You are on page 1of 26

Nonlinear Analysis 72 (2010) 2658–2683

Contents lists available at ScienceDirect

Nonlinear Analysis
journal homepage: www.elsevier.com/locate/na

Global existence and blow up of solutions to systems of nonlinear wave


equations with degenerate damping and source terms
Mohammad A. Rammaha a,∗ , Sawanya Sakuntasathien b
a
Department of Mathematics, University of Nebraska-Lincoln, Lincoln, NE 68588-0130, USA
b
Mathematics Department, Faculty of Science, Silpakorn University, Nakhonpathom, Thailand

article info abstract


Article history: We focus on the global well-posedness of the system of nonlinear wave equations
Received 12 August 2009
Accepted 3 November 2009 utt − ∆u + (d|u|k + e|v|l )|ut |m−1 ut = f1 (u, v)
vtt − ∆v + (d0 |v|θ + e0 |u|ρ )|vt |r −1 vt = f2 (u, v),
MSC:
primary 35L05 in a bounded domain Ω ⊂ Rn , n = 1, 2, 3, with Dirichlét boundary conditions. The
35L20 nonlinearities f1 (u, v) and f2 (u, v) act as a strong source in the system. Under some
secondary 58G16 restriction on the parameters in the system we obtain several results on the existence of
local solutions, global solutions, and uniqueness. In addition, we prove that weak solutions
Keywords:
Wave equations to the system blow up in finite time whenever the initial energy is negative and the
Damping and source terms exponent of the source term is more dominant than the exponents of both damping terms.
Weak solutions © 2009 Elsevier Ltd. All rights reserved.
Blow up of solutions
Energy identity

1. Introduction

1.1. The model

Many questions in physics and engineering give rise to problems that deal with coupled evolution equations. For instance,
in scattering theory and certain mechanical applications, such evolution equations come in the form of a system of nonlinear
wave equations. An important example of such systems goes back to Reed [1] in 1976 who proposed a system in three space
dimensions which is similar to system (1.1), but without the presence of any damping.
In this article, we study a system of nonlinear wave equations which features two competing forces. One force is a
degenerate damping term and the other is a strong source. In particular, we analyze the influence of these forces on the
long-time behavior of solutions.
Let F : R2 −→ R be the C 1 -function given by
p+1
F (u, v) = a |u + v|p+1 + 2b |uv| 2 ,
where p ≥ 3, a > 1 and b > 0. Throughout the paper, Ω is a bounded open connected set in Rn , n = 1, 2, 3, with a smooth
boundary Γ = ∂ Ω . We study the global well-posedness of the following initial boundary value problem:
utt − ∆u + (d|u|k + e|v|l )|ut |m−1 ut = f1 (u, v), in Ω × (0, T ) ≡ QT ,

∗ Corresponding author. Tel.: +1 402 472 7258; fax: +1 402 472 8466.
E-mail addresses: rammaha@math.unl.edu, mrammaha1@math.unl.edu (M.A. Rammaha), ssawanya@hotmail.com (S. Sakuntasathien).

0362-546X/$ – see front matter © 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.na.2009.11.013
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2659

vtt − ∆v + (d0 |v|θ + e0 |u|ρ )|vt |r −1 vt = f2 (u, v), in Ω × (0, T ) ≡ QT ,


u(x, 0) = u (x) ∈
0
H01 (Ω ), ut (x, 0) = u (x) ∈ L2 (Ω ),
1
(1.1)
v(x, 0) = v (x) ∈ (Ω ),
0
H01 vt (x, 0) = v (x) ∈ L2 (Ω ), 1

u = v = 0, on Γ × (0, T ),
where f1 (u, v) = ∂∂ Fu (u, v), f2 (u, v) = ∂∂vF (u, v) for (u, v) ∈ R2 , and d, e, d0 , e0 are nonnegative constants.
We note here that the nonlinearities f1 (u, v) and f2 (u, v) act as strong source terms in the system (1.1). In addition, the
functions f1 , f2 and F enjoy certain properties. First, it is easy to see that F (u, v) ≤ c1 (|u|p+1 + |v|p+1 ), for all (u, v) ∈ R2 ,
where c1 = 2p a + b. Moreover, a quick computation will show that for a fixed a, p > 1, there exists a constant c0 > 0
such that F (u, v) ≥ c0 (|u|p+1 + |v|p+1 ), for all (u, v) ∈ R2 , provided b is chosen large enough. Also, it is easy to see that
uf1 (u, v) + v f2 (u, v) = (p + 1)F (u, v) for all (u, v) ∈ R2 . Henceforth, the following conditions are assumed throughout the
paper.

Assumption 1.1. • 0 < m, r < 1; p ≥ 3 if n = 1, 2; p = 3 if n = 3.


• k, l, θ , ρ ≥ 1, and if n = 3 we require:
max{k, l} ≤ 3(1 − m) and max{θ , ρ} ≤ 3(1 − r ).
• u0 , v0 ∈ (Ω ), u1 , v1 ∈ L (Ω ), and d, d0 , e, e0 ≥ 0.
H01 2

• There exist constants c0 , c1 > 0 such that


c0 (|u|p+1 + |v|p+1 ) ≤ F (u, v) ≤ c1 (|u|p+1 + |v|p+1 ) for all (u, v) ∈ R2 . (1.2)
• In addition,
h p−3 p+1
i
f1 (u, v) = (p + 1) a|u + v|p−1 (u + v) + b|u| 2 |v| 2 u ,
h p−3 p+1
i
f2 (u, v) = (p + 1) a|u + v|p−1 (u + v) + b|v| 2 |u| 2 v , (1.3)

uf1 (u, v) + v f2 (u, v) = (p + 1)F (u, v) for all (u, v) ∈ R . 2

As in the case of a single wave equation [2–5], it is worth noting here that when the damping terms (d|u|k + e|v|l )|ut |m−1 ut
and (d0 |v|θ + e0 |u|ρ )|vt |r −1 vt are absent, then the strong source terms f1 (u, v) and f2 (u, v) should drive the solution of (1.1) to
blow up in finite time. In such a case, by appealing to a variety of methods (going back to the work of Glassey [2], Levine [3],
and others) one can show that most solutions to the system blow up in finite time. In addition, if the source terms f1 (u, v)
and f2 (u, v) are removed from the equations, then damping terms of various forms should yield the existence of global
solutions, (cf. [6–9]). However, when both damping and source terms are present, then the analysis of their interaction and
their influence on the global behavior of solutions becomes more difficult (see for instance [10–14,22,23] and the references
therein). In fact, one of the main goals of this paper is to determine whether or not such a degenerate damping can stabilize
the system and at the same time maintain Hadamard well-posedness of weak solutions.
It should be noted that the case of a non-degenerate damping, i.e., when k = l = θ = ρ = 0 and m, r ≥ 1, was
studied in [15] and subsequently in [16]. In this case, the presence of a locally Lipschitz source term from H 1 (Ω ) into L2 (Ω )
does not affect the classical arguments for establishing the existence of local solutions via perturbation theory of monotone
operators [7]. However, the situation is different when the damping term is degenerate, which leads to the degeneracy of
the classical monotonicity argument. In fact, if any of the exponents k, l, θ , ρ is positive, then (1.1) is no longer a locally
Lipschitz perturbation of a monotone problem, even though the source term is a locally Lipschitz from H 1 (Ω ) into L2 (Ω ).
The more difficult case when m or r > 1 will not be discussed in this article.
At this end we remark that the following notation will be used throughout the paper.
|u|s,Ω = kukH s (Ω ) , kukp = kukLp (Ω ) and hu, vi = hu, viL2 (Ω ) .
Also, the following Sobolev imbeddings will used frequently, and sometimes without mention:

H01 (Ω ) ,→ Lq (Ω ), for 1 ≤ q ≤ 6, n = 3,

(1.4)
H01 (Ω ) ,→ Lq (Ω ), for 1 ≤ q < ∞, n = 1, 2.
We finally note that Poincaré’s inequality implies that the norms kukH 1 (Ω ) and k∇ uk2 are equivalent norms on H01 (Ω ).
0

1.2. Main results

In order to state our main result we introduce the definition of a weak solution to (1.1). For simplicity, we shall assume
d = d0 = e = e0 = 1.

Definition 1.2. A pair of functions (u, v) is said to be a weak solution of (1.1) on [0, T ] if u, v ∈ Cw ([0, T ], H01 (Ω )),
ut , vt ∈ Cw ([0, T ], L2 (Ω )), (u(0), v(0)) = (u0 , v 0 ) ∈ H01 (Ω ) × H01 (Ω ), (ut (0), vt (0)) = (u1 , v 1 ) ∈ L2 (Ω ) × L2 (Ω );
2660 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

and (u, v) satisfies


Z t Z t
u (t ), φ
0
− u ,φ 1
h∇ u(τ ), ∇φiL2 (Ω ) dτ + (|u|k + |v|l )|u0 (τ )|m−1 u0 (τ ), φ dτ




L2 (Ω ) L2 (Ω )
+ L2 (Ω )
0 0
Z t
= hf1 (u(τ ), v(τ )), φiL2 (Ω ) dτ , (1.5)
0
Z t Z t

θ
v (t ), ψ L2 (Ω ) − v , ψ L2 (Ω ) + h∇v(τ ), ∇ψiL2 (Ω ) dτ + (|v| + |u|ρ )|v 0 (τ )|r −1 v 0 (τ ), ψ L2 (Ω ) dτ

0
1
0 0
Z t
= hf2 (u(τ ), v(τ )), ψiL2 (Ω ) dτ , (1.6)
0

for all test functions φ, ψ ∈ H01 (Ω ) and for almost all t ∈ [0, T ].
Our first theorem establishes the existence and uniqueness of a local weak solution to (1.1) that satisfies an energy
identity, without further restrictions on the parameters or the initial data. Specifically, we have the following result.

Theorem 1.3 (Local Weak Solutions). Assume the validity of Assumption 1.1. Then, there exists a unique local weak solution
(u, v) to (1.1) defined on [0, T0 ] for some T0 > 0. In addition, the said solution satisfies the energy identity
Z tZ Z tZ
E (t ) + (|u(τ )|k + |v(τ )|l )|u0 (τ )|m+1 dxdτ + (|v(τ )|θ + |u(τ )|ρ )|v 0 (τ )|r +1 dxdτ = E (0), (1.7)
0 Ω 0 Ω

where
1  0 2 0 2  Z
E (t ) := u (t ) + v (t ) + k∇ u(t )k2 + k∇v(t )k2 −
2 2 2 2 F (u(t ), v(t ))dx. (1.8)
2 Ω

The following theorem asserts that the weak solution furnished by Theorem 1.3 is a global solution provided p ≤ min{k +
m, l + m, θ + r , ρ + r } and without additional restriction on the initial data.

Theorem 1.4 (Global Weak Solutions). In addition to Assumption 1.1, assume n = 1, 2; and that p ≤ min{k + m, l + m, θ +
r , ρ + r }. Then, the said solution (u, v) in Theorem 1.3 is a global solution and T0 may be taken arbitrarily large.

Remark 1.5. In Theorem 1.4, all damping terms must be alive, i.e., the constants d, d0 , e, e0 must be all positive. We also note
here that the condition on the parameters in Theorem 1.4 is not possible when n = 3 (see Assumption 1.1). However, we
should point out here that all of our results in this paper are valid for a larger range of parameters; namely, p > 1, n ∈ N;
but at the expense of losing the uniqueness of solutions for the range 1 < p < 3.
Our next theorem provides an answer to the existence of a global solution to (1.1) when the condition p ≤ min{k + m, l +
m, θ + r , ρ + r } is violated. In order to state Theorem 1.6 below, we define
J (t ) := k∇ u(t )k22 + k∇v(t )k22 − 4G(t )
where G(t ) = Ω F (u(t ), v(t ))dx. Indeed, Theorem 1.6 asserts that the solution constructed in Theorem 1.3 is a global
R
solution; provided J (0) > 0 and the initial energy E (0) is sufficiently small. More specifically, we have the following result.
p p−1
Theorem 1.6 (Global Small Solutions). In addition to Assumption 1.1, assume that J (0) > 0 and 4 2 c0 E (0) 2 < 1, where c0 is
a computable positive constant that depends on p and Ω . Then, the said solution (u, v) in Theorem 1.3 is a global solution and T0
may be taken arbitrarily large.
Our next theorem addresses the issue of a strong source (large values of p when p > max{k + m, l + m, θ + r , ρ + r })
which may lead to a finite time blow up of solutions. Here, our result in Theorem 1.7 is inspired by the work of [10,11] for
their treatment of a single wave equation with a non-degenerate damping and a source term. Although, the basic calculus in
the proof of Theorem 1.7 draws from the ideas in [10,11], and also from [17,13] in their treatment of a single wave equation
equipped with a degenerate damping term and a source term, the proof here has to be significantly adjusted to accommodate
the coupling in (1.1).

Theorem 1.7 (Blow Up of Solutions). In addition to Assumption 1.1, assume that p > max{k + m, l + m, θ + r , ρ + r } and
E (0) < 0, where E (0) is the initial energy given by
Z
1  
E (0) := u1 2 + v 1 2 + ∇ u0 2 + ∇v 0 2 − F (u0 , v 0 )dx.

2 2 2 2
2 Ω

Then, the weak solution (u, v) in Theorem 1.3 blows up in finite time.
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2661

Remark 1.8. The condition on the parameters in Theorem 1.7 is always valid when n = 3. That is, when n = 3 every local
solution must blow up in finite time provided E (0) < 0. However, it is not clear what happens if
min{k + m, l + m, θ + r , ρ + r } < p ≤ max{k + m, l + m, θ + r , ρ + r }
without further restrictions on the initial data.
We conclude the introduction by reminding the reader of the following elementary inequalities which will be used in
various places in the paper. Specifically, we have

(|a|m−1 a − |b|m−1 b)(a − b) ≥ C |a − b|m+1 , (1.9)


for some constant C > 0, all m > 0, and all a, b ∈ R. In addition,
|a| − |b|k ≤ C |a − b| (|a|k−1 + |b|k−1 ),
k
(1.10)
for some constant C > 0, all k ≥ 1, and all a, b ∈ R. Also,
|a| a − |b|p b ≤ C |a − b| (|a|p + |b|p ),
p
(1.11)
for some constant C > 0, all p ≥ 0, and all a, b ∈ R.
We note here that one can use (1.10)–(1.11) to obtain the following useful inequalities enjoyed by f1 , f2 and F . Indeed,
straightforward computation yields
 
f1 (u, v) − f1 (ũ, ṽ) ≤ C0 u − ũ + |v − ṽ| |u|p−1 + |v|p−1 + ũ p−1 + |ṽ|p−1

  
p−1
 p−1 p−1
 p+1 p−3 p−3
+ C1 |v − ṽ| |u| 2 |v| 2 + |ṽ| 2 + u − ũ |ṽ| 2 |u| 2 + ũ 2 , (1.12)
 
f2 (u, v) − f2 (ũ, ṽ) ≤ C0 u − ũ + |v − ṽ| |u|p−1 + |v|p−1 + ũ p−1 + |ṽ|p−1


   
p−1 p−1 p−1 p+1  p−3 p−3
+ C1 u − ũ |v| 2 |u| 2 + ũ 2 + |v − ṽ| ũ 2 |v| 2 + |ṽ| 2 ,

(1.13)

F (u, v) − F (ũ, ṽ) ≤ C0 u − ũ + |v − ṽ| |u|p + |v|p + ũ p + |ṽ|p


 
 
p−1 p−1 p−1 p−1
+ C1 u − ũ |v| + ũ |v − ṽ| |u| ,
 2
2 |v| 2 + ũ
|ṽ| 2 (1.14)

for all u, v, ũ, ṽ ∈ R, and some positive constants C0 and C1 .

2. Local solutions and the proof of Theorem 1.3

This section is devoted to the proof of Theorem 1.3, which will be carried out in the following four subsections.

2.1. Approximate solutions

Let A = −∆ with its domain D (A) = H 2 (Ω ) ∩ H01 (Ω ). It is well known that A is positive, self-adjoint, and A is the
inverse of a compact operator. Moreover, A has the infinite sequence of positive eigenvalues {λj : j = 1, 2, . . .} and a
corresponding sequence of eigenfunctions {ej : j = 1, 2, . . .} that forms an orthonormal basis for L2 (Ω ). Also, the sequence
{ej : j = 1, 2, . . .} is an orthogonal basis for H01 (Ω ).
Let VN := the linear span of {e1 , . . . eN } and PN be the orthogonal projection of L2 (Ω ) onto VN . In order to establish the
existence of a local weak solution to the system (1.1) we shall use a standard Galerkin approximation scheme based on the
j=1 of the operator A = −∆. More precisely, let uN (t ) = j=1 uN ,j (t )ej and vN (t ) = j=1 vN ,j (t )ej be
PN PN
eigenfunctions {ej }∞
the approximate solutions in VN , i.e., uN (t ), vN (t ) satisfy the following system of ordinary differential equations (all inner
products in (2.1)–(2.2) below are L2 (Ω )-inner products):

u00N (t ), ej + ∇ uN (t ), ∇ ej + (|uN (t )|k + |vN (t )|l )|u0N (t )|m−1 u0N (t ), ej = f1 (uN (t ), vN (t )), ej ,





L2 (Ω )
(2.1)

vN00 (t ), ej + ∇vN (t ), ∇ ej + (|vN (t )|θ + |uN (t )|ρ )|vN0 (t )|r −1 vN0 (t ), ej = f2 (uN (t ), vN (t )), ej ,





L2 (Ω )
(2.2)

uN (0) = PN u0 , vN (0) = PN v 0 , u0N (0) = PN u1 , vN0 (0) = PN v 1 , (2.3)


for j = 1, 2 . . . , N. More specifically, (2.3) is equivalent to

uN ,j (0) = u0j , vN ,j (0) = vj0 , u0N ,j (0) = u1j , vN0 ,j (0) = vj1 , (2.4)

where u0j = u0 , ej L2 (Ω ) , vj0 = v 0 , ej L2 (Ω ) , u1j = u1 , ej L2 (Ω ) and vj1 = v 1 , ej L2 (Ω ) , for j = 1, 2 . . . , N.







2662 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

It is obvious that (2.1)–(2.3) is an initial value problem for a second order 2N × 2N system of ordinary differential
equations with continuous nonlinearities in the unknown functions uN ,j , vN ,j and their derivatives. Therefore, it follows
from the Cauchy–Peano Theorem that for every N ≥ 1, (2.1)–(2.3) has a solution uN ,j , vN ,j ∈ C 2 [0, TN ] for some TN > 0.
A priori estimates. In this step, we shall show that TN can be replaced by some T > 0, for all N ≥ 1.

Lemma 2.1. There exists a constant T > 0 such that the sequences of approximate solutions {uN } and {vN } satisfy the following:

• {uN }, {vN } are bounded sequences in L∞ (0, T ; H01 (Ω )).


• {u0N }, {vN0 } are bounded sequences in L∞ (0, T ; L2 (Ω )).
Rt R
• The sequences { 0 Ω (|uN (τ )|k + |vN (τ )|l )|u0N (τ )|m+1 dxdτ } and
Z t Z 
(|vN (τ )|θ + |uN (τ )|ρ )|vN0 (t )|r +1 dxdτ are bounded in L∞ (0, T ).
0 Ω

Proof. By multiplying (2.1) by u0N ,j (t ), (2.2) by vN0 ,j (t ), and summing for j = 1, . . . , N, we obtain

1 d  0  Z Z
u (t ) 2 + k∇ uN (t )k2 + (|uN (t )|k + |vN (t )|l )|u0 (t )|m+1 dx = f1 (uN (t ), vN (t ))u0N (t )dx,

N 2 2 N (2.5)
2 dt Ω Ω
Z Z
1 d  0 
v (t ) 2 + k∇vN (t )k2 + (|vN (t )|θ + |uN (t )|ρ )|vN0 (t )|r +1 dx = f2 (uN (t ), vN (t ))vN0 (t )dx.

N 2 2 (2.6)
2 dt Ω Ω

By adding (2.5) to (2.6) and integrating the resulting identity from 0 to t ≤ TN , we have

1  0  Z tZ
u (t ) 2 + v 0 (t ) 2 + k∇ uN (t )k2 + k∇vN (t )k2 + |uN (τ )|k + |vN (τ )|l |u0N (τ )|m+1 dxdτ

N 2 N 2 2 2
2 0 Ω
Z tZ
|vN (τ )|θ + |uN (τ )|ρ |vN0 (τ )|r +1 dxdτ

+
0 Ω
1  0 
u (0) 2 + v 0 (0) 2 + k∇ uN (0)k2 + k∇vN (0)k2

= N 2 N 2 2 2
2
Z tZ h i
+ f1 (uN (τ ), vN (τ ))u0N (τ ) + f2 (uN (τ ), vN (τ ))vN0 (τ ) dxdτ
0 Ω
Z tZ h i
≤ C (|u0 |1,Ω , |v 0 |1,Ω , |u1 |0,Ω , |v 1 |0,Ω ) + f1 (uN (τ ), vN (τ ))u0N (τ ) + f2 (uN (τ ), vN (τ ))vN0 (τ ) dxdτ , (2.7)
0 Ω

where we have used in (2.7) the fact that

uN (0) → u0 , vN (0) → v 0 strongly in H01 (Ω )

u0N (0) → u1 , vN0 (0) → v 1 strongly in L2 (Ω ). (2.8)

We estimate the last term in (2.7) as follows. By recalling (1.3) and using Hölder’s and Young’s inequalities, we have
Z Z  
p+1 p−1
f1 (uN , vN )u0 dx ≤ C v
p 0 0
N | u N + N | |u N | + |v N | 2 |u | 2 |u | dx
N N
Ω Ω

 p− 1 p+1

p p
≤ C (kuN k2p + kvN k2p ) uN 2 + kuN k3(p−1) kvN k 3(p+1) uN 2
0 2 2
0

2
 
2p 2p p−1 p+1
0 2
≤ C kuN k + kvN k + kuN k
2p 2p kvN k 3(p+1) + u
3(p−1) N 2
2
h 0 2 i
p−1 p+1
≤ C k∇ uN k2p
2 + k∇vN k 2p
2 + k∇ u N k 2 k∇v N k 2 + u ,
N 2 (2.9)

3(p+1)
where we have used in (2.9) the Sobolev imbeddings in (1.4) and the fact that when n = 3 then 2p = 3(p − 1) = 2
= 6.
Likewise, we have
Z Z  
p+1 p−1
f2 (uN , vN )u0 dx ≤ C |uN + vN |p |u0N | + |uN | 2 |vN | 2 |u0N | dx

N
Ω Ω

h 2 i
p−1
≤ C k∇ uN k2p 2p
2 + k∇vN k2 + k∇vN k2 k∇ uN kp2+1 + u0N 2 . (2.10)
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2663

2 2
Now, by putting yN (t ) := u0N (t ) 2 + vN0 (t ) 2 + k∇ uN (t )k22 + k∇vN (t )k22 then it follows from (2.7), (2.9)–(2.10) that

Z tZ Z tZ
yN (t ) + 2 (|uN (τ )|k + |vN (τ )|l )|u0N (τ )|m+1 dxdτ + 2 (|vN (τ )|θ + |uN (τ )|ρ )|vN0 (τ )|r +1 dxdτ
0 Ω 0 Ω
Z t
≤ C0 + C yN (τ )p dτ , (2.11)
0

where C0 = C (|u0 |1,Ω , |v 0 |1,Ω , |u1 |0,Ω , |v 1 |0,Ω ) > 0 and C > 0 is a generic constant. In particular, yN (t ) satisfies the
inequality
Z t
yN (t ) ≤ C0 + C yN (τ )p dτ . (2.12)
0

By using a standard comparison theorem (see for instance [18]), then (2.12) yields that yN (t ) ≤ z (t ), where z (t ) =
1
[C01−p − C (p − 1)t ]− p−1 is the solution of the Volterra integral equation
Z t
z (t ) = C0 + C z (τ )p dτ . (2.13)
0

Although z (t ) blows up in finite time (since p ≥ 3), nonetheless, there exists a time 0 < T < TN such that yN (t ) ≤ z (t ) ≤ C1
for all t ∈ [0, T ], where C1 is independent of N. Hence, for all N ≥ 1, one has yN (t ) ≤ C1 , for all t ∈ [0, T ], establishing the
first two parts of the lemma. The last part of the lemma immediately follows from (2.11). 

Remark 2.2. The proof of Lemma 2.1 does not depend on the assumption that the exponents of velocities are satisfying
0 < m, r < 1. Therefore, the statements in Lemma 2.1 are indeed valid for all values m, r > 0. In addition, it follows from
Lemma 2.1 (for all values m, r > 0) that there exists subsequences of uN and vN , which we still denote by uN and vN , and
functions u, v such that

uN → u and vN → v weakly∗ in L∞ (0, T ; H01 (Ω )),



(2.14)
u0N → u0 and vN0 → v 0 weakly∗ in L∞ (0, T ; L2 (Ω )).
However, the following lemma provides stronger information about the sequences of approximate solutions in the case
under consideration, namely 0 < m, r < 1. More precisely, we have the following result.

Lemma 2.3. The sequences of approximate solutions {uN } and {vN } satisfy the following:
• {uN } and {vN } are Cauchy sequences in L∞ (0, T ; H01 (Ω )).
• {u0N } and {vN0 } are Cauchy sequences in L∞ (0, T ; L2 (Ω )).
Proof. Let (uN , vN ) and (uL , vL ) be two pairs of approximate solutions, and without loss of generality, we assume N > L. Put
uNL (t ) := uN (t ) − uL (t ) = j=1 (uN ,j (t ) − uL,j (t ))ej and vNL := vN − vL = j=1 (vN ,j (t ) − vL,j (t ))ej , with the understanding
PN PN
that uL,j (t ) = vL,j (t ) ≡ 0 when j > L. Then, uNL and vNL satisfy (where the dependence on t will be suppressed some times):

u00NL , ej + ∇ uNL , ∇ ej + (|uN |k + |vN |l )|u0N |m−1 u0N − (|uL |k + |vL |l )|u0L |m−1 u0L , ej



= f1 (uN , vN ) − f1 (uL , vL ), ej ,


(2.15)
vNL , ej + ∇vNL , ∇ ej + (|vN |θ + |uN |ρ )|vN0 |r −1 vN0 − (|vL |θ + |uL |ρ )|vL0 |r −1 vL0 , ej

00

= f2 (uN , vN ) − f2 (uL , vL ), ej ,


(2.16)
uNL (0) = PN u − PL u , 0 0
vNL (0) = PN v − PL v ,
0 0

uNL (0) = PN u − PL u ,
0 1 1
vNL
0
(0) = PN v 1 − PL v 1 , (2.17)

where all inner products in (2.15)–(2.16) are L (Ω )-inner products. 2

Multiply (2.15) by u0N ,j (t ) − u0L,j (t ), (2.16) by vN0 ,j (t ) − vL0 ,j (t ), and sum for j = 1, . . . , N, one has

1 d  0 

u (t ) 2 + k∇ uNL (t )k2 + (|uN |k + |vN |l )|u0 |m−1 u0 − (|uL |k + |vL |l )|u0 |m−1 u0 , u0

NL 2 2 N N L L NL
2 dt
= f1 (uN (t ), vN (t )) − f1 (uL (t ), vL (t )), u0NL (t )


(2.18)
1 d  0 

v (t ) 2 + k∇vNL (t )k2 + (|vN |θ + |uN |ρ )|v 0 |r −1 v 0 − (|vL |θ + |uL |ρ )|v 0 |r −1 v 0 , v 0



NL 2 2 N N L L NL
2 dt
= f2 (uN (t ), vN (t )) − f2 (uL (t ), vL (t )), vNL
0
(t ) .


(2.19)
2664 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

By recalling (1.12)–(1.13) we estimate the last terms in (2.18) and (2.19) as follows.

f1 (uN (t ), vN (t )) − f1 (uL (t ), vL (t )), u0 (t ) ≤ I1 + I2 + I3 ,




NL (2.20)

where
Z
(|uNL | + |vNL |) |uN |p−1 + |vN |p−1 + |uL |p−1 + |vL |p−1 |u0NL |dx,

I1 = C0
ZΩ  
p−1 p−1 p−1
I2 = C1 |vNL | |uN | 2 |vN | 2 + |vL | 2 |u0NL |dx,

Z  
p+1 p−3 p−3
I3 = C1 |uNL | |vL | 2 |uN | 2 + |uL | 2 |u0NL |dx. (2.21)

Every term in I1 is estimated in the same way. In particular, for a typical term in I1 , we have
Z
(|uNL | + |vNL |) |uN |p−1 |u0NL |dx ≤ (kuNL k6 + kvNL k6 ) kuN kp3− 1
0
(p−1) uNL 2

≤ C (k∇ uNL k2 + k∇vNL k2 ) k∇ uN kp2−1 u0NL 2


≤ C (k∇ uNL k2 + k∇vNL k2 ) u0NL 2 ,



(2.22)

where we have used in (2.22) the Sobolev Imbedding Theorem, the fact that 3(p − 1) = 6 when n = 3, and the boundedness
statement in Lemma 2.1. By Young’s inequality, one has the estimate
 2 
I1 ≤ C k∇(uN − uL )k22 + k∇(vN − vL )k22 + u0NL 2 .

(2.23)

We also estimate a typical term in I2 as follows:


Z p−1 p−1
p−1 p−1
|vNL | |uN | 2 |vN | 2 |u0NL |dx ≤ kvNL k6 kuN k3(2p−1) kvN k3(2p−1) u0NL 2

p−1 p−1
≤ C k∇vNL k2 k∇ uN k2 2 k∇vN k2 2 u0NL 2
≤ C k∇vNL k2 u0NL 2 .

(2.24)

Therefore,
 2 
I2 ≤ C k∇vNL k22 + u0NL 2 .

(2.25)

By recalling p = 3 when n = 3, then a typical term in I3 is estimated as follows. If n = 3 then


Z p+1
p+1 p−3
|uNL | |vN | 2 |uN | 2 |u0NL |dx ≤ kuNL k6 kvN k 3(2p+1) u0NL 2
Ω 2

≤ C k∇ uNL k2 u0NL 2 .

(2.26)

Similarly, if n = 1, 2 then (2.26) is easily obtained and we have


 2 
I3 ≤ C k∇ uNL k22 + u0NL 2 .

(2.27)

It follows from (2.20), (2.23), (2.25), (2.27) that


 
f1 (uN (t ), vN (t )) − f1 (uL (t ), vL (t )), u0 (t ) ≤ C k∇ uNL k2 + k∇vNL k2 + u0 (t ) 2 .


NL 2 2 NL 2
(2.28)

Similarly, by using (1.13) one has


 
f2 (uN (t ), vN (t )) − f2 (uL (t ), vL (t )), v 0 (t ) ≤ C k∇ uNL k2 + k∇vNL k2 + v 0 (t ) 2 .


NL 2 2 NL 2
(2.29)

So, from (2.18)-(2.19) and (2.28)-(2.29) it follows that


1 d  0  Z 
u (t ) 2 + k∇ uNL (t )k2 + (|uN |k + |vN |l )|u0N |m−1 u0N − (|uL |k + |vL |l )|u0L |m−1 u0L u0NL dx

NL 2 2
2 dt Ω
 2 
≤ C k∇ uNL k22 + k∇vNL k22 + u0NL (t ) 2

(2.30)
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2665

and
1 d  0  Z 
v (t ) 2 + k∇vNL (t )k2 + (|vN |θ + |uN |ρ )|vN0 |r −1 vN0 − (|vL |θ + |uL |ρ )|vL0 |r −1 vL0 vNL
 0
NL 2 2 dx
2 dt Ω
 2 
≤ C k∇ uNL k22 + k∇vNL k22 + vNL (t ) 2 .
0
(2.31)

Now, by adding the term


Z
(|uL (t )|k + |vL (t )|l )|u0N (t )|m−1 u0N (t )u0NL (t )dx

to both sides of (2.30), we obtain


1 d  0  Z
u (t ) 2 + k∇ uNL (t )k2 + (|uL (t )|k + |vL (t )|l ) |u0 (t )|m−1 u0 (t ) − |u0 (t )|m−1 u0 (t ) u0 (t )dx

NL 2 2 N N L L NL
2 dt Ω
 2 
≤ C k∇ uNL k22 + k∇vNL k22 + u0NL (t ) 2 dx + J2 (t ),

(2.32)

where (suppressing the time variable t in (2.33))


Z
J2 (t ) = − |uN |k − |uL |k + |vN |l − |vL |l |u0N |m−1 u0N u0NL dx.
 
(2.33)

Let
Z
J1 (t ) = (|uL (t )|k + |vL (t )|l ) |u0N (t )|m−1 u0N (t ) − |u0L (t )|m−1 u0L (t ) u0NL (t )dx.

(2.34)

Then, by recalling the monotonicity inequality (1.9), we have


Z
m+1
J1 (t ) ≥ C (|uL (t )|k + |vL (t )|l ) u0NL (t ) dx.

(2.35)

We also note that


Z
|J2 (t )| ≤ |uN (t )|k − |uL (t )|k + |vN (t )|l − |vL (t )|l |u0 (t )|m |u0 (t )|dx,
 
N NL (2.36)

where both terms in (2.36) are estimated in the same way. To do this, we choose δ = 6−3m 6
−k
, and we note that { k−6 1 , 6, m
2
, δ}
are Hölder conjugate indices. Moreover, our assumption k ≤ 3(1 − m) when n = 3, and the fact that 0 < m < 1, imply that
1 < δ ≤ 2 (essential in obtaining estimate (2.37)). Therefore, by recalling (1.10), we have
Z Z
uN (t ) |k −|uL (t ) |k u0 (t )|m |u0 (t )|dx ≤ C |uN (t ) − uL (t )| |uN (t )|k−1 + |uL (t )|k−1 |u0N (t )|m |u0NL (t )|dx

N NL

Ω m
≤ C kuN (t )kk6−1 + kuL (t )kk6−1 kuNL (t )k6 u0N (t ) 2 u0NL (t ) δ


≤ C k∇ uNL (t )k2 u0NL (t ) 2 ,



(2.37)

where we have used Hölder’s inequality, the Sobolev Imbedding Theorem and the boundedness statement of Lemma 2.1.
The second term in (2.36) is estimated similarly by choosing δ = 6−3m
6
−l
. Indeed, we have
Z
vN (t ) |l −|vL (t ) |l u0 (t )|m |u0 (t )|dx ≤ C k∇vNL (t )k2 u0 (t ) .

N NL NL 2
(2.38)

Therefore, it follows from (2.36)–(2.38)


h 2 i
|J2 (t )| ≤ C1 k∇ uNL (t )k22 + k∇vNL (t )k22 + u0NL (t ) 2 .

(2.39)

By combining (2.32), (2.35) and (2.39), we obtain


Z
1 d  0 
u (t ) 2 + k∇ uNL (t )k2 + C (|uL (t )|k + |vL (t )|l )|u0 (t )|m+1 dx

NL 2 2 NL
2 dt Ω
 2 0 2 
≤ C2 k∇ uNL k22 + k∇vNL k22 + u0NL (t ) 2 + vNL (t ) 2 .

(2.40)
2666 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

As for (2.31), we can repeat the same steps above and obtain
Z
1 d  0 
v (t ) 2 + k∇vNL (t )k2 + C (|vL (t )|θ + |uL (t )|ρ )|v 0 (t )|m+1 dx

NL 2 2 NL
2 dt Ω
 2 0 2 
≤ C3 k∇ uNL k22 + k∇vNL k22 + u0NL (t ) 2 + vNL ( t ) 2 .

(2.41)

At this end, put


2 2
YNL (t ) := k∇ uNL (t )k22 + k∇vNL (t )k22 + u0NL (t ) 2 + vNL ( t ) 2 .
0

Then, it follows from (2.40)–(2.41) that


Z tZ Z tZ
YNL (t ) + C (|uL (t )|k + |vL (t )|l )|u0NL (t )|m+1 dxdτ + C (|vL (t )|θ + |uL (t )|ρ )|vNL
0
(t )|m+1 dxdτ
0 Ω 0 Ω
Z t
≤ YNL (0) + C 0 YNL (τ )dτ , (2.42)
0
2 2
where YNL (0) := k∇ uNL (0)k22 + k∇vNL (0)k22 + u0NL (0) 2 + vNL (0) 2 .
0

By Gronwall’s inequality, we have
YNL (t ) ≤ CT YNL (0), (2.43)
for all t ∈ [0, T ].
By recalling the strong convergence in (2.8), then YNL (0) → 0 as L, N → ∞. Hence, YNL (t ) −→ 0 as L, N −→ ∞, and the
proof of the lemma is complete. 

Remark 2.4. Let (uN , vN ) and (uL , vL ) be two pairs of approximate solutions as introduced in the proof of Lemma 2.3. Then,
it follows from (2.42) that
Z
lim (|uL (t )|k + |vL (t )|l )|u0NL (t )|m+1 dx = 0
N ,L→∞ Ω
Z
lim (|vL (t )|θ + |uL (t )|ρ )|vNL
0
(t )|r +1 dx = 0.
N ,L→∞ Ω

Moreover, there exists a pair of functions (u, v) such that

uN → u and vN → v strongly in L∞ (0, T ; H01 (Ω )),



(2.44)
u0N → u0 and vN0 → v 0 strongly in L∞ (0, T ; L2 (Ω )).

As a consequence of Lemma 2.3, we have two corollaries. The first corollary addresses the source term and the second
addresses the damping terms.

Corollary 2.5. The sequences of approximate solutions {uN } and {vN } satisfy the following:

f1 (uN , vN ) −→ f1 (u, v) strongly in L∞ (0, T ; L2 (Ω )),



(2.45)
f2 (uN , vN ) −→ f2 (u, v) strongly in L∞ (0, T ; L2 (Ω )).

Proof. The proofs of the statements in (2.45) are essentially the same, and so we only present the proof of the first statement.
By recalling (1.12), we have

kf1 (uN (t ), vN (t )) − f1 (u(t ), v(t ))k22 ≤ I1 + I2 + I3 + I4 , (2.46)


where
Z
|uN − u|2 |uN |2(p−1) + |vN |2(p−1) + |u|2(p−1) + |v|2(p−1) dx

I1 = C0
ZΩ
|vN − v|2 |uN |2(p−1) + |vN |2(p−1) + |u|2(p−1) + |v|2(p−1) dx

I2 = C0
ZΩ
|vN − v|2 |uN |p−1 |vN |p−1 + |v|p−1 dx

I3 = C1

Z
|uN − u|2 |v|p+1 |uN |p−3 + |u|p−3 dx.

I4 = C1 (2.47)

M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2667

We estimate a typical term in I1 and I2 as follows.


Z
|uN − u|2 |uN |2(p−1) dx ≤ kuN − uk26 kuN k23((pp− 1)
−1 )

≤ C k∇(uN − u)k22 k∇ uN k22(p−1) ≤ C k∇(uN − u)k22 , (2.48)


where we have used in (2.48) the Sobolev Imbedding Theorem, the fact that when n = 3 then 3(p − 1) = 6, and the bounds
furnished by Lemma 2.1. Hence, for all t ∈ [0, T ], we have

I1 + I2 ≤ C k∇(uN (t ) − u(t ))k22 + k∇(vN (t ) − v(t ))k22 .



(2.49)
Similarly, a typical term in I3 is estimated as follows.
Z
|vN − v|2 |uN |p−1 |vN |p−1 dx ≤ kvN − vk26 kuN k3p− 1 p−1
(p−1) kvN k3(p−1)

≤ C k∇(vN (t ) − v(t ))k22 . (2.50)


Also, a typical term in I4 is estimated in the same way, and one easily obtains

I3 + I4 ≤ C k∇(uN (t ) − u(t ))k22 + k∇(vN (t ) − v(t ))k22 .



(2.51)
Hence, the strong convergence furnished by Lemma 2.3 combined with (2.46), (2.49) and (2.51) prove the first convergence
in (2.45). 

Corollary 2.6. There exist subsequences of approximate solutions, which we still denote by {uN } and {vN }, that satisfy the
following:

|u |k |u0 |m−1 u0 → |u|k |u0 |m−1 u0 weakly in L2 (QT ),



 N l 0N m−1 0N

|vN | |uN | uN → |v|l |u0 |m−1 u0 weakly in L2 (QT ),

θ 0 r −1 0 θ 0 r −1 0 (2.52)
|vN |ρ |vN0 | r −1vN0 → |v| ρ|v 0| r −1v 0 weakly in L2 (QT ),
2

|uN | |vN | vN → |v| |v | v weakly in L (QT ).

Proof. By the strong convergence in Lemma 2.3 there exist subsequences of the approximate solutions, which we still denote
by {uN } and {vN }, that satisfy

uN → u and vN → v a.e. in QT ,

(2.53)
u0N → u0 and vN0 → v 0 a.e. in QT .

In particular, one has

|uN |k |u0N |m−1 u0N −→ |u|k |u0 |m−1 u0 a.e. in QT . (2.54)


Combining (2.54) with the fact that
|uN |k |u0 |m−1 u0 ≤ kuN kk 2k u0 m ≤ C k∇ uN kk u0 m ≤ CT ,

N N 2 N 2 2 N 2 (2.55)
1−m

for all N ≥ 1, then the first convergence in (2.52) follows from a standard result in analysis. The other statements in (2.52)
are proven similarly. 

2.2. Passage to the limit

In order to complete the proof of the existence statement in Theorem 1.3 we need to show several things. First, we
integrate (2.1) and (2.2) from 0 to t to obtain
Z t Z t
uN (t ), ej − uN (0), ej +
0 0
∇ uN (τ ), ∇ ej dτ + (|uN |k + |vN |l )|u0N |m−1 u0N , ej dτ





0 0
Z t
f1 (uN (τ ), vN (τ )), ej dτ ,


= (2.56)
0
Z t Z t
vN0 (t ), ej − vN0 (0), ej + ∇vN (τ ), ∇ ej dτ + (|vN |θ + |uN |ρ )|vN0 |r −1 vN0 , ej dτ





0 0
Z t
f2 (uN (τ ), vN (τ )), ej dτ ,


= (2.57)
0

for all t ∈ [0, T ], and where all inner products that appear in (2.56)–(2.57) are L2 (Ω )-inner products.
2668 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

By recalling the strong convergence in Lemma 2.3, Corollary 2.5 and (2.8); and the weak convergence in Corollary 2.6,
we can easily pass to the limit in (2.56)–(2.57). Indeed, we obtain the existence of a pair of functions (u, v) with u, v ∈
L∞ (0, T ; H01 (Ω )), ut , vt ∈ L∞ (0, T ; L2 (Ω )) and (u, v) verify the equalities (1.5)–(1.6). In addition, we have the following
result concerning the second order derivatives u00 , v 00 , 4u, 4v .

Lemma 2.7. Assume that u, v ∈ L∞ (0, T ; H01 (Ω )), ut , vt ∈ L∞ (0, T ; L2 (Ω )) and (u, v) verify the equalities (1.5)- (1.6). Then,

4u, 4v, u00 , v 00 ∈ L∞ (0, T ; H −1 (Ω )),


where H −1 (Ω ) is the dual space of H01 (Ω ).
Proof. Let h., .i and hh., .ii denote respectively the duality pairing between H −1 (Ω ) and H01 (Ω ) and between D 0 (Ω ) and
D (Ω ). We first note the inclusions H01 (Ω ) ⊂ L2 (Ω ) ⊂ H −1 (Ω ), where each space is dense in the following one and the
injections are continuous. In addition,
hf , φi = hf , φiL2 (Ω ) for all f ∈ L2 (Ω ) and all φ ∈ H01 (Ω ).

Also, since 4 ∈ L(H01 (Ω ), H −1 (Ω )) then 4u(t ) ∈ H −1 (Ω ). Moreover,


h4u(t ), φi = hh4u(t ), φii = −hh∇ u(t ), ∇φii = − h∇ u(t ), ∇φiL2 (Ω ) ,
for all φ ∈ D (Ω ). Since D (Ω ) is dense in H01 (Ω ), then

h4u(t ), φi = − h∇ u(t ), ∇φiL2 (Ω ) (2.58)

for all φ ∈ H01 (Ω ).


Now, the fact that 4u ∈ L∞ (0, T ; H −1 (Ω )) is trivial, since u ∈ L∞ (0, T ; H01 (Ω )) and

|h4u(t ), φi| = h∇ u(t ), ∇φiL2 (Ω ) ≤ ku(t )kH 1 (Ω ) kφkH 1 (Ω ) ≤ CT kφkH 1 (Ω )



(2.59)
0 0 0

for every φ ∈ H01 (Ω ) and all t ∈ [0, T ]. Similarly, 4v ∈ L∞ (0, T ; H −1 (Ω )).


By recalling (1.5), then for every φ ∈ H01 (Ω ) we have

hu (t ), φi = d hu0 (t ), φi = d u0 (t ), φ 2
00

dt dt L (Ω )



≤ h∇ u(t ), ∇φiL2 (Ω ) + (|u(t )|k + |v(t )|l )|u0 (t )|m−1 u0 (t ), φ L2 (Ω ) + hf1 (u(t ), v(t )), φiL2 (Ω )

≤ C k∇φk2 k∇ u(t )k2 + (|u(t )|k + |v(t )|l )|u0 (t )|m 2 + kf1 (u(t ), v(t ))k2 .
 
(2.60)
By Hölder’s inequality, we have
 
0 m
(|u(t )|k + |v(t )|l )|u0 (t )|m ≤ ku(t )kk 2k + kv(t )kl |u (t )|

2 2l 2
1−m 1−m
m
≤ C k∇ u(t )kk2 + k∇v(t )kl2 |u0 (t )| 2 ≤ CT ,

(2.61)
and
p−1

p+1
kf1 (u(t ), v(t ))k2 ≤ C |u(t ) + v(t )|p + |u(t )| 2 |v(t )| 2

2
 
p p p−1 p+1
≤ C ku(t )k2p + kv(t )k2p + ku(t )k3(p−1) kv(t )k 3(p+1)
2
 
p−1
p
2
p
≤ C k∇ u(t )k + k∇v(t )k + k∇ u(t )k
2 2 k∇v(t )kp2+1 ≤ CT , (2.62)

3(p+1)
where we have used in (2.62) the Sobolev Imbeddings in (1.4) and the fact that when n = 3 then 2p = 3(p − 1) = 2
= 6.
It follows from (2.60)–(2.62) that
hu (t ), φi ≤ CT k∇φk2
00
(2.63)

for all φ ∈ (Ω ).
H01
Hence, u00 ∈ L∞ (0, T ; H −1 (Ω )). Similarly, one has v 00 ∈ L∞ (0, T ; H −1 (Ω )). 

2.3. Additional regularity and the proof of the energy identity

We remind the reader at this point that the constructed weak solution (u, v) to (1.1) satisfies: u, v ∈ L∞ (0, T ; H01 (Ω )),
ut , vt ∈ L∞ (0, T ; L2 (Ω )), (u, v) verifies the equalities (1.5), (1.6), and (u, v) satisfies the conclusions of Lemma 2.7.
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2669

Therefore, it follows from standard results (Lemma 8.1–8.2, page 275–276, Lions and Magenes [19]) (after possibly a
modification on a set of measure zero) that

u, v ∈ Cw ([0, T ], H01 (Ω )) and ut , vt ∈ Cw ([0, T ], L2 (Ω )), (2.64)


as required in Definition 1.2.
We shall show in this section that any weak solution (u, v) to (1.1) enjoying the regularity in (2.64) must satisfy the
energy identity (1.7). Let us note here that ut , vt are not regular enough (we only have ut , vt ∈ Cw ([0, T ], L2 (Ω ))), and so, it
is not permissible to directly test the equations in (1.1) with ut and vt . In order to overcome this difficulty we shall use the
difference quotients Dh u and Dh v and their well known properties (see [20] for more details).
For any function y ∈ Cw ([0, t ], L2 (Ω )) and h > 0 we define

1
Dh y(τ ) = [ye (τ + h) − ye (τ − h)] , (2.65)
2h
where ye (τ ) denotes the extension of y(τ ) to R given by: ye (τ ) = y(τ ) for τ ∈ (0, t ); ye (τ ) = y(t ) for τ ≥ t; and
ye (τ ) = y(0) for τ ≤ 0.
Before getting to the proof of (1.7), the following elementary proposition will be needed.

Proposition 2.8. Let w1 , w2 ∈ H01 (Ω ), w3 ∈ L2 (Ω ), α ≥ 1, 0 < γ < 1, and if n = 3, further assume, α ≤ 3(1 − γ ). Then,

• fj (w1 , w2 ) ∈ L2 (Ω ) for j = 1, 2.
• |w1 |α |w3 |γ −1 w3 ∈ L2 (Ω ).
Proof. By Hölder’s inequality, Sobolev Imbedding Theorem, and the assumption that p = 3 when n = 3, one has
Z Z
|f1 (w1 , w2 )| dx ≤ C 2
|w1 |2p + |w2 |2p + |w1 |p−1 |w2 |p+1 dx
 

Ω 
p−1 p+1
≤ C kw1 k2p
2p + kw k 2p
2 2p + kw k
1 3(p−1) kw k
2 3(p+1)
2
h i
p−1
≤ C k∇w1 k2p 2p
2 + k∇w2 k2 + k∇w1 k2 k∇w2 kp2+1 ≤ C . (2.66)

Hence, f1 (w1 , w2 ) is in L2 (Ω ) and similarly, so is f2 (w1 , w2 ). The second statement is trivial and its proof is omitted. 

We are now in a position to prove the following result.

Lemma 2.9. Let u, v ∈ Cw ([0, T ], H01 (Ω )) and ut , vt ∈ Cw ([0, T ], L2 (Ω )) and (u, v) is a weak solution to (1.1) in the sense of
Definition 1.2. Then, (u, v) satisfies the energy identity:
Z tZ Z tZ
E (t ) + (|u(τ )| + |v(τ )| )|u (τ )|
k l 0 m+1
dxdτ + (|v(τ )|θ + |u(τ )|ρ )|v 0 (τ )|r +1 dxdτ = E (0), (2.67)
0 Ω 0 Ω

where
1  0 2 0 2  Z
E (t ) := u (t ) + v (t ) + k∇ u(t )k2 + k∇v(t )k2 −
2 2 2 2 F (u(t ), v(t ))dx. (2.68)
2 Ω

Proof. As we have noted earlier, the regularity of ut , vt does not allow us to directly test the equations in (1.1) with ut and
vt . In order to overcome this difficulty we shall use the difference quotients Dh u and Dh v and their well known properties
(see [20] for more details).
Throughout the proof, we fix t ∈ [0, T ]. Then, with the notation introduced in (2.65) for a fixed t ∈ [0, T ] and with the
given regularity of (u, v), the following properties have been established by Koch and Lasiecka [20]:
Z t
1
lim hut (τ ), Dh ut (τ )iL2 (Ω ) dτ = [kut (t )k2L2 (Ω ) − kut (0)k2L2 (Ω ) ],
h→0 0 2
Z t
1
lim hvt (τ ), Dh vt (τ )iL2 (Ω ) dτ = [kvt (t )k2L2 (Ω ) − kvt (0)k2L2 (Ω ) ],
h→0 0 2
Z t
1
lim h∇ u(τ ), Dh ∇ u(τ )iL2 (Ω ) dτ = [k∇ u(t )k2L2 (Ω ) − k∇ u(0)k2L2 (Ω ) ],
h→0 0 2
Z t
1
lim h∇v(τ ), Dh ∇v(τ )iL2 (Ω ) dτ = [k∇v(t )k2L2 (Ω ) − k∇v(0)k2L2 (Ω ) ]. (2.69)
h→0 0 2
2670 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

In addition,
lim Dh u = ut , lim Dh v = vt strongly in L2 (0, t ; L2 (Ω )),
h→0 h→0
1 1
lim Dh u(0) = ut (0), lim Dh v(0) = vt (0) weakly in L2 (Ω ),
h→0 2 h→0 2
1 1
lim Dh u(t ) = ut (t ), lim Dh v(t ) = vt (t ) weakly in L2 (Ω ), (2.70)
h→0 2 h→0 2
and for every τ ∈ (0, t ), we have

lim Dh u(τ ) = ut (τ ) and lim Dh v(τ ) = vt (τ ) weakly in L2 (Ω ). (2.71)


h→0 h→0

Now, multiply the equations in (1.1) respectively by Dh u and Dh v , then integrate over Ω × (0, t ) to obtain,
Z tZ Z tZ Z tZ
(utt − 4u)Dh udxdτ + (|u| + |v| )|ut |
k l m−1
ut Dh udxdτ = f1 (u, v)Dh udxdτ (2.72)
0 Ω 0 Ω 0 Ω
Z tZ Z tZ Z tZ
(vtt − 4v)Dh v dxdτ + (|v|θ + |u|ρ )|vt |r −1 vt Dh v dxdτ = f2 (u, v)Dh v dxdτ . (2.73)
0 Ω 0 Ω 0 Ω

Let us note that the regularity of (u, v) implies that fj (u, v), (|u|k + |v|l )|ut |m−1 ut , (|v|θ + |u|ρ )|vt |r −1 vt are all in
L (Ω × (0, T )) (see (2.52)). Therefore, passing to the limit in the nonlinear terms in (2.72)–(2.73) as h → 0 presents no
2

problem. We also note that u := utt − 4u and v := vtt − 4v are in L2 (Ω × (0, T )).
Let h., .i denote the standard duality pairing between H −1 (Ω ) and H01 (Ω ). Then, it follows from the regularity of u and
the results established by Koch and Lasiecka [20] that
Z t
1
lim hutt (τ ), Dh u(τ )idτ = [kut (t )k2L2 (Ω ) − kut (0)k2L2 (Ω ) ]. (2.74)
h→0 0 2
By recalling Lemma 2.7 and the fact that u ∈ L2 (Ω × (0, T )), then it follows form (2.69) and (2.74) that
Z tZ Z tZ
lim (utt − 4u) Dh u dxdτ = (utt − 4u) ut dxdτ
h→0 0 Ω Ω0
Z t Z t 
= lim hutt (τ ), Dh u(τ )idτ + h∇ u(τ ), Dh ∇ u(τ )iL2 (Ω ) dτ
h→0 0 0

1h i 1h i
= kut (t )k2L2 (Ω ) − kut (0)k2L2 (Ω ) + k∇ u(t )k2L2 (Ω ) − k∇ u(0)k2L2 (Ω ) . (2.75)
2 2
By the same exact steps we also have
Z tZ Z tZ
lim (vtt − 4v) Dh v dxdτ = (vtt − 4v) vt dxdτ
h→0 0 Ω 0 Ω
1h i 1h i
= kvt (t )k2L2 (Ω ) − kvt (0)k2L2 (Ω ) + k∇v(t )k2L2 (Ω ) − k∇v(0)k2L2 (Ω ) . (2.76)
2 2
Consequently, we can pass to the limit in (2.72)–(2.73) as h → 0 and obtain
1 h 0 2 i Z tZ
u (t ) + k∇ u(t )k2 + (|u(τ )|k + |v(τ )|l )|u0 (τ )|m+1 dxdτ
2 2
2 0 Ω
Z tZ
1 h i
f1 (u(τ ), v(τ ))u0 (τ )dxdτ + u1 2 + ∇ u0 2 ,

= 2 2
(2.77)
0 Ω 2
1 h 0 2 i Z tZ
v (t ) + k∇v(t )k2 +
2 2 (|v(τ )|θ + |u(τ )|ρ )|v 0 (τ )|r +1 dxdτ
2 0 Ω
Z tZ
1 h i
f2 (u(τ ), v(τ ))v 0 (τ )dxdτ + v 1 2 + ∇v 0 2 ,

= 2 2
(2.78)
0 Ω 2
for all t ∈ [0, T ].
By adding the identities (2.77)–(2.78) and using the fact that

(F (u(τ ), v(τ ))) = f1 (u(τ ), v(τ ))u0 (τ ) + f2 (u(τ ), v(τ ))v 0 (τ ),
∂τ
then the energy identity (2.67) follows. 
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2671

2.4. Continuity and the uniqueness statement in Theorem 1.3

Our goal in this section is to prove the uniqueness statement in Theorem 1.3. In addition, we aim to prove the additional
regularity statement in Lemma 2.11. The following elementary results will be needed.
2
Lemma 2.10. Assume that g1 , g2 , . . . , gk ∈ Cw ([0, T ], L2 (Ω )) and the mapping t 7→ j=1 gj (t ) 2 is continuous on [0, T ];
Pk
k ∈ N. Then, gj ∈ C 0 ([0, T ], L2 (Ω )) for every j = 1, 2, . . . , k.
2
Proof. Put φ(t ) = j=1 gj (t ) 2 for t ∈ [0, T ]. Then, φ is continuous on [0, T ] by assumption. For a fixed t ∈ [0, T ], let
Pk
{tN } be a sequence in [0, T ] such that tN → t as N → ∞. Then, for j = 1, 2, . . . , k, we have
gj (tN ) − gj (t ) 2 = gj (tN ) − gj (t ), gj (tN ) − gj (t ) 2


2 L (Ω )
2 2
= gj (tN ) 2 + gj (t ) 2 − 2 gj (tN ), gj (t ) L2 (Ω ) .


(2.79)

By summing from 1 to k in (2.79), we obtain


k k
gj (tN ) − gj (t ) 2 = φ(tN ) + φ(t ) − 2
X X
gj (tN ), gj (t ) L2 (Ω ) .


2
(2.80)
j=1 j=1

Now, the continuity of φ implies that φ(tN ) → φ(t ) as N → ∞. Also, the fact that gj ∈ Cw ([0, T ], L2 (Ω )) implies that
2
gj (tN ), gj (t ) → gj (t ), gj (t ) L2 (Ω ) = gj (t ) 2 for every j = 1, 2, . . . , k. Therefore, it follows from (2.80) that



L2 (Ω )

k k
gj (tN ) − gj (t ) 2 = 2φ(t ) − 2 gj (t ) 2 = 0.
X X
lim 2 2
(2.81)
N →∞
j =1 j =1

Hence, gj ∈ C 0 ([0, T ], L2 (Ω )) for every j = 1, 2, . . . , k. 


In addition to the regularity in (2.64) enjoyed by the constructed weak solution we have the following result.

Lemma 2.11. Assume u, v ∈ Cw ([0, T ], H01 (Ω )) and ut , vt ∈ Cw ([0, T ], L2 (Ω )), (u, v) verifies the equalities (1.5) and (1.6).
Then, after possibly a modification on a set of measure zero, the solution (u, v) satisfies

u, v ∈ C 0 ([0, T ], H01 (Ω )) and ut , vt ∈ C 0 ([0, T ], L2 (Ω )). (2.82)

Proof. We first note that since u, v ∈ Cw (0, T ; H01 (Ω )) and ut , vt ∈ Cw (0, T ; L2 (Ω )), then after possibly a modification on
a set of measure zero one has

u, v ∈ C 0 ([0, T ], L2 (Ω )). (2.83)


Let us first show that the mapping t 7→ G(t ) := Ω F (u(t ), v(t ))dx is continuous on [0, T ]. To show this, let t ∈ [0, T ] be
R
fixed and {tN } be a sequence in [0, T ] such that tN → t as N → ∞. By recalling (1.14), we have
Z
|G(tN ) − G(t )| = (F (u(tN ), v(tN )) − F (u(t ), v(t ))) dx

≤ C ( I1 + I2 + I3 + I4 ) , (2.84)
where
Z
|u(tN ) − u(t )| |u(tN )|p + |v(tN )|p + |u(t )|p + |v(t )|p dx

I1 =
ZΩ
|v(tN ) − v(t )| |u(tN )|p + |v(tN )|p + |u(t )|p + |v(t )|p dx

I2 =
ZΩ  p−1 p−1 p−1 p−1

I3 = |u(tN ) − u(t )| |v(tN )| |u(tN )| 2 |v(tN )| 2 + |u(t )| 2 |v(t )| 2 dx

Z  p−1 p−1 p−1 p−1

I4 = |v(tN ) − v(t )| |u(t )| |u(tN )| 2 |v(tN )| 2 + |u(t )| 2 |v(t )| 2 dx. (2.85)

All terms in I1 and I2 are estimated in the same way. For instance,
Z
|u(tN ) − u(t )| |u(tN )|p dx ≤ ku(tN ) − u(t )k2 ku(tN )(t )kp2p

≤ C ku(tN ) − u(t )k2 k∇ u(tN )kp2 ≤ C ku(tN ) − u(t )k2 , (2.86)


2672 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

where we have used in (2.86) the Sobolev Imbedding Theorem, the fact that 2p = 6 when n = 3, and the fact that
k∇ u(t )k2 ≤ C for all t ∈ [0, T ] (implied by the assumption u ∈ Cw (0, T ; H01 (Ω ))). Hence,
I1 + I2 ≤ C (ku(tN ) − u(t )k2 + kv(tN ) − v(t )k2 ) . (2.87)

Also, a typical term in I3 is estimated as follows.


Z p−1 p+1
p−1 p−1
|u(tN ) − u(t )| |v(tN )| |u(tN )| 2 |v(tN )| 2 dx ≤ ku(tN ) − u(t )k2 ku(tN )k3(2p−1) kv(tN )k 3(2p+1)
Ω 2

p−1 p+1
≤ ku(tN ) − u(t )k2 k∇ u(tN )k2 2
k∇v(tN )k2 2
≤ C ku(tN ) − u(t )k2 , (2.88)
3(p+1)
where we have used the fact that 3(p − 1) = = 6 when n = 3. 2
Similarly, a typical term in I4 is estimated in the same way, and one easily has

I3 + I4 ≤ C (ku(tN ) − u(t )k2 + kv(tN ) − v(t )k2 ) . (2.89)

Therefore, the continuity of G follows immediately from (2.84), (2.87) and (2.89). Finally, by recalling the energy identity
(1.7) we conclude that the function
1  0 2 0 2 
E0 (t ) := u (t ) + v (t ) + k∇ u(t )k2 + k∇v(t )k2
2 2 2 2
2
is continuous on [0, T ] and so, (2.82) follows from Lemma 2.10. 

Lemma 2.12. Assume (u, v) and (ũ, ṽ) are two weak solutions to the initial boundary value problem (1.1) on [0, T ] in the sense
of Definition 1.2. Then, (u, v) = (ũ, ṽ).

Proof. Let y = u − ũ and z = v − ṽ . Then, y and z satisfy

y − ∆y + (|u|k + |v|l )|ut |m−1 ut − (|ũ|k + |ṽ|l )|ũt |m−1 ũt = f1 (u, v) − f1 (ũ, ṽ), in Ω × (0, T ),

 tt
ztt − ∆z + (|v|θ + |u|ρ )|vt |r −1 vt − (|ṽ|θ + |ũ|ρ )|ṽt |r −1 ṽt = f2 (u, v) − f2 (ũ, ṽ), in Ω × (0, T ),

(2.90)
y(x, 0) = 0,
 yt (x, 0) = 0, z (x, 0) = 0, zt (x, 0) = 0, in Ω
y = z = 0, on Γ × (0, T ).

Again here, the regularity of yt , zt (we only have yt , zt ∈ C 0 ([0, T ], L2 (Ω ))) does not allow us to directly test the equations
in (2.90) with yt and zt and apply standard energy estimates to obtain the desired uniqueness result. In order to overcome
this difficulty we use the difference quotients Dh y and Dh z (as we defined in (2.65) for each fixed t ∈ [0, T ]), and their well
known properties listed in (2.69)–(2.71) and (2.74). For more details we refer the reader to [20].
Now, multiply the equations in (2.90) respectively by Dh y and Dh z, then integrate over Ω × (0, t ) to obtain,
Z tZ Z tZ
(ytt − 4y)Dh ydxdτ − f1 (u, v) − f1 (ũ, ṽ) Dh ydxdτ
 
0 Ω 0 Ω
Z tZ
(|u| + |v|l )|ut |m−1 ut − (|ũ|k + |ṽ|l )|ũt |m−1 ũt Dh ydxdτ ,
 k 
=− (2.91)
0 Ω
Z tZ Z tZ
(ztt − 4z )Dh zdxdτ − f2 (u, v) − f2 (ũ, ṽ) Dh zdxdτ
 
0 Ω 0 Ω
Z tZ
 θ
(|v| + |u|ρ )|vt |r −1 vt − (|ṽ|θ + |ũ|ρ )|ṽt |r −1 ṽt Dh zdxdτ .

=− (2.92)
0 Ω

Let us note that Proposition 2.8 implies that fj (u, v), fj (ũ, ṽ), (|u|k + |v|l )|ut |m−1 ut , (|ũ|k + |ṽ|l )|ũt |m−1 ũt , (|v|θ +
|u| )|vt |r −1 vt , (|ṽ|θ + |ũ|ρ )|ṽt |r −1 ṽt are all in L2 (Ω × (0, T )). Therefore, passing to the limit in the nonlinear terms in
ρ

(2.91)–(2.92) as h → 0 is straightforward. We also note that y := ytt − 4y and z := ztt − 4z are in L2 (Ω × (0, T )).
As we have done in obtaining the energy identity (2.67) and by using the properties of the difference quotient listed in
(2.69)–(2.71), (2.74); we can pass to the limit in (2.91)–(2.92) as h → 0 and obtain
Z tZ
1
kyt (t )k22 + k∇ y(t )k22 − f1 (u, v) − f1 (ũ, ṽ) yt dxdτ
  
2 0 Ω
Z tZ
(|u| + |v|l )|ut |m−1 ut − (|ũ|k + |ṽ|l )|ũt |m−1 ũt yt dxdτ ,
 k 
=− (2.93)
0 Ω
Z tZ
1
kzt (t )k2 + k∇ z (t )k22 −
2
f2 (u, v) − f2 (ũ, ṽ) zt dxdτ
  
2 0 Ω
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2673

Z tZ
 θ
(|v| + |u|ρ )|vt |r −1 vt − (|ṽ|θ + |ũ|ρ )|ṽt |r −1 ṽt zt dxdτ ,

=− (2.94)
0 Ω

where we have used the fact that y(x, 0) = yt (x, 0) = z (x, 0) = zt (x, 0) = 0.
By adding respectively the terms
Z tZ Z tZ
(|ũ| + |ṽ| )|ut |
k l m−1
ut yt dxdτ , (|ṽ|θ + |ũ|ρ )|vt |r −1 vt zt dxdτ
0 Ω 0 Ω

to both sides of (2.93) and (2.94), one finds


Z t
1
kyt (t )k22 + k∇ y(t )k22 + J1 (τ )dτ

2 0
Z tZ Z t
f1 (u(τ ), v(τ )) − f1 (ũ(τ ), ṽ(τ )) yt (τ )dxdτ + J2 (τ )dτ ,
 
= (2.95)
0 Ω 0
Z t
1
kzt (t )k22 + k∇ z (t )k22 + K1 (τ )dτ

2 0
Z tZ Z t
f2 (u(τ ), v(τ )) − f2 (ũ(τ ), ṽ(τ )) zt (τ )dxdτ + K2 (τ )dτ ,
 
= (2.96)
0 Ω 0

where
Z
J1 (τ ) = (|ũ|k + |ṽ|l ) |ut |m−1 ut − |ũt |m−1 ũt yt dx,
 

Z
J2 (τ ) = − |u| − |ũ|k + |v|l − |ṽ|l |ut |m−1 ut yt dx,
 k 

Z
K1 (τ ) = (|ṽ|θ + |ũ|ρ ) |vt |r −1 vt − |ṽt |r −1 ṽt zt dx,
 

Z
 θ
K2 (τ ) = − |v| − |ṽ|θ + |u|ρ − |ũ|ρ |vt |r −1 vt zt dx.

(2.97)

By recalling the monotonicity problem (1.9), we have


Z
J1 (τ ) ≥ c0 (|ũ|k + |ṽ|l )|yt |m+1 dx,

Z
K1 (τ ) ≥ c0 (|ṽ|θ + |ũ|ρ )|zt |r +1 dx. (2.98)

Also, by appealing to the proof of Lemma 2.3, particularly by replacing (uN , vN ) by (u, v), (uL , vL ) by (ũ, ṽ), and (uNL , vNL ) by
(y, z ) in (2.28)–(2.29) and (2.39), we obtain
Z
( (τ ), v(τ )) (ũ (τ ), ṽ(τ )) (τ ) ≤ C k∇ y(τ )k2 + k∇ z (τ )k2 + kyt (τ )k2 ,
  
f 1 u − f 1 yt dx 2 2 2


Z
f2 (u(τ ), v(τ )) − f2 (ũ(τ ), ṽ(τ )) zt (τ )dx ≤ C k∇ y(τ )k22 + k∇ z (τ )k22 + kzt (τ )k22 ,
  
(2.99)

and

|J2 (τ )| ≤ C k∇ y(τ )k22 + k∇ z (τ )k22 + kyt (τ )k22 ,




|K2 (τ )| ≤ C k∇ y(τ )k22 + k∇ z (τ )k22 + kzt (τ )k22 .



(2.100)

Now put

Y (t ) := kyt (t )k22 + kzt (t )k22 + k∇ y(t )k22 + k∇ z (t )k22 .

Then, it follows from (2.95)–(2.100) that


Z tZ Z t
Y (t ) + c0 (|ũ|k + |ṽ|l )|yt |m+1 + (|ṽ|θ + |ũ|ρ )|zt |r +1 dxdτ ≤ C Y (τ )dτ .
 
(2.101)
0 Ω 0

By Gronwall’s inequality, Y (t ) = 0 for all t ∈ [0, T ]. Hence, (u, v) = (ũ, ṽ), which completes the proof. 
2674 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

3. Global solutions

We devote this section to the proofs of Theorems 1.4 and 1.6.

3.1. Proof of Theorem 1.4

Proof. Let (u, v) be a weak solution to the initial boundary value problem (1.1) defined on [0, T ] as furnished by Theorem 1.3.
Assume that n = 1, 2,
p ≤ min{k + m, l + m, θ + r , ρ + r }
and define
1  0 2 0 2 
E0 (t ) := u (t ) + v (t ) + k∇ u(t )k2 + k∇v(t )k2 ,
2 2 2 2
2 Z
E1 (t ) := E0 (t ) + F (u(t ), v(t ))dx.

Our aim is to prove that, for all t ∈ [0, T ], the following inequality holds:
Z tZ Z tZ
E1 ( t ) + (|u(τ )|k + |v(τ )|l )|u0 (τ )|m+1 dxdτ + (|v(τ )|θ + |u(τ )|ρ )|v 0 (τ )|r +1 dxdτ ≤ CT , (3.1)
0 Ω 0 Ω

where CT depends on |u0 |1,Ω , |v 0 |1,Ω , |u1 |0,Ω , |v 1 |0,Ω , and T > 0 is being arbitrary.
Once (3.1) is established, then the existence of a global weak solution to (1.1), as stated in Theorem 1.4, follows from a
standard continuation argument. By recalling (1.2), we have
  Z  
p+1 p+1 p+1 p+1
c0 ku(t )kp+1 + kv(t )kp+1 ≤ F (u(t ), v(t ))dx ≤ c1 ku(t )kp+1 + kv(t )kp+1

 
≤ C k∇ u(t )kp2+1 + k∇v(t )kp2+1 . (3.2)

Therefore,
 p+1

E0 (t ) ≤ E1 (t ) ≤ C E0 (t ) + E0 (t ) 2 , (3.3)

and

ku(t )kpp+ 1 p+1


+1 + kv(t )kp+1 ≤ CE1 (t ). (3.4)

Put Qt := Ω × (0, t ). Then, the energy identity (1.7) can be written as follows.
Z Z
E0 ( t ) + (|u(τ )|k + |v(τ )|l )|u0 (τ )|m+1 dxdτ + (|v(τ )|θ + |u(τ )|ρ )|v 0 (τ )|r +1 dxdτ
Qt Qt


Z
= E0 (0) + F (u(τ ), v(τ ))dxdτ . (3.5)
Qt ∂τ
After adding the term

Z tZ Z Z
(F (u(τ ), v(τ ))) dxdτ = F (u(t ), v(t ))dx − F (u0 , v 0 )dx
0 Ω ∂τ Ω Ω

to both sides of (3.5), we obtain


Z Z
E1 ( t ) + (|u(τ )|k + |v(τ )|l )|u0 (τ )|m+1 dxdτ + (|v(τ )|θ + |u(τ )|ρ )|v 0 (τ )|r +1 dxdτ
Qt Qt


Z
= E1 (0) + 2 F (u(τ ), v(τ ))dxdτ . (3.6)
Qt ∂τ
In order to estimate the last term in (3.6), we put
Q11 := {(x, τ ) ∈ Qt : |u(x, τ )| ≤ 1, |v(x, τ )| ≤ 1} ,
Q12 := {(x, τ ) ∈ Qt : |u(x, τ )| ≤ 1, |v(x, τ )| > 1} ,
Q21 := {(x, τ ) ∈ Qt : |u(x, τ )| > 1, |v(x, τ )| ≤ 1} ,
Q22 := {(x, τ ) ∈ Qt : |u(x, τ )| > 1, |v(x, τ )| > 1} . (3.7)
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2675

Let us first notice that



Z Z
(F (u(τ ), v(τ ))) dxdτ ≤ 2 |f1 (u, v)| u0 + |f2 (u, v)| v 0 dxdτ

2
Qt ∂τ Qt

≤ I (t ) + J (t ), (3.8)

where
Z  
p−1 p+1
I (t ) = C |u|p + |v|p + |u| 2 |v| 2 u0 dxdτ
Qt
Z  
p−1 p+1
J (t ) = C |u|p + |v|p + |v| 2 |u| 2 v 0 dxdτ . (3.9)
Qt

In order to estimate I (t ) and J (t ) we write I (t ) = I11 + I12 + I21 + I22 , J (t ) = J11 + J12 + J21 + J22 , where
Z  
p−1 p+1
Iij (t ) = C |u|p + |v|p + |u| 2 |v| 2 u0 dQij
Qij
Z  
p−1 p+1
Jij (t ) = C |u|p + |v|p + |v| 2 |u| 2 v 0 dQij , i, j = 1, 2; (3.10)
Qij

and estimate each Iij (t ) and Jij (t ). It is easy to see that


Z Z
I11 (t ) ≤ C |u (τ )|dQ11 ≤ δ|Qt | + Cδ
0
|u0 (τ )|2 dQ11
Q11 Q11
Z t
≤ δ|Qt | + Cδ E1 (τ )dτ , (3.11)
0

for some δ > 0 and where |Qt | denotes the Lebesgue measure of Qt . Similarly,
Z t
J11 (t ) ≤ δ|Qt | + Cδ E1 (τ )dτ . (3.12)
0

We estimate
Z  
p−1 p+1
I22 (t ) = C |u(τ )|p + |v(τ )|p + |u(τ )| 2 |v(τ )| 2 |u0 (τ )|dQ22 , (3.13)
Q22

as follows. By noting u, v > 1 on Q22 , the first two terms in I22 (t ) are estimated in the same way. With s =
p−m
m+1
and noting
m(p+1)
that p − s = m+1
, then we have
Z Z
|v(τ )|p u0 (τ ) dQ22 = |v(τ )|p−s |v(τ )|s |u0 (τ )|dQ22

Q22 Q22
Z  mm+1 Z  m+1 1
≤ |v(τ )| p+1
dQ22 |v(τ )| p−m
|u (t )|
0 m+1
dQ22
Q22 Q22

Z  mm+1 Z  m+1 1
≤ |v(τ )|p+1 dQ22 |v(τ )|l |u0 (τ )|m+1 dQ22 , (3.14)
Q22 Q22

where we have used in (3.14) Hölder’s inequality and the fact p ≤ l + m and |v| > 1 on Q22 . Therefore, by Young’s inequality,
we have
Z Z Z
|v(τ )|p u0 (τ ) dQ22 ≤  |v(τ )|l |u0 (τ )|m+1 dQ22 + C |v(τ )|p+1 dQ22

C
Q22 Q22 Q22
Z Z t
≤ |v(τ )| |u (τ )|
l 0 m+1
dxdτ + C E1 (τ )dτ , (3.15)
Qt 0

where  > 0 that will be chosen later. Similarly, we have


Z Z Z t
|u(τ )|p u0 (τ ) dQ22 ≤  |u(τ )|k |u0 (τ )|m+1 dxdτ + C E1 (τ )dτ .

C (3.16)
Q22 Qt 0
2676 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

As for the last term in (3.13) we use (3.15)–(3.16) after a small adjustment, as follows:
Z Z
p−1 p+1 p−1 p−1 p+1 p+1
|u(τ )| 2 |v(τ )| 2 |u (τ )|dQ22 =
0
|u(τ )| 2 |u0 (τ )| 2p |v(τ )| 2 |u0 (τ )| 2p dQ22
Q22 Q 22
Z −1 Z
 p2p +1
 p2p
≤ |u(τ )| |u (τ )|dQ22
p 0
|v(τ )| |u (τ )|dQ22
p 0
Q Q22
Z22 Z 
≤C |u(τ )| |u (τ )|dQ22 +
p 0
|v(τ )| |u (τ )|dQ22
p 0
Q22 Q22
Z Z Z t
≤ |u(τ )|k |u0 (τ )|m+1 dxdτ +  |v(τ )|l |u0 (τ )|m+1 dxdτ + 2C E1 (τ )dτ . (3.17)
Qt Qt 0

Therefore, it follows from (3.15)–(3.17) that


Z Z Z t
I22 (t ) ≤ 2 |u(τ )|k |u0 (τ )|m+1 dxdτ + 2 |v(τ )|l |u0 (τ )|m+1 dxdτ + 4C E1 (τ )dτ . (3.18)
Qt Qt 0

Similarly, we obtain
Z Z Z t
θ ρ
J22 (t ) ≤ 2 |v(τ )| |v (τ )| 0 r +1
dxdτ + 2 |u(τ )| |v (τ )|
0 r +1
dxdτ + 4C E1 (τ )dτ . (3.19)
Qt Qt 0

Finally, we turn our attention to Iij for i 6= j. By noting that u ≤ 1 and v > 1 on Q12 , we have
Z  p+1

I12 (t ) ≤ C 1 + |v(τ )|p + |v(τ )| u0 (τ ) dQ12

2

Q12
Z Z
0 2
≤ δ |Qt | + Cδ u (τ ) dQ12 + C |v(τ )|p u0 (τ ) dQ12

Q12 Q12
Z t Z
≤ δ |Qt | + Cδ E1 (τ )dτ + C |v(τ )|p u0 (τ ) dQ12 .

(3.20)
0 Q12

Again, since v > 1 on Q12 , then by invoking (3.15), one has


Z t Z
I12 (t ) ≤ δ|Qt | + (Cδ + C ) E1 (τ )dτ +  |v(τ )|l |u0 (τ )|m+1 dxdτ . (3.21)
0 Qt

We estimate J21 (t ) by repeating the same steps in (3.20)–(3.21) with switching u and v , using s = r +1 , and using the fact
p−r

that p ≤ θ + r and |u| > 1 on Q21 . One easily has


Z t Z
J21 (t ) ≤ δ|Qt | + (Cδ + C ) E1 (τ )dτ +  |u(τ )|θ |v 0 (τ )|r +1 dxdτ . (3.22)
0 Qt

The estimates for I21 and J12 are similar and they are omitted. Indeed, we have
Z t Z
I21 (t ) ≤ δ|Qt | + (Cδ + C ) E1 (τ )dτ +  |u(τ )|k |u0 (τ )|m+1 dxdτ ,
0 Qt
Z t Z
J12 (t ) ≤ δ|Qt | + (Cδ + C ) E1 (τ )dτ +  |v(τ )|ρ |v 0 (τ )|r +1 dxdτ . (3.23)
0 Qt

By combining (3.9)–(3.12), (3.18)–(3.19), and (3.21)–(3.23), we obtain


Z t Z
I (t ) + J (t ) ≤ 6δ|Qt | + Cδ, E1 (τ )dτ + 3 |u(τ )|k + |v(τ )|l |u0 (τ )|m+1 dxdτ

0 Qt
Z
+ 3 |v(τ )|θ + |u(τ )| ρ
|v 0 (τ )|r +1 dxdτ .

(3.24)
Qt

By choosing  > 0 small enough, then it follows from (3.6), (3.8) and (3.24) that
Z tZ Z tZ
E1 (t ) + c |u(τ )|k + |v(τ )|l |u0 (τ )|m+1 dxdτ + c |u(τ )|θ + |v(τ )|ρ |v 0 (τ )|r +1 dxdτ
 
0 Ω 0 Ω
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2677

Z t
≤ E1 (0) + 6δ|Qt | + Cδ, E1 (τ )dτ , (3.25)
0

for some constant c > 0. By Gronwall’s inequality, E1 (t ) ≤ (E1 (0) + 6δ|Qt |) eCt , where C is some positive constant. Finally,
(3.25) leads to the inequality
Z tZ Z tZ
E1 ( t ) + |u(τ )|k + |v(τ )|l |u0 (τ )|m+1 dxdτ + |u(τ )|θ + |v(τ )|ρ |v 0 (τ )|r +1 dxdτ
 
0 Ω 0 Ω
≤ CT (E1 (0) + 6δ|Qt |) , (3.26)

where (3.26) is valid for all 0 < t ≤ T , where T is being arbitrary. Hence, the proof of Theorem 1.4 is now complete. 

3.2. Proof of Theorem 1.6

Proof. Let (u, v) be a weak solution to the initial boundary value problem (1.1) defined on [0, T ] as furnished by Theorem 1.3.
For the convenience of the reader, we recall
Z
J (t ) := k∇ u(t )k22 + k∇v(t )k22 − 4G(t ) where G(t ) := F (u(t ), v(t ))dx.

p p−1
Assume that J (0) > 0 and 4 2 c0 E (0) 2 < 1, where c0 is a computable positive constant. Here, we aim to prove that J (t ) > 0
on [0, T ] and for all t ∈ [0, T ], the following inequality holds:
Z tZ
1  0 2 0 2 
u (t ) + v (t ) + k∇ u(t )k2 + k∇v(t )k2 + G(t ) + |u(τ )|k + |v(τ )|l |u0 (τ )|m+1 dxdτ

2 2 2 2
4 0 Ω
Z tZ
|u(τ )|θ + |v(τ )|ρ |v 0 (τ )|r +1 dxdτ ≤ 2E (0),

+ (3.27)
0 Ω

where T > 0 is being arbitrary.


Once (3.27) is established, then the existence of a global weak solution to (1.1), as stated in Theorem 1.6, follows from a
standard continuation argument. By recalling Lemma 2.11 and the continuity of the function G on [0, T ], we note that J and
E are both continuous on [0, T ], where

1  0 2 0 2 
E (t ) := u (t ) + v (t ) + k∇ u(t )k2 + k∇v(t )k2 − G(t ).
2 2 2 2 (3.28)
2
Let t0 be given by

t0 := sup{t ∈ [0, T ] : J (τ ) > 0 on [0, t )}.

By the continuity of J and the assumption J (0) > 0, we know that t0 ∈ (0, T ]. We shall show below that t0 = T .
From the definitions of J (t ) and E (t ), we have
1  0 2 0 2 
E (0) ≥ E (t ) = u (t ) + v (t ) + 1 k∇ u(t )k2 + k∇v(t )k2 + 1 J (t )

2 2 2 2
2 4 4
1
> k∇ u(t )k22 + k∇v(t )k22 , for 0 ≤ t < t0 .

(3.29)
4
By recalling (3.2) and using (3.29), then for all t ∈ [0, t0 ) we have
 
p+1 p+1 p+1
4G(t ) ≤ 4c1 (ku(t )kp+1 + kv(t )kp+1 ) ≤ c0 k∇ u(t )k2 + k∇v(t )kp2+1
 p+2 1 p−1
≤ 2c0 k∇ u(t )k22 + k∇v(t )k22 ≤ 2c0 k∇ u(t )k22 + k∇v(t )k22 (4E (0)) 2


p p−1
= 4 2 c0 E (0) k∇ u(t )k22 + k∇v(t )k22 < k∇ u(t )k22 + k∇v(t )k22 ,
 
2 (3.30)

where the constant c0 in (3.30) is a positive constant that depends on c1 and the imbedding constant in (1.4), which itself
depends on Ω and p.
p p−1
By assumption, we write 4 2 c0 E (0) 2 = 1 − 0 for some 0 < 0 < 1. Therefore, it follows from (3.30), the continuity of
J , E, and the definition of t0 that t0 = T , i.e., J (t ) > 0 on [0, T ). Moreover,

J (t ) ≥ 0 k∇ u(t )k22 + k∇v(t )k22 on [0, T ].



(3.31)
2678 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

Finally, we use the energy identity (1.7), and (3.29)–(3.30) to obtain the following estimate
Z tZ
1  0 2 0 2 
u (t ) + v (t ) + 1 k∇ u(t )k2 + k∇v(t )k2 + G(t ) + (|u(τ )|k + |v(τ )|l )|u0 (τ )|m+1 dxdτ

2 2 2 2
2 4 0 Ω
Z tZ
+ (|v(τ )|θ + |u(τ )|ρ )|v 0 (τ )|r +1 dxdτ
0 Ω
Z tZ Z tZ
= E (t ) + (|u(τ )|k + |v(τ )|l )|u0 (τ )|m+1 dxdτ + (|v(τ )|θ + |u(τ )|ρ )|v 0 (τ )|r +1 dxdτ
0 Ω 0 Ω
1
+ 2G(t ) − k∇ u(t )k22 + k∇v(t )k2 2

4
1
≤ E (0) + k∇ u(t )k22 + k∇v(t )k22 ≤ 2E (0),

(3.32)
4
for all t ∈ [0, T ], where T > 0 is being arbitrary. Hence, (3.27) follows and the proof is complete. 

4. Proof of Theorem 1.7—Blow-up of solutions

Proof. Let (u, v) be a weak solution to (1.1) in the sense of Definition 1.2. Throughout the proof we assume that p >
max{k + m, l + m, θ + r , ρ + r } and E (0) < 0. We define the life span T of such a solution (u, v) to be the supremum
of all T ∗ > 0 such that (u, v) is a solution to (1.1) in the sense of Definition 1.2 on [0, T ∗ ]. Our goal is to show that T is
necessarily finite.
For t ∈ [0, T ), define
Z
N (t ) = ku(t )k22 + kv(t )k22 , G(t ) = F (u(t ), v(t ))dx, (4.1)

1  0 2 0 2 
H (t ) = −E (t ) = − u (t ) + v (t ) + k∇ u(t )k2 + k∇v(t )k2 + G(t ).
2 2 2 2 (4.2)
2
We should mention here that although our argument below draws from ideas in [15,17,10,13,14] in their treatment of the
single wave equation; our proof has to be significantly adjusted to accommodate the strong coupling in the system and how,
indeed, one shows that the source does dominate the degenerate damping terms in the system.
By assumption, H (0) > 0. Also, the energy identity (1.7) implies that
Z Z
H 0 (t ) = (|u(t )|k + |v(t )|l )|u0 (t )|m+1 dx + (|v(t )|θ + |u(t )|ρ )|v 0 (t )|r +1 dx ≥ 0. (4.3)
Ω Ω

Therefore, by recalling (1.2), one has


 
p+1 p+1
0 < H (0) ≤ H (t ) ≤ G(t ) ≤ c1 ku(t )kp+1 + kv(t )kp+1 , (4.4)

for t ∈ [0, T ). In addition, (1.2) yields

c0 (ku(t )kp+1 + kv(t )kp+1 ) ≤ G(t ), 0 ≤ t < T. (4.5)

Throughout the proof, we fix


p − (k + m) p − (l + m) p − (ρ + r ) p − (θ + r )
 
p−1
0 < α < min , , , , .
(m + 1)(p + 1) (m + 1)(p + 1) (r + 1)(p + 1) (r + 1)(p + 1) 2(p + 1)
In particular, 0 < α < 1
2
. In order to simplify the notation, we introduce the following constants:

p−1 p−(k+m) p−1 p−(l+m)


δ1 = H (0) (m+1)(p+1) , δ2 = H (0) (m+1)(p+1) ,
8 8
p−1 p−(ρ+r ) p−1 p−(θ+r )
δ3 = H (0) (r +1)(p+1) , δ4 = H (0) (r +1)(p+1) , (4.6)
8 8
and
k+m+1
− (p+1)(m+1)
p−(k+m) − (p+l+1m +1
)(m+1)
p−(l+m)
K1 = c0 |Ω | (p+1)(m+1) , K2 = c0 |Ω | (p+1)(m+1) ,
ρ+r +1 p−(ρ+r ) θ +r +1 p−(θ+r )
− (p+1)(r +1)
− (p+ 1)(r +1)
K3 = c0 |Ω | (p+1)(r +1) , K4 = c0 |Ω | (p+1)(r +1) . (4.7)
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2679

where c0 is the constant that appeared in (4.5) and |Ω | denotes the Lebesgue measure of Ω . Since 0 < α < 1
2
, we can choose
0 <  ≤ 1 small enough so that
 m+1 1 m+1 −1
−m k+m−p l+m−p
+α +α
A := 1 − α − 2 K1 m δ1 H (0) (m+1)(p+1) + K2 m δ2m H (0) (m+1)(p+1)

r +1
−1 ρ+r −p r +1
−1 θ +r −p
+α +α
+ K3 r δ3 r H (0) (r +1)(p+1) + K4 r δ4 r H (0) (r +1)(p+1) ≥ 0. (4.8)

We may need to adjust  again.


In the remainder of the proof, most generic constants will be denoted by c , C0 , C , . . ., they may depend on various
parameters; but they are totally independent from  and the initial data, and they may change from line to line.
We note here that the smallness condition on  in (4.8) implies that
h k+m−p l+m−p ρ+r −p θ +r −p i
+α +α +α +α
1 − α ≥ C  H (0) m(p+1) + H (0) m(p+1) + H (0) r (p+1) + H (0) r (p+1)

≥ C  H (0)−ζ , (4.9)
where ζ > 0 is defined by

p − (k + m) p − (l + m) p − (ρ + r ) p − (θ + r )
  
 −α + max , , , ; if H (0) ≥ 1,
 m(p + 1) m(p + 1) r (p + 1) r (p + 1)


ζ = (4.10)
p − (k + m) p − (l + m) p − (ρ + r ) p − (θ + r )

, , , if 0 < H (0) < 1.

−α + min
 ;
m(p + 1) m(p + 1) r (p + 1) r (p + 1)
ζ
Therefore, (4.8) will be valid provided we choose  = cH (0) for a sufficiently small generic constant c > 0. Hence,  is
chosen as
 = min 1, cH (0)ζ

(4.11)
where our interest lies in the case when H (0) > 0 is sufficiently small.
Now, from the definition of N (t ) and the given regularity of (u, v), we have
Z
N 0 (t ) = 2 [u0 (t )u(t ) + v 0 (t )v(t )]dx. (4.12)

In addition, by recalling (1.5)–(1.6), one has
d d
0
hu00 (t ), φi = hu0 (t ), φi = u (t ), φ L2 (Ω ) = − h∇ u(t ), ∇φiL2 (Ω )

dt dt
− [|u(t )| + |v(t )|l ]|u0 (t )|m−1 u0 (t ), φ L2 (Ω ) + hf1 (u(t ), v(t )), φiL2 (Ω ) ,
k


(4.13)
d 0 d
0
hv 00 (t ), ψi = hv (t ), ψi = v (t ), ψ L2 (Ω ) = − h∇v(t ), ∇ψiL2 (Ω )

dt dt
− [|u(t )|θ + |v(t )|ρ ]|v 0 (t )|r −1 v 0 (t ), ψ L2 (Ω ) + hf2 (u(t ), v(t )), ψiL2 (Ω ) ,


(4.14)

for all φ, ψ ∈ H01 (Ω ), and where in (4.13)–(4.14) and later, h., .i denotes the standard duality pairing between H −1 (Ω ) and
H01 (Ω ).
It follows from the regularity of u and a standard result in analysis (for example, see [21], Lemma 1.2, page 260) that
Z
d 2
u0 (t )u(t )dx = u0 (t ) 2 + hu00 (t ), u(t )i,

(4.15)
dt Ω
and so, by using (4.13), we have
Z
d 2 2
u0 (t )u(t )dx = u0 (t ) 2 + hu00 (t ), u(t )i = u0 (t ) 2 − k∇ u(t )k22

dt Ω
Z Z
− [|u(t )| + |v(t )| ]|u (t )|
k l 0 m−1 0
u (t )u(t )dx + f1 (u(t ), v(t ))u(t )dx. (4.16)
Ω Ω
Similarly,
Z
d 2 2
v 0 (t )v(t )dx = v 0 (t ) 2 + hv 00 (t ), v(t )i = v 0 (t ) 2 − k∇v(t )k22

dt Ω
Z Z
− [|u(t )|θ + |v(t )|ρ ]|v 0 (t )|r −1 v 0 (t )v(t )dx + f2 (u(t ), v(t ))v(t )dx. (4.17)
Ω Ω
2680 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

Therefore, by using (4.12), (4.16)–(4.17) and recalling (1.3), we obtain


 
2 2
N 00 (t ) = 2 u0 (t ) 2 + v 0 (t ) 2 − 2 k∇ u(t )k22 + k∇v(t )k22

Z Z
−2 [|u(t )| + |v(t )| ]|u (t )|
k l 0 m −1 0
u (t )u(t )dx − 2 [|u(t )|θ + |v(t )|ρ ]|v 0 (t )|r −1 v 0 (t )v(t )dx
Ω Ω
Z
+ 2(p + 1) F (u(t ), v(t ))dx. (4.18)

As in [15,17,10,13] we let

Y (t ) = H (t )1−α +  N 0 (t ), (4.19)
where 0 <  ≤ 1 is as chosen in (4.8). It follows from (4.18)–(4.19) that
 
2 2
Y 0 (t ) = (1 − α)H (t )−α H 0 (t ) + 2 u0 (t ) 2 + v 0 (t ) 2

Z
− 2 k∇ u(t )k + k∇v(t )k 2 2
− 2 (|u(t )|k + |v(t )|l )|u0 (t )|m−1 u0 (t )u(t )dx

2 2

Z
− 2 (|u(t )|θ + |v(t )|ρ )|v 0 (t )|r −1 v 0 (t )v(t )dx + 2(p + 1)G(t ). (4.20)

Now, the definition of H (t ) in (4.2) implies


 2 2 
k∇ u(t )k22 + k∇v(t )k22 = −2H (t ) − u0 (t ) 2 + v 0 (t ) 2 + 2G(t ). (4.21)

Therefore, (4.20)–(4.21) yield


 
2 2
Y 0 (t ) = (1 − α)H (t )−α H 0 (t ) + 4 u0 (t ) 2 + v 0 (t ) 2 + 4 H (t )

Z
+ 2(p − 1)G(t ) − 2 (|u(t )|k + |v(t )|l )|u0 (t )|m−1 u0 (t )u(t )dx

Z
− 2 (|u(t )|θ + |v(t )|ρ )|v 0 (t )|r −1 v 0 (t )v(t )dx. (4.22)

We estimate the last two terms that involve damping in (4.22) as follows.
Z Z Z
(|u(t )|k + |v(t )|l )|u0 (t )|m−1 u0 (t )u(t )dx ≤ ( )| ( )| |v(t )|l |u(t )||u0 (t )|m dx := I1 + I2 . (4.23)
k+1 0 m
| u t | u t dx +
Ω Ω Ω

By using Hölder’s Inequality and the assumption that p > k + m, we have
Z
km km
I1 = |u(t )|k+1− m+1 |u(t )| m+1 |u0 (t )|m dx

Z  mm+1 Z  m+1 1
≤ |u(t )|k |u0 (t )|m+1 dx |u(t )|k+m+1 dx
Ω Ω
m k+m+1
≤ H 0 (t ) m+1 ku(t )kk+m+1 m+1

m k+m+1 p−(k+m)
≤ H 0 (t ) m+1 ku(t )kp+m1+1 |Ω | (p+1)(m+1) . (4.24)

By recalling 0 < H (0) ≤ H (t ) ≤ G(t ), k + m − p < 0, and (4.5); then it follows from Young’s Inequality that
p−(k+m) k+m+1
− (p+1)(m+1)
k+m+1 m
I1 ≤ |Ω | (p+1)(m+1) c0 G(t ) (p+1)(m+1) H 0 (t ) m+1
k+m−p 1 m
= K1 G(t ) (p+1)(m+1) G(t ) m+1 H 0 (t ) m+1
k+m−p
 1 m

≤ H (t ) (p+1)(m+1) K1 G(t ) m+1 H 0 (t ) m+1
 m+1

k+m−p −1
≤ H (t ) (p+1)(m+1) δ1 G(t ) + K1 m δ1 m H 0 (t )
m+1
k+m−p −1 k+m−p
≤ δ1 H (0) (p+1)(m+1) G(t ) + K1 m δ1 m H (t ) (p+1)(m+1) H 0 (t ). (4.25)
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2681

By noting that (p+1)(m+1) + α < 0, implied by the definition of α , we obtain


k+m−p

m+1
k+m−p −1 k+m−p

I1 ≤ δ1 H (0) (p+1)(m+1) G(t ) + K1 m
δ1 m H (t ) (p+1)(m+1) H (t )−α H 0 (t )
m+1
k+m−p −1 k+m−p

≤ δ1 H (0) (p+1)(m+1) G(t ) + K1 m δ1 m H (0) (p+1)(m+1) H (t )−α H 0 (t ). (4.26)

As for I2 , we also use Hölder’s Inequality with a slight adjustment as follows:


Z
I2 = |v(t )|l |u0 (t )|m |u(t )|dx

Z n on o
lm lm
= |v(t )| m+1 |u0 (t )|m |v(t )|l− m+1 |u(t )| dx

Z  mm+1 Z  m+1 1
≤ |v(t )| |u (t )|
l 0 m +1
dx |v(t )| |u(t )|
l m+1
dx
Ω Ω
m l p−(l+m)
≤ H 0 (t ) m+1 kv(t )kpm++11 ku(t )kp+1 |Ω | (m+1)(p+1)
m l+m+1
≤ K2 H 0 (t ) m+1 G(t ) (p+1)(m+1) , (4.27)

where we have used (4.5).


Now we continue exactly as in (4.25) (replacing k with l) to obtain
m+1
l+m−p −1 l+m−p

I2 ≤ δ2 H (0) (p+1)(m+1) G(t ) + K2 m
δ2 m H (0) (p+1)(m+1) H (t )−α H 0 (t ). (4.28)

The last term in (4.22) is estimated in a similar way:


Z Z Z
(|u(t )|θ + |v(t )|ρ )|v 0 (t )|r −1 v 0 (t )v(t )dx ≤ ( )|θ 0
( )| )| |v(t )|ρ+1 |v 0 (t )|r dx := J1 + J2 , (4.29)
r
|u t |v t |v(t dx +
Ω Ω Ω

where J2 , J1 are estimated as we have done for I1 and I2 ; respectively, with replacing k by ρ , m by r, θ by l, u by v , and v by u.
Indeed, we have
ρ+r −p r +1
−1 ρ+r −p

J2 ≤ δ3 H (0) (p+1)(r +1) G(t ) + K3 r
δ3 r H (0) (p+1)(r +1) H (t )−α H 0 (t ), (4.30)

and
θ +r −p r +1
−1 θ +r −p

J1 ≤ δ4 H (0) (p+1)(r +1) G(t ) + K4 r
δ4 r H (0) (p+1)(r +1) H (t )−α H 0 (t ). (4.31)

It follows from (4.22)–(4.23), (4.26), (4.28)–(4.31) that


 
2 2
Y 0 (t ) ≥ (1 − α)H (t )−α H 0 (t ) + 4 u0 (t ) 2 + v 0 (t ) 2 + 4 H (t ) + 2(p − 1)G(t ) − 2[I1 + I2 + J2 + J1 ]

 2 2 
≥ AH (t )−α H 0 (t ) + 4 u0 (t ) 2 + v 0 (t ) 2 + 4 H (t )
h n k+m−p l+m−p ρ+r −p θ +r −p oi
+ 2 G(t ) (p − 1) − δ1 H (0) (p+1)(m+1) + δ2 H (0) (p+1)(m+1) + δ3 H (0) (p+1)(r +1) + δ4 H (0) (p+1)(r +1) , (4.32)

where A is defined in (4.8). By recalling the definitions of δ1 , . . . , δ4 , we note that

k+m−p l+m−p ρ+r −p θ +r −p p−1


δ1 H (0) (p+1)(m+1) + δ2 H (0) (p+1)(m+1) + δ3 H (0) (p+1)(r +1) + δ4 H (0) (p+1)(r +1) = , (4.33)
2
and since A ≥ 0, then it follows from (4.32) that
 
2 2
Y 0 (t ) ≥ 4 H (t ) + 4 u0 (t ) 2 + v 0 (t ) 2 + (p − 1)G(t )

h 2 2 i
+1 p+1
≥ C  H (t ) + u0 (t ) 2 + v 0 (t ) 2 + ku(t )kpp+ 1 + kv(t )kp+1 ,

(4.34)

for t ∈ [0, T ) and where C > 0 is a constant that does not depend on  . In particular (4.34) shows that Y (t ) is increasing on
[0, T ), with

Y (t ) = H (t )1−α +  N 0 (t ) ≥ H (0)1−α +  N 0 (0). (4.35)


2682 M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683

If N 0 (0) ≥ 0, then no further condition on  is needed. However, if N 0 (0) < 0, then we further adjust  so that 0 <  ≤
(0) 1−α
− H2N 0 (0) . In any case, one has

1
Y (t ) ≥ H (0)1−α > 0 for t ∈ [0, T ). (4.36)
2
Finally, we prove that

Y 0 (t ) ≥  1+σ CY (t )η for t ∈ [0, T ), (4.37)


where
 
1 1 2
η := , σ := 1− ,
1−α ζ (1 − 2α)(p + 1)
and C > 0 is a generic constant. It is important to note that the definition of α implies that 1 < η < 2 and σ > 0.
If N 0 (t ) ≤ 0 for some t ∈ [0, T ), then for such values of t we have

Y (t )η = [H (t )1−α +  N 0 (t )]η ≤ H (t ), (4.38)


and in this case, (4.34) and (4.38) yield

Y 0 (t ) ≥ C  H (t ) ≥ C  1+σ H (t ) ≥ C  1+σ Y (t )η . (4.39)


Hence, (4.37) holds for all t ∈ [0, T ) for which N (t ) ≤ 0. However, if t ∈ [0, T ) is such that N (t ) > 0, then (4.37) is still
0 0

valid, but with some more work. First we note that

Y (t )η ≤ 2η−1 [H (t ) + N 0 (t )η ]. (4.40)
η
We estimate N (t ) as follows. By using Hölder’s and Young’s inequalities and noting that 1 < η < 2, we have
0
Z η
N 0 (t )η = 2η (u(t )u0 (t ) + v(t )v 0 (t ))dx


2η ku(t )k2 u0 (t ) 2 + kv(t )k2 v 0 (t ) 2



C ku(t )kp+1 u0 (t ) 2 + kv(t )kp+1 v 0 (t ) 2


η η η η 
C ku(t )kp+1 u0 (t ) 2 + kv(t )kp+1 v 0 (t ) 2


2η 2η
 
2 2
C ku(t )kp+1 + u0 (t ) 2 + kv(t )kp+1 + v 0 (t ) 2 .
2−η
2−η

≤ (4.41)

From the definition of α , namely, α < 2(p+1) , it is easy to see that


p−1

2η 2
−1= − 1 < 0.
(2 − η)(p + 1) (1 − 2α)(p + 1)
Therefore, by recalling (4.4) we have
2η  2η 2η
1 (2−η)(p+1)

−1
ku(t )kp2+−η1 = ku(t )kpp+
+1 ≤ CG(t ) G(t ) (2−η)(p+1)

−1
≤ CG(t ) H (0) (2−η)(p+1) ≤ C  −σ G(t ), (4.42)
where we have used (4.9) and the definition of σ . Similarly,

kv(t )kp2+−η1 ≤ C  −σ G(t ). (4.43)


It follows from (4.41)–(4.43) that
 
N 0 (t )η ≤ C  −σ u0 (t ) 2 + v 0 (t ) 2 + G(t ) .
2 2

(4.44)

Finally, (4.34), (4.40)–(4.41) and (4.44) allow us to conclude that


h 2 2 i
Y 0 (t ) ≥ C  H (t ) + u0 (t ) 2 + v 0 (t ) 2 + G(t )

≥ C  1+σ Y (t )η , (4.45)
M.A. Rammaha, S. Sakuntasathien / Nonlinear Analysis 72 (2010) 2658–2683 2683

for all values of t ∈ [0, T ) for which N 0 (t ) > 0. Hence, (4.37) is valid and therefore, Y (t ) blows up in finite time T , where

T < C  −1−σ Y (0)−α/(1−α) . (4.46)


In addition, (4.36) and (4.46) yield the following upper bound for the life span of the solution
−α/(1−α)
T < C  −1−σ H (0)1−α +  N 0 (0) ≤ C  −1−σ H (0)−α .

(4.47)
Finally, by recalling (4.11) and the special case when H (0) > 0 is sufficiently small, we have

T < CH (0)−[α+ζ (1+σ )] .  (4.48)

References

[1] M. Reed, Abstract Non Linear Wave equations, Lecture Notes in Mathematics, Springer-Verlag, 1976.
[2] R.T. Glassey, Blow-up theorems for nonlinear wave equations, Math. Z. 132 (1973) 183–203.
[3] H.A. Levine, Instability and nonexistence of global solutions of nonlinear wave equations of the form Putt = Au + F (u), Trans. Amer. Math. Soc. 192
(1974) 1–21.
[4] L.E. Payne, D. Sattinger, Saddle points and instability of nonlinear hyperbolic equations, Israel Math. J. 22 (1981) 273–303.
[5] H. Tsutsumi, On solutions of semilinear differential equations in a Hilbert space, Math. Japonicea 17 (1972) 173–193.
[6] K. Agre, M.A. Rammaha, Global solutions to boundary value problems for a nonlinear wave equation in high space dimensions, Differential Integral
Equations 14 (11) (2001) 1315–1331.
[7] V. Barbu, Nonlinear Differential Equations in Banach Spaces, Nordhoff, 1976.
[8] Dang Dinh Ang, A. Pham Ngoc Dinh, Mixed problem for some semi-linear wave equation with a nonhomogeneous condition, Nonlinear Anal. MA 12
(1988) 581–592.
[9] A. Haraux, Nonlinear Evolution Equations—Global Behaviour of Solutions, Springer Verlag, 1981.
[10] V. Georgiev, G. Todorova, Existence of a solution of the wave equation with nonlinear damping and source terms, J. Differential Equations 109 (1994)
295–308.
[11] H.A. Levine, J. Serrin, Global nonexistence theorems for quasilinear evolution equations with dissipation, Arch. Ration. Mech. Anal. 137 (1997) 341–361.
[12] H.A. Levine, S.R. Park, J.M. Serrin, Global existence and global nonexistence of solutions of the Cauchy problem for a nonlinearly damped wave equation,
J. Math. Anal. Appl. 228 (1998) 181–205.
[13] D.R. Pitts, M.A. Rammaha, Global existence and non-existence theorems for nonlinear wave equations, Indiana Univ. Math. J. 51 (6) (2002) 1479–1509.
[14] M.A. Rammaha, T.A. Strei, Global existence and nonexistence for nonlinear wave equations with damping and source terms, Trans. Amer. Math. Soc.
354 (9) (2002) 3621–3637.
[15] K. Agre, M.A. Rammaha, Systems of nonlinear wave equations with damping and source terms, Differential Integral Equations 19 (11) (2006)
1235–1270.
[16] C.O. Alves, M.M. Cavalcanti, V.N. Domingos, M. Rammaha, D. Toundykov, On existence, uniform decay rates and blow up for solutions of systems of
nonlinear wave equations with damping and source terms, Discrete Contin. Dyn. Syst-S 2 (2009) 583–608.
[17] V. Barbu, I. Lasiecka, M. Rammaha, On nonlinear wave equations with degenerate damping and source terms, Trans. Amer. Math. Soc. 357 (2005)
2571–2611.
[18] V. Lakshmikantham, S. Leela, Differential and Integral Inequalities: Theory and Applications, Vol I: Ordinary Differential Equations, Academic Press,
New York, 1969.
[19] J.L. Lions, E. Magenes, Non-Homogeneous Boundary Value Problems and Applications I, II, Springer-Verlag, New York, Heidelberg, Berlin, 1972.
[20] H. Koch, I. Lasiecka, Hadamard wellposedness of weak solutions in nonlinear dynamic elasticity-full Von Karman systems, in: Evolution Equations,
Semigroups and Functional Analysis, vol. 50, Birkhauser, 2002, pp. 197–211.
[21] R. Temam, Navier–Stokes Equations, Theory and Numerical Analysis, North-Holland, 1984.
[22] V. Barbu, I. Lasiecka, M. Rammaha, Existence and uniqueness of solutions to wave equations with nonlinear degenerate damping and source terms,
Control Cybernet. 34 (3) (2005) 665–687.
[23] V. Barbu, I. Lasiecka, M. Rammaha, Blow-up of generalized solutions to wave equations with nonlinear degenerate damping and source terms, Indiana
Univ. Math. J. 56 (3) (2007) 995–1022.

Further reading

[1] R.A. Adams, Sobolev Spaces, Academic Press, New York, 1975.

You might also like