, we have

Z Z
d
E2[w](t) = 2 ∇wt (t, x) · ∇w(t, x)dx + 2 wtt (t, x)wt (t, x)dx
dt Rn Z ZR
n

+ 2τ wttt (t, x)wt (t, x)dx + 2τ wtt (t, x)wtt (t, x)dx
Rn Rn
= −2βk∇wt (t, ·)k2L2 (Rn ) + 2τ kwtt (t, ·)k2L2 (Rn )
Z
+ 2τ (∇vt (t, x) + β∇vtt (t, x)) · ∇wt (t, x)dx.
Rn

Let k be a positive parameter to be fixed later. Hence, by applying

Z Z
−2 ∆w(t, x)wt (t, x)dx = 2 ∇w(t, x) · ∇wt (t, x)dx
Rn Rn
2 !
1 1

= β ∇w(t, ·) + ∇wt (t, ·) − k∇w(t, ·)k2L2(Rn ) + βk∇wt(t, ·)k2L2 (Rn )
β 2 β
L (Rn )

and
Z  
2τ k wtt (t, x)wt (t, x)dx = τ kwtt (t, ·) + kwt (t, ·)k2L2 (Rn ) − τ kwtt (t, ·)k2L2 (Rn ) + k 2 kwt (t, ·)k2L2 (Rn ) ,
Rn

one may rewrite the total energy by

2
1

E1 [w](t) + k E2[w](t) = β ∇wt (t, ·) + ∇w(t, ·) + τ kwtt (t, ·) + kwt (t, ·)k2L2 (Rn )
β 2 n
L (R )
!
1
+ k(1 − τ k)kwt (t, ·)k2L2(Rn ) + k− k∇w(t, ·)k2L2(Rn ) .
β
h i
1 1
To guarantee the non-negativity of the above combined energy, we need to restrict k ∈ ,
β τ
. Here,
we should underline in advance that

E1 [w](0) + k E2[w](0) = τ ku2 − ∆u0 − β∆u1 k2L2 (Rn ) ,

since w(0, x) = wt (0, x) = 0 and wtt (0, x) = u2 (x) − ∆u0 (x) − β∆u1 (x).
 23

Furthermore, the application of Cauchy’s inequality indicates that there exists a small constant
ε1 > 0 such that
Z Z
− 2τ (∆vt (t, x) + β∆vtt (t, x))wtt (t, x)dx + 2kτ (∇vt (t, x) + β∇vtt (t, x)) · ∇wt (t, x)dx
Rn Rn
2τ 2   
6 k∆vt (t, ·)k2L2 (Rn ) + β 2 k∆vtt (t, ·)k2L2 (Rn ) + k 2 k∇vt (t, ·)k2L2(Rn ) + β 2 k∇vtt (t, ·)k2L2 (Rn )
ε1
 
+ ε1 kwtt (t, ·)k2L2 (Rn ) + k∇wt (t, ·)k2L2(Rn ) .

We summarize the derived inequalities, which lead to

d
( E1[w](t) + k E2[w](t)) + (2 − ε1 − 2τ k)kwtt (t, ·)k2L2 (Rn ) + (2βk − ε1 − 2)k∇wt (t, ·)k2L2(Rn )
dt
2τ 2   
6 k∆vt (t, ·)k2L2 (Rn ) + k 2 k∇vt (t, ·)k2L2 (Rn ) + β 2 k∆vtt (t, ·)k2L2 (Rn ) + k 2 k∇vtt (t, ·)k2L2 (Rn ) .
ε1
h i
2+ε1 2−ε1
By choosing k ∈ 2β
, 2τ , we found that 2 − ε1 − 2τ k > 0 and 2βk − ε1 − 2 > 0. To guarantee
 i
the non-empty set of k, we restrict ourselves ε1 ∈ 0, 2β−2τ
β+τ
.
2
Using the derived L estimates (29), we see
d
( E1[w](t) + k E2 [w](t)) + (2 − ε1 − 2τ k)kwtt (t, ·)k2L2 (Rn ) + (2βkε1 − 2)k∇wt (t, ·)k2L2 (Rn )
dt 
Cτ 2 (1 + t)−1 k(u , u )k2 4 n if u1 6= 0,
0 1 H (R )×H 4 (Rn )
6
Cτ 2 (1 + t)−2 ku k2 4 n if u1 = 0,
0 H (R )

where C is a positive constant independent of τ . Finally, integrating the above inequality over [0, t],
one gets our desired inequality.

4.2 Singular limit for the solution itself

Let us turn to the single limit for the solution itself, which is not a trivial generalization of the last
result because the L2 norm for the solution itself is not included in an energy of the MGT equation
in the dissipative case. Motivated by [25, 29], we will use Hardy’s inequality associated with a new
variable to overcome the difficulty.
Theorem 4.2. Let n > 3. Let us assume (u0, u1 , u2 ) ∈ H 2 (Rn ) ×H 2 (Rn ) ×L2 (Rn ) and additionally
|x|u2 ∈ L2 (Rn ), where u0 and u1 are not zero simultaneously. Then, the difference w(t, x) defined
in (31) fulfills the following estimates for small τ such that 0 < τ ≪ β:
Z t Z t
C̄kw(t, ·)k2L2(Rn ) + (2k̃β − ε2 − 2) k∇w(η, ·)k2L2(Rn ) dη + (2 − ε2 − 2k̃τ ) kwη (η, ·)k2L2 (Rn ) dη
0  0
  Cτ 2 tk(u 2
2 0 , u1 )kH 2 (Rn )×H 2 (Rn ) if u1 6= 0,
6 Cτ ku2 k2L2 (Rn ) + k |x|u2k2L2 (Rn ) + Cτ 2 ln(e + t)ku k2 2 n
0 H (R ) if u1 = 0,
h i  i
where C, C̄ are positive constants independent of τ , and k̃ ∈ 2+ε2 2−ε2
2β
, 2τ carrying ε2 ∈ 0, 2β−2τ
β+τ
and ε2 < 2.
 24

Remark 4.4. In the case when u1 6= 0, by using Theorem 2.1 in [24], we still can provide the
estimate with ln(e+t) rather than t, where we need to assume the additional condition k(1+|x|)(u1 −
∆u0 )kL2 (Rn ) < ∞.
Remark 4.5. In the remaindering case for n = 1, 2, we may use the integral formula w(t, x) =
Rt
0 wη (η, x)dη with w(0, x) = 0. Then, by applying Minkowski’s integral inequality and the derived
inequality in Theorem 4.1, we have
Z Z t
2 Z t 2
 
kw(t, ·)k2L2(Rn ) = 


wη (η, x)dη  dx 6 kwη (η, ·)kL (R ) dη
2 n
R n 0 0
 R 2
1
Cτ 2 t
(ln(e + η)) 2 dη k(u , u 2
0 1 )kH 4 (Rn )×H 4 (Rn ) if u1 6= 0,
6 Cτ t2 ku2 − ∆u0 − β∆u1 k2L2 (Rn ) +  0
Cτ 2 t2 ku0k2H 4 (Rn ) if u1 = 0,

providing that we assume (u0 , u1, u2 ) ∈ H 4(Rn ) × H 4 (Rn ) × L2 (Rn ).

Proof. To begin with the proof, we introduce W = W (t, x) fulfilling
Z t
W (t, x) := w(η, x)dη.
0

Then, carrying out direct computations, we find that the new variable W (t, x) also fulfills a kind of
inhomogeneous linear MGT equation in the dissipative case. Precisely, it holds
Z t
τ Wttt + Wtt − ∆W − β∆Wt = τ wtt + wt − ∆w(η, x)dη − β∆w
0
Z t
= τ wtt + wt − (τ wηηη + wηη − β∆wη + τ (∆vη + β∆vηη ))(η, x)dη − β∆w,
0

where we applied (33). In other words, one has

τ Wttt + Wtt − ∆W − β∆Wt = τ u2 − τ ∆v − τ β∆vt , (34)

since w(0, x) = wt (0, x) = 0 and wtt (0, x) = u2 (x) − ∆u0 (x) − β∆u1 (x).
Let us set two auxiliary energies as follows:
Z
e [W ](t)
E := τ kWtt (t, ·)k2L2 (Rn ) + βk∇Wt (t, ·)k2L2 (Rn ) +2 ∇W (t, x) · ∇Wt (t, x)dx,
1
Z Rn
e [W ](t) := kW (t, ·)k2 2 n + k∇W (t, ·)k2 2 n + 2τ
E Wtt (t, x)Wt (t, x)dx.
2 t L (R ) L (R )
Rn

Clearly, from integration by parts and Cauchy’s inequality, we see

Z Z
− 2τ (∆v(t, x) + β∆vt (t, x))Wtt (t, x)dx − 2k̃τ (∆v(t, x) + β∆vt (t, x))Wt (t, x)dx
Rn Rn
2τ 2   
6 k∆v(t, ·)k2L2(Rn ) + k̃ 2 k∇v(t, ·)k2L2(Rn ) + β 2 k∆vt (t, ·)k2L2 (Rn ) + k̃ 2 k∇vt (t, ·)k2L2 (Rn )
ε2
 
+ ε2 kWtt (t, ·)k2L2 (Rn ) + k∇Wt (t, ·)k2L2 (Rn ) ,
 25

 i
where we set ε2 ∈ 0, 2β−2τ
β+τ
. Here, k̃ is a positive constant to be restricted later. We now apply
the similar procedure to those in the proof of Theorem 4.1, then from (34) we may obtain
2
1
e [W ](t) + k̃ E
e [W ](t) =
E1 2 β ∇Wt (t, ·) + ∇W (t, ·) + τ kWtt (t, ·) + k̃Wt (t, ·)k2L2(Rn )
β 2 n
L (R )
!
1
+ k̃(1 − τ k̃)kWt (t, ·)k2L2 (Rn ) + k̃ − k∇W (t, ·)k2L2(Rn ) ,
β
and
d e e [W ](t)

E1 [W ](t) + k̃ E2
dt
6 (2k̃τ + ε2 − 2)kWtt (t, ·)k2L2 (Rn ) + (2 + ε2 − 2k̃β)k∇Wt (t, ·)k2L2 (Rn )
2τ 2   
+ k∆v(t, ·)k2L2 (Rn ) + k̃ 2 k∇v(t, ·)k2L2(Rn ) + β 2 k∆vt (t, ·)k2L2 (Rn ) + k̃ 2 k∇vt (t, ·)k2L2 (Rn )
ε2
Z 
d
+ 2τ u2 (x)(Wt (t, x) + k̃W (t, x))dx .
dt Rn
Due to the estimates (30), we observe that
d e 
e [W ](t) + (2 − ε − 2k̃τ )kW (t, ·)k2 2 n + (2k̃β − ε − 2)k∇W (t, ·)k2 2 n
E1 [W ](t) + k̃ E2 2 tt L (R ) 2 t L (R )
dt 
Z  
d Cτ 2 k(u0 , u1)k2H 2 (Rn )×H 2 (Rn ) if u1 6= 0,
6 2τ u2 (x)(Wt (t, x) + k̃W (t, x))dx +
dt Rn Cτ 2 (1 + t)−1 ku k2 2 n if u1 = 0.
0 H (R )

e [W ](0) + k̃ E
According to Wtt (0, x) = wt (0, x) = 0, we get E e [W ](0) = 0. Integrating the previous
1 2
inequality over [0, t] yields
2
1

β ∇Wt (t, ·) + ∇W (t, ·) + τ kWtt (t, ·) + k̃Wt (t, ·)k2L2(Rn ) + k̃(1 − τ k̃)kWt (t, ·)k2L2 (Rn )
β L2 (Rn )
! Z
1 t
+ k̃ − k∇W (t, ·)k2L2(Rn ) + (2 − ε2 − 2k̃τ ) kWηη (η, ·)k2L2(Rn ) dη
β 0
Z t
+ (2k̃β − ε2 − 2) k∇Wη (η, ·)k2L2(Rn ) dη
0 
Z Cτ 2 tk(u 2
0 , u1 )kH 2 (Rn )×H 2 (Rn ) if u1 6= 0,
6 2τ u2 (x)(Wt (t, x) + k̃W (t, x))dx + (35)
Rn Cτ 2 ln(e + t)ku k2 2 n if u1 = 0.
0 H (R )

Let us now estimate the first term on the right-hand side of (35). For one thing, there exists a
positive constant ε3 such that
Z
τ2
2τ u2 (x)Wt (t, x)dx 6 ku2 k2L2 (Rn ) + ε3 kWt (t, ·)k2L2 (Rn ) .
Rn ε3
For another, making use of Hardy’s inequality for n > 3, we get
Z Z
k̃τ 2 2 |W (t, x)|2
2k̃τ u2 (x)W (t, x)dx 6 k |x|u2kL2 (Rn ) + ε3 k̃ dx
Rn ε3 Rn |x|2
k̃τ 2 n
6 k |x|u2k2L2 (Rn ) + ε3 k̃k∇W (t, ·)k2L2(Rn ) .
ε3 n−2
 26

All in all, we derive

2
1

β ∇Wt (t, ·) + ∇W (t, ·) + τ kWtt (t, ·) + k̃Wt (t, ·)k2L2(Rn )
β L2 (Rn )
!
2 1 n
+ (k̃ − τ k̃ − ε3 )kWt (t, ·)k2L2 (Rn )
+ k̃ − − ε3 k̃ k∇W (t, ·)k2L2(Rn )
β n−2
 Z t  Z t
ε2 2 ε2
+ 2 1 − − k̃τ kWηη (η, ·)kL2(Rn ) dη + 2 k̃β − −1 k∇Wη (η, ·)k2L2 (Rn ) dη
2 0
 2 0

Cτ 2   Cτ 2 tk(u , u )k2 2 n if u1 6= 0,

2 2 0 1 H (R )×H 2 (Rn )
6 ku2 kL2 (Rn ) + k |x|u2kL2 (Rn ) +
ε3 Cτ 2 ln(e + t)ku k2 2 n if u1 = 0.
0 H (R )

Eventually, we just need to discuss the nonnegativity of coefficients for the norms. In the above,
we need to restrict k̃ such that
" #
ε2 ε2 1 + ε2 /2 1 − ε2 /2
1− − k̃τ > 0 and k̃β − − 1 > 0, iff k̃ ∈ , .
2 2 β τ
Let us choose a small constant ε3 > 0 satisfying
 
2 n 1
k̃ − τ k̃ − ε3 > 0 and 1− ε3 k̃ − > 0.
n−2 β
Namely, we can determine small constant ε3 such that
" #  q 
1 + ε2 /2 1 − ε2 /2 (n − 2)/(n − 2 − nε3 ) 1/2 + 1/4 − τ ε3
k̃ ∈ , ⊂ , .
β τ β τ

So, the set of k̃ is not empty, providing that additional assumption ε2 < 2 and
( )
n−2 1 ε2 2ε2 − ε22
0 < ε3 < min , , , ,
n 4τ 2 + ε2 4τ
hold for n > 3. Indeed, the choice for these parameters can be independent of τ . Let us give an
example. Similarly to Remark 4.2, in the case of small τ such that 0 < τ 6 min{39β/41, 1}, we
may choose ε2 = 1/20, k̃ = 41/(40β) and ε3 = 1/1600. Providing that τ → 0+ , one may enlarge
the choices of k̃, ε2 , ε3 . Recalling the relation
Wt (t, x) = w(t, x) = uτ (t, x) − v(t, x),
we immediately conclude our result.

5 Global (in time) existence of small data Sobolev solutions

5.1 Philosophy of the proof
According to Section 2, we may represent the solution to the linear MGT equation in dissipative
case by the form
ulin(t, x) := K(t, x) ∗(x) u2 (x),
 27

where the partial Fourier transform of K(t, x) with respect to x was defined in (5). Furthermore,
some L2 estimates have been obtained. In Theorem 2.2, by choosing m = 1, we see

k |D|s ulin(t, ·)kL2 (Rn ) . g̃n,s (t)ku2 kL2 (Rn )∩L1 (Rn ) ,

where the time-dependent coefficient is given by

 1

 (ln(e + t)) 2 if n = 2, s = 0,




(1 + t) 1−5s
6 if n = 2, s ∈ (0, 1/2),
g̃n,s(t) := − 2s

 (1 + t) if n = 2, s ∈ [1/2, 2],




(1 + t) 12 − 2s − n4 if n > 3, s ∈ [0, 2].

Particularly, we denote gn (t) := g̃n,0(t). Moreover, from Theorem 2.1, one observes

k |D|s ulin(t, ·)kL2 (Rn ) . hs (t)ku2 kL2 (Rn ) ,

where the time-dependent coefficient is given by


(1 + t)1− 2s if s ∈ [0, 1),
hs (t) :=  1
−s
(1 + t) 2 2 if s ∈ [1, 2].
For T > 0, we introduce the operator N such that

N : u ∈ Xs (T ) → Nu(t, x) := ulin(t, x) + unon (t, x),

where Xs (T ) is an evolution space such that

Xs (T ) := C([0, T ], H s (Rn )), (36)

with some suitable positive constants s to be fixed later, and the integral operator is denoted by
Z t
non
u (t, x) := K(t − σ, x) ∗(x) |u(σ, x)|p dσ,
0

which is motivated by Duhamel’s principle.

In the forthcoming parts, we are going to demonstrate global (in time) existence of small data
Sobolev solutions to the semilinear MGT equation (3) by proving a fixed point of operator N which
means Nu ∈ Xs (T ). In other words, the next two crucial inequalities:

kNukXs (T ) . ku2 kL2 (Rn )∩L1 (Rn ) + kukpXs(T ) , (37)

 
kNu − NvkXs (T ) . ku − vkXs (T ) kukp−1 p−1
Xs (T ) + kvkXs (T ) , (38)

will be proved. Throughout this section, u and v are two solutions to the semilinear MGT equation
(3). Precisely, if we assume ku2 kL2 (Rn )∩L1 (Rn ) = ǫ to be a sufficiently small constant, then we together
(37) with (38) to conclude that there exists a uniquely determined local (in time) large data and
global (in time) small data solution u∗ = u∗(t, x) belonging to the Sobolev space Xs (T ) by using
Banach’s fixed point theorem.
To end this subsection, we recall the fractional Gagliardo-Nirenberg inequality, whose proof can
be found in [22].
 28

Lemma 5.1. Let p, p0 , p1 ∈ (1, ∞) and κ ∈ [0, s) with s ∈ (0, ∞). Then, it holds for all f ∈
Lp0 (Rn ) ∩ Ḣps1 (Rn )
kf kḢpκ(Rn ) . kf kL1−γ γ
p0 (Rn ) kf k s
Ḣ (Rn )
,
p1

   h i
1
where γ = p0
− p1 + κ
n
1
p0
− 1
p1
+ s
n
∈ κ
s
,1 .

5.2 Lower regular Sobolev solution

It is well-known that the study of lower regular Sobolev solution is more challenging than the study
of higher regular one since the Sobolev embedding theory does not work well. In this part, we will
study global (in time) existence of small data Sobolev solutions with low regularity in the evolution
space Xs (T ), in which we will focus on the cases s ∈ [1/2, 2] if n = 2, and s ∈ (0, 2] if n > 3.
Theorem 5.1. Let τ ∈ (0, β). Let us consider s ∈ [1/2, 2] if n = 2, and s ∈ (0, 2] if n > 3. We
suppose that p > 2, and p 6 2n/(n − s) if s < n 6 3s, p 6 n/(n − 2s) if 3s < n 6 4s. If

>
 5 if n = 2, s ∈ [1/2, 1),





 > s+3 if n = 2, s ∈ [1, 2],




> (n + 3)/(n − 1) if 3 6 n 6 6, s ∈ (0, 1),
p (39)

 > (n + 2)/(n − 1) if 3 6 n 6 6, s ∈ [1, 2],





 > max{3n/2 − 1, n + 3}/(n − 1) if n > 7, s ∈ (0, 1),




> (3n/2 − 1)/(n − 1) if n > 7, s ∈ [1, 2],

there exists a sufficiently small constant ǫ > 0 such that for u2 ∈ L2 (Rn ) ∩ L1 (Rn ) satisfying the
assumption ku2kL2 (Rn )∩L1 (Rn ) 6 ǫ, there is a uniquely determined global (in time) Sobolev solution

u ∈ C([0, ∞), H s(Rn ))

to the semilinear MGT equation (3). Furthermore, the solution fulfills the following estimates:

ku(t, ·)kL2(Rn ) . gn (t)ku2 kL2 (Rn )∩L1 (Rn ) ,

k |D|su(t, ·)kL2(Rn ) . g̃n,s (t)ku2kL2 (Rn )∩L1 (Rn ) .

Example 5.1. Let us consider s = 2. Then, the observation of Theorem 5.1 with s = 2 shows the
global (in time) small data Sobolev solution (in the classical energy sense)

u ∈ C([0, ∞), H 2(Rn ))

to the semilinear MGT equation (3) with τ ∈ (0, β) uniquely exists providing that
• when n = 2, we assume p > 5;
• when n = 3, 4, we assume (n + 2)/(n − 1) < p 6 2n/(n − 2);
• when n = 5, 6, we assume 2 6 p 6 2n/(n − 2);
• when n = 7, 8, we assume 2 6 p 6 n/(n − 4).
 29

Remark 5.1. Comparing the result of the linearized problem in Theorem 2.2 with m = 1, the
estimates stated in Theorem 5.1 are no loss of decay with respect to the corresponding linear Cauchy
problem.
Remark 5.2. Indeed, one may also follow the proof of Theorem 5.1 to prove global (in time)
existence results for other regularity assumptions on initial data. By considering u2 ∈ L2 (Rn ) ∩
Lm (Rn ) for m ∈ (1, 2), one just need to use Theorem 2.2, and the lower bound of the exponent
p > 2 will be replaced by p > 2/m due to the application of the fractional Gagliardo-Nirenberg
inequality.
Remark 5.3. From the restriction n 6 4s in Theorem 5.1, we should control the dimension sat-
isfying n 6 8 due to s ∈ (0, 2]. For the global (in time) existence result in high-dimensional space
n > 9 with additional L1 data, one may study higher regular Sobolev solution, i.e.
u ∈ C([0, ∞), H s(Rn )) with s ∈ (2, ∞).
We should emphasize that due to s ∈ (2, ∞) in Theorem 2.2, it is necessary to estimate
k |u(σ, ·)|pkḢ s−2 (Rn ) and k |u(σ, ·)|p − |v(σ, ·)|pkḢ s−2 (Rn ) .
To estimate the first norm, one may apply the fractional chain rule with the additional restriction
p > ⌈s − 2⌉. While in the estimate of the second norm, the main tools are the fractional Leibniz rule
and the fractional chain rule (see [21] and [46]) carrying a stronger condition p > 1 + ⌈s − 2⌉ > 2.
Furthermore, if s − 2 > n/2 and p > s − 1, one may apply the fractional powers rule to estimate
the last mentioned norms to prove existence of large regular (s > n/2 + 2) Sobolev solutions.
Proof. To begin with the proof, we construct the time-weighted norm for the evolution space Xs (T )
with s ∈ (0, 2] for T > 0 by
 
kukXs (T ) := sup (gn (t))−1 ku(t, ·)kL2(Rn ) + (g̃n,s (t))−1 ku(t, ·)kḢ s(Rn ) .
t∈[0,T ]

From Theorem 2.2 with m = 1, we easily get

kulinkXs (T ) . ku2kL2 (Rn )∩L1 (Rn ) .
Thus, we may claim that ulin ∈ Xs (T ) for any s ∈ (0, 2]. In the view of the desired inequality (37),
we just have to justify the next one:
kunon kXs (T ) . kukpXs(T ) .
Initially, we apply the derived (L2 ∩ L1 ) − L2 estimate stated in Theorem 2.2 in the interval [0, t]
to get
Z t
kunon (t, ·)kL2 (Rn ) . gn (t − σ)k |u(σ, ·)|pkL2 (Rn )∩L1 (Rn ) dσ.
0

To estimate the power nonlinear term in the norm, we employ the fractional Gagliardo-Nirenberg
inequality that
k |u(σ, ·)|pkL1 (Rn ) = ku(σ, ·)kpLp (Rn ) . (gn (σ))(1−γ1 )p (g̃n,s (σ))γ1 p kukpXs(σ) ,
k |u(σ, ·)|pkL2 (Rn ) = ku(σ, ·)kpL2p (Rn ) . (gn (σ))(1−γ2 )p (g̃n,s (σ))γ2 p kukpXs(σ) ,
 30

   
where the parameters are γ1 := ns 1
2
− 1
p
∈ [0, 1] and γ2 := n
s
1
2
− 1
2p
∈ [0, 1].
The previous restrictions lead to


 < ∞ if 1 < n 6 s,


2 6 p 6 2n/(n − s) if s < n 6 3s, (40)



6 n/(n − 2s) if 3s < n 6 4s.

Here, the restriction for n 6 4s comes from the nonempty set of p ∈ [2, n/(n − 2s)].
Obviously, we know that
 s 1
g̃n,s (σ) (1 + σ)− 2 (ln(e + σ))− 2 if n = 2, s ∈ [1/2, 2],
1> =
gn (σ) (1 + σ)− 2s if n > 3, s ∈ (0, 2],

which implies from γ1 < γ2 ,

k |u(σ, ·)|pkL2 (Rn )∩L1 (Rn ) . (gn (σ))(1−γ1 )p (g̃n,s(σ))γ1 p kukpXs(σ) .

For one thing, by some direct computations, they yield


(1 + σ)− p2 +1 (ln(e + σ)) (s−1)p+2
2s if n = 2, s ∈ [1/2, 2],
(gn (σ))(1−γ1 )p (g̃n,s (σ))γ1 p = (n−1)p
+n
(1 + σ) − 2 2 if n > 3, s ∈ (0, 2],

and

(1 + σ)− p2 + 21 (ln(e + σ)) (s−1)p+1
2s if n = 2, s ∈ [1/2, 2],
(gn (σ))(1−γ2 )p (g̃n,s (σ))γ2 p =
(1 + σ)− (n−1)p
2
+n
4 if n > 3, s ∈ (0, 2].

According to the assumption



>max{s + 3, 4} if n = 2, s ∈ [1/2, 2],


p > (n + 2)/(n − 1) if 3 6 n 6 6, s ∈ (0, 2], (41)



> (3n − 2)/(2n − 2) if n > 7, s ∈ (0, 2],

it is true that
Z t/2
(gn (σ))(1−γ1 )p (g̃n,s (σ))γ1 p dσ . 1,
0

and
 p 3 (s−1)p+s+2

 (1 + t)− 2 + 2 (ln(e + t)) 2s if n = 2,


Z 
 (n−1)p 3 n
t  (1 + t)− 2
+ + 2 4 if 3 6 n 6 5,
(gn (t))(1−γ1 )p (g̃n,s(t))γ1 p gn (t − σ)dσ . − 5p
t/2 
 (1 + t) 2
+3
ln(e + t) if n = 6,




(1 + t)− (n−1)p
2
+n
2 if n > 7,
. gn (t).
 31

Then, by dividing [0, t] into [0, t/2] and [t/2, t], one may immediately arrive at
Z t/2
ku non
(t, ·)kL2 (Rn ) . gn (t)kukpXs (T ) (gn (σ))(1−γ1 )p (g̃n,s(σ))γ1 p dσ
0
Z t
+ (gn (t)) (1−γ1 )p
(g̃n,s (t)) γ1 p
kukpXs (T ) gn (t − σ)dσ
t/2

. gn (t)kukpXs (T ) ,

where we used kukXs(σ) . kukXs(T ) for any σ ∈ [0, T ] and taking account into the fact that

(1 + t − σ) ≈ (1 + t) for σ ∈ [0, t/2] and (1 + σ) ≈ (1 + t) for σ ∈ [t/2, t].

Next, we will estimate the solution in the Ḣ s norm. At this time, we employ the obtained
(L2 ∩ L1 ) − L2 estimate in [0, t/2], and L2 − L2 estimate in [t/2, t] leading to
Z t/2 Z t
non p
ku (t, ·)kḢ s (Rn ) . g̃n,s (t − σ)k |u(σ, ·)| kL2 (Rn )∩L1 (Rn ) dσ + hs (t − σ)k |u(σ, ·)|pkL2 (Rn ) dσ
0 t/2
Z t/2
. g̃n,s (t)kukpXs(T ) (gn (σ))(1−γ1 )p (g̃n,s (σ))γ1 p dσ
0
+ (gn (t))(1−γ2 )p (g̃n,s (t))γ2 p (1 + t)hs (t)kukpXs(T )
. g̃n,s (t)kukpXs(T ) ,

where we used our assumption (41) and additionally,



 >
5 if n = 2, s ∈ [1/2, 1)




>4 if n = 2, s ∈ [1, 2],
p (42)

 >(n + 3)/(n − 1) if n > 3, s ∈ (0, 1),




> (n + 2)/(n − 1) if n > 3, s ∈ [1, 2],

to derive

1 & (gn (t))(1−γ2 )p (g̃n,s (t))γ2 p (1 + t)hs (t)(g̃n,s (t))−1

 (s−1)p+1
 − p2 + 52



(1 + t) (ln(e + t)) 2s if n = 2, s ∈ [1/2, 1),

 p (s−1)p+1
(1 + t)− 2 +2 (ln(e + t)) 2s if n = 2, s ∈ [1, 2],
= (n−1)p
− 2 +n + 32

 (1 + t) 2 if n > 3, s ∈ (0, 1),



 (n−1)p
 − + n +1
(1 + t) 2 2 if n > 3, s ∈ [1, 2].

By assuming (40), (41), (42) and summarizing the derived estimates, it is proved that the operator
N maps Xs (T ) into itself, namely, Nu ∈ Xs (T ).
Finally, with the aim of proving the Lipschitz condition, we may take two solutions u, v ∈ Xs (T ).
From the derived result of (37), it is clear that Nu, Nv ∈ Xs (T ). Therefore, we have
Z t

p p
kNu − NvkXs (T ) = K(t − σ, x) ∗(x) (|u(σ, x)| − |v(σ, x)| )dσ .

0 Xs (T )
 32

We assume that (40) and (39) hold. For the estimate in the L2 norm, we apply Hölder’s inequality
and the fractional Gagliardo-Nirenberg inequality to arrive at
k(Nu − Nv)(t, ·)kL2 (Rn )
Z t
. gn (t − σ)k |u(σ, ·)|p − |v(σ, ·)|pkL2 (Rn )∩L1 (Rn ) dσ
0
Z t  
. gn (t − σ)ku(σ, ·) − v(σ, ·)kLp (Rn ) ku(σ, ·)kp−1 p−1
Lp (Rn ) + kv(σ, ·)kLp (Rn ) dσ
0
Z t  
+ gn (t − σ)ku(σ, ·) − v(σ, ·)kL2p (Rn ) ku(σ, ·)kp−1 p−1
L2p (Rn ) + kv(σ, ·)kL2p (Rn ) dσ
0
Z t  
. gn (t − σ)(gn (σ))(1−γ1 )p (g̃n,s (σ))γ1 p dσ ku − vkXs (T ) kukp−1 p−1
Xs (T ) + kvkXs (T )
0
 
. gn (t)ku − vkXs (T ) kukp−1 p−1
Xs (T ) + kvkXs (T ) .

By repeating the same approach as before, we conclude

 
k(Nu − Nv)(t, ·)kḢ s (Rn ) . g̃n,s (t)ku − vkXs (T ) kukp−1 p−1
Xs (T ) + kvkXs (T ) .

Therefore, the crucial estimates (37) and (38) are valid. By using Banach’s fixed point theorem,
there exists a unique determined global (in time) low regular Sobolev solution to the semilinear
MGT equation (3). The proof is complete.

6 Nonexistence of global (in time) weak solutions

Before showing our main result on blow-up of solutions, let us first give a definition of weak solutions
to the semilinear MGT equation (3).
Definition 6.1. Let p > 1. We say u ∈ Lploc ([0, ∞) × Rn ) is a global (in time) weak solution to the
semilinear MGT equation (3) if the integral equality
Z ∞ Z
u(t, x)(−τ ψttt (t, x) + ψtt (t, x) − ∆ψ(t, x) + β∆ψt (t, x))dxdt
0 n
ZR Z ∞ Z
=τ u2 (x)ψ(0, x)dx + |u(t, x)|p ψ(t, x)dxdt (43)
Rn 0 Rn

holds for any ψ ∈ C0∞ ([0, ∞) × Rn ).

Theorem 6.1. Let τ ∈ (0, β). Let us assume that u2 ∈ L1 (Rn ) and
Z
u2 (x)dx > 0.
Rn

Then, there exist no any the global (in time) weak solutions to the semilinear MGT equation (3) in
the sense of Definition 6.1 providing that the exponent of nonlinearity satisfies

< ∞ if n = 1,
1<p
6 (n + 1)/(n − 1) if n > 2.
Remark 6.1. In the one-dimensional case, every weak solution according to Definition 6.1 blows
up for any 1 < p < ∞, which means that the result in 1D is optimal.
 33

Remark 6.2. We may derive blow-up results for other regularity assumptions on initial data. Let
us assume u2 ∈ Lm (Rn ) with m ∈ (1, 2) and
n
u2 (x) & |x|− m (ln(1 + |x|))−1 for |x| ≫ 1.
Then, one may also prove blow-up of weak solutions to the semilinear MGT equation (3) providing
that 1 < p < ∞ if n = 1, and 1 < p < (n + m)/(n − m) if n > 2. The proof is strictly following
those of Theorem 4.1 in [10].
Proof. Let us now introduce two bump functions η ∈ C0∞ ([0, ∞)) and φ ∈ C0∞ (Rn ) such that
η = η(t) is decreasing with η = 1 on [0, 1/2] and supp η ⊂ [0, 1]; φ = φ(x) is radial symmetric,
decreasing with respect to |x| with φ = 1 on B1/2 and supp φ ⊂ B1 . Moreover, we assume
p′  ′ ′ ′

(η(t))− p |η ′′′(t)|p + |η ′′ (t)|p + |η ′ (t)|p 6 C, (44)
p′ ′
(φ(x))− p |∆φ(x)|p 6 C, (45)
where p′ is the conjugate of p, i.e. 1/p + 1/p′ = 1, and C is a positive constant, with η, φ ∈ [0, 1].
To begin with, we define a test function
ψR (t, x) := ηR (t)φR (x) := η(t/R)φ(x/R),
where R ∈ [1, ∞) is a large parameter. Furthermore, we may introduce
Z ∞ Z
IR := |u(t, x)|p ψR (t, x)dxdt.
0 Rn

By considering (43) in the definition of weak solution with the test function ψ(t, x) = ψR (t, x), one
immediately has
Z
IR + τ u2 (x)φR (x)dx
Rn
Z ∞ Z  
= u(t, x) −τ ∂t3 ψR (t, x) + ∂t2 ψR (t, x) − ∆ψR (t, x) + β∂t ∆ψR (t, x) dxdt
0 Rn
Z p′ 
Z 
1 1 ∞ ′ ′ ′
6 IR + ′ (ηR (t)φR (x))− p τ p |d3t ηR (t)φR (x)|p + |d2t ηR (t)φR (x)|p dxdt
p p 0 Rn
Z ∞Z p′  
1 ′ ′ ′
+ ′ (ηR (t)φR (x))− p |ηR (t)∆φR (x)|p + β p |dt ηR (t)∆φR (x)|p dxdt,
p 0 Rn
′
where we employed Young’s inequality ab 6 ap /p + bp /p′ .
Due to the fact that
∆φR (x) = R−2 ∆φ(x/R), and dkt ηR (t) = R−k dkt η(t/R) for k = 1, 2, 3,
we are able to deduce
Z
1
IR . ′ IR + τ u2 (x)φR (x)dx
p Rn
Z ∞ Z  
′ p′ ′ p′ ′
. R−2p (η(t/R))− p + |η ′′(t/R)|p φ(x/R) + η(t/R)(φ(x/R))− p |∆φ(x/R)|p dxdt
0 Rn
Z ∞Z  
′ p′ ′ ′ p′ ′
+ R−3p (η(t/R))− p |η ′′′ (t/R)|p φ(x/R) + |η ′ (t/R)|p (φ(x/R))− p |∆φ(x/R)|p dxdt
0 Rn
−2p′ +1+n −3p′ +1+n ′
.R +R . R−2p +1+n ,
 34

where the conditions for test functions in (44) and (45) were used. Moreover, we applied our
assumption on initial data such that
Z Z
τ u2 (x)dx > 0 ⇒ τ u2 (x)φR (x)dx > 0
Rn Rn

for any R ≫ 1 because of the fact that

Z Z
lim u2 (x)φR (x)dx = u2 (x)dx.
R→∞ Rn Rn

According to the condition on p leading to −2p′ + 1 + n < 0, and letting R → ∞, we get

limR→∞ IR = 0, which leads to u = 0 a.e., however, this contradicts to our assumption. In other
words, the global (in time) weak solution does not exist.
To prove the blow-up result in the limit case when p = (n + 1)/(n − 1) if n > 2, we can also
conclude the contradiction that limR→∞ IR = 0 by following the approach in [55], i.e. the monotone
convergence theorem and the dominant convergence theorem. All in all, the proof is completed.

Acknowledgments
The second author was supported in part by Grant-in-Aid for scientific Research (C) 20K03682 of
JSPS. The authors thank Michael Reissig (TU Bergakademie Freiberg) for the suggestions in the
preparation of the paper.

References
[1] M.O. Alves, A.H. Caixeta, M.A.J. Silva, J.H. Rodrigues, Moore-Gibson-Thompson equation
with memory in a history framework: a semigroup approach, Z. Angew. Math. Phys. 69 (2018)
106.
[2] J. Barrera, H. Volkmer, Asymptotic expansion of the L2 -norm of a solution of the strongly
damped wave equation in space dimension 1 and 2, Asymptot. Anal. (in press) (2020).
[3] J. Barrera, H. Volkmer, Asymptotic expansion of the L2 -norm of a solution of the strongly
damped wave equation, J. Differential Equations 267 (2019) 902–937.
[4] F. Bucci, M. Eller, The Cauchy-Dirichlet problem for the Moore-Gibson-Thompson equation,
Preprint (2020). arxiv.org/abs/2004.11167
[5] F. Bucci, I. Lasiecka, Feedback control of the acoustic pressure in ultrasonic wave propagation,
Optimization 68 (2019) 1811–1854.
[6] F. Bucci, L. Pandolfi, On the regularity of solutions to the Moore-Gibson-Thompson equation:
a perspective via wave equations with memory, J. Evol. Equ. (in press) (2019).
[7] A.H. Caixeta, I. Lasiecka, N.V.D. Cavalcanti, On long time behavior of Moore-Gibson-
Thompson equation with molecular relaxation, Evol. Equ. Control Theory 5 (2016) 661–676.
[8] R.C. Charão, R. Ikehata, A note on decay rates of the local energy for wave equations with
Lipschitz wavespeeds, J. Math. Anal. Appl. 483 (2020) 123636.
 35

[9] W. Chen, Dissipative structure and diffusion phenomena for doubly dissipative elastic waves
in two space dimensions, J. Math. Anal. Appl. 486 (2020) 123922.
[10] W. Chen, T.A. Dao, On the Cauchy problem for semilinear regularity-loss-type σ-evolution
models with memory term, Preprint (2020). arxiv.org/abs/2003.10137
[11] W. Chen, A. Palmieri, Nonexistence of global solutions for the semilinear Moore – Gibson –
Thompson equation in the conservative case, Discrete Contin. Dyn. Syst. 40 (2020) 5513–5540.
[12] W. Chen, A. Palmieri, A blow – up result for the semilinear Moore – Gibson – Thompson
equation with nonlinearity of derivative type in the conservative case, Evol. Equ. Control
Theory (in press) (2021).
[13] R. Chill, A. Haraux, An optimal estimate for the time singular limit of an abstract wave
equation, Funkcial. Ekvac. 47 (2004) 277–290.
[14] M. D’Abbicco, M. Reissig, Semilinear structural damped waves, Math. Methods Appl. Sci. 37
(2014) 1570–1592.
[15] F. Dell’Oro, I. Lasiecka, V. Pata, A note on the Moore-Gibson-Thompson equation with
memory of type II, J. Evol. Equ. (in press) (2019).
[16] F. Dell’Oro, I. Lasiecka, V. Pata, The Moore-Gibson-Thompson equation with memory in the
critical case, J. Differential Equations 261 (2016) 4188–4222.
[17] F. Dell’Oro, V. Pata, On the Moore-Gibson-Thompson equation and its relation to linear
viscoelasticity, Appl. Math. Optim. 76 (2017) 641–655.
[18] M.R. Ebert, M. Reissig, Methods for Partial Differential Equations, Birkhäuser Basel, Germany,
2018.
[19] M. Ghisi, M. Gobbino, Hyperbolic-parabolic singular perturbation for nondegenerate Kirchhoff
equations with critical weak dissipation, Math. Ann. 354 (2012) 1079–1102.
[20] G.C. Gorain, Stabilization for the vibrations modeled by the ‘standard linear model’ of
viscoelasticity, Proc. Indian Acad. Sci. Math. Sci. 120 (2010) 495–506.
[21] L. Grafakos, S. Oh, The Kato-Ponce inequality, Comm. Partial Differential Equations 39
(2014) 1128-1157.
[22] H. Hajaiej, L. Molinet, T. Ozawa, B. Wang, Necessary and sufficient conditions for the
fractional Gagliardo-Nirenberg inequalities and applications to Navier-Stokes and generalized
boson equations, Harmonic Analysis and Nonlinear Partial Differential Equations 159-175,
RIMS Kôkyûroku Bessatsu, B26, Res. Inst. Math. Sci. (RIMS), Kyoto, 2011.
[23] H. Hashimoto, T. Yamazaki, Hyperbolic-parabolic singular perturbation for quasilinear equa-
tions of Kirchhoff type, J. Differential Equations 237 (2007) 491–525.
[24] R. Ikehata, Decay estimates of solutions for the wave equations with strong damping terms in
unbounded domains, Math. Methods Appl. Sci. 24 (2001) 659–670.
 36

[25] R. Ikehata, L2 -convergence results for linear dissipative wave equations in unbounded domains,
Asymptot. Anal. 36 (2003) 63–74.
[26] R. Ikehata, New decay estimates for linear damped wave equations and its application to
nonlinear problem, Math. Methods Appl. Sci. 27 (2004) 865–889.
[27] R. Ikehata, Asymptotic profiles for wave equations with strong damping, J. Differential
Equations 257 (2014) 2159–2177.
[28] R. Ikehata, M. Natsume, Energy decay estimates for wave equations with a fractional damping,
Differential Integral Equations 25 (2012) 939–956.
[29] R. Ikehata, K. Nishihara, Diffusion phenomenon for second order linear evolution equations,
Studia Mathematica 158 (2003) 153–161.
[30] R. Ikehata, M. Onodera, Remarks on large time behavior of the L2 -norm of solutions to
strongly damped wave equations, Differential Integral Equations 30 (2017) 505–520.
[31] R. Ikehata, M. Sobajima, Singular limit problem of abstract second order evolution equations,
Preprint (2019). arxiv.org/abs/1912.10181
[32] R. Ikehata, G. Todorova, B. Yordanov, Wave equations with strong damping in Hilbert spaces,
J. Differential Equations 254 (2013) 3352–3368.
[33] P.M. Jordan, Second-sound phenomena in inviscid, thermally relaxing gases, Discrete Contin.
Dyn. Syst. Ser. B 19 (2014) 2189–2205.
[34] V. Kalantarov, A. Tiryaki, On the stability results for third order differential-operator equa-
tions, Turk. J. Math. 21 (1997) 179–186.
[35] B. Kaltenbacher, I. Lasiecka, Exponential decay for low and higher energies in the third order
linear Moore-Gibson-Thompson equation with variable viscosity, Palest. J. Math. 1 (2012)
1–10.
[36] B. Kaltenbacher, I. Lasiecka, R. Marchand, Wellposedness and exponential decay rates for the
Moore-Gibson-Thompson equation arising in high intensity ultrasound, Control Cybernet. 40
(2011) 971–988.
[37] J. Kisyński, Sur les équations hyerboliques avec petit paramètre, Colloq. Math. 10 (1963)
331–343.
[38] I. Lasiecka, Global solvability of Moore-Gibson-Thompson equation with memory arising in
nonlinear acoustics, J. Evol. Equ. 17 (2017) 411–441.
[39] I. Lasiecka, X. Wang, Moore-Gibson-Thompson equation with memory, part I: exponential
decay of energy, Z. Angew. Math. Phys. 67 (2016) 17.
[40] I. Lasiecka, X. Wang, Moore-Gibson-Thompson equation with memory, part II: General decay
of energy, J. Differential Equations 259 (2015) 7610–7635.
 37

[41] R. Marchand, T. McDevitt, R. Triggiani, An abstract semigroup approach to the third-

order Moore-Gibson-Thompson partial differential equation arising in high-intensity ultrasound:
structural decomposition, spectral analysis, exponential stability, Math. Methods Appl. Sci. 35
(2012) 1896–1929.
[42] F.K. Moore, W.E. Gibson, Propagation of weak disturbances in a gas subject to relaxation
effect, J. Aero/Space Sci. 27 (1960) 117–127.
[43] V. Nikolić, B. Said-Houari, Mathematical analysis of memory effects and thermal relaxation
in nonlinear sound waves on unbounded domains, Preprint (2020). arxiv.org/abs/2003.11840
[44] V. Nikolić, B. Said-Houari, On the Jordan-Moore-Gibson-Thompson wave equation in heredi-
tary fluids with quadratic gradient nonlinearity, Preprint (2020). arxiv.org/abs/2005.07245
[45] K. Nishihara, Lp -Lq estimates of solutions to the damped wave equation in 3-dimensional
space and their application, Math. Z. 244 (2003) 631–649.
[46] A. Palmieri, M. Reissig, Semi-linear wave models with power non-linearity and scale-invariant
time-dependent mass and dissipation, II, Math. Nachr. 291 (2018) 1859–1892.
[47] M. Pellicer, B. Said-Houari, On the Cauchy problem for the standard linear solid model with
heat conduction: Fourier versus Cattaneo, Preprint (2019). arxiv.org/abs/1903.10181
[48] M. Pellicer, B. Said-Houari, Wellposedness and decay rates for the Cauchy problem of the
Moore-Gibson-Thompson equation arising in high intensity ultrasound, Appl. Math. Optim.
80 (2019) 447–478.
[49] M. Pellicer, J. Solá-Morales, Optimal scalar products in the Moore-Gibson-Thompson equation,
Evol. Equ. Control Theory 8 (2019) 203–220.
[50] R. Quintanilla, Moore-Gibson-Thompson thermoelasticity, Math. Mech. Solids 24 (2019)
4020–4031.
[51] R. Racke, B. Said-Houari, Global well-posedness of the Cauchy problem for the 3-D Jordan-
Moore-Gibson-Thompson equation, Preprint (2019).
[52] Y. Shibata, On the rate of decay of solutions to linear viscoelastic equation, Math. Methods
Appl. Sci. 23 (2000) 203–226.
[53] M.E. Taylor, Pseudodifferential Operators, Princeton University Press, 1981.
[54] P.A. Thompson, Compressible fluid dynamics, McGraw-Hill, New York, 1972.
[55] Q.S. Zhang, A blow-up result for a nonlinear wave equation with damping: The critical case,
C. R. Acad. Sci. Paris Sér. I Math. 333 (2001) 109–114.

View publication stats