Introduction
This paper is devoted to the following non-autonomous weakly damped Korteweg-de Vries-Burgers (KdV-Burgers for short) equation: for τ ∈ R \tau\in\mathbb{R} τ ∈ R ,
{ u t + u x x x + u u x + λ u − μ u x x = g ( t , x ) , x ∈ R , t > τ , u ( τ , x ) = u τ ( x ) , x ∈ R , (1) \left\{
\begin{aligned}
u_t+u_{xxx}+uu_x+\lambda u-\mu u_{xx}&=g(t,x), \quad x\in\mathbb{R},\quad t>\tau,\\
u(\tau,x)&=u_\tau(x), \quad x\in\mathbb{R},
\end{aligned}
\right. \tag*{(1)} { u t + u xxx + u u x + λ u − μ u xx u ( τ , x ) = g ( t , x ) , x ∈ R , t > τ , = u τ ( x ) , x ∈ R , (1)
where λ > 0 \lambda>0 λ > 0 and μ ∈ ( 0 , 1 ] \mu\in(0,1] μ ∈ ( 0 , 1 ] , the initial data u τ u_\tau u τ belongs to L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , and the time-dependent external force g ∈ W l o c 1 , 2 ( R ; L 2 ( R ) ) g\in W^{1,2}_{\mathrm{loc}}(\mathbb{R};L^2(\mathbb{R})) g ∈ W loc 1 , 2 ( R ; L 2 ( R )) satisfies
∥ g ( t ) ∥ L 2 ( R ) 2 + ∥ g ′ ( t ) ∥ L 2 ( R ) 2 ≤ M e γ ∣ t ∣ , t ∈ R , (2) \lVert g(t)\rVert_{L^2(\mathbb{R})}^2+\lVert g'(t)\rVert_{L^2(\mathbb{R})}^2\le Me^{\gamma|t|},\quad t\in\mathbb{R}, \tag*{(2)} ∥ g ( t ) ∥ L 2 ( R ) 2 + ∥ g ′ ( t ) ∥ L 2 ( R ) 2 ≤ M e γ ∣ t ∣ , t ∈ R , (2)
with γ < λ \gamma<\lambda γ < λ and M M M is independent of t t t . Note that, when μ = 0 \mu=0 μ = 0 , equation (1) becomes the well-known weakly damped forced KdV equation.
The KdV-Burgers equation has been derived as an asymptotic model for the propagation of weakly nonlinear dispersive long waves in certain physical situations when dissipative and damping effects occur, such as waves in an elastic tube, ion-sound waves damped by ion-neutron
2010 Mathematics Subject Classification. 35B41, 35K65, 35D05.
Key words and phrases. Korteweg-de Vries-Burgers equation; pullback attractor; asymptotic smoothing effect; energy equation method; splitting of the solutions. collisions, waves in a tube filled with viscous fluid, flow of a fluid containing gas bubbles, turbulence, or crystal lattice theory (see e.g. [15 , 16 , 22 , 24 ] ) and as a model for propagation of tsunamis (see [9 ] ). The well-posedness of homogeneous KdV-Burgers equations, i.e. with g ≡ 0 g \equiv0 g ≡ 0 , has been thoroughly studied (see e.g. [20 , 21 , 4 ] and references therein). The large time dynamics of KdV-Burgers in the phase spaces H 1 ( R ) H^1(\mathbb{R}) H 1 ( R ) or H 2 ( R ) H^2(\mathbb{R}) H 2 ( R ) have also been extensively investigated, see e.g. [7 DS10] for the autonomous case and [1 ] for the non-autonomous case. Concerning the phase space L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) and autonomous external force g ≡ g ( x ) g \equiv g(x) g ≡ g ( x ) , the existence, regularity and finite fractal dimension of a global attractor are obtained in [3 ] , [14 ] and [10 ] , respectively. It is also worth mentioning the related works with low regular initial data such as [Goy22, McC23]. In practice, the external force is likely to be time dependent, i.e. g ≡ g ( t , x ) g \equiv g(t,x) g ≡ g ( t , x ) , leading to the non-autonomous equation (1) . The long-time behavior of solutions this equation in phase space L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) has not been studied, and this serves as a motivation for the current paper.
The first objective of this paper is to study the existence and regularity of a pullback attractor in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) for the non-autonomous process associated to problem (1) . More precisely, we show that the pullback attractor of the associated process exists in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) and is in fact compact in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) ; thus, obtaining an asymptotic smoothing effect of solutions which generally belong only to L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . Note that the result obtained are of optimal order and holds for both the KdV-Burgers equation (i.e., when μ > 0 \mu> 0 μ > 0 ) and the weakly damped KdV equation (i.e., when μ = 0 \mu= 0 μ = 0 ); moreover, when μ = 0 \mu= 0 μ = 0 and g ≡ g ( x ) ∈ L 2 ( R ) g \equiv g(x) \in L^2(\mathbb{R}) g ≡ g ( x ) ∈ L 2 ( R ) , we recover the results of [13 ] for the autonomous weakly damped forced KdV equation. Since the initial data u τ u_\tau u τ is in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , studying well-posedness and estimates of the solutions requires Bourgain-space techniques. Furthermore, the problem under consideration is posed in the real line R \mathbb{R} R , which makes Sobolev embeddings H 1 ( R ) ↪ L 2 ( R ) H^1(\mathbb{R}) \hookrightarrow L^2(\mathbb{R}) H 1 ( R ) ↪ L 2 ( R ) no longer compact. These lead to certain difficulties for proving the pullback asymptotic compactness of solutions and therefore in studying the existence and regularity of pullback attractors for the KdV-Burgers equation. In this paper, we deal with these issues by first using the energy equation method introduced by Ball in [2 ] to get the existence of a family of bounded pullback absorbing sets in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . To obtain the higher regularity of the pullback attractor, following the general lines of the approach in [11 , 13 ] , we split the solutions into two parts possessing different frequencies, in which the first one decays exponentially in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) while the other one is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . Once the boundedness in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) of the regular part of the solution is obtained, the compactness of the pullback attractor in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) the follows by applying again the energy equation method, now for the equation u t u_t u t along trajectories lying in the pullback attractor. Delicate issues arise due to the lack of compact Sobolev embeddings which are compensated by exploiting the dispersive regularization property of the equation under study in Bourgain spaces.
The second objective of this paper is to study the upper semi-continuity of the pullback attractors { A μ } μ ∈ [ 0 , 1 ] \{\mathcal{A}_\mu\}_{\mu\in[0,1]} { A μ } μ ∈ [ 0 , 1 ] at μ = 0 \mu= 0 μ = 0 and their finite fractal dimensions in H 1 ( R ) H^1(\mathbb{R}) H 1 ( R ) . In other words, we show that when μ → 0 + \mu\to0^+ μ → 0 + , the pullback attractor A μ \mathcal{A}_\mu A μ of the KdV-Burgers equation converges to the pullback attractor A 0 \mathcal{A}_0 A 0 of the weakly damped KdV equation in the sense of the Hausdorff semi-distance in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . This result is new even in the autonomous case. This is done by showing the compactness of the pullback attractors uniformly in μ ≥ 0 \mu\geq0 μ ≥ 0 using some tail estimates technique. Finally, we show that the pullback attractor has a finite fractal dimension in H 1 ( R ) H^1(\mathbb{R}) H 1 ( R ) for any μ > 0 \mu> 0 μ > 0 . In order to do this, we prove a result for finite fractal dimensions for non-autonomous problem, which is an extension of the autonomous case in [5 Theorem 4.15] . Since this extension is well applicable for (38) on the whole line R \mathbb{R} R and avoids using finite dimensional projection, we expect that it can be used in other non-autonomous problems in unbounded domains. Furthermore, by exploiting the dissipation by the term − μ u x x -\mu u_{xx} − μ u xx , we can get our result without having to pursue higher regularity of the pullback attractors, as in [WH20], which requires to work with associated Bourgain spaces and likely imposes higher regularity of the external force.
As a final remark, we believe that the results of this paper can be extended to show that if g ( t ) ∈ H k ( R ) g(t) \in H^k(\mathbb{R}) g ( t ) ∈ H k ( R ) , for some k ∈ N k \in\mathbb{N} k ∈ N , then equation (1) generates a process in the phase space H k ( R ) H^k(\mathbb{R}) H k ( R ) which has a pullback attractor in H k ( R ) H^k(\mathbb{R}) H k ( R ) , and the pullback attractor is compact in H k + 3 ( R ) H^{k+3}(\mathbb{R}) H k + 3 ( R ) . Therefore, this paper, in a sense, completes the study of existence and regularity of pullback attractors for equation (1) in the scale of Sobolev spaces H k ( R ) H^k(\mathbb{R}) H k ( R ) for integers k ≥ 0 k \ge0 k ≥ 0 . The details are left for the interested reader.
The paper is organized as follows: In section 2 show the global existence of (1) in the phase space L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . Section 3 is devoted to the the existence and regularity of pullback attractor A μ \mathcal{A}_\mu A μ for the for the process generated by (1) . In the last two sections, we prove the upper semicontinuity of A μ \mathcal{A}_\mu A μ at μ = 0 \mu= 0 μ = 0 and their finite fractal dimensions, respectively.
Throughout this paper, we use the following notation:
The usual norm in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) is written as ∥ ⋅ ∥ \| \cdot\| ∥ ⋅ ∥ or convenience.
S ( R ) \mathcal{S}(\mathbb{R}) S ( R ) denotes the Schwartz space on R \mathbb{R} R and S ′ ( R ) \mathcal{S}'(\mathbb{R}) S ′ ( R ) is the corresponding space of distributions. For f ∈ S ′ f \in\mathcal{S}' f ∈ S ′ , f ^ \widehat{f} f or F ( f ) \mathcal{F}(f) F ( f ) is its Fourier transform, i.e.,
f ^ ( ξ ) = ∫ R n e − i z ⋅ ξ f ( z ) d z . \widehat{f}(\xi) = \int_{\mathbb{R}^{n}} e^{-iz\cdot\xi} f(z)\, dz. f ( ξ ) = ∫ R n e − i z ⋅ ξ f ( z ) d z .
The usual Sobolev space is denoted by H s ( R ) H^s(\mathbb{R}) H s ( R ) for s ∈ R s \in\mathbb{R} s ∈ R with norm
∥ u ∥ H s = ∥ ⟨ ξ ⟩ s f ^ ( ξ ) ∥ L 2 ( R ) . \|u\|_{H^s} = \|\langle\xi\rangle^s \widehat{f}(\xi)\|_{L^2(\mathbb{R})}. ∥ u ∥ H s = ∥ ⟨ ξ ⟩ s f ( ξ ) ∥ L 2 ( R ) .
where ⟨ ξ ⟩ = 1 + ∣ ξ ∣ 2 \langle\xi\rangle= \sqrt{1+|\xi|^2} ⟨ ξ ⟩ = 1 + ∣ ξ ∣ 2 .
The Bourgain space-time function space X s , b X^{s,b} X s , b for s , b ∈ R s,b \in\mathbb{R} s , b ∈ R is defined as the completion of the Schwartz space S ( R 2 ) \mathcal{S}(\mathbb{R}^2) S ( R 2 ) with respect to the norm
∥ u ∥ X s , b = ∥ ⟨ i ( τ − ξ 3 ) + ξ 2 ⟩ b ⟨ ξ ⟩ s u ^ ( τ , ξ ) ∥ L 2 ( R 2 ) = ∥ ⟨ ∣ τ − ξ 3 ∣ + ∣ ξ 2 ∣ ⟩ b ⟨ ξ ⟩ s u ^ ( τ , ξ ) ∥ L 2 ( R 2 ) . (3) \|u\|_{X^{s,b}} = \|\langle i(\tau-\xi^3)+\xi^2\rangle^b\langle\xi\rangle^s\widehat{u}(\tau,\xi)\|_{L^2(\mathbb{R}^2)}
= \|\langle|\tau-\xi^3|+|\xi^2|\rangle^b\langle\xi\rangle^s\widehat{u}(\tau,\xi)\|_{L^2(\mathbb{R}^2)}. \tag*{(3)} ∥ u ∥ X s , b = ∥ ⟨ i ( τ − ξ 3 ) + ξ 2 ⟩ b ⟨ ξ ⟩ s u ( τ , ξ ) ∥ L 2 ( R 2 ) = ∥ ⟨ ∣ τ − ξ 3 ∣ + ∣ ξ 2 ∣ ⟩ b ⟨ ξ ⟩ s u ( τ , ξ ) ∥ L 2 ( R 2 ) . (3)
We have the continuous embedding
X s , b ⊂ C b ( R ; H s ( R x ) ) . X^{s,b} \subset C_b(\mathbb{R};H^s(\mathbb{R}_x)). X s , b ⊂ C b ( R ; H s ( R x )) .
for any s ∈ R s \in\mathbb{R} s ∈ R and b > 1 / 2 b > 1/2 b > 1/2 . For an interval I ⊆ R I \subseteq\mathbb{R} I ⊆ R , we define the space X I s , b X_I^{s,b} X I s , b with norm
∥ u ∥ X I s , b = inf w ∈ X s , b { ∥ w ∥ X s , b : w ≡ u on I } . \|u\|_{X_I^{s,b}} = \inf_{w\in X^{s,b}}\{\|w\|_{X^{s,b}}: w \equiv u \text{ on } I\}. ∥ u ∥ X I s , b = w ∈ X s , b inf { ∥ w ∥ X s , b : w ≡ u on I } .
For a Banach space ( X , ∥ ⋅ ∥ ) (X,\|\cdot\|) ( X , ∥ ⋅ ∥ ) , B ( X ) \mathcal{B}(X) B ( X ) denotes the set of bounded subsets of X X X , and for A , B ⊆ X A,B \subseteq X A , B ⊆ X , the Hausdorff semi-distance between A A A and B B B is defined by
dist X ( A , B ) = sup x ∈ A inf y ∈ B ∥ x − y ∥ . \operatorname{dist}_{X}(A,B) = \sup_{x\in A}\inf_{y\in B}\|x-y\|. dist X ( A , B ) = x ∈ A sup y ∈ B inf ∥ x − y ∥.
Global existence of solutions
We start with some results on the local well-posedness of the KdV-Burgers equation in spaces of low regularity. The results presented here follow from [19 ] . Denote by W W W the group associated with the free evolution of (1) , i.e. u t + u x x x − μ u x x = 0 u_t+u_{xxx}-\mu u_{xx}=0 u t + u xxx − μ u xx = 0 ,
F ( W ( t ) φ ) ( ξ ) = exp ( − μ ∣ ξ ∣ 2 t + i ξ 3 t ) φ ^ ( ξ ) , φ ∈ S ′ ( R ) , ∀ t ≥ 0. \mathcal{F}(W(t)\varphi)(\xi)=\exp(-\mu|\xi|^2t+i\xi^3t)\widehat{\varphi}(\xi),\qquad\varphi\in\mathcal{S}'(\mathbb{R}),\ \forall t\ge0. F ( W ( t ) φ ) ( ξ ) = exp ( − μ ∣ ξ ∣ 2 t + i ξ 3 t ) φ ( ξ ) , φ ∈ S ′ ( R ) , ∀ t ≥ 0.
and we extend W W W to the whole real axis by
F ( W ( t ) φ ) ( ξ ) = exp ( − μ ∣ ξ ∣ 2 ∣ t ∣ + i ξ 3 t ) φ ^ ( ξ ) , φ ∈ S ′ ( R ) , ∀ t ∈ R . \mathcal{F}(W(t)\varphi)(\xi)=\exp(-\mu|\xi|^2|t|+i\xi^3t)\widehat{\varphi}(\xi),\qquad\varphi\in\mathcal{S}'(\mathbb{R}),\ \forall t\in\mathbb{R}. F ( W ( t ) φ ) ( ξ ) = exp ( − μ ∣ ξ ∣ 2 ∣ t ∣ + i ξ 3 t ) φ ( ξ ) , φ ∈ S ′ ( R ) , ∀ t ∈ R .
Fix 0 ≤ s ≤ 23 / 48 0 \le s \le23/48 0 ≤ s ≤ 23/48 and 1 / 2 < b < 25 / 48 1/2 < b < 25/48 1/2 < b < 25/48 . Assume that λ > 0 \lambda> 0 λ > 0 and g = g ( x , t ) ∈ X [ τ − T g , τ + T g ] − s , b − 1 g = g(x,t) \in X^{-s,b-1}_{[\tau-T_g,\tau+T_g]} g = g ( x , t ) ∈ X [ τ − T g , τ + T g ] − s , b − 1 for some T g > 0 T_g > 0 T g > 0 . We look for a fixed-point of the mapping
M ( u ) ( t ) = ψ T g ( t ) ( W ( t ) u τ − χ R τ ( t ) 2 ∫ τ t W ( t − s ) [ ( ψ T g ( s ) u ( s ) ) x 2 + 2 λ ψ T g ( s ) u ( s ) − 2 ψ T g ( s ) g ] d s ) (4) \mathcal{M}(u)(t)=\psi_{T_g}(t)\left(W(t)u_\tau-\frac{\chi_{\mathbb{R}_\tau}(t)}{2}\int_\tau^t W(t-s)\left[(\psi_{T_g}(s)u(s))^2_x+2\lambda\psi_{T_g}(s)u(s)-2\psi_{T_g}(s)g\right]\,ds\right) \tag*{(4)} M ( u ) ( t ) = ψ T g ( t ) ( W ( t ) u τ − 2 χ R τ ( t ) ∫ τ t W ( t − s ) [ ( ψ T g ( s ) u ( s ) ) x 2 + 2 λ ψ T g ( s ) u ( s ) − 2 ψ T g ( s ) g ] d s ) (4)
where R τ = [ τ , + ∞ ) \mathbb{R}_\tau=[\tau,+\infty) R τ = [ τ , + ∞ ) and χ R τ \chi_{\mathbb{R}_\tau} χ R τ is its characteristic function, ψ \psi ψ is a time cut-off function satisfying
ψ ∈ C 0 ∞ ( R ) , supp ψ ⊆ [ τ − 1 , τ + 1 ] , ψ ≡ 1 on [ τ − 1 2 , τ + 1 2 ] , \psi\in C^\infty_0(\mathbb{R}),\quad\operatorname{supp}\psi\subseteq[\tau-1,\tau+1],\quad\psi\equiv1\text{ on }\left[\tau-\frac{1}{2},\tau+\frac{1}{2}\right], ψ ∈ C 0 ∞ ( R ) , supp ψ ⊆ [ τ − 1 , τ + 1 ] , ψ ≡ 1 on [ τ − 2 1 , τ + 2 1 ] ,
and ψ T ( ⋅ ) = ψ ( ⋅ / T ) \psi_T(\cdot)=\psi(\cdot/T) ψ T ( ⋅ ) = ψ ( ⋅ / T ) . Indeed, if u u u solves (4) on [ τ − T , τ + T ] [\tau-T,\tau+T] [ τ − T , τ + T ] , then u u u is a solution of (1) on [ τ , τ + T / 2 ] [\tau,\tau+T/2] [ τ , τ + T /2 ] .
Theorem 2.1 ([13 ] ). Let 0 ≤ s ≤ 23 / 48 0 \le s \le23/48 0 ≤ s ≤ 23/48 and 1 / 2 < b < 25 / 48 1/2 < b < 25/48 1/2 < b < 25/48 . For any u τ ∈ L 2 ( R ) u_\tau\in L^2(\mathbb{R}) u τ ∈ L 2 ( R ) , there exists T > 0 T > 0 T > 0 such that the problem (1) is uniquely solvable in X [ τ , τ + T ] − s , b X^{-s,b}_{[\tau,\tau+T]} X [ τ , τ + T ] − s , b . In particulars, the solution u ∈ C ( [ τ , T ] ; L 2 ( R ) ) u \in C([\tau,T];L^2(\mathbb{R})) u ∈ C ([ τ , T ] ; L 2 ( R )) . Moreover, this solution can be extended globally on R τ \mathbb{R}_\tau R τ .
Remark 2.2. It is in fact possible to show sharper local well-posedness results of (1) , e.g. (1) is well-posed in H s ( R ) H^s(\mathbb{R}) H s ( R ) for any s > − 1 s>-1 s > − 1 . The interested reader is referred to [19 ] for more details.
Proof. The proof of local well-posedness follows from [19 ] with slight modifications, so we omit it here. To show the global existence, we take the inner product of (1) with 2 u ( t ) 2u(t) 2 u ( t ) in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) to get
d d t ∥ u ( t ) ∥ 2 + 2 λ ∥ u ( t ) ∥ 2 + 2 μ ∥ u x ( t ) ∥ 2 = 2 ( g ( t ) , u ( t ) ) . (5) \frac{d}{dt}\lVert u(t)\rVert^2+2\lambda\lVert u(t)\rVert^2+2\mu\lVert u_x(t)\rVert^2=2(g(t),u(t)). \tag*{(5)} d t d ∥ u ( t ) ∥ 2 + 2 λ ∥ u ( t ) ∥ 2 + 2 μ ∥ u x ( t ) ∥ 2 = 2 ( g ( t ) , u ( t )) . (5)
Applying the Cauchy and Young’s inequalities, we see that
d d t ∥ u ( t ) ∥ 2 + 3 λ 2 ∥ u ( t ) ∥ 2 + 2 μ ∥ u x ( t ) ∥ 2 ≤ 1 2 λ ∥ g ( t ) ∥ 2 . \frac{d}{dt}\lVert u(t)\rVert^2+\frac{3\lambda}{2}\lVert u(t)\rVert^2+2\mu\lVert u_x(t)\rVert^2\le\frac{1}{2\lambda}\lVert g(t)\rVert^2. d t d ∥ u ( t ) ∥ 2 + 2 3 λ ∥ u ( t ) ∥ 2 + 2 μ ∥ u x ( t ) ∥ 2 ≤ 2 λ 1 ∥ g ( t ) ∥ 2 .
Therefore,
∥ u ( t ) ∥ 2 + ∫ τ t e − λ ( t − s ) ( λ 2 ∥ u ( s ) ∥ 2 + 2 μ ∥ u x ( s ) ∥ 2 ) d s ≤ e − λ ( t − τ ) ∥ u τ ∥ 2 + e − λ t 2 λ ∫ τ t e λ s ∥ g ( s ) ∥ 2 d s . (6) \lVert u(t)\rVert^2+\int_\tau^t e^{-\lambda(t-s)}\left(\frac{\lambda}{2}\lVert u(s)\rVert^2+2\mu\lVert u_x(s)\rVert^2\right)\,ds
\le e^{-\lambda(t-\tau)}\lVert u_\tau\rVert^2+\frac{e^{-\lambda t}}{2\lambda}\int_\tau^t e^{\lambda s}\lVert g(s)\rVert^2\,ds. \tag*{(6)} ∥ u ( t ) ∥ 2 + ∫ τ t e − λ ( t − s ) ( 2 λ ∥ u ( s ) ∥ 2 + 2 μ ∥ u x ( s ) ∥ 2 ) d s ≤ e − λ ( t − τ ) ∥ u τ ∥ 2 + 2 λ e − λ t ∫ τ t e λ s ∥ g ( s ) ∥ 2 d s . (6)
This implies that the local solution can be extended globally on R τ \mathbb{R}_\tau R τ .
Existence and regularity of pullback attractors
This section is devoted to the existence and regularity of pullback attractors for KdV-Burgers equation, that is (1) with μ > 0 \mu>0 μ > 0 . We first recall an useful bilinear estimate which is crucial in proving the regularity of the attractors. Next, the existence of the pullback attractor in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) is shown in subsection 3.1 3.1 3.1 by an energy equation estimate. The boundedness and compactness of the attractor in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) are proved in subsections 3.2 3.2 3.2 and 3.3 3.3 3.3 respectively.
We first briefly recall the basic concepts concerning pullback attractors fo non-autonomous dynamical systems. For more details, the interested reader is referred to the book [6 ] . Let { U ( t , τ ) : t ≥ τ , τ ∈ R } \{U(t,\tau):t\ge\tau,\ \tau\in\mathbb{R}\} { U ( t , τ ) : t ≥ τ , τ ∈ R } be a process in X X X , i.e., U ( t , τ ) : X → X U(t,\tau):X\to X U ( t , τ ) : X → X satisfy U ( τ , τ ) = Id X U(\tau,\tau)=\operatorname{Id}_X U ( τ , τ ) = Id X and U ( t , s ) U ( s , τ ) = U ( t , τ ) U(t,s)U(s,\tau)=U(t,\tau) U ( t , s ) U ( s , τ ) = U ( t , τ ) for all t ≥ s ≥ τ t\ge s\ge\tau t ≥ s ≥ τ , τ ∈ R \tau\in\mathbb{R} τ ∈ R . This process is called continuous (weakly continuous) if for each t > τ t>\tau t > τ the mapping x ↦ U ( t , τ ) x x\mapsto U(t,\tau)x x ↦ U ( t , τ ) x is continuous (weakly continuous).
Definition 3.1. A family A = { A ( t ) : t ∈ R } ⊂ B ( X ) \mathcal{A}=\{A(t):t\in\mathbb{R}\}\subset\mathcal{B}(X) A = { A ( t ) : t ∈ R } ⊂ B ( X ) is said to be a pullback attractor for the process { U ( t , τ ) } \{U(t,\tau)\} { U ( t , τ )} if
A ( t ) A(t) A ( t ) is compact for all t ∈ R t\in\mathbb{R} t ∈ R ,
A \mathcal{A} A is invariant, i.e.,
U ( t , τ ) A ( τ ) = A ( t ) for all t ≥ τ . U(t,\tau)A(\tau)=A(t)\quad\text{for all }t\ge\tau. U ( t , τ ) A ( τ ) = A ( t ) for all t ≥ τ .
(3) A \mathcal{A} A is pullback attracting, i.e.,
lim τ → − ∞ dist X ( U ( t , τ ) D , A ( t ) ) = 0 for all D ∈ B ( X ) and all t ∈ R . (7) \lim_{\tau\to-\infty}\operatorname{dist}_{X}(U(t,\tau)D,A(t))=0\quad\text{for all }D\in\mathcal{B}(X)\text{ and all }t\in\mathbb{R}. \tag*{(7)} τ → − ∞ lim dist X ( U ( t , τ ) D , A ( t )) = 0 for all D ∈ B ( X ) and all t ∈ R . (7)
(4) If { C ( t ) : t ∈ R } \{C(t):t\in\mathbb{R}\} { C ( t ) : t ∈ R } is another family of closed pullback attracting sets, then A ( t ) ⊂ C ( t ) A(t)\subset C(t) A ( t ) ⊂ C ( t ) for all t ∈ R t\in\mathbb{R} t ∈ R .
Theorem 3.2 ([17 , 6 ] ). Let { U ( t , τ ) } \{U(t,\tau)\} { U ( t , τ )} be a process satisfying the following conditions:
(i) { U ( t , τ ) } \{U(t,\tau)\} { U ( t , τ )} is continuous or weakly continuous;
(ii) there exists a family of pullback absorbing sets B = { B ( t ) : t ∈ R } \mathcal{B}=\{B(t):t\in\mathbb{R}\} B = { B ( t ) : t ∈ R } , i.e., for any t ∈ R t\in\mathbb{R} t ∈ R and any D ∈ B ( X ) D\in\mathcal{B}(X) D ∈ B ( X ) , there is a τ 0 = τ 0 ( D , t ) ≤ t \tau_0=\tau_0(D,t)\leq t τ 0 = τ 0 ( D , t ) ≤ t such that
⋃ τ ≤ τ 0 U ( t , τ ) D ⊆ B ( t ) ; \bigcup_{\tau\leq\tau_0}U(t,\tau)D\subseteq B(t); τ ≤ τ 0 ⋃ U ( t , τ ) D ⊆ B ( t ) ; (iii) { U ( t , τ ) } \{U(t,\tau)\} { U ( t , τ )} is pullback asymptotically compact, i.e. for any t ∈ R t\in\mathbb{R} t ∈ R , any D ∈ B ( X ) D\in\mathcal{B}(X) D ∈ B ( X ) , any sequence τ n → − ∞ \tau_n\to-\infty τ n → − ∞ , and any sequence { x n } ⊂ D \{x_n\}\subset D { x n } ⊂ D , the sequence { U ( t , τ n ) x n } \{U(t,\tau_n)x_n\} { U ( t , τ n ) x n } is relatively compact.
Then { U ( t , τ ) } \{U(t,\tau)\} { U ( t , τ )} has a unique pullback attractor A = { A ( t ) : t ∈ R } \mathcal{A}=\{A(t):t\in\mathbb{R}\} A = { A ( t ) : t ∈ R } , and
A ( t ) = ⋂ s ≤ t ⋃ τ ≤ s U ( t , τ ) B ( τ ) ‾ X . A(t)=\bigcap_{s\leq t}\overline{\bigcup_{\tau\leq s}U(t,\tau)B(\tau)}^{X}. A ( t ) = s ≤ t ⋂ τ ≤ s ⋃ U ( t , τ ) B ( τ ) X . where A ‾ X \overline{A}^{X} A X denotes the closure of A A A with respect to the norm topology of X X X .
Pullback attractors in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R )
Thanks to Theorem 2.1 , for each μ ≥ 0 \mu\geq0 μ ≥ 0 , we can define a process { U μ ( t , τ ) } \{U_\mu(t,\tau)\} { U μ ( t , τ )} in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) as follows: for any u τ ∈ L 2 ( R ) u_\tau\in L^2(\mathbb{R}) u τ ∈ L 2 ( R ) , U μ ( t , τ ) u τ U_\mu(t,\tau)u_\tau U μ ( t , τ ) u τ is the solution to (1) established in Theorem 2.1 at time t > τ t>\tau t > τ with initial data u τ u_\tau u τ .
Proposition 3.3. There exists a family of pullback absorbing sets B = { B ( t ) : t ∈ R } \mathcal{B}=\{B(t):t\in\mathbb{R}\} B = { B ( t ) : t ∈ R } in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) for U μ ( t , τ ) U_\mu(t,\tau) U μ ( t , τ ) .
Proof. Denote by B ( t ) B(t) B ( t ) the ball in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) centered at origin and has the radius
r ( t ) = ( e − λ t λ ∫ − ∞ t e λ s ∥ g ( s ) ∥ 2 d s ) 1 / 2 . (8) r(t)=\left(\frac{e^{-\lambda t}}{\lambda}\int_{-\infty}^{t}e^{\lambda s}\lVert g(s)\rVert^2\,\mathrm{d}s\right)^{1/2}. \tag*{(8)} r ( t ) = ( λ e − λ t ∫ − ∞ t e λ s ∥ g ( s ) ∥ 2 d s ) 1/2 . (8)
Then the estimate (6) implies that for any bounded set B ⊂ L 2 ( R ) B\subset L^2(\mathbb{R}) B ⊂ L 2 ( R ) there exists τ 0 ∈ R \tau_0\in\mathbb{R} τ 0 ∈ R such that
U μ ( t , τ ) u τ ∈ B ( t ) ∀ τ ≤ τ 0 , ∀ u τ ∈ B . U_\mu(t,\tau)u_\tau\in B(t)\quad\forall\tau\leq\tau_0,\quad\forall u_\tau\in B. U μ ( t , τ ) u τ ∈ B ( t ) ∀ τ ≤ τ 0 , ∀ u τ ∈ B .
We prove in the next lemma the weak continuity of the process { U μ ( t , τ ) } \{U_\mu(t,\tau)\} { U μ ( t , τ )} , which is not only needed for the abstract Theorem 3.2 but also is very useful in showing the asymptotic compactness via an energy equation technique.
Lemma 3.4. The process { U μ ( t , τ ) } \{U_\mu(t,\tau)\} { U μ ( t , τ )} is weakly continuous on L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) in the sense that, if u τ ε ⇀ u τ u^\varepsilon_\tau\rightharpoonup u_\tau u τ ε ⇀ u τ in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , then U μ ( t , τ ) u τ ε ⇀ U μ ( t , τ ) u τ U_\mu(t,\tau)u^\varepsilon_\tau\rightharpoonup U_\mu(t,\tau)u_\tau U μ ( t , τ ) u τ ε ⇀ U μ ( t , τ ) u τ in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) for all t ≥ τ t\geq\tau t ≥ τ .
Proof. The proof goes similarly to [11 Section 6] . Without loss of generality, we consider the case λ = 0 \lambda=0 λ = 0 , g = 0 g=0 g = 0 , and μ = 1 \mu=1 μ = 1 . We fix N N N and define the projection P N P_N P N by
F ( P N u ) ( ξ ) = u ^ ( ξ ) χ ( ξ N ) , (9) \mathcal{F}(P_Nu)(\xi)=\widehat{u}(\xi)\chi\left(\frac{\xi}{N}\right), \tag*{(9)} F ( P N u ) ( ξ ) = u ( ξ ) χ ( N ξ ) , (9)
where χ \chi χ is the characteristic function of [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] . Let v N ε ( t ) v_N^\varepsilon(t) v N ε ( t ) be the solution of
∂ t v N ε + ∂ x x x v N ε + v N ε ∂ x v N ε − ∂ x x v N ε = 0 , (10) \partial_t v_N^\varepsilon+\partial_{xxx}v_N^\varepsilon+v_N^\varepsilon\partial_xv_N^\varepsilon-\partial_{xx}v_N^\varepsilon=0, \tag*{(10)} ∂ t v N ε + ∂ xxx v N ε + v N ε ∂ x v N ε − ∂ xx v N ε = 0 , (10)
with initial data v N ε ( τ ) = P N u τ ε v_N^\varepsilon(\tau)=P_Nu_\tau^\varepsilon v N ε ( τ ) = P N u τ ε . Since P N u τ ε P_Nu_\tau^\varepsilon P N u τ ε is smooth and
∥ P N u τ ε ∥ H 2 ( R ) ≤ K N 2 , \lVert P_Nu_\tau^\varepsilon\rVert_{H^2(\mathbb{R})}\leq KN^2, ∥ P N u τ ε ∥ H 2 ( R ) ≤ K N 2 ,
where K K K is independent of ϵ \epsilon ϵ , we conclude that v N ϵ v_N^\epsilon v N ϵ is bounded in
C ( [ τ − T , τ + T ] ; H 2 ( R ) ) ∩ C 1 ( [ τ − T , τ + T ] ; H − 1 ( R ) ) , C\left([\tau-T,\tau+T];H^2(\mathbb{R})\right)\cap C^1\left([\tau-T,\tau+T];H^{-1}(\mathbb{R})\right), C ( [ τ − T , τ + T ] ; H 2 ( R ) ) ∩ C 1 ( [ τ − T , τ + T ] ; H − 1 ( R ) ) ,
as a result of [25 ] . Thus,
v N ϵ → v N strongly in C ( [ τ − T , τ + T ] ; L l o c 2 ( R ) ) , v_N^\epsilon\to v_N\text{ strongly in }C\left([\tau-T,\tau+T];L^2_{\mathrm{loc}}(\mathbb{R})\right), v N ϵ → v N strongly in C ( [ τ − T , τ + T ] ; L loc 2 ( R ) ) ,
because H 2 ( R ) ↪ L l o c 2 ( R ) H^2(\mathbb{R})\hookrightarrow L^2_{\mathrm{loc}}(\mathbb{R}) H 2 ( R ) ↪ L loc 2 ( R ) with a compact embedding. Hence, in equation (10) we can pass to the limit to find that v N v_N v N is a solution of (10) with initial data v N ( τ ) = P N u τ v_N(\tau)=P_Nu_\tau v N ( τ ) = P N u τ . On the other hand,
v N ∈ L ∞ ( [ τ − T , τ + T ] ; H 2 ( R ) ) and ∂ t v N ∈ L ∞ ( [ τ − T , τ + T ] ; H − 1 ( R ) ) . v_N\in L^\infty\left([\tau-T,\tau+T];H^2(\mathbb{R})\right)\text{ and }\partial_tv_N\in L^\infty\left([\tau-T,\tau+T];H^{-1}(\mathbb{R})\right). v N ∈ L ∞ ( [ τ − T , τ + T ] ; H 2 ( R ) ) and ∂ t v N ∈ L ∞ ( [ τ − T , τ + T ] ; H − 1 ( R ) ) .
We conclude that v N ∈ C ( [ τ − T , τ + T ] ; H s ( R ) ) v_N\in C\left([\tau-T,\tau+T];H^s(\mathbb{R})\right) v N ∈ C ( [ τ − T , τ + T ] ; H s ( R ) ) for all s < 2 s<2 s < 2 . Now, let w N ϵ ( t ) = u ϵ ( t ) − v N ϵ ( t ) w_N^\epsilon(t)=u^\epsilon(t)-v_N^\epsilon(t) w N ϵ ( t ) = u ϵ ( t ) − v N ϵ ( t ) , where u ϵ ( t ) = U μ ( t , τ ) u τ ϵ u^\epsilon(t)=U_\mu(t,\tau)u_\tau^\epsilon u ϵ ( t ) = U μ ( t , τ ) u τ ϵ . We get
∂ t w N ϵ + ∂ x x x w N ϵ + w N ϵ ∂ x w N ϵ + ( v N ϵ w N ϵ ) x − ∂ x x w N ϵ = 0 , (11) \partial_tw_N^\epsilon+\partial_{xxx}w_N^\epsilon+w_N^\epsilon\partial_xw_N^\epsilon+(v_N^\epsilon w_N^\epsilon)_x-\partial_{xx}w_N^\epsilon=0, \tag*{(11)} ∂ t w N ϵ + ∂ xxx w N ϵ + w N ϵ ∂ x w N ϵ + ( v N ϵ w N ϵ ) x − ∂ xx w N ϵ = 0 , (11)
with initial data w N ϵ ( τ ) = ( Id − P N ) u τ ϵ w_N^\epsilon(\tau)=(\operatorname{Id}-P_N)u_\tau^\epsilon w N ϵ ( τ ) = ( Id − P N ) u τ ϵ (including the case ϵ = 0 \epsilon=0 ϵ = 0 ). From the well-posedness of (11) in X − s , 0 X^{-s,0} X − s , 0 , s ∈ [ 0 , 3 / 4 ) s\in[0,3/4) s ∈ [ 0 , 3/4 ) , we obtain that on bounded time intervals,
∥ w N ϵ ( t ) ∥ H − s ( R ) ≤ K ∥ w N ϵ ( τ ) ∥ H − s ( R ) , (12) \left\|w_N^\epsilon(t)\right\|_{H^{-s}(\mathbb{R})}\leq K\left\|w_N^\epsilon(\tau)\right\|_{H^{-s}(\mathbb{R})}, \tag*{(12)} ∥ w N ϵ ( t ) ∥ H − s ( R ) ≤ K ∥ w N ϵ ( τ ) ∥ H − s ( R ) , (12)
where K K K depends on T T T , but is independent of N N N and ϵ \epsilon ϵ . On the other hand, since { u τ ϵ } \{u_\tau^\epsilon\} { u τ ϵ } is bounded in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) ,
∥ w N ϵ ( τ ) ∥ H − s ( R ) = ∥ ( Id − P N ) u τ ϵ ∥ H − s ( R ) ≤ ∥ u τ ϵ ∥ L 2 ( R ) N s ≤ K N − s . \left\|w_N^\epsilon(\tau)\right\|_{H^{-s}(\mathbb{R})}=\left\|(\operatorname{Id}-P_N)u_\tau^\epsilon\right\|_{H^{-s}(\mathbb{R})}\leq\frac{\left\|u_\tau^\epsilon\right\|_{L^2(\mathbb{R})}}{N^s}\leq KN^{-s}. ∥ w N ϵ ( τ ) ∥ H − s ( R ) = ∥ ( Id − P N ) u τ ϵ ∥ H − s ( R ) ≤ N s ∥ u τ ϵ ∥ L 2 ( R ) ≤ K N − s .
Now, for a test function ψ \psi ψ that is smooth, compactly supported, and satisfies ∥ ψ ∥ L 2 ( R ) = 1 \|\psi\|_{L^2(\mathbb{R})}=1 ∥ ψ ∥ L 2 ( R ) = 1 ,
∣ ( u ( t ) − u ϵ ( t ) , ψ ) ∣ = ∣ ( w N 0 ( t ) + v N ( t ) − v N ϵ ( t ) + w N ϵ ( t ) , ψ ) ∣ \left|(u(t)-u^\epsilon(t),\psi)\right|=\left|(w_N^0(t)+v_N(t)-v_N^\epsilon(t)+w_N^\epsilon(t),\psi)\right| ∣ ( u ( t ) − u ϵ ( t ) , ψ ) ∣ = ( w N 0 ( t ) + v N ( t ) − v N ϵ ( t ) + w N ϵ ( t ) , ψ )
≤ 2 K N − s + ∣ ( v N ( t ) − v N ϵ ( t ) , ψ ) ∣ \leq2KN^{-s}+\left|(v_N(t)-v_N^\epsilon(t),\psi)\right| ≤ 2 K N − s + ∣ ( v N ( t ) − v N ϵ ( t ) , ψ ) ∣
because of (12) . From the estimates on { v N ϵ } \{v_N^\epsilon\} { v N ϵ } , letting ϵ → 0 \epsilon\to0 ϵ → 0 with N N N fixed we obtain
lim sup ϵ → 0 ∣ ( v N ( t ) − v N ϵ ( t ) , ψ ) ∣ ≤ 2 K N − s . \limsup_{\epsilon\to0}\left|(v_N(t)-v_N^\epsilon(t),\psi)\right|\leq2KN^{-s}. ϵ → 0 lim sup ∣ ( v N ( t ) − v N ϵ ( t ) , ψ ) ∣ ≤ 2 K N − s .
Let N → + ∞ N\to+\infty N → + ∞ to conclude that u ϵ ( t ) u^\epsilon(t) u ϵ ( t ) converges to u ( t ) u(t) u ( t ) in the distributional sense. Since w N ϵ w_N^\epsilon w N ϵ is a bounded sequence in L ∞ ( [ τ − T , τ + T ] ; L 2 ( R ) ) L^\infty([\tau-T,\tau+T];L^2(\mathbb{R})) L ∞ ([ τ − T , τ + T ] ; L 2 ( R )) , a density argument completes the proof. □
Theorem 3.5. The process { U μ ( t , τ ) } \{U_\mu(t,\tau)\} { U μ ( t , τ )} possesses a pullback attractor A μ = { A μ ( t ) : t ∈ R } \mathcal{A}_\mu=\{\mathcal{A}_\mu(t):t\in\mathbb{R}\} A μ = { A μ ( t ) : t ∈ R } in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) .
Proof. By Lemma 3.4 and Proposition 3.3, { U μ ( t , τ ) } \{U_\mu(t,\tau)\} { U μ ( t , τ )} is weakly continuous and possesses a family of bounded pullback absorbing set B = { B ( t ) : t ∈ R } \mathcal{B}=\{B(t):t\in\mathbb{R}\} B = { B ( t ) : t ∈ R } . Thanks to Theorem 3.2, it remains to show the pullback asymptotic compactness of { U μ ( t , τ ) } \{U_\mu(t,\tau)\} { U μ ( t , τ )} . Let { x n } \{x_n\} { x n } be a bounded sequence in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) and { τ n } ⊂ R \{\tau_n\}\subset\mathbb{R} { τ n } ⊂ R be a sequence such that τ n → − ∞ \tau_n\to-\infty τ n → − ∞ . We will prove that { U μ ( t , τ n ) x n } \{U_\mu(t,\tau_n)x_n\} { U μ ( t , τ n ) x n } is precompact in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) .
From Proposition 3.3, we have that { U μ ( t , τ n ) x n } \{U_\mu(t,\tau_n)x_n\} { U μ ( t , τ n ) x n } is bounded in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . Thus there exists a w 0 w_0 w 0 belonging to L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) such that
U μ ( t , τ n ) x n ⇀ w 0 as n → + ∞ . U_\mu(t,\tau_n)x_n\rightharpoonup w_0\text{ as }n\to+\infty. U μ ( t , τ n ) x n ⇀ w 0 as n → + ∞.
For fixed T > 0 T>0 T > 0 , by a similar argument, we get a w T ∈ L 2 ( R ) w_T\in L^2(\mathbb{R}) w T ∈ L 2 ( R ) such that
U μ ( t − T , τ n ) x n ⇀ w T as n → + ∞ . U_\mu(t-T,\tau_n)x_n\rightharpoonup w_T\text{ as }n\to+\infty. U μ ( t − T , τ n ) x n ⇀ w T as n → + ∞.
Then, because of the weak continuity of { U μ ( t , τ ) } \{U_\mu(t,\tau)\} { U μ ( t , τ )} , for any T > 0 T>0 T > 0 ,
w 0 = w − l i m U μ ( t , τ n ) x n = w − l i m U μ ( t , t − T ) U μ ( t − T , τ n ) x n = U μ ( t , t − T ) w − l i m U μ ( t − T , τ n ) x n = U μ ( t , t − T ) w T , \begin{aligned}
w_0=\mathop{\mathrm{w-lim}} U_\mu(t,\tau_n)x_n=\mathop{\mathrm{w-lim}} U_\mu(t,t-T)U_\mu(t-T,\tau_n)x_n \\
&=U_\mu(t,t-T)\mathop{\mathrm{w-lim}} U_\mu(t-T,\tau_n)x_n=U_\mu(t,t-T)w_T,
\end{aligned} w 0 = w − lim U μ ( t , τ n ) x n = w − lim U μ ( t , t − T ) U μ ( t − T , τ n ) x n = U μ ( t , t − T ) w − lim U μ ( t − T , τ n ) x n = U μ ( t , t − T ) w T ,
where w-lim denotes the weak limit in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) as n → + ∞ n \to+\infty n → + ∞ . From (72) , denoting E ( u ) = λ ∥ u ∥ 2 + 2 μ ∥ u x ∥ 2 E(u)=\lambda\|u\|^2+2\mu\|u_x\|^2 E ( u ) = λ ∥ u ∥ 2 + 2 μ ∥ u x ∥ 2 , we get
∥ U μ ( t , τ ) u τ ∥ 2 + ∫ τ t e − λ ( t − s ) E ( u ) d s = e − λ ( t − τ ) ∥ u τ ∥ 2 + 2 ∫ τ t e − λ ( t − s ) ( g ( s ) , u ( s ) ) d s . (13) \|U_\mu(t,\tau)u_\tau\|^2+\int_\tau^t e^{-\lambda(t-s)}E(u)\,ds=e^{-\lambda(t-\tau)}\|u_\tau\|^2+2\int_\tau^t e^{-\lambda(t-s)}(g(s),u(s))\,ds. \tag*{(13)} ∥ U μ ( t , τ ) u τ ∥ 2 + ∫ τ t e − λ ( t − s ) E ( u ) d s = e − λ ( t − τ ) ∥ u τ ∥ 2 + 2 ∫ τ t e − λ ( t − s ) ( g ( s ) , u ( s )) d s . (13)
Applying (13) to w 0 = U μ ( t , t − T ) w T w_0=U_\mu(t,t-T)w_T w 0 = U μ ( t , t − T ) w T and U μ ( t , τ n ) x n = U μ ( t , t − T ) U μ ( t − T , τ n ) x n U_\mu(t,\tau_n)x_n=U_\mu(t,t-T)U_\mu(t-T,\tau_n)x_n U μ ( t , τ n ) x n = U μ ( t , t − T ) U μ ( t − T , τ n ) x n respectively, we get
∥ w 0 ∥ 2 + ∫ t − T t e − λ ( t − s ) E ( U μ ( s , t − T ) w T ) d s = e − λ T ∥ w T ∥ 2 + 2 ∫ t − T t e − λ ( t − s ) ( g ( s ) , U μ ( s , t − T ) w T ) d s (14) \begin{aligned}
\|w_0\|^2+\int_{t-T}^t e^{-\lambda(t-s)}E(U_\mu(s,t-T)w_T)\,ds
&=e^{-\lambda T}\|w_T\|^2\\
&\quad+2\int_{t-T}^t e^{-\lambda(t-s)}\left(g(s),U_\mu(s,t-T)w_T\right)\,ds \tag*{(14)}
\end{aligned} ∥ w 0 ∥ 2 + ∫ t − T t e − λ ( t − s ) E ( U μ ( s , t − T ) w T ) d s = e − λ T ∥ w T ∥ 2 + 2 ∫ t − T t e − λ ( t − s ) ( g ( s ) , U μ ( s , t − T ) w T ) d s (14)
and
∥ U μ ( t , τ n ) x n ∥ 2 + ∫ t − T t e − λ ( t − s ) E ( U μ ( s , τ n ) x n ) d s = e − λ T ∥ U μ ( t − T , τ n ) x n ∥ 2 + 2 ∫ t − T t e − λ ( t − s ) ( g ( s ) , U μ ( s , τ n ) x n ) d s . \begin{aligned}
\|U_\mu(t,\tau_n)x_n\|^2+\int_{t-T}^t e^{-\lambda(t-s)}E(U_\mu(s,\tau_n)x_n)\,ds
&=e^{-\lambda T}\|U_\mu(t-T,\tau_n)x_n\|^2\\
&\quad+2\int_{t-T}^t e^{-\lambda(t-s)}\left(g(s),U_\mu(s,\tau_n)x_n\right)\,ds.
\end{aligned} ∥ U μ ( t , τ n ) x n ∥ 2 + ∫ t − T t e − λ ( t − s ) E ( U μ ( s , τ n ) x n ) d s = e − λ T ∥ U μ ( t − T , τ n ) x n ∥ 2 + 2 ∫ t − T t e − λ ( t − s ) ( g ( s ) , U μ ( s , τ n ) x n ) d s .
It is easy to see that
u ↦ ( ∫ t − T t e − λ ( t − s ) E ( u ) d s ) 1 / 2 u\mapsto\left(\int_{t-T}^t e^{-\lambda(t-s)}E(u)\,ds\right)^{1/2} u ↦ ( ∫ t − T t e − λ ( t − s ) E ( u ) d s ) 1/2
is a norm in L 2 ( t − T , t ; H 1 ( R ) ) L^2(t-T,t;H^1(\mathbb{R})) L 2 ( t − T , t ; H 1 ( R )) that is equivalent to the usual one. Then, from (6) , we get as n → + ∞ n\to+\infty n → + ∞
U μ ( ⋅ , τ n ) x n ⇀ U μ ( ⋅ , t − T ) w T weakly in L 2 ( t − T , t ; H 1 ( R ) ) U_\mu(\cdot,\tau_n)x_n\rightharpoonup U_\mu(\cdot,t-T)w_T\quad\text{weakly in }L^2(t-T,t;H^1(\mathbb{R})) U μ ( ⋅ , τ n ) x n ⇀ U μ ( ⋅ , t − T ) w T weakly in L 2 ( t − T , t ; H 1 ( R ))
and consequently
∫ t − T t e − λ ( t − s ) E ( U μ ( ⋅ , t − T ) w T ) d s ≤ lim sup n → ∞ ∫ t − T t e − λ ( t − s ) E ( U μ ( ⋅ , τ n ) x n ) d s . (15) \int_{t-T}^t e^{-\lambda(t-s)}E(U_\mu(\cdot,t-T)w_T)\,ds\leq\limsup_{n\to\infty}\int_{t-T}^t e^{-\lambda(t-s)}E(U_\mu(\cdot,\tau_n)x_n)\,ds. \tag*{(15)} ∫ t − T t e − λ ( t − s ) E ( U μ ( ⋅ , t − T ) w T ) d s ≤ n → ∞ lim sup ∫ t − T t e − λ ( t − s ) E ( U μ ( ⋅ , τ n ) x n ) d s . (15)
From (14) and (3.8, and noting that U μ ( ⋅ , τ n ) x n ⇀ U μ ( ⋅ , t − T ) w T U_\mu(\cdot,\tau_n)x_n\rightharpoonup U_\mu(\cdot,t-T)w_T U μ ( ⋅ , τ n ) x n ⇀ U μ ( ⋅ , t − T ) w T in L 2 ( t − T , t ; L 2 ( R ) ) L^2(t-T,t;L^2(\mathbb{R})) L 2 ( t − T , t ; L 2 ( R )) , we get that
∥ w 0 ∥ 2 ≥ lim sup n → ∞ ∥ U μ ( t , τ n ) x n ∥ 2 − lim sup n → ∞ e − λ T ∥ U μ ( t − T , τ n ) x n ∥ 2 ≥ lim sup n → ∞ ∥ U μ ( t , τ n ) x n ∥ 2 + e − λ T r ( t − T ) 2 \begin{aligned}
\|w_0\|^2\geq\limsup_{n\to\infty}\|U_\mu(t,\tau_n)x_n\|^2-\limsup_{n\to\infty}e^{-\lambda T}\|U_\mu(t-T,\tau_n)x_n\|^2 \\
&\geq\limsup_{n\to\infty}\|U_\mu(t,\tau_n)x_n\|^2+e^{-\lambda T}r(t-T)^2
\end{aligned} ∥ w 0 ∥ 2 ≥ n → ∞ lim sup ∥ U μ ( t , τ n ) x n ∥ 2 − n → ∞ lim sup e − λ T ∥ U μ ( t − T , τ n ) x n ∥ 2 ≥ n → ∞ lim sup ∥ U μ ( t , τ n ) x n ∥ 2 + e − λ T r ( t − T ) 2
where r ( t ) r(t) r ( t ) is defined as in (8) . Hence, thanks to lim T → + ∞ e − λ T r ( t − T ) 2 = 0 \lim_{T\to+\infty}e^{-\lambda T}r(t-T)^2=0 lim T → + ∞ e − λ T r ( t − T ) 2 = 0 , we obtain
∥ w 0 ∥ 2 ≥ lim sup n → ∞ ∥ U μ ( t , τ n ) x n ∥ 2 . \|w_0\|^2\geq\limsup_{n\to\infty}\|U_\mu(t,\tau_n)x_n\|^2. ∥ w 0 ∥ 2 ≥ n → ∞ lim sup ∥ U μ ( t , τ n ) x n ∥ 2 .
Combined with the fact that U μ ( t , τ n ) x n ⇀ w 0 U_\mu(t,\tau_n)x_n\rightharpoonup w_0 U μ ( t , τ n ) x n ⇀ w 0 in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , we have
U μ ( t , τ n ) x n → w 0 strongly in L 2 ( R ) , U_\mu(t,\tau_n)x_n\to w_0\quad\text{strongly in }L^2(\mathbb{R}), U μ ( t , τ n ) x n → w 0 strongly in L 2 ( R ) ,
when passing to a subsequence if necessary. This completes the proof. □
Boundedness in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R )
In this subsection, we will prove that the L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) pullback attractor A μ \mathcal{A}_\mu A μ of (1) obtained in subsection 3.1 is in fact bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . In order to do that, we will employ the method of splitting solution. Roughly speaking, we will divide the solution into two parts in which one part decays exponentially in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) while the other part is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) .
Splitting solutions
We write u = v + w u = v + w u = v + w , where v , w v, w v , w are solutions of the equations
v t + v x x x − μ v x x + λ v + v v x = g ( t ) − P N ( ( v w ) x + w w x ) , x ∈ R , t > τ (16) v_t + v_{xxx} - \mu v_{xx} + \lambda v + vv_x = g(t) - P_N((vw)_x + ww_x), \quad x \in\mathbb{R},\ t > \tau \tag*{(16)} v t + v xxx − μ v xx + λ v + v v x = g ( t ) − P N (( v w ) x + w w x ) , x ∈ R , t > τ (16)
w t + w x x x − μ w x x + λ w + Q N ( w w x ) = − Q N ( v w ) x , x ∈ R , t > τ (17) w_t + w_{xxx} - \mu w_{xx} + \lambda w + Q_N(ww_x) = -Q_N(vw)_x, \quad x \in\mathbb{R},\ t > \tau \tag*{(17)} w t + w xxx − μ w xx + λ w + Q N ( w w x ) = − Q N ( v w ) x , x ∈ R , t > τ (17)
with initial data v ( τ ) = P N u τ v(\tau) = P_Nu_\tau v ( τ ) = P N u τ and w ( τ ) = Q N u τ w(\tau) = Q_Nu_\tau w ( τ ) = Q N u τ , where we recall that P N P_N P N is the projection defined in (9) and Q N = Id − P N Q_N = \operatorname{Id} - P_N Q N = Id − P N .
Since (17) is a dissipative equation without forces, its solution w w w is expected to be exponentially decay. While the solution to (16) is expected to be smooth, i.e. belong to H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) , thanks to the regularity of initial data v ( τ ) = P N u τ v(\tau) = P_Nu_\tau v ( τ ) = P N u τ .
The analysis of the rest of this section frequently uses the following bilinear estimate.
Lemma 3.6. Let s ∈ ( − 1 / 2 , 0 ] s \in(-1/2, 0] s ∈ ( − 1/2 , 0 ] be given. Then there exists a constant c s c_s c s such that, for b ′ ∈ ( 1 / 2 , 9 / 16 ] b' \in(1/2, 9/16] b ′ ∈ ( 1/2 , 9/16 ] , b ∈ [ 1 / 2 , min { b ′ , 1 + s } ] b \in[1/2,\min\{b', 1+s\}] b ∈ [ 1/2 , min { b ′ , 1 + s }] , and any function u u u in X s , b X^{s,b} X s , b ,
∥ D ( u 2 ) ∥ X s , b ′ − 1 ≤ c s ∥ u ∥ X s , b ∥ u ∥ X − 1 / 2 , b . (18) \|D(u^2)\|_{X^{s,b'-1}} \le c_s\|u\|_{X^{s,b}}\|u\|_{X^{-1/2,b}}. \tag*{(18)} ∥ D ( u 2 ) ∥ X s , b ′ − 1 ≤ c s ∥ u ∥ X s , b ∥ u ∥ X − 1/2 , b . (18) As a corollary, if u ∈ X s , b u \in X^{s,b} u ∈ X s , b is a function satisfying that the Fourier transform u ^ ( ξ , τ ) \widehat{u}(\xi,\tau) u ( ξ , τ ) of u u u is supported in { ( ξ , τ ) : ∣ ξ ∣ ≥ N } \{(\xi,\tau): |\xi| \ge N\} {( ξ , τ ) : ∣ ξ ∣ ≥ N } , where N > 0 N > 0 N > 0 . Then, for 1 / 2 < b ≤ 9 / 16 1/2 < b \le9/16 1/2 < b ≤ 9/16 ,
∥ D ( u 2 ) ∥ X 0 , b − 1 ≤ C N ∥ u ∥ X 0 , b 2 . (19) \|D(u^2)\|_{X^{0,b-1}} \le\frac{C}{\sqrt{N}}\|u\|_{X^{0,b}}^2. \tag*{(19)} ∥ D ( u 2 ) ∥ X 0 , b − 1 ≤ N C ∥ u ∥ X 0 , b 2 . (19)
Proof. The estimate (18) is similar to [11 Proposition 2.2] for the KdV equation. However, we emphasize that we currently use different Bourgain spaces, which are based on the symbol of the free KdV-Burgers equation instead of the free KdV equation. The proof of (18) then follows instead similarly from [19 Proposition 3.2] with slight modifications. □
Well-posedness and decay of the part w w w
Since v = u − w v = u - w v = u − w , we can rewrite (17) as
w t + w x x x + λ w − μ w x x − Q N ( w w x ) = − Q N ( u w ) x (20) w_t + w_{xxx} + \lambda w - \mu w_{xx} - Q_N(ww_x) = -Q_N(uw)_x \tag*{(20)} w t + w xxx + λ w − μ w xx − Q N ( w w x ) = − Q N ( u w ) x (20)
with w ( τ ) = Q N u τ w(\tau) = Q_Nu_\tau w ( τ ) = Q N u τ . The local well-posedness of (20) follows as for (1) . Applying arguments of Section 2 to the equation
w ( t ) = ψ 1 ( t ) W ( t ) w ( τ ) + 1 2 ψ T ( t ) ∫ τ t W ( t − s ) [ − 2 λ ψ 1 ( s ) w ( s ) − ( Q N ( ψ T ( s ) u ( s ) ψ T ( s ) w ( s ) ) ) x + ( Q N ( ψ T ( s ) w ( s ) ) 2 ) x ] d s . \begin{aligned}
w(t) ={}& \psi_1(t)W(t)w(\tau) \\
&+ \frac{1}{2}\psi_T(t)\int_\tau^t W(t-s)\left[-2\lambda\psi_1(s)w(s) - \left(Q_N(\psi_T(s)u(s)\psi_T(s)w(s))\right)_x \right.\\
&\left.\qquad\qquad\qquad\qquad+ \left(Q_N(\psi_T(s)w(s))^2\right)_x\right]\,ds.
\end{aligned} w ( t ) = ψ 1 ( t ) W ( t ) w ( τ ) + 2 1 ψ T ( t ) ∫ τ t W ( t − s ) [ − 2 λ ψ 1 ( s ) w ( s ) − ( Q N ( ψ T ( s ) u ( s ) ψ T ( s ) w ( s )) ) x + ( Q N ( ψ T ( s ) w ( s ) ) 2 ) x ] d s .
Recall that ψ T ( ⋅ ) = ψ ( ⋅ / T ) \psi_T(\cdot) = \psi(\cdot/T) ψ T ( ⋅ ) = ψ ( ⋅ / T ) is a time cut-off function satisfying
ψ ∈ C 0 ∞ ( R ) , supp ψ ⊆ [ τ − 1 , τ + 1 ] , ψ ≡ 1 on [ τ − 1 2 , τ + 1 2 ] . \psi\in C^\infty_0(\mathbb{R}),\quad\operatorname{supp}\psi\subseteq[\tau- 1,\tau+ 1],\quad\psi\equiv1\ \text{on}\ \left[\tau- \frac{1}{2},\tau+ \frac{1}{2}\right]. ψ ∈ C 0 ∞ ( R ) , supp ψ ⊆ [ τ − 1 , τ + 1 ] , ψ ≡ 1 on [ τ − 2 1 , τ + 2 1 ] .
Then, we get that, for T T T sufficiently small,
∥ w ∥ X 0 , b ≤ C ∥ w ( τ ) ∥ , \|w\|_{X^{0,b}} \le C\|w(\tau)\|, ∥ w ∥ X 0 , b ≤ C ∥ w ( τ ) ∥ ,
where 1 / 2 < b < 25 / 48 1/2 < b < 25/48 1/2 < b < 25/48 . By change the of initial time we can write
∥ w ∥ X [ t 0 − T , t 0 + T ] 0 , b ≤ C ∥ w ( t 0 ) ∥ (21) \|w\|_{X^{0,b}_{[t_0-T,t_0+T]}} \le C\|w(t_0)\| \tag*{(21)} ∥ w ∥ X [ t 0 − T , t 0 + T ] 0 , b ≤ C ∥ w ( t 0 ) ∥ (21)
for every t 0 t_0 t 0 . Now, taking the inner product of equation (20) with 2 w 2w 2 w in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , we find that
d d t ∥ w ( t ) ∥ 2 + 2 λ ∥ w ( t ) ∥ 2 + 2 μ ∥ w x ( t ) ∥ 2 = ( u x ( t ) , w 2 ( t ) ) . (22) \frac{d}{dt}\|w(t)\|^2 + 2\lambda\|w(t)\|^2 + 2\mu\|w_x(t)\|^2 = (u_x(t),w^2(t)). \tag*{(22)} d t d ∥ w ( t ) ∥ 2 + 2 λ ∥ w ( t ) ∥ 2 + 2 μ ∥ w x ( t ) ∥ 2 = ( u x ( t ) , w 2 ( t )) . (22)
Multiplying (22) by e λ t e^{\lambda t} e λ t yields
d d t ( e λ t ∥ w ( t ) ∥ 2 ) + 2 μ e λ t ∥ w x ( t ) ∥ 2 = e λ t ( ( u x ( t ) , w 2 ( t ) ) − λ ∥ w ( t ) ∥ 2 ) . (23) \frac{d}{dt}\left(e^{\lambda t}\|w(t)\|^2\right) + 2\mu e^{\lambda t}\|w_x(t)\|^2 = e^{\lambda t}\left((u_x(t),w^2(t)) - \lambda\|w(t)\|^2\right). \tag*{(23)} d t d ( e λ t ∥ w ( t ) ∥ 2 ) + 2 μ e λ t ∥ w x ( t ) ∥ 2 = e λ t ( ( u x ( t ) , w 2 ( t )) − λ ∥ w ( t ) ∥ 2 ) . (23)
We integrate (23) from τ \tau τ to t t t to obtain
∥ w ( t ) ∥ 2 + 2 μ ∫ τ t e − λ ( t − s ) ∥ w x ( s ) ∥ 2 d s = ∥ w ( τ ) ∥ 2 e − λ ( t − τ ) + ∫ τ t e − λ ( t − s ) ( ( u x ( s ) , w 2 ( s ) ) − λ ∥ w ( s ) ∥ 2 ) d s . (24) \|w(t)\|^2+2\mu\int_{\tau}^{t}e^{-\lambda(t-s)}\|w_x(s)\|^2\,ds
=\|w(\tau)\|^2e^{-\lambda(t-\tau)}+\int_{\tau}^{t}e^{-\lambda(t-s)}\left((u_x(s),w^2(s))-\lambda\|w(s)\|^2\right)\,ds. \tag*{(24)} ∥ w ( t ) ∥ 2 + 2 μ ∫ τ t e − λ ( t − s ) ∥ w x ( s ) ∥ 2 d s = ∥ w ( τ ) ∥ 2 e − λ ( t − τ ) + ∫ τ t e − λ ( t − s ) ( ( u x ( s ) , w 2 ( s )) − λ ∥ w ( s ) ∥ 2 ) d s . (24)
We aim now to prove that the last term on the right hand side of (24) is negative for large enough N > 0 N>0 N > 0 , which implies the decay of w w w ,
∥ w ( t ) ∥ 2 ≤ ∥ w ( τ ) ∥ 2 e − λ ( t − τ ) . \|w(t)\|^2\leq\|w(\tau)\|^2e^{-\lambda(t-\tau)}. ∥ w ( t ) ∥ 2 ≤ ∥ w ( τ ) ∥ 2 e − λ ( t − τ ) .
Consider the truncation function ψ ˉ = ψ [ τ , t ] \bar{\psi}=\psi_{[\tau,t]} ψ ˉ = ψ [ τ , t ] and the characteristic function χ = χ [ τ , t ] \chi=\chi_{[\tau,t]} χ = χ [ τ , t ] . Let b b b and b ′ b' b ′ such that 1 / 2 < b < 25 / 48 1/2<b<25/48 1/2 < b < 25/48 and 1 − b < b ′ < 1 / 2 1-b<b'<1/2 1 − b < b ′ < 1/2 . Then, by duality, we find that
∫ τ t e − λ ( t − s ) ( u x ( s ) , w 2 ( s ) ) d s = ∫ − ∞ ∞ e − λ ( t − s ) χ ( s ) ( ψ ˉ ( s ) u ( s ) , ∂ x ( ( ψ ˉ ( s ) w ( s ) ) 2 ) ) d s ≤ ∥ χ ψ ˉ u ∥ X 0 , b ′ ∥ ∂ x ( ψ ˉ w ) 2 ∥ X 0 , − b ′ . (25) \int_{\tau}^{t}e^{-\lambda(t-s)}(u_x(s),w^2(s))\,ds
=\int_{-\infty}^{\infty}e^{-\lambda(t-s)}\chi(s)(\bar{\psi}(s)u(s),\partial_x((\bar{\psi}(s)w(s))^2))\,ds
\leq\|\chi\bar{\psi}u\|_{X^{0,b'}}\|\partial_x(\bar{\psi}w)^2\|_{X^{0,-b'}}. \tag*{(25)} ∫ τ t e − λ ( t − s ) ( u x ( s ) , w 2 ( s )) d s = ∫ − ∞ ∞ e − λ ( t − s ) χ ( s ) ( ψ ˉ ( s ) u ( s ) , ∂ x (( ψ ˉ ( s ) w ( s ) ) 2 )) d s ≤ ∥ χ ψ ˉ u ∥ X 0 , b ′ ∥ ∂ x ( ψ ˉ w ) 2 ∥ X 0 , − b ′ . (25)
By the choice of b , b ′ b,b' b , b ′ and by applying the bilinear estimate (19) , we have
∥ ∂ x ( ψ ˉ w ) 2 ∥ X 0 , b ′ ≤ ∥ ∂ x ( ψ ˉ w ) 2 ∥ X 0 , b − 1 ≤ C N ∥ w ∥ X [ τ , t ] 0 , b 2 . (26) \|\partial_x(\bar{\psi}w)^2\|_{X^{0,b'}}\leq\|\partial_x(\bar{\psi}w)^2\|_{X^{0,b-1}}\leq\frac{C}{\sqrt{N}}\|w\|_{X^{0,b}_{[\tau,t]}}^2. \tag*{(26)} ∥ ∂ x ( ψ ˉ w ) 2 ∥ X 0 , b ′ ≤ ∥ ∂ x ( ψ ˉ w ) 2 ∥ X 0 , b − 1 ≤ N C ∥ w ∥ X [ τ , t ] 0 , b 2 . (26)
On the other hand, since the characteristic function χ ∈ H t b ′ ∩ L t ∞ \chi\in H_t^{b'}\cap L_t^\infty χ ∈ H t b ′ ∩ L t ∞ (recall b ′ < 1 / 2 b'<1/2 b ′ < 1/2 ) and ψ ˉ u \bar{\psi}u ψ ˉ u belong to X 0 , b X^{0,b} X 0 , b which is contained in L t ∞ L x 0 L_t^\infty L_x^0 L t ∞ L x 0 (see (1)), one shows that
∥ χ ψ ˉ u ∥ X 0 , b ′ ≤ ∥ χ ∥ L t ∞ ∥ ψ ˉ u ∥ X 0 , b ′ + ∥ χ ∥ H t b ′ ∥ ψ ˉ u ∥ X 0 , b ≤ C ( χ ) ∥ u ∥ X [ τ , t ] 0 , b . (27) \|\chi\bar{\psi}u\|_{X^{0,b'}}\leq\|\chi\|_{L_t^\infty}\|\bar{\psi}u\|_{X^{0,b'}}+\|\chi\|_{H_t^{b'}}\|\bar{\psi}u\|_{X^{0,b}}\leq C(\chi)\|u\|_{X^{0,b}_{[\tau,t]}}. \tag*{(27)} ∥ χ ψ ˉ u ∥ X 0 , b ′ ≤ ∥ χ ∥ L t ∞ ∥ ψ ˉ u ∥ X 0 , b ′ + ∥ χ ∥ H t b ′ ∥ ψ ˉ u ∥ X 0 , b ≤ C ( χ ) ∥ u ∥ X [ τ , t ] 0 , b . (27)
where we have used the definition of the Bourgain spaces. Here, C ( χ ) C(\chi) C ( χ ) depends on χ \chi χ , but is independent of N N N . Applying (26) and (27) in (25) yields
∫ τ t e − λ ( t − s ) ( u x ( s ) , w 2 ( s ) ) d s ≤ C ( χ ) N ∥ u ∥ X [ τ , t ] 0 , b ∥ w ∥ X [ τ , t ] 0 , b 2 ≤ C ( χ ) N ∥ u ∥ X [ τ , t ] 0 , b 1 t − τ ∫ τ t ∥ w ∥ X [ τ , t ] 0 , b 2 d s ≤ C ( χ ) N ∥ u ∥ X [ τ , t ] 0 , b 1 t − τ ∫ τ t ∥ w ( s ) ∥ 2 d s . (28) \begin{aligned}
\int_{\tau}^{t}e^{-\lambda(t-s)}(u_x(s),w^2(s))\,ds
&\leq\frac{C(\chi)}{\sqrt{N}}\|u\|_{X^{0,b}_{[\tau,t]}}\|w\|_{X^{0,b}_{[\tau,t]}}^2\\
&\leq\frac{C(\chi)}{\sqrt{N}}\|u\|_{X^{0,b}_{[\tau,t]}}\frac{1}{t-\tau}\int_{\tau}^{t}\|w\|_{X^{0,b}_{[\tau,t]}}^2\,ds\\
&\leq\frac{C(\chi)}{\sqrt{N}}\|u\|_{X^{0,b}_{[\tau,t]}}\frac{1}{t-\tau}\int_{\tau}^{t}\|w(s)\|^2\,ds. \tag*{(28)}
\end{aligned} ∫ τ t e − λ ( t − s ) ( u x ( s ) , w 2 ( s )) d s ≤ N C ( χ ) ∥ u ∥ X [ τ , t ] 0 , b ∥ w ∥ X [ τ , t ] 0 , b 2 ≤ N C ( χ ) ∥ u ∥ X [ τ , t ] 0 , b t − τ 1 ∫ τ t ∥ w ∥ X [ τ , t ] 0 , b 2 d s ≤ N C ( χ ) ∥ u ∥ X [ τ , t ] 0 , b t − τ 1 ∫ τ t ∥ w ( s ) ∥ 2 d s . (28)
here we have used (21) at the last step. Thus, from (24)
∥ w ( t ) ∥ 2 ≤ ∥ w ( τ ) ∥ 2 e − λ ( t − τ ) + ∫ τ t ( C ( χ ) N ( t − τ ) ∥ u ∥ X [ τ , t ] 0 , b − λ e − λ ( t − s ) ) ∥ w ( s ) ∥ 2 d s . (29) \|w(t)\|^2\leq\|w(\tau)\|^2e^{-\lambda(t-\tau)}+\int_{\tau}^{t}\left(\frac{C(\chi)}{\sqrt{N}(t-\tau)}\|u\|_{X^{0,b}_{[\tau,t]}}-\lambda e^{-\lambda(t-s)}\right)\|w(s)\|^2\,ds. \tag*{(29)} ∥ w ( t ) ∥ 2 ≤ ∥ w ( τ ) ∥ 2 e − λ ( t − τ ) + ∫ τ t ( N ( t − τ ) C ( χ ) ∥ u ∥ X [ τ , t ] 0 , b − λ e − λ ( t − s ) ) ∥ w ( s ) ∥ 2 d s . (29)
for t t t is close enough to τ \tau τ . Naturally, this argument can be repeated to obtain (29) for all t ≥ τ t\geq\tau t ≥ τ . Hence, for N N N large enough,
∥ w ( t ) ∥ 2 ≤ ∥ w ( τ ) ∥ 2 e − λ ( t − τ ) = ∥ Q N u τ ∥ 2 e − λ ( t − τ ) . (30) \|w(t)\|^2\leq\|w(\tau)\|^2e^{-\lambda(t-\tau)}=\|Q_Nu_\tau\|^2e^{-\lambda(t-\tau)}. \tag*{(30)} ∥ w ( t ) ∥ 2 ≤ ∥ w ( τ ) ∥ 2 e − λ ( t − τ ) = ∥ Q N u τ ∥ 2 e − λ ( t − τ ) . (30)
Letting τ → − ∞ \tau\to-\infty τ → − ∞ , we deduce that w w w decays exponentially.
Regularity of v v v
First, v v v is defined globally in time, since v = u − w v=u-w v = u − w and u , w u,w u , w are defined globally in time. Next, we will prove the boundedness of v v v in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) .
For y ( t ) = P N v ( t ) = P N u ( t ) y(t)=P_Nv(t)=P_Nu(t) y ( t ) = P N v ( t ) = P N u ( t ) , we have that y y y is smooth and satisfies the estimate, thanks to Proposition 3.3,
lim τ → − ∞ ∥ y ( t ) ∥ H 3 ( R ) ≤ r ( t ) N 3 , \lim_{\tau\to-\infty}\|y(t)\|_{H^3(\mathbb{R})}\leq r(t)N^3, τ → − ∞ lim ∥ y ( t ) ∥ H 3 ( R ) ≤ r ( t ) N 3 ,
where r ( t ) r(t) r ( t ) is the radius of the pullback absorbing set defined in (8) .
It remains to show that Z = Q N v Z=Q_Nv Z = Q N v , which solves
Z t + Z x x x + Q N ( ( y + Z ) ( y + Z ) x ) + λ Z − μ Z x x = Q N g ( t ) Z_t+Z_{xxx}+Q_N((y+Z)(y+Z)_x)+\lambda Z-\mu Z_{xx}=Q_Ng(t) Z t + Z xxx + Q N (( y + Z ) ( y + Z ) x ) + λ Z − μ Z xx = Q N g ( t )
with Z ( τ ) = 0 Z(\tau)=0 Z ( τ ) = 0 , is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) .
It is easy to see that Z = Q N v Z=Q_Nv Z = Q N v remains bounded in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . Now, by using the Agmon’s inequality
∥ f ∥ L ∞ ( R ) ≤ ∥ u ∥ 1 / 2 ∥ ∂ x u ∥ 1 / 2 \|f\|_{L^\infty(\mathbb{R})}\leq\|u\|^{1/2}\|\partial_xu\|^{1/2} ∥ f ∥ L ∞ ( R ) ≤ ∥ u ∥ 1/2 ∥ ∂ x u ∥ 1/2
which implies
∥ f 2 ∥ H − 1 ( R ) ≤ ∥ f ∥ 2 \|f^2\|_{H^{-1}(\mathbb{R})}\leq\|f\|^2 ∥ f 2 ∥ H − 1 ( R ) ≤ ∥ f ∥ 2
we get
∥ Z x x x ∥ ≤ ∥ Z t ∥ + 1 2 ∥ ( y + Z ) 2 ∥ H − 1 ( R ) + λ ∥ Z ∥ + μ ∥ Z x x ∥ + ∥ g ( t ) ∥ ≤ ∥ Z t ∥ + 1 4 ∥ y ∥ 2 + 1 4 ∥ Z ∥ 2 + λ ∥ Z ∥ + μ ∥ Z x x ∥ + ∥ g ( t ) ∥ . (31) \begin{aligned}
\|Z_{xxx}\|&\leq\|Z_t\|+\frac{1}{2}\|(y+Z)^2\|_{H^{-1}(\mathbb{R})}+\lambda\|Z\|+\mu\|Z_{xx}\|+\|g(t)\|\\
&\leq\|Z_t\|+\frac{1}{4}\|y\|^2+\frac{1}{4}\|Z\|^2+\lambda\|Z\|+\mu\|Z_{xx}\|+\|g(t)\|. \tag*{(31)}
\end{aligned} ∥ Z xxx ∥ ≤ ∥ Z t ∥ + 2 1 ∥ ( y + Z ) 2 ∥ H − 1 ( R ) + λ ∥ Z ∥ + μ ∥ Z xx ∥ + ∥ g ( t ) ∥ ≤ ∥ Z t ∥ + 4 1 ∥ y ∥ 2 + 4 1 ∥ Z ∥ 2 + λ ∥ Z ∥ + μ ∥ Z xx ∥ + ∥ g ( t ) ∥. (31)
We observe from (31) and interpolation inequalities that it is sufficient to show the boundedness of Z t Z_t Z t in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . The equation for Z ′ ( t ) = Z t ( t ) Z'(t)=Z_t(t) Z ′ ( t ) = Z t ( t ) is
Z t ′ + Z x x x ′ + Q N ∂ x ( ( y + Z ) Z ′ ) + λ Z ′ − μ Z x x ′ = − Q N ∂ x ( ( y + Z ) y t ) + g ′ ( t ) (32) Z'_t+Z'_{xxx}+Q_N\partial_x((y+Z)Z')+\lambda Z'-\mu Z'_{xx}=-Q_N\partial_x((y+Z)y_t)+g'(t) \tag*{(32)} Z t ′ + Z xxx ′ + Q N ∂ x (( y + Z ) Z ′ ) + λ Z ′ − μ Z xx ′ = − Q N ∂ x (( y + Z ) y t ) + g ′ ( t ) (32)
with initial condition Z ′ ( τ ) = Q N g ( τ ) − Q N y ( τ ) y x ( τ ) Z'(\tau)=Q_Ng(\tau)-Q_Ny(\tau)y_x(\tau) Z ′ ( τ ) = Q N g ( τ ) − Q N y ( τ ) y x ( τ ) in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . Multiplying equation (32) by 2 Z ′ 2Z' 2 Z ′ , integrating over R \mathbb{R} R , and using Young’s inequality, we obtain
d d t ∥ Z ′ ∥ 2 + 3 λ 2 ∥ Z ′ ∥ 2 + 2 μ ∥ Z x ′ ∥ 2 ≤ C ( ∥ ∂ x ( y t ( y + Z ) ) ∥ 2 + ∫ R ( y + Z ) ∂ x ( Z ′ ) 2 d x + ∥ g ′ ( t ) ∥ 2 ) . (33) \frac{d}{dt}\|Z'\|^2+\frac{3\lambda}{2}\|Z'\|^2+2\mu\|Z'_x\|^2
\leq C\left(\|\partial_x(y_t(y+Z))\|^2+\int_{\mathbb{R}}(y+Z)\partial_x(Z')^2\,dx+\|g'(t)\|^2\right). \tag*{(33)} d t d ∥ Z ′ ∥ 2 + 2 3 λ ∥ Z ′ ∥ 2 + 2 μ ∥ Z x ′ ∥ 2 ≤ C ( ∥ ∂ x ( y t ( y + Z )) ∥ 2 + ∫ R ( y + Z ) ∂ x ( Z ′ ) 2 d x + ∥ g ′ ( t ) ∥ 2 ) . (33)
Using Young’s inequality, then inverse (Poincaré-type) inequalities (for functions with bounded spatial frequency) we find that
∥ ∂ x ( y t ( y + Z ) ) ∥ 2 ≤ ∥ y t ∥ H 1 2 ∥ y + Z ∥ H 1 2 ≤ C ( N ) ( 1 + ∥ g ( t ) ∥ 2 + ∥ Z ′ ∥ 2 ) , \|\partial_x(y_t(y+Z))\|^2\leq\|y_t\|_{H^1}^2\|y+Z\|_{H^1}^2\leq C(N)\left(1+\|g(t)\|^2+\|Z'\|^2\right), ∥ ∂ x ( y t ( y + Z )) ∥ 2 ≤ ∥ y t ∥ H 1 2 ∥ y + Z ∥ H 1 2 ≤ C ( N ) ( 1 + ∥ g ( t ) ∥ 2 + ∥ Z ′ ∥ 2 ) ,
where
∥ ∂ x ( y t ( y + Z ) ) ∥ 2 = ∥ ∂ x y t ( y + Z ) + y t ∂ x ( y + Z ) ∥ 2 ≤ ∥ ∂ x y t ∥ ∥ y + Z ∥ L ∞ + ∥ y t ∥ L ∞ ∥ ∂ x ( y + Z ) ∥ , ≤ 2 ∥ y t ∥ H 1 ∥ y + Z ∥ H 1 . \begin{aligned}
\|\partial_x(y_t(y+Z))\|^2&=\|\partial_xy_t(y+Z)+y_t\partial_x(y+Z)\|^2\\
&\leq\|\partial_xy_t\|\|y+Z\|_{L^\infty}+\|y_t\|_{L^\infty}\|\partial_x(y+Z)\|,\\
&\leq2\|y_t\|_{H^1}\|y+Z\|_{H^1}.
\end{aligned} ∥ ∂ x ( y t ( y + Z )) ∥ 2 = ∥ ∂ x y t ( y + Z ) + y t ∂ x ( y + Z ) ∥ 2 ≤ ∥ ∂ x y t ∥∥ y + Z ∥ L ∞ + ∥ y t ∥ L ∞ ∥ ∂ x ( y + Z ) ∥ , ≤ 2∥ y t ∥ H 1 ∥ y + Z ∥ H 1 .
Hence, (33) implies that
∥ Z ′ ( t ) ∥ 2 ≤ ∥ Z ′ ( τ ) ∥ 2 e − λ ( t − τ ) + C ∫ τ t e − λ ( t − s ) ( ∫ R ( y + Z ) ∂ x ( ( Z ′ ) 2 ) + 1 + ∣ g ( s ) ∣ 2 + ∣ g ′ ( s ) ∣ 2 ) d x d s . \|Z'(t)\|^2\leq\|Z'(\tau)\|^2e^{-\lambda(t-\tau)}
+C\int_\tau^t e^{-\lambda(t-s)}\left(\int_{\mathbb{R}}(y+Z)\partial_x((Z')^2)+1+|g(s)|^2+|g'(s)|^2\right)\,dx\,ds. ∥ Z ′ ( t ) ∥ 2 ≤ ∥ Z ′ ( τ ) ∥ 2 e − λ ( t − τ ) + C ∫ τ t e − λ ( t − s ) ( ∫ R ( y + Z ) ∂ x (( Z ′ ) 2 ) + 1 + ∣ g ( s ) ∣ 2 + ∣ g ′ ( s ) ∣ 2 ) d x d s .
We proceed as in (25) –(30) , using the bilinear estimate (19) and the local well-posedness of (32) in L 2 L^2 L 2 , to obtain
∥ Z ′ ( t ) ∥ 2 ≤ C ( 1 + ∫ − ∞ t e − λ ( t − s ) ( ∥ g ( s ) ∥ 2 + ∥ g ′ ( s ) ∥ 2 ) d s + 1 N ∥ Z ′ ∥ X τ , t 0 , b 2 ) , \|Z'(t)\|^2\leq C\left(1+\int_{-\infty}^t e^{-\lambda(t-s)}\left(\|g(s)\|^2+\|g'(s)\|^2\right)\,ds+\frac{1}{\sqrt{N}}\|Z'\|^2_{X_{\tau,t}^{0,b}}\right), ∥ Z ′ ( t ) ∥ 2 ≤ C ( 1 + ∫ − ∞ t e − λ ( t − s ) ( ∥ g ( s ) ∥ 2 + ∥ g ′ ( s ) ∥ 2 ) d s + N 1 ∥ Z ′ ∥ X τ , t 0 , b 2 ) ,
for t − τ t-\tau t − τ small enough, 1 / 2 < b < 25 / 48 1/2<b<25/48 1/2 < b < 25/48 , and consequently
∥ Z ′ ( t ) ∥ 2 ≤ C ( 1 + ∫ − ∞ t e − λ ( t − s ) ( ∥ g ( s ) ∥ 2 + ∥ g ′ ( s ) ∥ 2 ) d s ) , ∀ τ ≤ τ 0 , (34) \|Z'(t)\|^2\leq C\left(1+\int_{-\infty}^t e^{-\lambda(t-s)}\left(\|g(s)\|^2+\|g'(s)\|^2\right)\,ds\right),\qquad\forall\tau\leq\tau_0, \tag*{(34)} ∥ Z ′ ( t ) ∥ 2 ≤ C ( 1 + ∫ − ∞ t e − λ ( t − s ) ( ∥ g ( s ) ∥ 2 + ∥ g ′ ( s ) ∥ 2 ) d s ) , ∀ τ ≤ τ 0 , (34)
for some τ 0 ≤ t \tau_0\leq t τ 0 ≤ t . This gives the boundedness of v v v in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) .
Proposition 3.7. The pullback attractor A μ = { A μ ( t ) : t ∈ R } \mathcal{A}_\mu=\{\mathcal{A}_\mu(t):t\in\mathbb{R}\} A μ = { A μ ( t ) : t ∈ R } for Eq. (1) is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) , i.e., for any t ∈ R t\in\mathbb{R} t ∈ R , A μ ( t ) \mathcal{A}_\mu(t) A μ ( t ) is a bounded subset of H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) .
Proof. Let τ n → − ∞ \tau_n\to-\infty τ n → − ∞ and let { u τ n } \{u_{\tau_n}\} { u τ n } be a bounded sequence in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . Let v n v_n v n and w n w_n w n be the solutions of (16) and (17) respectively with initial data u τ n u_{\tau_n} u τ n . From 3.2.3, we have
{ v m ( t , τ n + ⋅ ) } n is bounded in C ( [ − T , T ] ; H 3 ( R ) ) , \{v_m(t,\tau_n+\cdot)\}_n\text{ is bounded in }C([-T,T];H^3(\mathbb{R})), { v m ( t , τ n + ⋅ ) } n is bounded in C ([ − T , T ] ; H 3 ( R )) ,
and, for the time derivative,
{ d v m d t ( t , τ n + ⋅ ) } n is bounded in C ( [ − T , T ] ; L 2 ( R ) ) , \left\{\frac{dv_m}{dt}(t,\tau_n+\cdot)\right\}_n\text{ is bounded in }C([-T,T];L^2(\mathbb{R})), { d t d v m ( t , τ n + ⋅ ) } n is bounded in C ([ − T , T ] ; L 2 ( R )) ,
for each T > 0 T>0 T > 0 (and starting with n n n sufficiently large so that t − T ≥ τ n t-T\geq\tau_n t − T ≥ τ n ). Taking the inner product of (16) with 2 v 2v 2 v in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) gives us an energy equality for v v v ,
∥ v ( t ) ∥ 2 + 2 μ ∫ τ t e − 2 λ ( t − s ) ∥ v x ( s ) ∥ 2 d s = e − 2 λ ( t − τ ) ∥ v ( τ ) ∥ 2 + ∫ τ t e − 2 λ ( t − s ) ( 2 ( g ( s ) , v ( s ) ) + ( P N ( 2 v ( s ) w ( s ) + w 2 ( s ) ) , v x ( s ) ) ) d s . \begin{aligned}
\|v(t)\|^2+2\mu\int_{\tau}^{t}e^{-2\lambda(t-s)}\|v_x(s)\|^2\,ds \\
&=e^{-2\lambda(t-\tau)}\|v(\tau)\|^2+\int_{\tau}^{t}e^{-2\lambda(t-s)}\left(2(g(s),v(s))+(P_N(2v(s)w(s)+w^2(s)),v_x(s))\right)\,ds.
\end{aligned} ∥ v ( t ) ∥ 2 + 2 μ ∫ τ t e − 2 λ ( t − s ) ∥ v x ( s ) ∥ 2 d s = e − 2 λ ( t − τ ) ∥ v ( τ ) ∥ 2 + ∫ τ t e − 2 λ ( t − s ) ( 2 ( g ( s ) , v ( s )) + ( P N ( 2 v ( s ) w ( s ) + w 2 ( s )) , v x ( s )) ) d s .
Hence,
∥ v ( t ) ∥ 2 + 2 μ ∫ t − T t e − 2 λ ( t − s ) ∥ v x ( s ) ∥ 2 d s = e − 2 λ T ∥ v ( t − T ) ∥ 2 + ∫ t − T t e − 2 λ ( t − s ) ( 2 ( g ( s ) , v ( s ) ) + ( P N ( 2 v ( s ) w ( s ) + w 2 ( s ) ) , v x ( s ) ) ) d s . (35) \|v(t)\|^2+2\mu\int_{t-T}^{t}e^{-2\lambda(t-s)}\|v_x(s)\|^2\,ds=e^{-2\lambda T}\|v(t-T)\|^2
+\int_{t-T}^{t}e^{-2\lambda(t-s)}\left(2(g(s),v(s))+(P_N(2v(s)w(s)+w^2(s)),v_x(s))\right)\,ds. \tag*{(35)} ∥ v ( t ) ∥ 2 + 2 μ ∫ t − T t e − 2 λ ( t − s ) ∥ v x ( s ) ∥ 2 d s = e − 2 λ T ∥ v ( t − T ) ∥ 2 + ∫ t − T t e − 2 λ ( t − s ) ( 2 ( g ( s ) , v ( s )) + ( P N ( 2 v ( s ) w ( s ) + w 2 ( s )) , v x ( s )) ) d s . (35)
Note that { v n } \{v_n\} { v n } is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) and w n → 0 w_n\to0 w n → 0 in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) as n → 0 n\to0 n → 0 . Thus, by using (35) , v n → v 0 v_n\to v_0 v n → v 0 in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) for some v 0 v_0 v 0 in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) (since { v n } \{v_n\} { v n } is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) ). On the other hand, by Theorem 3.5 ,
u n = v n + w n → w 0 in L 2 ( R ) as n → ∞ . u_n=v_n+w_n\to w_0\quad\text{in }L^2(\mathbb{R})\text{ as }n\to\infty. u n = v n + w n → w 0 in L 2 ( R ) as n → ∞.
This implies that v 0 = w 0 v_0=w_0 v 0 = w 0 and thus w 0 w_0 w 0 belongs to H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . Hence we can conclude that the pullback attractor A μ \mathcal{A}_\mu A μ is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . □
Compactness in H3(R)
Thanks to boundedness in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) and the compactness in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , we can use interpolation inequalities to get that the pullback attractor A μ \mathcal{A}_\mu A μ for (1) is compact in H s ( R ) H^s(\mathbb{R}) H s ( R ) for all s < 3 s<3 s < 3 , and it is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . We will prove that A μ \mathcal{A}_\mu A μ is indeed compact in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . To this end, we work with equations satisfied by u n ′ = d u n d t u_n'=\frac{du_n}{dt} u n ′ = d t d u n and show, by using the energy equation method applied to u n ′ u_n' u n ′ , some compactness property of the u n ′ u_n' u n ′ and then deduce the compactness of pullback attractor in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) .
Theorem 3.8. The pullback attractor A μ \mathcal{A}_\mu A μ corresponding to problem (1) is compact in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) .
Proof. We can now restrict the process under study to A μ \mathcal{A}_\mu A μ . Since A μ \mathcal{A}_\mu A μ is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) , we take initial data { u τ n } \{u_{\tau_n}\} { u τ n } being a bounded sequence in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) , where τ n → − ∞ \tau_n\to-\infty τ n → − ∞ . From equation (1) and the fact that A μ \mathcal{A}_\mu A μ is compact in H s ( R ) H^s(\mathbb{R}) H s ( R ) for s < 3 s<3 s < 3 , it is sufficient to prove the compactness of the sequence { u n ′ = d u n d t } \left\{u_n'=\frac{du_n}{dt}\right\} { u n ′ = d t d u n } in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) .
We use the equation for u n ′ u_n' u n ′ ,
d u n ′ d t + ∂ x x x u n ′ + ( u n u n ′ ) x + λ u n ′ − μ ∂ x x u n ′ = g ′ ( t ) . (36) \frac{du_n'}{dt}+\partial_{xxx}u_n'+(u_nu_n')_x+\lambda u_n'-\mu\partial_{xx}u_n'=g'(t). \tag*{(36)} d t d u n ′ + ∂ xxx u n ′ + ( u n u n ′ ) x + λ u n ′ − μ ∂ xx u n ′ = g ′ ( t ) . (36)
Since { u n ( t ) } n \{u_n(t)\}_n { u n ( t ) } n is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) and converges strongly to u 0 ( t ) u_0(t) u 0 ( t ) in C ( [ τ − T , τ + T ] ; L 2 ( R ) ) C([\tau-T,\tau+T];L^2(\mathbb{R})) C ([ τ − T , τ + T ] ; L 2 ( R )) , we see that
u n ′ ( t ) → u 0 ′ ( t ) strongly in C ( [ − T , T ] ; H − s ( R ) ) u_n'(t)\to u_0'(t)\text{ strongly in }C([-T,T];H^{-s}(\mathbb{R})) u n ′ ( t ) → u 0 ′ ( t ) strongly in C ([ − T , T ] ; H − s ( R ))
and weakly star in L ∞ ( [ − T , T ] ; L 2 ( R ) ) L^\infty([-T,T];L^2(\mathbb{R})) L ∞ ([ − T , T ] ; L 2 ( R )) . (3.30)
Now, consider the energy equality for u n ′ u_n' u n ′ by multiplying (36) with u n ′ u_n' u n ′ in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) then integrating over ( t − T , t ) (t-T,t) ( t − T , t )
∥ u n ′ ( t ) ∥ 2 + μ ∫ t − T t e − 2 λ ( t − s ) ∥ ∂ x u n ′ ( s ) ∥ 2 d s = e − 2 λ T ∥ u n ′ ( t − T ) ∥ 2 − ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u n , u n ′ 2 ) d s + ∫ t − T t e − 2 λ ( t − s ) ( g ′ ( s ) , u n ′ ( s ) ) d s . (37) \|u_n'(t)\|^2+\mu\int_{t-T}^{t}e^{-2\lambda(t-s)}\|\partial_xu_n'(s)\|^2\,ds=e^{-2\lambda T}\|u_n'(t-T)\|^2
-\int_{t-T}^{t}e^{-2\lambda(t-s)}(\partial_xu_n,u_n'^2)\,ds+\int_{t-T}^{t}e^{-2\lambda(t-s)}(g'(s),u_n'(s))\,ds. \tag*{(37)} ∥ u n ′ ( t ) ∥ 2 + μ ∫ t − T t e − 2 λ ( t − s ) ∥ ∂ x u n ′ ( s ) ∥ 2 d s = e − 2 λ T ∥ u n ′ ( t − T ) ∥ 2 − ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u n , u n ′2 ) d s + ∫ t − T t e − 2 λ ( t − s ) ( g ′ ( s ) , u n ′ ( s )) d s . (37)
We want to pass to the limit as n → + ∞ n\to+\infty n → + ∞ in (37) . On one hand, using (3.30), as n → ∞ n\to\infty n → ∞ , we get
∫ t − T t e − 2 λ ( t − s ) ( ∂ x u n − ∂ x u 0 , u 0 ′ 2 ) d s → 0. \int_{t-T}^{t}e^{-2\lambda(t-s)}\left(\partial_xu_n-\partial_xu_0,u_0'^2\right)\,ds\to0. ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u n − ∂ x u 0 , u 0 ′2 ) d s → 0.
On the other hand, for 0 < s ≤ 23 / 48 0 < s \le23/48 0 < s ≤ 23/48 and 1 / 2 < b < 25 / 48 1/2 < b < 25/48 1/2 < b < 25/48 , we take b ′ ′ b'' b ′′ such that 1 − b < b ′ ′ < 1 / 2 1-b < b'' < 1/2 1 − b < b ′′ < 1/2 and proceed as in (25) –(28) (but with − s -s − s and using Lemma 3.6) to find
∣ ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u n , u n ′ 2 − u 0 ′ 2 ) d s ∣ ≤ c ∥ u n ∥ X [ τ , τ + T ] s , b ∥ u n ′ − u 0 ′ ∥ X [ τ , τ + T ] − s , b 2 . \left|\int_{t-T}^{t} e^{-2\lambda(t-s)}\left(\partial_x u_n,u_n^{\prime2}-u_0^{\prime2}\right)\,ds\right| \le c\|u_n\|_{X^{s,b}_{[\tau,\tau+T]}}\|u_n'-u_0'\|_{X^{-s,b}_{[\tau,\tau+T]}}^2 . ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u n , u n ′2 − u 0 ′2 ) d s ≤ c ∥ u n ∥ X [ τ , τ + T ] s , b ∥ u n ′ − u 0 ′ ∥ X [ τ , τ + T ] − s , b 2 .
From (3.30) and the well-posedness of (36) in X [ τ , τ + T ] − s , b X^{-s,b}_{[\tau,\tau+T]} X [ τ , τ + T ] − s , b , we can pass to the limit as n n n goes to infinity to obtain
lim n → ∞ ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u n ( t + s ) , u n ′ ( t + s ) 2 ) d s = ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u 0 ( s ) , u 0 ′ ( s ) 2 ) d s . \lim_{n\to\infty}\int_{t-T}^{t}e^{-2\lambda(t-s)}\left(\partial_xu_n(t+s),u_n'(t+s)^2\right)\,ds=\int_{t-T}^{t}e^{-2\lambda(t-s)}\left(\partial_xu_0(s),u_0'(s)^2\right)\,ds. n → ∞ lim ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u n ( t + s ) , u n ′ ( t + s ) 2 ) d s = ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u 0 ( s ) , u 0 ′ ( s ) 2 ) d s .
By (3.30) once again, we get
lim n → ∞ ∫ t − T t e − 2 λ ( t − s ) ( g ′ ( s ) , u n ′ ( s ) ) d s = ∫ t − T t e − 2 λ ( t − s ) ( g ′ ( s ) , u 0 ′ ( s ) ) d s . \lim_{n\to\infty}\int_{t-T}^{t}e^{-2\lambda(t-s)}\left(g'(s),u_n'(s)\right)\,ds=\int_{t-T}^{t}e^{-2\lambda(t-s)}\left(g'(s),u_0'(s)\right)\,ds. n → ∞ lim ∫ t − T t e − 2 λ ( t − s ) ( g ′ ( s ) , u n ′ ( s ) ) d s = ∫ t − T t e − 2 λ ( t − s ) ( g ′ ( s ) , u 0 ′ ( s ) ) d s .
Then, from (37) ,
lim sup n → ∞ ( ∥ u n ′ ( t ) ∥ 2 + μ ∫ t − T t e − 2 λ ( t − s ) ∥ ∂ x u n ′ ( s ) ∥ 2 d s ) ≤ C e − 2 λ T − ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u 0 ( s ) , u 0 ′ ( s ) 2 ) d s + ∫ t − T t e − 2 λ ( t − s ) ( g ′ ( s ) , u 0 ′ ( s ) ) d s . \limsup_{n\to\infty}\left(\|u_n'(t)\|^2+\mu\int_{t-T}^{t}e^{-2\lambda(t-s)}\|\partial_xu_n'(s)\|^2\,ds\right)
\le Ce^{-2\lambda T}-\int_{t-T}^{t}e^{-2\lambda(t-s)}\left(\partial_xu_0(s),u_0'(s)^2\right)\,ds+\int_{t-T}^{t}e^{-2\lambda(t-s)}\left(g'(s),u_0'(s)\right)\,ds. n → ∞ lim sup ( ∥ u n ′ ( t ) ∥ 2 + μ ∫ t − T t e − 2 λ ( t − s ) ∥ ∂ x u n ′ ( s ) ∥ 2 d s ) ≤ C e − 2 λ T − ∫ t − T t e − 2 λ ( t − s ) ( ∂ x u 0 ( s ) , u 0 ′ ( s ) 2 ) d s + ∫ t − T t e − 2 λ ( t − s ) ( g ′ ( s ) , u 0 ′ ( s ) ) d s .
By employing a corresponding L 2 L^2 L 2 energy equality for u 0 ′ u_0' u 0 ′ and using
lim sup n → ∞ μ ∫ t − T t e − 2 λ ( t − s ) ∥ ∂ x u n ′ ( s ) ∥ 2 d s ≤ ∫ t − T t e − 2 λ ( t − s ) ∥ ∂ x u 0 ′ ( s ) ∥ 2 d s , \limsup_{n\to\infty}\mu\int_{t-T}^{t}e^{-2\lambda(t-s)}\|\partial_xu_n'(s)\|^2\,ds\le\int_{t-T}^{t}e^{-2\lambda(t-s)}\|\partial_xu_0'(s)\|^2\,ds, n → ∞ lim sup μ ∫ t − T t e − 2 λ ( t − s ) ∥ ∂ x u n ′ ( s ) ∥ 2 d s ≤ ∫ t − T t e − 2 λ ( t − s ) ∥ ∂ x u 0 ′ ( s ) ∥ 2 d s ,
we find that
lim sup n → ∞ ∥ u n ′ ( t ) ∥ 2 ≤ C e − 2 λ T + ∥ u 0 ′ ( t ) ∥ 2 . \limsup_{n\to\infty}\|u_n'(t)\|^2\le Ce^{-2\lambda T}+\|u_0'(t)\|^2. n → ∞ lim sup ∥ u n ′ ( t ) ∥ 2 ≤ C e − 2 λ T + ∥ u 0 ′ ( t ) ∥ 2 .
We let T T T go off to infinity to obtain
lim sup n → ∞ ∥ u n ′ ( t ) ∥ 2 ≤ ∥ u 0 ′ ( t ) ∥ 2 . \limsup_{n\to\infty}\|u_n'(t)\|^2\le\|u_0'(t)\|^2. n → ∞ lim sup ∥ u n ′ ( t ) ∥ 2 ≤ ∥ u 0 ′ ( t ) ∥ 2 .
Therefore,
u n ′ ( t ) → u 0 ′ ( t ) u_n'(t)\to u_0'(t) u n ′ ( t ) → u 0 ′ ( t ) strongly in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) .
This complete the proof.
Upper semicontinuity at μ = 0 \mu= 0 μ = 0
In this section, we will prove that, as μ → 0 + \mu\to0^+ μ → 0 + , the pullback attractor A μ \mathcal{A}_\mu A μ of the KdV-Burgers equation converges to the pullback attractor A 0 \mathcal{A}_0 A 0 of the weakly damped KdV equation in the sense of the Hausdorff semi-distance. To this end, we first show
Theorem 4.1. Assume that the external force g g g satisfies (2) , then there exists a unique pullback attractor A 0 = { A 0 ( t ) } \mathcal{A}_0=\{\mathcal{A}_0(t)\} A 0 = { A 0 ( t )} in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) for the equation
{ u t + u x x x + u u x + λ u = g ( t , x ) , x ∈ R , t > 0 , u ( τ , x ) = u τ ( x ) , x ∈ R . (38) \begin{cases}
u_t+u_{xxx}+uu_x+\lambda u=g(t,x),\quad x\in\mathbb{R},\quad t>0,\\
u(\tau,x)=u_\tau(x),\quad x\in\mathbb{R}. \tag*{(38)}
\end{cases} { u t + u xxx + u u x + λ u = g ( t , x ) , x ∈ R , t > 0 , u ( τ , x ) = u τ ( x ) , x ∈ R . (38) Moreover, A 0 ( t ) \mathcal{A}_0(t) A 0 ( t ) is compact in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) for all t ∈ R t\in\mathbb{R} t ∈ R .
Proof. The arguments of the proof of this Theorem follow from the previous section, except the fact that we need to work in the Bourgain spaces corresponding to the free equation of (38) instead of the spaces defined by (3) . More precisely, we use the Bourgain space
X s , b = { u : R → R : ∥ u ∥ X s , b < + ∞ } X^{s,b}=\{u:\mathbb{R}\to\mathbb{R}:\|u\|_{X^{s,b}}<+\infty\} X s , b = { u : R → R : ∥ u ∥ X s , b < + ∞ }
where
∥ u ∥ X s , b = ∥ ⟨ τ − ξ 3 ⟩ ⟨ ξ ⟩ s u ^ ( τ , ξ ) ∥ L 2 ( R 2 ) . \|u\|_{X^{s,b}}=\left\|\langle\tau-\xi^3\rangle\langle\xi\rangle^s\widehat{u}(\tau,\xi)\right\|_{L^2(\mathbb{R}^2)}. ∥ u ∥ X s , b = ⟨ τ − ξ 3 ⟩ ⟨ ξ ⟩ s u ( τ , ξ ) L 2 ( R 2 ) .
The crucial estimate (18) in this case is replaced by
∥ D ( u 2 ) ∥ X s , b ′ − 1 ≤ c s ∥ u ∥ X s , b ∥ u ∥ X − 1 / 2 , b , \|D(u^2)\|_{X^{s,b'-1}}\le c_s\|u\|_{X^{s,b}}\|u\|_{X^{-1/2,b}}, ∥ D ( u 2 ) ∥ X s , b ′ − 1 ≤ c s ∥ u ∥ X s , b ∥ u ∥ X − 1/2 , b ,
which was proved in [11 Proposition 2.2] .
Repeat the arguments in the previous section we first get the existence of the pullback attractor A 0 = { A 0 ( t ) } \mathcal{A}_0=\{\mathcal{A}_0(t)\} A 0 = { A 0 ( t )} in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , then by splitting solution technique (see (16) and (17) ) we obtain the compactness of A 0 \mathcal{A}_0 A 0 in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) .
Thanks to the compactness of A μ \mathcal{A}_\mu A μ and A 0 \mathcal{A}_0 A 0 in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) , we can from now on assume that the initial data u τ u_\tau u τ belongs to H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . We also additionally assume that the external force g g g belongs to L l o c 2 ( R ; H 1 ( R ) ) L^2_{\mathrm{loc}}(\mathbb{R};H^1(\mathbb{R})) L loc 2 ( R ; H 1 ( R )) and satisfies
lim t → − ∞ e − λ t ∫ − ∞ t e λ s ∥ g ( s ) ∥ H 1 ( R ) 2 d s < + ∞ . (39) \lim_{t\to-\infty} e^{-\lambda t}\int_{-\infty}^{t}e^{\lambda s}\lVert g(s)\rVert_{H^1(\mathbb{R})}^{2}\,\mathrm{d}s<+\infty. \tag*{(39)} t → − ∞ lim e − λ t ∫ − ∞ t e λ s ∥ g ( s ) ∥ H 1 ( R ) 2 d s < + ∞. (39)
We need the following lemmas.
Lemma 4.2 ([23 ] ). Let V V V be a Banach space, H H H be a Hilbert space, and V ⊆ H V\subseteq H V ⊆ H be continuous. Suppose that { u n } \{u_n\} { u n } is bounded in L ∞ ( τ , T ; V ) L^\infty(\tau,T;V) L ∞ ( τ , T ; V ) ,
ess sup t ∈ [ τ , T ] ∥ u n ( t ) ∥ V ≤ C , \mathop{\operatorname{ess\,sup}}_{t\in[\tau,T]}\lVert u_n(t)\rVert_V\le C, ess sup t ∈ [ τ , T ] ∥ u n ( t ) ∥ V ≤ C , and that u n ⇀ u u_n\rightharpoonup u u n ⇀ u in L 2 ( τ , T ; V ) L^2(\tau,T;V) L 2 ( τ , T ; V ) , where, in addition, u ∈ C ( [ τ , T ] ; H ) u\in C([\tau,T];H) u ∈ C ([ τ , T ] ; H ) . Then
sup t ∈ [ τ , T ] ∥ u ( t ) ∥ V ≤ C . \sup_{t\in[\tau,T]}\lVert u(t)\rVert_V\le C. t ∈ [ τ , T ] sup ∥ u ( t ) ∥ V ≤ C .
Lemma 4.3. Let u ( t ) u(t) u ( t ) be a solution of (1) with initial data u τ ∈ H 3 ( R ) u_\tau\in H^3(\mathbb{R}) u τ ∈ H 3 ( R ) . Then u u u is bounded in L ∞ ( τ , t ; H 2 ( R ) ) L^\infty(\tau,t;H^2(\mathbb{R})) L ∞ ( τ , t ; H 2 ( R )) uniformly in μ ∈ [ 0 , 1 ] \mu\in[0,1] μ ∈ [ 0 , 1 ] , i.e.,
∥ u ∥ L ∞ ( τ , t ; H 2 ( R ) ) ≤ C , (40) \lVert u\rVert_{L^\infty(\tau,t;H^2(\mathbb{R}))}\le C, \tag*{(40)} ∥ u ∥ L ∞ ( τ , t ; H 2 ( R )) ≤ C , (40) where C C C depends only on τ \tau τ , t t t , and ∥ u τ ∥ H 3 ( R ) \lVert u_\tau\rVert_{H^3(\mathbb{R})} ∥ u τ ∥ H 3 ( R ) , but is independent of μ \mu μ .
Proof. As in [DS10], we consider the regularized equation
{ u t ε + ε u x x x x ε + u x x x ε + u ε u x ε + λ u ε − μ u x x ε = g ( t ) , u ∣ t = τ = u τ ∈ H 3 ( R ) . \begin{cases}
u^\varepsilon_t+\varepsilon u^\varepsilon_{xxxx}+u^\varepsilon_{xxx}+u^\varepsilon u^\varepsilon_x+\lambda u^\varepsilon-\mu u^\varepsilon_{xx}=g(t),\\
u|_{t=\tau}=u_\tau\in H^3(\mathbb{R}).
\end{cases} { u t ε + ε u xxxx ε + u xxx ε + u ε u x ε + λ u ε − μ u xx ε = g ( t ) , u ∣ t = τ = u τ ∈ H 3 ( R ) . By results of [DS10],
{ u ε } ε ∈ ( 0 , 1 ] is bounded in L ∞ ( τ , t ; H 2 ( R ) ) uniformly in μ \{u^\varepsilon\}_{\varepsilon\in(0,1]}\text{ is bounded in }L^\infty(\tau,t;H^2(\mathbb{R}))\text{ uniformly in }\mu { u ε } ε ∈ ( 0 , 1 ] is bounded in L ∞ ( τ , t ; H 2 ( R )) uniformly in μ and
u ε ⇀ u in L 2 ( τ , t ; H 2 ( R ) ) as ε → 0 + . u^\varepsilon\rightharpoonup u\text{ in }L^2(\tau,t;H^2(\mathbb{R}))\text{ as }\varepsilon\to0^+. u ε ⇀ u in L 2 ( τ , t ; H 2 ( R )) as ε → 0 + . Using Lemma 4.2, we immediately get (40) .
Lemma 4.4. For any t ∈ R t\in\mathbb{R} t ∈ R , the set
⋃ τ ≤ t ⋃ μ ∈ [ 0 , 1 ] A μ ( t ) \bigcup_{\tau\le t}\bigcup_{\mu\in[0,1]}\mathcal{A}_\mu(t) τ ≤ t ⋃ μ ∈ [ 0 , 1 ] ⋃ A μ ( t ) is bounded in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) .
Proof. From Proposition 3.3, { U μ ( t , τ ) } \{U_\mu(t,\tau)\} { U μ ( t , τ )} has a pullback absorbing set B μ = { B μ ( t ) } t ∈ R \mathcal{B}_\mu=\{\mathcal{B}_\mu(t)\}_{t\in\mathbb{R}} B μ = { B μ ( t ) } t ∈ R in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , where B μ ( t ) \mathcal{B}_\mu(t) B μ ( t ) is the ball centered at the origin with radius
R ( t ) = ( e − λ t λ ∫ − ∞ t e λ s ∥ g ( s ) ∥ 2 d s ) 1 / 2 . R(t)=\left(\frac{e^{-\lambda t}}{\lambda}\int_{-\infty}^{t}e^{\lambda s}\lVert g(s)\rVert^2\,\mathrm{d}s\right)^{1/2}. R ( t ) = ( λ e − λ t ∫ − ∞ t e λ s ∥ g ( s ) ∥ 2 d s ) 1/2 . Note that R ( t ) R(t) R ( t ) is independent of μ \mu μ . Combining this with A μ ( t ) ⊆ B μ ( t ) \mathcal{A}_\mu(t)\subseteq\mathcal{B}_\mu(t) A μ ( t ) ⊆ B μ ( t ) , we get
⋃ μ ∈ [ 0 , 1 ] A μ ( t ) ⊆ B 0 ( t ) . (41) \bigcup_{\mu\in[0,1]}\mathcal{A}_\mu(t)\subseteq\mathcal{B}_0(t). \tag*{(41)} μ ∈ [ 0 , 1 ] ⋃ A μ ( t ) ⊆ B 0 ( t ) . (41) On the other hand, by assumption (39) on g g g ,
lim sup t → − ∞ R ( t ) < + ∞ . \limsup_{t\to-\infty}R(t)<+\infty. t → − ∞ lim sup R ( t ) < + ∞.
This implies that
⋃ τ ≤ t B 0 ( τ ) is bounded in L 2 ( R ) . (42) \bigcup_{\tau\le t} B_0(\tau) \text{ is bounded in } L^2(\mathbb{R}). \tag*{(42)} τ ≤ t ⋃ B 0 ( τ ) is bounded in L 2 ( R ) . (42)
From (41) and (42) , we get the desired result. □
Lemma 4.5. For any t ∈ R t\in\mathbb{R} t ∈ R , the set ⋃ μ ∈ [ 0 , 1 ] A μ ( t ) \bigcup_{\mu\in[0,1]} A_\mu(t) ⋃ μ ∈ [ 0 , 1 ] A μ ( t ) is precompact in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) .
Proof. By the invariance property of A μ A_\mu A μ ,
A μ ( t ) = U μ ( t , τ ) A μ ( τ ) , ∀ t ≥ τ . A_\mu(t)=U_\mu(t,\tau)A_\mu(\tau),\qquad\forall t\ge\tau. A μ ( t ) = U μ ( t , τ ) A μ ( τ ) , ∀ t ≥ τ .
In the previous section, we proved that A μ ( t ) A_\mu(t) A μ ( t ) is compact in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . Thus, any element u ( t ) u(t) u ( t ) of A μ ( t ) A_\mu(t) A μ ( t ) can be considered as a solution of (1.1) at time t t t with initial data belonging to H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) . Hence, the result follows directly from [1 Theorem 5.2] . For the reader’s convenience, we sketch the proof here. Namely, we prove that, for any η > 0 \eta>0 η > 0 , there exists a finite covering of the set ⋃ μ ∈ ( 0 , 1 ] A μ ( t ) \bigcup_{\mu\in(0,1]} A_\mu(t) ⋃ μ ∈ ( 0 , 1 ] A μ ( t ) in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) by balls of radius less than η \eta η . First, using the boundedness of u u u from Lemma 4.3, we multiply equation (1.1) by θ ( x 2 / k 2 ) u \theta(x^2/k^2)u θ ( x 2 / k 2 ) u , where θ : [ 0 , ∞ ) → [ 0 , 1 ] \theta:[0,\infty)\to[0,1] θ : [ 0 , ∞ ) → [ 0 , 1 ] is the smooth cut-off function satisfying θ ∣ [ 0 , 1 ] = 0 \theta|_{[0,1]}=0 θ ∣ [ 0 , 1 ] = 0 and θ ∣ [ 2 , + ∞ ) = 1 \theta|_{[2,+\infty)}=1 θ ∣ [ 2 , + ∞ ) = 1 , to obtain that
∫ ∣ x ∣ ≥ k ∣ u ∣ 2 d x ≤ η 4 , ∀ u ∈ ⋃ μ ∈ ( 0 , 1 ] A μ ( t ) , (43) \int_{\lvert x\rvert\ge k}|u|^2\,dx\le\frac{\eta}{4},\qquad\forall u\in\bigcup_{\mu\in(0,1]}A_\mu(t), \tag*{(43)} ∫ ∣ x ∣ ≥ k ∣ u ∣ 2 d x ≤ 4 η , ∀ u ∈ μ ∈ ( 0 , 1 ] ⋃ A μ ( t ) , (43)
for some k > 0 k>0 k > 0 .
Next, from the boundedness of solutions in H 2 ( R ) H^2(\mathbb{R}) H 2 ( R ) , we conclude that
∥ u ∥ H 1 ( − k , k ) ≤ C , ∀ u ∈ ⋃ μ ∈ ( 0 , 1 ] A μ ( t ) . (44) \lVert u\rVert_{H^1(-k,k)}\le C,\qquad\forall u\in\bigcup_{\mu\in(0,1]}A_\mu(t). \tag*{(44)} ∥ u ∥ H 1 ( − k , k ) ≤ C , ∀ u ∈ μ ∈ ( 0 , 1 ] ⋃ A μ ( t ) . (44)
Because H 1 ( − k , k ) ↪ L 2 ( − k , k ) H^1(-k,k)\hookrightarrow L^2(-k,k) H 1 ( − k , k ) ↪ L 2 ( − k , k ) compactly, we obtain from (43) and (44) that ⋃ μ ∈ ( 0 , 1 ] A μ ( t ) \bigcup_{\mu\in(0,1]}A_\mu(t) ⋃ μ ∈ ( 0 , 1 ] A μ ( t ) has the desired finite covering in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) by balls of radius less than η \eta η . This completes the proof. □
The following result is the key in proving the upper semicontinuity.
Lemma 4.6. Let { x n } ⊂ H 3 ( R ) \{x_n\}\subset H^3(\mathbb{R}) { x n } ⊂ H 3 ( R ) and { μ n } ⊂ [ 0 , 1 ] \{\mu_n\}\subset[0,1] { μ n } ⊂ [ 0 , 1 ] be two sequences such that { x n } \{x_n\} { x n } is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) , x n → x 0 x_n\to x_0 x n → x 0 in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) and μ n → 0 \mu_n\to0 μ n → 0 . Then, for any t ≥ τ t\ge\tau t ≥ τ ,
∥ U μ n ( t , τ ) x n − U 0 ( t , τ ) x 0 ∥ → 0 as n → 0. (45) \lVert U_{\mu_n}(t,\tau)x_n-U_0(t,\tau)x_0\rVert\to0\quad\text{as }n\to0. \tag*{(45)} ∥ U μ n ( t , τ ) x n − U 0 ( t , τ ) x 0 ∥ → 0 as n → 0. (45)
Proof. Denote u n ( t ) = U μ n ( t , τ ) x n u_n(t)=U_{\mu_n}(t,\tau)x_n u n ( t ) = U μ n ( t , τ ) x n and v ( t ) = U 0 ( t , τ ) x 0 v(t)=U_0(t,\tau)x_0 v ( t ) = U 0 ( t , τ ) x 0 . Then, u n u_n u n and v v v solve the equations
{ ∂ t u n + ∂ x x x u n + u n ∂ x u n + λ u n − μ n ∂ x x u n = g ( t , x ) , u n ( τ ) = x n , \begin{cases}
\partial_tu_n+\partial_{xxx}u_n+u_n\partial_xu_n+\lambda u_n-\mu_n\partial_{xx}u_n=g(t,x),\\
u_n(\tau)=x_n,
\end{cases} { ∂ t u n + ∂ xxx u n + u n ∂ x u n + λ u n − μ n ∂ xx u n = g ( t , x ) , u n ( τ ) = x n ,
and
{ v t + v x x x + v v x + λ v = g ( t , x ) , v ( τ ) = x 0 , \begin{cases}
v_t+v_{xxx}+vv_x+\lambda v=g(t,x),\\
v(\tau)=x_0,
\end{cases} { v t + v xxx + v v x + λ v = g ( t , x ) , v ( τ ) = x 0 ,
respectively. Let w n ( t ) = u n ( t ) − v ( t ) w_n(t)=u_n(t)-v(t) w n ( t ) = u n ( t ) − v ( t ) . Then we have
∂ t w n + ∂ x x x w n + ( u n ∂ x u n − v v x ) + λ w n − μ n ∂ x x u n = 0 (46) \partial_tw_n+\partial_{xxx}w_n+(u_n\partial_xu_n-vv_x)+\lambda w_n-\mu_n\partial_{xx}u_n=0 \tag*{(46)} ∂ t w n + ∂ xxx w n + ( u n ∂ x u n − v v x ) + λ w n − μ n ∂ xx u n = 0 (46)
with initial data w n ( τ ) = x n − x 0 w_n(\tau)=x_n-x_0 w n ( τ ) = x n − x 0 . Multiplying (46) by w n w_n w n and integrating over R \mathbb{R} R , we obtain
1 2 d d t ∥ w n ∥ 2 + ∫ R ( u n ∂ x u n − v v x ) w n d x + λ ∥ w n ∥ 2 + μ n ∫ R ∂ x u n ∂ x w n d x = 0. (47) \frac{1}{2}\frac{d}{dt}\lVert w_n\rVert^2+\int_{\mathbb{R}}(u_n\partial_xu_n-vv_x)w_n\,dx+\lambda\lVert w_n\rVert^2+\mu_n\int_{\mathbb{R}}\partial_xu_n\partial_xw_n\,dx=0. \tag*{(47)} 2 1 d t d ∥ w n ∥ 2 + ∫ R ( u n ∂ x u n − v v x ) w n d x + λ ∥ w n ∥ 2 + μ n ∫ R ∂ x u n ∂ x w n d x = 0. (47)
Integrating by parts and using interpolation inequalities, we find that
∣ ∫ R ( u n ∂ x u n − v v x ) w n d x ∣ = 1 2 ∣ ∫ R ( u n 2 − v 2 ) w n d x ∣ = 1 2 ∣ ∫ R ( u n + v ) w n ∂ x w n d x ∣ ≤ ∫ R ( ∣ ∂ x u n ∣ + ∣ v x ∣ ) w n 2 d x ≤ ( ∥ ∂ x u n ∥ L ∞ ( R ) + ∥ v x ∥ L ∞ ( R ) ) ∥ w n ∥ 2 ≤ ( ∥ u n ∥ H 2 ( R ) + ∥ v ∥ H 2 ( R ) ) ∥ w n ∥ 2 . (48) \begin{aligned}
\left|\int_{\mathbb{R}}(u_n\partial_xu_n-vv_x)w_n\,dx\right|
&=\frac{1}{2}\left|\int_{\mathbb{R}}(u_n^2-v^2)w_n\,dx\right|\\
&=\frac{1}{2}\left|\int_{\mathbb{R}}(u_n+v)w_n\partial_xw_n\,dx\right|\\
&\leq\int_{\mathbb{R}}(|\partial_xu_n|+|v_x|)w_n^2\,dx\\
&\leq\left(\|\partial_xu_n\|_{L^\infty(\mathbb{R})}+\|v_x\|_{L^\infty(\mathbb{R})}\right)\|w_n\|^2\\
&\leq\left(\|u_n\|_{H^2(\mathbb{R})}+\|v\|_{H^2(\mathbb{R})}\right)\|w_n\|^2. \tag*{(48)}
\end{aligned} ∫ R ( u n ∂ x u n − v v x ) w n d x = 2 1 ∫ R ( u n 2 − v 2 ) w n d x = 2 1 ∫ R ( u n + v ) w n ∂ x w n d x ≤ ∫ R ( ∣ ∂ x u n ∣ + ∣ v x ∣ ) w n 2 d x ≤ ( ∥ ∂ x u n ∥ L ∞ ( R ) + ∥ v x ∥ L ∞ ( R ) ) ∥ w n ∥ 2 ≤ ( ∥ u n ∥ H 2 ( R ) + ∥ v ∥ H 2 ( R ) ) ∥ w n ∥ 2 . (48)
On the other hand, by Young’s inequality,
∣ μ n ∫ R ∂ x u n ∂ x w n d x ∣ ≤ μ n ∣ ∫ R ∂ x u n ∂ x w n d x ∣ ≤ μ n ∥ ∂ x u n ∥ L 2 ( R ) ∥ ∂ x w n ∥ L 2 ( R ) ≤ μ n ∥ u n ∥ H 2 ( R ) ∥ w n ∥ H 2 ( R ) . (49) \begin{aligned}
\left|\mu_n\int_{\mathbb{R}}\partial_xu_n\partial_xw_n\,dx\right|
&\leq\mu_n\left|\int_{\mathbb{R}}\partial_xu_n\partial_xw_n\,dx\right|\\
&\leq\mu_n\|\partial_xu_n\|_{L^2(\mathbb{R})}\|\partial_xw_n\|_{L^2(\mathbb{R})}\\
&\leq\mu_n\|u_n\|_{H^2(\mathbb{R})}\|w_n\|_{H^2(\mathbb{R})}. \tag*{(49)}
\end{aligned} μ n ∫ R ∂ x u n ∂ x w n d x ≤ μ n ∫ R ∂ x u n ∂ x w n d x ≤ μ n ∥ ∂ x u n ∥ L 2 ( R ) ∥ ∂ x w n ∥ L 2 ( R ) ≤ μ n ∥ u n ∥ H 2 ( R ) ∥ w n ∥ H 2 ( R ) . (49)
From (47) –(49) , we obtain
d d t ∥ w n ∥ 2 ≤ ( ∥ u n ∥ H 2 ( R ) + ∥ v ∥ H 2 ( R ) ) ∥ w n ∥ 2 + μ n ∥ u n ∥ H 2 ( R ) ∥ w n ∥ H 2 ( R ) . \frac{d}{dt}\|w_n\|^2\leq\left(\|u_n\|_{H^2(\mathbb{R})}+\|v\|_{H^2(\mathbb{R})}\right)\|w_n\|^2+\mu_n\|u_n\|_{H^2(\mathbb{R})}\|w_n\|_{H^2(\mathbb{R})}. d t d ∥ w n ∥ 2 ≤ ( ∥ u n ∥ H 2 ( R ) + ∥ v ∥ H 2 ( R ) ) ∥ w n ∥ 2 + μ n ∥ u n ∥ H 2 ( R ) ∥ w n ∥ H 2 ( R ) .
Integrating from τ \tau τ to t t t and using Lemma 4.3 yields
∥ w n ( t ) ∥ 2 ≤ ∥ w n ( τ ) ∥ 2 + ∫ τ t ( ∥ u n ( s ) ∥ H 2 ( R ) + ∥ v ( s ) ∥ H 2 ( R ) ) ∥ w n ( s ) ∥ 2 d s + μ n ∫ τ t ∥ u n ( s ) ∥ H 2 ( R ) ∥ w n ( s ) ∥ H 2 ( R ) d s ≤ ∥ w n ( τ ) ∥ 2 + ( ∥ u n ∥ L ∞ ( τ , t ; H 2 ( R ) ) + ∥ w n ∥ L ∞ ( τ , t ; H 2 ( R ) ) ) ∫ τ t ∥ w n ( s ) ∥ 2 d s + μ n ∥ u n ∥ L 2 ( τ , t ; H 2 ( R ) ) ∥ w n ∥ L 2 ( τ , t ; H 2 ( R ) ) . \begin{aligned}
\|w_n(t)\|^2
&\leq\|w_n(\tau)\|^2+\int_\tau^t\left(\|u_n(s)\|_{H^2(\mathbb{R})}+\|v(s)\|_{H^2(\mathbb{R})}\right)\|w_n(s)\|^2\,ds\\
&\quad+\mu_n\int_\tau^t\|u_n(s)\|_{H^2(\mathbb{R})}\|w_n(s)\|_{H^2(\mathbb{R})}\,ds\\
&\leq\|w_n(\tau)\|^2+\left(\|u_n\|_{L^\infty(\tau,t;H^2(\mathbb{R}))}+\|w_n\|_{L^\infty(\tau,t;H^2(\mathbb{R}))}\right)\int_\tau^t\|w_n(s)\|^2\,ds\\
&\quad+\mu_n\|u_n\|_{L^2(\tau,t;H^2(\mathbb{R}))}\|w_n\|_{L^2(\tau,t;H^2(\mathbb{R}))}.
\end{aligned} ∥ w n ( t ) ∥ 2 ≤ ∥ w n ( τ ) ∥ 2 + ∫ τ t ( ∥ u n ( s ) ∥ H 2 ( R ) + ∥ v ( s ) ∥ H 2 ( R ) ) ∥ w n ( s ) ∥ 2 d s + μ n ∫ τ t ∥ u n ( s ) ∥ H 2 ( R ) ∥ w n ( s ) ∥ H 2 ( R ) d s ≤ ∥ w n ( τ ) ∥ 2 + ( ∥ u n ∥ L ∞ ( τ , t ; H 2 ( R )) + ∥ w n ∥ L ∞ ( τ , t ; H 2 ( R )) ) ∫ τ t ∥ w n ( s ) ∥ 2 d s + μ n ∥ u n ∥ L 2 ( τ , t ; H 2 ( R )) ∥ w n ∥ L 2 ( τ , t ; H 2 ( R )) .
Applying Gronwall’s lemma, we find that
∥ w n ( t ) ∥ 2 ≤ ( ∥ w n ( τ ) ∥ 2 + μ n ∥ u n ∥ L 2 ( τ , t ; H 2 ( R ) ) ∥ w n ∥ L 2 ( τ , t ; H 2 ( R ) ) ) × exp ( ( ∥ u n ∥ L ∞ ( τ , t ; H 2 ( R ) ) + ∥ w n ∥ L ∞ ( τ , t ; H 2 ( R ) ) ) ( t − τ ) ) . \begin{aligned}
\|w_n(t)\|^2
&\leq\left(\|w_n(\tau)\|^2+\mu_n\|u_n\|_{L^2(\tau,t;H^2(\mathbb{R}))}\|w_n\|_{L^2(\tau,t;H^2(\mathbb{R}))}\right)\\
&\quad\times\exp\left(\left(\|u_n\|_{L^\infty(\tau,t;H^2(\mathbb{R}))}+\|w_n\|_{L^\infty(\tau,t;H^2(\mathbb{R}))}\right)(t-\tau)\right).
\end{aligned} ∥ w n ( t ) ∥ 2 ≤ ( ∥ w n ( τ ) ∥ 2 + μ n ∥ u n ∥ L 2 ( τ , t ; H 2 ( R )) ∥ w n ∥ L 2 ( τ , t ; H 2 ( R )) ) × exp ( ( ∥ u n ∥ L ∞ ( τ , t ; H 2 ( R )) + ∥ w n ∥ L ∞ ( τ , t ; H 2 ( R )) ) ( t − τ ) ) .
Now let n → ∞ n\to\infty n → ∞ and note that w n ( τ ) = x n − x 0 w_n(\tau)=x_n-x_0 w n ( τ ) = x n − x 0 , then use (40) and x n → x 0 x_n\to x_0 x n → x 0 in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) , to get the desired result. □ \square □
We can now prove the main result of this section.
Theorem 4.7. Assume that g ∈ L l o c 2 ( R ; H 1 ( R ) ) g\in L^2_{\mathrm{loc}}(\mathbb{R};H^1(\mathbb{R})) g ∈ L loc 2 ( R ; H 1 ( R )) satisfies (39) . Then the family of pullback attractors { A μ } μ ∈ [ 0 , 1 ] \{A_\mu\}_{\mu\in[0,1]} { A μ } μ ∈ [ 0 , 1 ] corresponding to Eq. (1) is upper semi-continuous in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) at μ = 0 \mu=0 μ = 0 , i.e.,
lim μ → 0 + dist L 2 ( R ) ( A μ ( t ) , A 0 ( t ) ) = 0 for all t ∈ R , \lim_{\mu\to0^+}\operatorname{dist}_{L^2(\mathbb{R})}(A_\mu(t),A_0(t))=0\quad\text{for all }t\in\mathbb{R}, μ → 0 + lim dist L 2 ( R ) ( A μ ( t ) , A 0 ( t )) = 0 for all t ∈ R , where dist L 2 ( R ) \operatorname{dist}_{L^2(\mathbb{R})} dist L 2 ( R ) denotes the Hausdorff semi-distance in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) .
Proof. We prove this by contradiction. Assume that there exists a t 0 ∈ R t_0\in\mathbb{R} t 0 ∈ R such that
dist L 2 ( R ) ( A μ ( t 0 ) , A 0 ( t 0 ) ) ↛ 0 as μ → 0 + . \operatorname{dist}_{L^2(\mathbb{R})}(A_\mu(t_0),A_0(t_0))\not\to0\quad\text{as }\mu\to0^+. dist L 2 ( R ) ( A μ ( t 0 ) , A 0 ( t 0 )) → 0 as μ → 0 + .
By the compactness of A μ A_\mu A μ , we can choose a positive constant δ > 0 \delta>0 δ > 0 , a sequence μ n → 0 \mu_n\to0 μ n → 0 , and a sequence { φ n } \{\varphi_n\} { φ n } such that φ n ∈ A μ n ( t 0 ) \varphi_n\in A_{\mu_n}(t_0) φ n ∈ A μ n ( t 0 ) and
dist L 2 ( R ) ( φ n , A 0 ( t 0 ) ) ≥ δ for all n ≥ 1. (50) \operatorname{dist}_{L^2(\mathbb{R})}(\varphi_n,A_0(t_0))\geq\delta\quad\text{for all }n\geq1. \tag*{(50)} dist L 2 ( R ) ( φ n , A 0 ( t 0 )) ≥ δ for all n ≥ 1. (50)
By Lemma 4.4, the set ⋃ τ ≤ t 0 ⋃ μ ∈ [ 0 , 1 ] A μ ( τ ) \bigcup_{\tau\le t_0}\bigcup_{\mu\in[0,1]} A_\mu(\tau) ⋃ τ ≤ t 0 ⋃ μ ∈ [ 0 , 1 ] A μ ( τ ) is bounded in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . Then, by the pullback attracting property of { U 0 ( t , τ ) } \{U_0(t,\tau)\} { U 0 ( t , τ )} , we can choose a τ 0 ≤ t 0 \tau_0\le t_0 τ 0 ≤ t 0 satisfying
dist L 2 ( R ) ( U 0 ( t 0 , τ 0 ) ψ , A 0 ( t ) ) ≤ δ 4 \operatorname{dist}_{L^2(\mathbb{R})}(U_0(t_0,\tau_0)\psi,A_0(t))\le\frac{\delta}{4} dist L 2 ( R ) ( U 0 ( t 0 , τ 0 ) ψ , A 0 ( t )) ≤ 4 δ
for all ψ ∈ ⋃ τ ≤ t 0 ⋃ μ ∈ [ 0 , 1 ] A μ ( τ 0 ) \psi\in\bigcup_{\tau\le t_0}\bigcup_{\mu\in[0,1]} A_\mu(\tau_0) ψ ∈ ⋃ τ ≤ t 0 ⋃ μ ∈ [ 0 , 1 ] A μ ( τ 0 ) .
Since A μ n A_{\mu_n} A μ n is invariant, for each n ≥ 1 n\ge1 n ≥ 1 ,
U μ n ( t 0 , τ 0 ) A μ n ( τ 0 ) = A μ n ( t 0 ) . U_{\mu_n}(t_0,\tau_0)A_{\mu_n}(\tau_0)=A_{\mu_n}(t_0). U μ n ( t 0 , τ 0 ) A μ n ( τ 0 ) = A μ n ( t 0 ) .
Thus, there exist ψ n ∈ A μ n ( τ 0 ) \psi_n\in A_{\mu_n}(\tau_0) ψ n ∈ A μ n ( τ 0 ) such that
φ n = U μ n ( t 0 , τ 0 ) ψ n . (51) \varphi_n=U_{\mu_n}(t_0,\tau_0)\psi_n. \tag*{(51)} φ n = U μ n ( t 0 , τ 0 ) ψ n . (51)
The sequence { ψ n } \{\psi_n\} { ψ n } is contained in the set ⋃ n ≥ 1 A μ n ( τ 0 ) ⊆ ⋃ μ ∈ [ 0 , 1 ] A μ ( τ 0 ) \bigcup_{n\ge1}A_{\mu_n}(\tau_0)\subseteq\bigcup_{\mu\in[0,1]}A_\mu(\tau_0) ⋃ n ≥ 1 A μ n ( τ 0 ) ⊆ ⋃ μ ∈ [ 0 , 1 ] A μ ( τ 0 ) . By Lemma 4.5, we can assume that ψ n → ψ 0 \psi_n\to\psi_0 ψ n → ψ 0 in L 2 ( R ) L^2(\mathbb{R}) L 2 ( R ) . Applying Lemma 4.6, we have
φ n = U μ n ( t 0 , τ 0 ) ψ n → U 0 ( t 0 , τ 0 ) ψ 0 in L 2 ( R ) as n → ∞ . \varphi_n=U_{\mu_n}(t_0,\tau_0)\psi_n\to U_0(t_0,\tau_0)\psi_0\quad\text{in }L^2(\mathbb{R})\text{ as }n\to\infty. φ n = U μ n ( t 0 , τ 0 ) ψ n → U 0 ( t 0 , τ 0 ) ψ 0 in L 2 ( R ) as n → ∞.
Choose N N N so large that
∥ U μ n ( t 0 , τ 0 ) ψ n − U 0 ( t 0 , τ 0 ) ψ 0 ∥ ≤ δ 4 for all n ≥ N . (52) \left\|U_{\mu_n}(t_0,\tau_0)\psi_n-U_0(t_0,\tau_0)\psi_0\right\|\le\frac{\delta}{4}\quad\text{for all }n\ge N. \tag*{(52)} ∥ U μ n ( t 0 , τ 0 ) ψ n − U 0 ( t 0 , τ 0 ) ψ 0 ∥ ≤ 4 δ for all n ≥ N . (52)
Then, by (50) , (51) , and (52) ,
0 < δ ≤ dist L 2 ( R ) ( U μ N ( t 0 , τ 0 ) ψ N , A 0 ( t 0 ) ) ≤ ∥ U μ N ( t 0 , τ 0 ) ψ N − U 0 ( t 0 , τ 0 ) ψ 0 ∥ + dist L 2 ( R ) ( U 0 ( t 0 , τ 0 ) ψ 0 , A 0 ( t 0 ) ) ≤ δ 4 + δ 4 = δ 2 . \begin{aligned}
0<\delta&\le\operatorname{dist}_{L^2(\mathbb{R})}(U_{\mu_N}(t_0,\tau_0)\psi_N,A_0(t_0))\\
&\le\left\|U_{\mu_N}(t_0,\tau_0)\psi_N-U_0(t_0,\tau_0)\psi_0\right\|+\operatorname{dist}_{L^2(\mathbb{R})}(U_0(t_0,\tau_0)\psi_0,A_0(t_0))\\
&\le\frac{\delta}{4}+\frac{\delta}{4}=\frac{\delta}{2}.
\end{aligned} 0 < δ ≤ dist L 2 ( R ) ( U μ N ( t 0 , τ 0 ) ψ N , A 0 ( t 0 )) ≤ ∥ U μ N ( t 0 , τ 0 ) ψ N − U 0 ( t 0 , τ 0 ) ψ 0 ∥ + dist L 2 ( R ) ( U 0 ( t 0 , τ 0 ) ψ 0 , A 0 ( t 0 )) ≤ 4 δ + 4 δ = 2 δ .
This is a contradiction which thus completes the proof. □
Finite fractal dimensions
In this section, we consider the case μ > 0 \mu>0 μ > 0 and prove that the pullback attractor { A μ ( t ) } t ∈ R \{A_\mu(t)\}_{t\in\mathbb{R}} { A μ ( t ) } t ∈ R has finite fractal dimensions in H 1 ( R ) H^1(\mathbb{R}) H 1 ( R ) . This will be achieved by proving and applying an extension of the splitting method, see e.g. [5 Theorem 4.15] . Therein the abstract results were written for global attractors of a semigroup. Here we prove an generalized version of that result for pullback attractors of non-autonomous processes. We start with the definition of fractal dimension: A subset M M M of a Banach space X X X is called to have finite fractal dimension in X X X if
dim f X ( M ) : = lim sup ε → 0 ln n ( M , ε ) ln ( 1 / ε ) < + ∞ \dim_{\mathrm{f}}^X(M):=\limsup_{\varepsilon\to0}\frac{\ln n(M,\varepsilon)}{\ln(1/\varepsilon)}<+\infty dim f X ( M ) := ε → 0 lim sup ln ( 1/ ε ) ln n ( M , ε ) < + ∞
where n ( M , ε ) n(M,\varepsilon) n ( M , ε ) is the minimal number of closed balls in X X X of radius ε \varepsilon ε which covers the set M M M .
Theorem 5.1. Let X X X be a Banach space and { A ( t ) } t ∈ R \{A(t)\}_{t\in\mathbb{R}} { A ( t ) } t ∈ R be a family of compact sets in X X X . Assume that there exists a family of operators V ( t , s ) : A ( s ) → X \mathcal{V}(t,s):A(s)\to X V ( t , s ) : A ( s ) → X , t ≥ s t\ge s t ≥ s , such that
(i) A ( t ) ⊂ V ( t , s ) A ( s ) A(t)\subset\mathcal{V}(t,s)A(s) A ( t ) ⊂ V ( t , s ) A ( s ) , for t ≥ s t\ge s t ≥ s .
(ii) V ( t , s ) \mathcal{V}(t,s) V ( t , s ) is Lipschitz on A ( s ) A(s) A ( s ) , i.e, there exists L t , s > 0 L_{t,s}>0 L t , s > 0 such that
∥ V ( t , s ) v 1 − V ( t , s ) v 2 ∥ ≤ L t , s ∥ v 1 − v 2 ∥ , ∀ v 1 , v 2 ∈ A ( s ) , \|\mathcal{V}(t,s)v_1-\mathcal{V}(t,s)v_2\|\le L_{t,s}\|v_1-v_2\|,\qquad\forall v_1,v_2\in A(s), ∥ V ( t , s ) v 1 − V ( t , s ) v 2 ∥ ≤ L t , s ∥ v 1 − v 2 ∥ , ∀ v 1 , v 2 ∈ A ( s ) , where the family of constants { L t , s } t ≥ s \{L_{t,s}\}_{t\ge s} { L t , s } t ≥ s satisfies
lim sup s → − ∞ L t , s < + ∞ . \limsup_{s\to-\infty}L_{t,s}<+\infty. s → − ∞ lim sup L t , s < + ∞. (iii) There exist compact semi-norms n 1 , n 2 n_1,n_2 n 1 , n 2 on X X X such that
∥ V ( t , s ) v 1 − V ( t , s ) v 2 ∥ ≤ η t , s ∥ v 1 − v 2 ∥ + K t , s ( n 1 ( v 1 − v 2 ) + n 2 ( V ( t , s ) v 1 − V ( t , s ) v 2 ) ) , \|\mathcal{V}(t,s)v_1-\mathcal{V}(t,s)v_2\|\le\eta_{t,s}\|v_1-v_2\|+K_{t,s}\bigl(n_1(v_1-v_2)+n_2(\mathcal{V}(t,s)v_1-\mathcal{V}(t,s)v_2)\bigr), ∥ V ( t , s ) v 1 − V ( t , s ) v 2 ∥ ≤ η t , s ∥ v 1 − v 2 ∥ + K t , s ( n 1 ( v 1 − v 2 ) + n 2 ( V ( t , s ) v 1 − V ( t , s ) v 2 ) ) , for any v 1 , v 2 ∈ A ( s ) v_1,v_2\in A(s) v 1 , v 2 ∈ A ( s ) , where the family of constants η t , s > 0 \eta_{t,s}>0 η t , s > 0 , K t , s > 0 K_{t,s}>0 K t , s > 0 satisfy
lim sup s → − ∞ η t , s ∈ ( 0 , 1 / 2 ) and lim sup s → − ∞ K t , s < + ∞ . \limsup_{s\to-\infty}\eta_{t,s}\in(0,1/2)\qquad\text{and}\qquad\limsup_{s\to-\infty}K_{t,s}<+\infty. s → − ∞ lim sup η t , s ∈ ( 0 , 1/2 ) and s → − ∞ lim sup K t , s < + ∞.
Then, { A ( t ) } t ∈ R \{\mathcal{A}(t)\}_{t\in\mathbb{R}} { A ( t ) } t ∈ R has finite fractal dimensions in X X X , that is
dim f X ( A ( t ) ) < + ∞ for each t ∈ R . \dim_{\mathrm{f}}^{X}(\mathcal{A}(t))<+\infty\quad\text{for each }t\in\mathbb{R}. dim f X ( A ( t )) < + ∞ for each t ∈ R .
Proof. Fix t ∈ R t\in\mathbb{R} t ∈ R . Define
η ~ t : = lim sup s → − ∞ η t , s , K ~ t : = lim sup s → − ∞ K t , s , L ~ t : = lim sup s → − ∞ L t , s . \widetilde{\eta}_{t}:=\limsup_{s\to-\infty}\eta_{t,s},\qquad\widetilde{K}_{t}:=\limsup_{s\to-\infty}K_{t,s},\qquad\widetilde{L}_{t}:=\limsup_{s\to-\infty}L_{t,s}. η t := s → − ∞ lim sup η t , s , K t := s → − ∞ lim sup K t , s , L t := s → − ∞ lim sup L t , s .
By assumptions 0 < η ~ t < 1 / 2 0<\widetilde{\eta}_{t}<1/2 0 < η t < 1/2 , K ~ t < ∞ \widetilde{K}_{t}<\infty K t < ∞ and L ~ t < ∞ \widetilde{L}_{t}<\infty L t < ∞ . Choose η , K , L \eta,K,L η , K , L such that η ~ t < η < 1 / 2 \widetilde{\eta}_{t}<\eta<1/2 η t < η < 1/2 , K ~ t < K \widetilde{K}_{t}<K K t < K , and L ~ t < L \widetilde{L}_{t}<L L t < L . Then, there exists S ≤ t S\le t S ≤ t such that for all s ≤ S s\le S s ≤ S ,
η t , s ≤ η , K t , s ≤ K , L t , s ≤ L , \eta_{t,s}\le\eta,\qquad K_{t,s}\le K,\qquad L_{t,s}\le L, η t , s ≤ η , K t , s ≤ K , L t , s ≤ L ,
and for every v 1 , v 2 ∈ A ( s ) v_{1},v_{2}\in\mathcal{A}(s) v 1 , v 2 ∈ A ( s ) ,
∥ V t , s v 1 − V t , s v 2 ∥ ≤ η ∥ v 1 − v 2 ∥ + K [ n 1 ( v 1 − v 2 ) + n 2 ( V t , s v 1 − V t , s v 2 ) ] . (53) \left\|\mathcal{V}_{t,s}v_{1}-\mathcal{V}_{t,s}v_{2}\right\|\le\eta\left\|v_{1}-v_{2}\right\|+K\left[n_{1}(v_{1}-v_{2})+n_{2}\left(\mathcal{V}_{t,s}v_{1}-\mathcal{V}_{t,s}v_{2}\right)\right]. \tag*{(53)} ∥ V t , s v 1 − V t , s v 2 ∥ ≤ η ∥ v 1 − v 2 ∥ + K [ n 1 ( v 1 − v 2 ) + n 2 ( V t , s v 1 − V t , s v 2 ) ] . (53)
For any s ≤ S s\le S s ≤ S we define a pseudo-metric ϱ t , s \varrho_{t,s} ϱ t , s on A ( s ) \mathcal{A}(s) A ( s ) by
ϱ t , s ( x , y ) : = K [ n 1 ( x − y ) + n 2 ( V t , s x − V t , s y ) ] , x , y ∈ A ( s ) . \varrho_{t,s}(x,y):=K\left[n_{1}(x-y)+n_{2}\left(\mathcal{V}_{t,s}x-\mathcal{V}_{t,s}y\right)\right],\qquad x,y\in\mathcal{A}(s). ϱ t , s ( x , y ) := K [ n 1 ( x − y ) + n 2 ( V t , s x − V t , s y ) ] , x , y ∈ A ( s ) .
Because n 1 , n 2 n_{1},n_{2} n 1 , n 2 are compact semi-norms and A ( s ) \mathcal{A}(s) A ( s ) is bounded, ϱ t , s \varrho_{t,s} ϱ t , s is compact. From (iii), we have
∥ V ( t , s ) x − V ( t , s ) y ∥ ≤ η ∥ x − y ∥ + ϱ t , s ( x , y ) , ∀ x , y ∈ A ( s ) . (54) \left\|\mathcal{V}(t,s)x-\mathcal{V}(t,s)y\right\|\le\eta\left\|x-y\right\|+\varrho_{t,s}(x,y),\qquad\forall x,y\in\mathcal{A}(s). \tag*{(54)} ∥ V ( t , s ) x − V ( t , s ) y ∥ ≤ η ∥ x − y ∥ + ϱ t , s ( x , y ) , ∀ x , y ∈ A ( s ) . (54)
Fix δ ∈ ( 0 , 1 − 2 η ) \delta\in(0,1-2\eta) δ ∈ ( 0 , 1 − 2 η ) and set q : = 2 η + δ < 1 q:=2\eta+\delta<1 q := 2 η + δ < 1 . For any ε > 0 \varepsilon>0 ε > 0 and any B ⊂ A ( s ) B\subset\mathcal{A}(s) B ⊂ A ( s ) with diam B ≤ 2 ε \operatorname{diam}B\le2\varepsilon diam B ≤ 2 ε , let m ϱ t , s ( B , δ ε ) m_{\varrho_{t,s}}(B,\delta\varepsilon) m ϱ t , s ( B , δ ε ) be the maximal cardinality of a ( δ ε , ϱ t , s ) (\delta\varepsilon,\varrho_{t,s}) ( δ ε , ϱ t , s ) -distinguishable subset of B B B (where a subset Q ⊆ B Q\subseteq B Q ⊆ B is defined as ( δ ε , ϱ t , s ) (\delta\varepsilon,\varrho_{t,s}) ( δ ε , ϱ t , s ) -distinguishable if for any two distinct elements x , y ∈ Q x,y\in Q x , y ∈ Q , the pseudo-metric satisfies ϱ t , s ( x , y ) > δ ε \varrho_{t,s}(x,y)>\delta\varepsilon ϱ t , s ( x , y ) > δ ε ). Define the local capacity
C ϱ t , s ( A ( s ) ; ε , δ ε ) : = sup { ln m ϱ t , s ( B , δ ε ) : B ⊆ A ( s ) , diam ( B ) ≤ 2 ε } . C_{\varrho_{t,s}}(\mathcal{A}(s);\varepsilon,\delta\varepsilon):=\sup\{\ln m_{\varrho_{t,s}}(B,\delta\varepsilon):B\subseteq\mathcal{A}(s),\operatorname{diam}(B)\le2\varepsilon\}. C ϱ t , s ( A ( s ) ; ε , δ ε ) := sup { ln m ϱ t , s ( B , δ ε ) : B ⊆ A ( s ) , diam ( B ) ≤ 2 ε } .
Now, we estimate C ϱ t , s C_{\varrho_{t,s}} C ϱ t , s . For any x , y ∈ B x,y\in B x , y ∈ B , using the Lipschitz property (ii) with constant L L L ,
∥ x − y ∥ 2 + ∥ V t , s x − V t , s y ∥ 2 ≤ ( 1 + L 2 ) ∥ x − y ∥ 2 ≤ ( 1 + L 2 ) ( 2 ε ) 2 = 4 ( 1 + L 2 ) ε 2 . \left\|x-y\right\|^{2}+\left\|\mathcal{V}_{t,s}x-\mathcal{V}_{t,s}y\right\|^{2}\le(1+L^{2})\left\|x-y\right\|^{2}\le(1+L^{2})(2\varepsilon)^{2}=4(1+L^{2})\varepsilon^{2}. ∥ x − y ∥ 2 + ∥ V t , s x − V t , s y ∥ 2 ≤ ( 1 + L 2 ) ∥ x − y ∥ 2 ≤ ( 1 + L 2 ) ( 2 ε ) 2 = 4 ( 1 + L 2 ) ε 2 .
Thus, the points Z = ( x , V t , s x ) Z=(x,\mathcal{V}_{t,s}x) Z = ( x , V t , s x ) lie in a ball of radius R = 2 1 + L 2 ε R=2\sqrt{1+L^{2}}\varepsilon R = 2 1 + L 2 ε in X × X X\times X X × X . The condition ϱ t , s ( x , y ) > δ ε \varrho_{t,s}(x,y)>\delta\varepsilon ϱ t , s ( x , y ) > δ ε is equivalent to
n 1 ( x − y ) + n 2 ( V t , s x − V t , s y ) > δ ε K . n_{1}(x-y)+n_{2}\left(\mathcal{V}_{t,s}x-\mathcal{V}_{t,s}y\right)>\frac{\delta\varepsilon}{K}. n 1 ( x − y ) + n 2 ( V t , s x − V t , s y ) > K δ ε .
By defining λ = K / ( δ ε ) \lambda=K/(\delta\varepsilon) λ = K / ( δ ε ) , the scaled points Z ~ = λ Z \widetilde{Z}=\lambda Z Z = λ Z lie in the ball of radius
K δ ε R = 2 K 1 + L 2 δ \frac{K}{\delta\varepsilon}R=\frac{2K\sqrt{1+L^{2}}}{\delta} δ ε K R = δ 2 K 1 + L 2
and satisfy η 1 ( x ~ − y ~ ) + η 2 ( V t , s x ~ − V t , s y ~ ) > 1 \eta_{1}(\widetilde{x}-\widetilde{y})+\eta_{2}(\mathcal{V}_{t,s}\widetilde{x}-\mathcal{V}_{t,s}\widetilde{y})>1 η 1 ( x − y ) + η 2 ( V t , s x − V t , s y ) > 1 for all x ~ , y ~ ∈ Z ~ \widetilde{x},\widetilde{y}\in\widetilde{Z} x , y ∈ Z . Hence the maximal number of such points is at most
m 0 ( 2 K 1 + L 2 δ ) , m_{0}\left(\frac{2K\sqrt{1+L^{2}}}{\delta}\right), m 0 ( δ 2 K 1 + L 2 ) ,
where m 0 ( R ) m_{0}(R) m 0 ( R ) is the maximal cardinality of pairs ( x i , y i ) ∈ X × X (x_{i},y_{i})\in X\times X ( x i , y i ) ∈ X × X with ∥ x i ∥ 2 + ∥ y i ∥ 2 ≤ R 2 \|x_{i}\|^{2}+\|y_{i}\|^{2}\le R^{2} ∥ x i ∥ 2 + ∥ y i ∥ 2 ≤ R 2 and η 1 ( x i − x j ) + η 2 ( y i − y j ) > 1 \eta_{1}(x_{i}-x_{j})+\eta_{2}(y_{i}-y_{j})>1 η 1 ( x i − x j ) + η 2 ( y i − y j ) > 1 for i ≠ j i\ne j i = j . Therefore
m ϱ t , s ( B , δ ε ) ≤ m 0 ( 2 K 1 + L 2 δ ) , m_{\varrho_{t,s}}(B,\delta\varepsilon)\le m_{0}\left(\frac{2K\sqrt{1+L^{2}}}{\delta}\right), m ϱ t , s ( B , δ ε ) ≤ m 0 ( δ 2 K 1 + L 2 ) ,
and taking the supremum over all such B B B gives
C ϱ t , s ( A ( s ) ; ε , δ ε ) ≤ ln m 0 ( 2 K 1 + L 2 δ ) . (55) C_{\varrho_{t,s}}(\mathcal{A}(s);\varepsilon,\delta\varepsilon)\le\ln m_{0}\left(\frac{2K\sqrt{1+L^{2}}}{\delta}\right). \tag*{(55)} C ϱ t , s ( A ( s ) ; ε , δ ε ) ≤ ln m 0 ( δ 2 K 1 + L 2 ) . (55)
Now we prove the dimension bound. For any map V V V satisfying (54) and any sets M ⊂ V N M\subset V N M ⊂ V N ,
n ( M , q ε ) ≤ n ( N , ε ) exp ( C ϱ t , s ( N ; ε , δ ε ) ) . (56) n(M,q\varepsilon)\le n(N,\varepsilon)\exp\left(C_{\varrho_{t,s}}(N;\varepsilon,\delta\varepsilon)\right). \tag*{(56)} n ( M , q ε ) ≤ n ( N , ε ) exp ( C ϱ t , s ( N ; ε , δ ε ) ) . (56)
where q = 2 η + δ < 1 q=2\eta+\delta<1 q = 2 η + δ < 1 . Choose a decreasing sequence { s k } k ≥ 0 \{s_{k}\}_{k\ge0} { s k } k ≥ 0 with s 0 = S s_{0}=S s 0 = S and s k + 1 ≤ s k s_{k+1}\le s_{k} s k + 1 ≤ s k . Then the estimates (53) hold for each map V ( s k , s k + 1 ) \mathcal{V}(s_{k},s_{k+1}) V ( s k , s k + 1 ) , since all s k ≤ S s_{k}\le S s k ≤ S . By condition (i),
A ( s k ) ⊂ V ( s k , s k + 1 ) A ( s k + 1 ) , A ( t ) ⊂ V ( t , S ) A ( S ) . \mathcal{A}(s_{k})\subset\mathcal{V}(s_{k},s_{k+1})\mathcal{A}(s_{k+1}),\qquad\mathcal{A}(t)\subset\mathcal{V}(t,S)\mathcal{A}(S). A ( s k ) ⊂ V ( s k , s k + 1 ) A ( s k + 1 ) , A ( t ) ⊂ V ( t , S ) A ( S ) .
Apply (56) to V = V ( t , S ) V=\mathcal{V}(t,S) V = V ( t , S ) with N = A ( S ) N=\mathcal{A}(S) N = A ( S ) and M = A ( t ) M=\mathcal{A}(t) M = A ( t ) :
n ( A ( t ) , q ε ) ≤ n ( A ( S ) , ε ) exp ( C ϱ t , S ( A ( S ) ; ε , δ ε ) ) . (57) n(\mathcal{A}(t),q\varepsilon)\le n(\mathcal{A}(S),\varepsilon)\exp\left(C_{\varrho_{t,S}}(\mathcal{A}(S);\varepsilon,\delta\varepsilon)\right). \tag*{(57)} n ( A ( t ) , q ε ) ≤ n ( A ( S ) , ε ) exp ( C ϱ t , S ( A ( S ) ; ε , δ ε ) ) . (57)
For each k ≥ 0 k \ge0 k ≥ 0 , apply (56) to V = V ( s k , s k + 1 ) V=\mathcal{V}(s_k,s_{k+1}) V = V ( s k , s k + 1 ) with N = A ( s k + 1 ) N=\mathcal{A}(s_{k+1}) N = A ( s k + 1 ) and M = A ( s k ) M=\mathcal{A}(s_k) M = A ( s k ) :
n ( A ( s k ) , q ε ) ≤ n ( A ( s k + 1 ) , ε ) exp ( C ϱ s k , s k + 1 ( A ( s k + 1 ) ; ε , δ ε ) ) . (58) n(\mathcal{A}(s_k),q\varepsilon)\le n(\mathcal{A}(s_{k+1}),\varepsilon)\exp\left(C_{\varrho s_k,s_{k+1}}(\mathcal{A}(s_{k+1});\varepsilon,\delta\varepsilon)\right). \tag*{(58)} n ( A ( s k ) , q ε ) ≤ n ( A ( s k + 1 ) , ε ) exp ( C ϱ s k , s k + 1 ( A ( s k + 1 ) ; ε , δ ε ) ) . (58)
Iterating (58) for k = 0 , 1 , … , m − 1 k=0,1,\ldots,m-1 k = 0 , 1 , … , m − 1 and combining with (57) yields
n ( A ( t ) , q m + 1 ε ) ≤ n ( A ( s m ) , ε ) exp ( ∑ j = 0 m C j ) , (59) n(\mathcal{A}(t),q^{m+1}\varepsilon)\le n(\mathcal{A}(s_m),\varepsilon)\exp\left(\sum_{j=0}^{m}C_j\right), \tag*{(59)} n ( A ( t ) , q m + 1 ε ) ≤ n ( A ( s m ) , ε ) exp ( j = 0 ∑ m C j ) , (59)
where
C 0 = C ϱ t , S ( A ( S ) ; ε , δ ε ) , C j + 1 = C ϱ s j , s j + 1 ( A ( s j + 1 ) ; q j ε , δ q j ε ) . C_0=C_{\varrho t,S}(\mathcal{A}(S);\varepsilon,\delta\varepsilon),\qquad C_{j+1}=C_{\varrho s_j,s_{j+1}}(\mathcal{A}(s_{j+1});q^j\varepsilon,\delta q^j\varepsilon). C 0 = C ϱt , S ( A ( S ) ; ε , δ ε ) , C j + 1 = C ϱ s j , s j + 1 ( A ( s j + 1 ) ; q j ε , δ q j ε ) .
By (55) , each C j ≤ ln m 0 ( 2 K 1 + L 2 δ ) C_j\le\ln m_0\left(\frac{2K\sqrt{1+L^2}}{\delta}\right) C j ≤ ln m 0 ( δ 2 K 1 + L 2 ) . Hence the sum is at most ( m + 1 ) ln m 0 ( 2 K 1 + L 2 δ ) (m+1)\ln m_0\left(\frac{2K\sqrt{1+L^2}}{\delta}\right) ( m + 1 ) ln m 0 ( δ 2 K 1 + L 2 ) .
Since q < 1 q<1 q < 1 , for any sufficiently small ε > 0 \varepsilon>0 ε > 0 we can choose m m m such that q m + 1 ε q^{m+1}\varepsilon q m + 1 ε is arbitrarily small; moreover m + 1 ≤ ln ( 1 / ε ) ln ( 1 / q ) + 1 m+1\le\frac{\ln(1/\varepsilon)}{\ln(1/q)}+1 m + 1 ≤ l n ( 1/ q ) l n ( 1/ ε ) + 1 . Taking logarithms in (59) and letting ε → 0 \varepsilon\to0 ε → 0 gives
dim f X A ( t ) ≤ ln m 0 ( 2 K 1 + L 2 δ ) ln ( 1 / q ) . \dim_f^X\mathcal{A}(t)\le\frac{\ln m_0\left(\frac{2K\sqrt{1+L^2}}{\delta}\right)}{\ln(1/q)}. dim f X A ( t ) ≤ ln ( 1/ q ) ln m 0 ( δ 2 K 1 + L 2 ) .
To have more explicit bounds, we can choose δ = 1 − 2 η 2 \delta=\frac{1-2\eta}{2} δ = 2 1 − 2 η . Then q = 1 + 2 η 2 q=\frac{1+2\eta}{2} q = 2 1 + 2 η and 2 δ = 4 1 − 2 η \frac{2}{\delta}=\frac{4}{1-2\eta} δ 2 = 1 − 2 η 4 . Substituting into above yields
dim f X A ( t ) ≤ [ ln 2 1 + 2 η ] − 1 ln m 0 ( 4 K 1 + L 2 1 − 2 η ) < ∞ . \dim_f^X\mathcal{A}(t)\le\left[\ln\frac{2}{1+2\eta}\right]^{-1}\ln m_0\left(\frac{4K\sqrt{1+L^2}}{1-2\eta}\right)<\infty. dim f X A ( t ) ≤ [ ln 1 + 2 η 2 ] − 1 ln m 0 ( 1 − 2 η 4 K 1 + L 2 ) < ∞.
Since η , K , L \eta,K,L η , K , L can be taken arbitrarily close to η ~ t , K ~ , L ~ \tilde{\eta}_t,\widetilde{K},\widetilde{L} η ~ t , K , L , the bound is finite. Hence dim f X A ( t ) < ∞ \dim_f^X\mathcal{A}(t)<\infty dim f X A ( t ) < ∞ for every fixed t t t . □
For the rest of this section, we will prove all the assumptions in Theorem 5.1 for the attractors A μ ( t ) \mathcal{A}_\mu(t) A μ ( t ) obtained in the previous sections and obtain that they have finite fractional dimensions. We first impose an additional assumption on the external force
∥ g ∥ : = sup t ∈ R ( ∥ g ( t ) ∥ L 2 ( R ) + ∥ g ′ ( t ) ∥ L 2 ( R ) ) < + ∞ . (60) \lVert g\rVert:=\sup_{t\in\mathbb{R}}\left(\lVert g(t)\rVert_{L^2(\mathbb{R})}+\lVert g'(t)\rVert_{L^2(\mathbb{R})}\right)<+\infty. \tag*{(60)} ∥ g ∥ := t ∈ R sup ( ∥ g ( t ) ∥ L 2 ( R ) + ∥ g ′ ( t ) ∥ L 2 ( R ) ) < + ∞. (60)
Fix μ ∈ ( 0 , 1 ] \mu\in(0,1] μ ∈ ( 0 , 1 ] and let u 1 ( t ) u_1(t) u 1 ( t ) and u 2 ( t ) u_2(t) u 2 ( t ) be two orbits on the attractor { A μ ( t ) } \{\mathcal{A}_\mu(t)\} { A μ ( t )} for all t ≥ τ t\ge\tau t ≥ τ . We set U ( t ) = u 1 ( t ) − u 2 ( t ) U(t)=u_1(t)-u_2(t) U ( t ) = u 1 ( t ) − u 2 ( t ) . It follows that
{ U t + U x x x + ( U u ˉ ) x − μ U x x + λ U = 0 , x ∈ R , t > τ , U ( τ , x ) = u 1 ( τ , x ) − u 2 ( τ , x ) , x ∈ R , (61) \begin{cases}
U_t+U_{xxx}+(U\bar{u})_x-\mu U_{xx}+\lambda U=0, & x\in\mathbb{R},\ t>\tau,\\
U(\tau,x)=u_1(\tau,x)-u_2(\tau,x), & x\in\mathbb{R}, \tag*{(61)}
\end{cases} { U t + U xxx + ( U u ˉ ) x − μ U xx + λ U = 0 , U ( τ , x ) = u 1 ( τ , x ) − u 2 ( τ , x ) , x ∈ R , t > τ , x ∈ R , (61)
where u ˉ = u 1 + u 2 2 \bar{u}=\dfrac{u_1+u_2}{2} u ˉ = 2 u 1 + u 2 .
Lemma 5.2. Let U U U be the solution of (61) . There exist k 0 ≥ 1 k_0\ge1 k 0 ≥ 1 and a constant c 1 > 0 c_1>0 c 1 > 0 , depending only on λ \lambda λ , μ \mu μ and ∥ g ∥ \lVert g\rVert ∥ g ∥ , such that for all k ≥ k 0 k\ge k_0 k ≥ k 0 and t > τ t>\tau t > τ ,
∥ U ( t ) ∥ H 1 2 ≤ e − λ ( t − τ ) ∥ U ( τ ) ∥ H 1 2 + c 1 ∫ τ t ∫ ∣ x ∣ ≤ k U 2 ( s , x ) d x d s . (62) \lVert U(t)\rVert_{H^1}^2\le e^{-\lambda(t-\tau)}\lVert U(\tau)\rVert_{H^1}^2+c_1\int_\tau^t\int_{|x|\le k}U^2(s,x)\,dx\,ds. \tag*{(62)} ∥ U ( t ) ∥ H 1 2 ≤ e − λ ( t − τ ) ∥ U ( τ ) ∥ H 1 2 + c 1 ∫ τ t ∫ ∣ x ∣ ≤ k U 2 ( s , x ) d x d s . (62)
Proof. From estimates (6) and (34) in Sections 2 and 3, and thanks to (60) we see that
sup t ∈ R ∥ u ˉ ( t ) ∥ H 3 ≤ C 0 , \sup_{t\in\mathbb{R}}\lVert\bar{u}(t)\rVert_{H^3}\le\mathcal{C}_0, t ∈ R sup ∥ u ˉ ( t ) ∥ H 3 ≤ C 0 ,
where C 0 \mathcal{C}_0 C 0 depends on λ \lambda λ , μ \mu μ and ∥ g ∥ \lVert g\rVert ∥ g ∥ . Multiplying (61) with ( U − β U x x ) (U-\beta U_{xx}) ( U − β U xx ) , where β = min { 1 , μ / C 0 } \beta=\min\{1,\mu/\mathcal{C}_0\} β = min { 1 , μ / C 0 } , and integrating over R \mathbb{R} R , we find
1 2 d d t ∫ R ( β U x 2 + U 2 ) d x + μ ∫ R ( β U x x 2 + U x 2 ) d x + λ ∫ R ( β U x 2 + U 2 ) d x = ∫ R ( U u ˉ ) x ( − U + β U x x ) d x . (63) \frac{1}{2}\frac{d}{dt}\int_{\mathbb{R}}(\beta U_x^2+U^2)\,dx+\mu\int_{\mathbb{R}}(\beta U_{xx}^2+U_x^2)\,dx+\lambda\int_{\mathbb{R}}(\beta U_x^2+U^2)\,dx
=\int_{\mathbb{R}}(U\bar{u})_x(-U+\beta U_{xx})\,dx. \tag*{(63)} 2 1 d t d ∫ R ( β U x 2 + U 2 ) d x + μ ∫ R ( β U xx 2 + U x 2 ) d x + λ ∫ R ( β U x 2 + U 2 ) d x = ∫ R ( U u ˉ ) x ( − U + β U xx ) d x . (63)
By integration by parts, we have
− ∫ R U ( U u ˉ ) x d x = − 1 2 ∫ R U 2 u ˉ x d x ≤ 1 2 ∥ u ˉ x ∥ L ∞ ( ∣ x ∣ ≥ k ) ∫ R U 2 d x + 1 2 C 0 ∫ ∣ x ∣ ≤ k U 2 d x ≤ λ 4 ∫ R U 2 d x + 1 2 C 0 ∫ ∣ x ∣ ≤ k U 2 d x . (64) \begin{aligned}
-\int_{\mathbb{R}} U(U\bar{u})_x\,dx=-\frac{1}{2}\int_{\mathbb{R}}U^2\bar{u}_x\,dx\leq\frac{1}{2}\|\bar{u}_x\|_{L^\infty(|x|\geq k)}\int_{\mathbb{R}}U^2\,dx+\frac{1}{2}\mathcal{C}_0\int_{|x|\leq k}U^2\,dx \\
&\leq\frac{\lambda}{4}\int_{\mathbb{R}}U^2\,dx+\frac{1}{2}\mathcal{C}_0\int_{|x|\leq k}U^2\,dx. \tag*{(64)}
\end{aligned} − ∫ R U ( U u ˉ ) x d x = − 2 1 ∫ R U 2 u ˉ x d x ≤ 2 1 ∥ u ˉ x ∥ L ∞ ( ∣ x ∣ ≥ k ) ∫ R U 2 d x + 2 1 C 0 ∫ ∣ x ∣ ≤ k U 2 d x ≤ 4 λ ∫ R U 2 d x + 2 1 C 0 ∫ ∣ x ∣ ≤ k U 2 d x . (64)
provided k k k is sufficiently large. For the second term on the right of (63) ,
β ∫ R U x x ( U u ˉ ) x d x = − β 2 ∫ R U x 2 u ˉ x d x + β ∫ R U x x U u ˉ x d x ≤ β 2 C 0 ∫ R U x 2 d x + β μ 2 ∫ R U x x 2 d x + β 2 μ ∫ R U 2 ∣ u ˉ x ∣ 2 d x , ≤ β 2 C 0 ∫ R U x 2 d x + β μ 2 ∫ R U x x 2 d x + β C 0 2 2 μ ∫ ∣ x ∣ ≤ k U 2 d x + β 2 μ ∥ u ˉ x ∥ L ∞ ( ∣ x ∣ ≥ k ) 2 ∫ R U 2 d x , ≤ β 2 C 0 ∫ R U x 2 d x + β μ 2 ∫ R U x x 2 d x + β C 0 2 2 μ ∫ ∣ x ∣ ≤ k U 2 d x + λ 4 ∫ R U 2 d x . (65) \begin{aligned}
\beta\int_{\mathbb{R}}U_{xx}(U\bar{u})_x\,dx=-\frac{\beta}{2}\int_{\mathbb{R}}U_x^2\bar{u}_x\,dx+\beta\int_{\mathbb{R}}U_{xx}U\bar{u}_x\,dx \\
&\leq\frac{\beta}{2}\mathcal{C}_0\int_{\mathbb{R}}U_x^2\,dx+\frac{\beta\mu}{2}\int_{\mathbb{R}}U_{xx}^2\,dx+\frac{\beta}{2\mu}\int_{\mathbb{R}}U^2|\bar{u}_x|^2\,dx, \\
&\leq\frac{\beta}{2}\mathcal{C}_0\int_{\mathbb{R}}U_x^2\,dx+\frac{\beta\mu}{2}\int_{\mathbb{R}}U_{xx}^2\,dx+\frac{\beta\mathcal{C}_0^2}{2\mu}\int_{|x|\leq k}U^2\,dx+\frac{\beta}{2\mu}\|\bar{u}_x\|_{L^\infty(|x|\geq k)}^2\int_{\mathbb{R}}U^2\,dx, \\
&\leq\frac{\beta}{2}\mathcal{C}_0\int_{\mathbb{R}}U_x^2\,dx+\frac{\beta\mu}{2}\int_{\mathbb{R}}U_{xx}^2\,dx+\frac{\beta\mathcal{C}_0^2}{2\mu}\int_{|x|\leq k}U^2\,dx+\frac{\lambda}{4}\int_{\mathbb{R}}U^2\,dx. \tag*{(65)}
\end{aligned} β ∫ R U xx ( U u ˉ ) x d x = − 2 β ∫ R U x 2 u ˉ x d x + β ∫ R U xx U u ˉ x d x ≤ 2 β C 0 ∫ R U x 2 d x + 2 β μ ∫ R U xx 2 d x + 2 μ β ∫ R U 2 ∣ u ˉ x ∣ 2 d x , ≤ 2 β C 0 ∫ R U x 2 d x + 2 β μ ∫ R U xx 2 d x + 2 μ β C 0 2 ∫ ∣ x ∣ ≤ k U 2 d x + 2 μ β ∥ u ˉ x ∥ L ∞ ( ∣ x ∣ ≥ k ) 2 ∫ R U 2 d x , ≤ 2 β C 0 ∫ R U x 2 d x + 2 β μ ∫ R U xx 2 d x + 2 μ β C 0 2 ∫ ∣ x ∣ ≤ k U 2 d x + 4 λ ∫ R U 2 d x . (65)
for large enough k k k . Since β C 0 ≤ μ \beta\mathcal{C}_0\leq\mu β C 0 ≤ μ , the terms β 2 C 0 ∫ U x 2 \frac{\beta}{2}\mathcal{C}_0\int U_x^2 2 β C 0 ∫ U x 2 and β μ 2 ∫ U x x 2 \frac{\beta\mu}{2}\int U_{xx}^2 2 β μ ∫ U xx 2 are absorbed by the second term on the left of (63) . By inserting (65) and (64) into (63) , we obtain
d d t ∫ R ( U 2 + β U x 2 ) d x + λ ∫ R ( U 2 + β U x 2 ) d x ≤ c 1 ∫ ∣ x ∣ ≤ k U 2 d x , (66) \frac{d}{dt}\int_{\mathbb{R}}(U^2+\beta U_x^2)\,dx+\lambda\int_{\mathbb{R}}(U^2+\beta U_x^2)\,dx\leq c_1\int_{|x|\leq k}U^2\,dx, \tag*{(66)} d t d ∫ R ( U 2 + β U x 2 ) d x + λ ∫ R ( U 2 + β U x 2 ) d x ≤ c 1 ∫ ∣ x ∣ ≤ k U 2 d x , (66)
where c 1 = C 0 + β C 0 2 / μ ≤ 2 C 0 c_1=\mathcal{C}_0+\beta\mathcal{C}_0^2/\mu\leq2\mathcal{C}_0 c 1 = C 0 + β C 0 2 / μ ≤ 2 C 0 . By using the equivalent norm ∥ f ∥ H 1 2 ∼ ∥ f ∥ L 2 2 + β ∥ ∂ x f ∥ L 2 2 \|f\|_{H^1}^2\sim\|f\|_{L^2}^2+\beta\|\partial_xf\|_{L^2}^2 ∥ f ∥ H 1 2 ∼ ∥ f ∥ L 2 2 + β ∥ ∂ x f ∥ L 2 2 , and applying Gronwall’s lemma, we get
∥ U ( t ) ∥ H 1 2 ≤ e − λ ( t − τ ) ∥ U ( τ ) ∥ H 1 2 + c 1 ∫ τ t ∫ ∣ x ∣ ≤ k U 2 ( s , x ) d x d s . (67) \|U(t)\|_{H^1}^2\leq e^{-\lambda(t-\tau)}\|U(\tau)\|_{H^1}^2+c_1\int_\tau^t\int_{|x|\leq k}U^2(s,x)\,dx\,ds. \tag*{(67)} ∥ U ( t ) ∥ H 1 2 ≤ e − λ ( t − τ ) ∥ U ( τ ) ∥ H 1 2 + c 1 ∫ τ t ∫ ∣ x ∣ ≤ k U 2 ( s , x ) d x d s . (67)
Let k ≥ 1 k\geq1 k ≥ 1 . We consider the cutoff function φ k ∈ C ∞ ( R ) \varphi_k\in C^\infty(\mathbb{R}) φ k ∈ C ∞ ( R ) , which satisfies 0 ≤ φ k ≤ 1 0\leq\varphi_k\leq1 0 ≤ φ k ≤ 1 , φ k = 1 \varphi_k=1 φ k = 1 for ∣ x ∣ ≤ k |x|\leq k ∣ x ∣ ≤ k , φ k = 0 \varphi_k=0 φ k = 0 for ∣ x ∣ ≥ 2 k |x|\geq2k ∣ x ∣ ≥ 2 k , and
∣ ∂ x j φ k ( x ) ∣ ≤ C j k − j , j = 0 , 1 , 2 , 3. (68) |\partial_x^j\varphi_k(x)|\leq C_jk^{-j},\qquad j=0,1,2,3. \tag*{(68)} ∣ ∂ x j φ k ( x ) ∣ ≤ C j k − j , j = 0 , 1 , 2 , 3. (68)
Denote by U k = φ k U U_k=\varphi_kU U k = φ k U . Then U k U_k U k satisfies
{ ( U k ) t + ( U k ) x x x + ( U k u ˉ ) x + λ U k − μ ( U k ) x x = m , U k ( τ , x ) = φ k U ( τ , x ) , (69) \begin{cases}
(U_k)_t+(U_k)_{xxx}+(U_k\bar{u})_x+\lambda U_k-\mu(U_k)_{xx}=m,\\
U_k(\tau,x)=\varphi_kU(\tau,x), \tag*{(69)}
\end{cases} { ( U k ) t + ( U k ) xxx + ( U k u ˉ ) x + λ U k − μ ( U k ) xx = m , U k ( τ , x ) = φ k U ( τ , x ) , (69)
where
m = 3 ∂ x ( ∂ x U ∂ x φ k ) + U ∂ x 3 φ k + U u ˉ ∂ x φ k − 2 μ ∂ x U ∂ x φ k − μ U ∂ x 2 φ k . (70) m=3\partial_x(\partial_xU\partial_x\varphi_k)+U\partial_x^3\varphi_k+U\bar{u}\partial_x\varphi_k-2\mu\partial_xU\partial_x\varphi_k-\mu U\partial_x^2\varphi_k. \tag*{(70)} m = 3 ∂ x ( ∂ x U ∂ x φ k ) + U ∂ x 3 φ k + U u ˉ ∂ x φ k − 2 μ ∂ x U ∂ x φ k − μU ∂ x 2 φ k . (70)
Lemma 5.3. There exist constants c 6 , c 7 c_6,c_7 c 6 , c 7 such that for all t ≥ τ t\geq\tau t ≥ τ ,
∫ τ t ∥ U k ( s ) ∥ L 2 2 d s ≤ c 6 ∥ U k ( τ ) ∥ L 2 2 + c 7 k 2 ∥ U ( τ ) ∥ H 1 2 . (71) \int_\tau^t\|U_k(s)\|_{L^2}^2\,ds\leq c_6\|U_k(\tau)\|_{L^2}^2+\frac{c_7}{k^2}\|U(\tau)\|_{H^1}^2. \tag*{(71)} ∫ τ t ∥ U k ( s ) ∥ L 2 2 d s ≤ c 6 ∥ U k ( τ ) ∥ L 2 2 + k 2 c 7 ∥ U ( τ ) ∥ H 1 2 . (71)
Proof. Since u ˉ \bar{u} u ˉ is bounded in H 3 ( R ) H^3(\mathbb{R}) H 3 ( R ) , we have ∥ u ˉ ∥ L ∞ + ∥ ∂ x u ˉ ∥ L ∞ ≤ 2 C 0 \|\bar{u}\|_{L^\infty}+\|\partial_x\bar{u}\|_{L^\infty}\leq2\mathcal{C}_0 ∥ u ˉ ∥ L ∞ + ∥ ∂ x u ˉ ∥ L ∞ ≤ 2 C 0 . Multiplying (69) by U k U_k U k and integrating over R \mathbb{R} R , we get
1 2 d d t ∥ U k ∥ L 2 2 + λ ∥ U k ∥ L 2 2 + μ ∥ ( U k ) x ∥ L 2 2 = − 1 2 ∫ R u ˉ x U k 2 d x + ∫ R m U k d x . \frac{1}{2}\frac{d}{dt}\|U_k\|_{L^2}^2+\lambda\|U_k\|_{L^2}^2+\mu\|(U_k)_x\|_{L^2}^2=-\frac{1}{2}\int_{\mathbb{R}}\bar{u}_xU_k^2\,dx+\int_{\mathbb{R}}mU_k\,dx. 2 1 d t d ∥ U k ∥ L 2 2 + λ ∥ U k ∥ L 2 2 + μ ∥ ( U k ) x ∥ L 2 2 = − 2 1 ∫ R u ˉ x U k 2 d x + ∫ R m U k d x .
The first term on the right is bounded by C ∥ U k ∥ L 2 2 C\|U_k\|_{L^2}^2 C ∥ U k ∥ L 2 2 . From (70) , we write
∫ R m U k d x = I 1 + I 2 + I 3 + I 4 + I 5 \int_{\mathbb{R}}mU_k\,dx=I_1+I_2+I_3+I_4+I_5 ∫ R m U k d x = I 1 + I 2 + I 3 + I 4 + I 5
and estimate each term separately. Integrating by parts and using Young’s inequality and (68) ,
I 1 = 3 ∫ R ∂ x ( ∂ x U ∂ x φ k ) U k d x = − 3 ∫ R U x ∂ x φ k ( U k ) x d x , ≤ μ 2 ∥ ( U k ) x ∥ L 2 2 + 9 2 μ ∫ R ∣ U x ∣ 2 ∣ ∂ x φ k ∣ 2 d x , ≤ μ 2 ∥ ( U k ) x ∥ L 2 2 + C k 2 ∥ U ∥ H 1 2 . \begin{aligned}
I_1=3\int_{\mathbb{R}}\partial_x(\partial_xU\partial_x\varphi_k)U_k\,dx=-3\int_{\mathbb{R}}U_x\partial_x\varphi_k(U_k)_x\,dx, \\
&\leq\frac{\mu}{2}\lVert(U_k)_x\rVert_{L^2}^2+\frac{9}{2\mu}\int_{\mathbb{R}}|U_x|^2|\partial_x\varphi_k|^2\,dx, \\
&\leq\frac{\mu}{2}\lVert(U_k)_x\rVert_{L^2}^2+\frac{C}{k^2}\lVert U\rVert_{H^1}^2.
\end{aligned} I 1 = 3 ∫ R ∂ x ( ∂ x U ∂ x φ k ) U k d x = − 3 ∫ R U x ∂ x φ k ( U k ) x d x , ≤ 2 μ ∥( U k ) x ∥ L 2 2 + 2 μ 9 ∫ R ∣ U x ∣ 2 ∣ ∂ x φ k ∣ 2 d x , ≤ 2 μ ∥( U k ) x ∥ L 2 2 + k 2 C ∥ U ∥ H 1 2 .
For the remaining terms, we use (68) and Young’s inequality:
I 2 = ∫ R U ∂ x 3 φ k U k d x ≤ C k − 3 ∥ U ∥ L 2 ∥ U k ∥ L 2 ≤ λ 8 ∥ U k ∥ L 2 2 + C k − 6 ∥ U ∥ L 2 2 , I_2=\int_{\mathbb{R}}U\partial_x^3\varphi_kU_k\,dx\leq Ck^{-3}\lVert U\rVert_{L^2}\lVert U_k\rVert_{L^2}\leq\frac{\lambda}{8}\lVert U_k\rVert_{L^2}^2+Ck^{-6}\lVert U\rVert_{L^2}^2, I 2 = ∫ R U ∂ x 3 φ k U k d x ≤ C k − 3 ∥ U ∥ L 2 ∥ U k ∥ L 2 ≤ 8 λ ∥ U k ∥ L 2 2 + C k − 6 ∥ U ∥ L 2 2 ,
I 3 = ∫ R U u ‾ ∂ x φ k U k d x ≤ C k − 1 ∥ u ∥ L ∞ ∥ U ∥ L 2 ∥ U k ∥ L 2 ≤ λ 8 ∥ U k ∥ L 2 2 + C k − 2 ∥ U ∥ L 2 2 , I_3=\int_{\mathbb{R}}U\overline{u}\partial_x\varphi_kU_k\,dx\leq Ck^{-1}\lVert u\rVert_{L^\infty}\lVert U\rVert_{L^2}\lVert U_k\rVert_{L^2}\leq\frac{\lambda}{8}\lVert U_k\rVert_{L^2}^2+Ck^{-2}\lVert U\rVert_{L^2}^2, I 3 = ∫ R U u ∂ x φ k U k d x ≤ C k − 1 ∥ u ∥ L ∞ ∥ U ∥ L 2 ∥ U k ∥ L 2 ≤ 8 λ ∥ U k ∥ L 2 2 + C k − 2 ∥ U ∥ L 2 2 ,
I 4 = − 2 μ ∫ R ∂ x U ∂ x φ k U k d x ≤ 2 C μ k − 1 ∥ ∂ x U ∥ L 2 ∥ U k ∥ L 2 ≤ λ 8 ∥ U k ∥ L 2 2 + C k − 2 ∥ U ∥ H 1 2 , I_4=-2\mu\int_{\mathbb{R}}\partial_xU\partial_x\varphi_kU_k\,dx\leq2C\mu k^{-1}\lVert\partial_xU\rVert_{L^2}\lVert U_k\rVert_{L^2}\leq\frac{\lambda}{8}\lVert U_k\rVert_{L^2}^2+Ck^{-2}\lVert U\rVert_{H^1}^2, I 4 = − 2 μ ∫ R ∂ x U ∂ x φ k U k d x ≤ 2 C μ k − 1 ∥ ∂ x U ∥ L 2 ∥ U k ∥ L 2 ≤ 8 λ ∥ U k ∥ L 2 2 + C k − 2 ∥ U ∥ H 1 2 ,
I 5 = − μ ∫ R U ∂ x 2 φ k U k d x ≤ C μ k − 2 ∥ U ∥ L 2 ∥ U k ∥ L 2 ≤ λ 8 ∥ U k ∥ L 2 2 + C k − 4 ∥ U ∥ L 2 2 . I_5=-\mu\int_{\mathbb{R}}U\partial_x^2\varphi_kU_k\,dx\leq C\mu k^{-2}\lVert U\rVert_{L^2}\lVert U_k\rVert_{L^2}\leq\frac{\lambda}{8}\lVert U_k\rVert_{L^2}^2+Ck^{-4}\lVert U\rVert_{L^2}^2. I 5 = − μ ∫ R U ∂ x 2 φ k U k d x ≤ C μ k − 2 ∥ U ∥ L 2 ∥ U k ∥ L 2 ≤ 8 λ ∥ U k ∥ L 2 2 + C k − 4 ∥ U ∥ L 2 2 .
Since k ≥ 1 k\geq1 k ≥ 1 , we have k − 4 , k − 6 ≤ k − 2 k^{-4},k^{-6}\leq k^{-2} k − 4 , k − 6 ≤ k − 2 , hence
∫ R m U k d x ≤ μ 2 ∥ ( U k ) x ∥ L 2 2 + λ 2 ∥ U k ∥ L 2 2 + C k 2 ∥ U ∥ H 1 2 . \int_{\mathbb{R}}mU_k\,dx\leq\frac{\mu}{2}\lVert(U_k)_x\rVert_{L^2}^2+\frac{\lambda}{2}\lVert U_k\rVert_{L^2}^2+\frac{C}{k^2}\lVert U\rVert_{H^1}^2. ∫ R m U k d x ≤ 2 μ ∥( U k ) x ∥ L 2 2 + 2 λ ∥ U k ∥ L 2 2 + k 2 C ∥ U ∥ H 1 2 .
Substituting this above, absorbing μ 2 ∥ ( U k ) x ∥ L 2 2 \frac{\mu}{2}\lVert(U_k)_x\rVert_{L^2}^2 2 μ ∥( U k ) x ∥ L 2 2 into μ ∥ ( U k ) x ∥ L 2 2 \mu\lVert(U_k)_x\rVert_{L^2}^2 μ ∥( U k ) x ∥ L 2 2 on the left, we get
d d t ∥ U k ∥ L 2 2 + λ ∥ U k ∥ L 2 2 ≤ C 4 k 2 ∥ U ( t ) ∥ H 1 2 . \frac{d}{dt}\lVert U_k\rVert_{L^2}^2+\lambda\lVert U_k\rVert_{L^2}^2\leq\frac{C_4}{k^2}\lVert U(t)\rVert_{H^1}^2. d t d ∥ U k ∥ L 2 2 + λ ∥ U k ∥ L 2 2 ≤ k 2 C 4 ∥ U ( t ) ∥ H 1 2 .
Integrating this on ( τ , t ) (\tau,t) ( τ , t ) yields
∫ τ t ∥ U k ( s ) ∥ L 2 2 d s ≤ λ − 1 ∥ U k ( τ ) ∥ L 2 2 + C 5 k 2 ∫ τ t ∥ U ( s ) ∥ H 1 2 d s . (72) \int_\tau^t\lVert U_k(s)\rVert_{L^2}^2\,ds\leq\lambda^{-1}\lVert U_k(\tau)\rVert_{L^2}^2+\frac{C_5}{k^2}\int_\tau^t\lVert U(s)\rVert_{H^1}^2\,ds. \tag*{(72)} ∫ τ t ∥ U k ( s ) ∥ L 2 2 d s ≤ λ − 1 ∥ U k ( τ ) ∥ L 2 2 + k 2 C 5 ∫ τ t ∥ U ( s ) ∥ H 1 2 d s . (72)
At the same time, by integrating (66) on ( τ , t ) (\tau,t) ( τ , t ) , we get, again using ∥ f ∥ H 1 2 ∼ ∥ f ∥ L 2 2 + β ∥ ∂ x f ∥ L 2 2 \lVert f\rVert_{H^1}^2\sim\lVert f\rVert_{L^2}^2+\beta\lVert\partial_xf\rVert_{L^2}^2 ∥ f ∥ H 1 2 ∼ ∥ f ∥ L 2 2 + β ∥ ∂ x f ∥ L 2 2 ,
λ ∫ τ t ∥ U ( s ) ∥ H 1 2 d s ≤ ∥ U ( τ ) ∥ H 1 2 + c 1 ∫ τ t ∥ U k ( s ) ∥ L 2 2 d s \lambda\int_\tau^t\lVert U(s)\rVert_{H^1}^2\,ds\leq\lVert U(\tau)\rVert_{H^1}^2+c_1\int_\tau^t\lVert U_k(s)\rVert_{L^2}^2\,ds λ ∫ τ t ∥ U ( s ) ∥ H 1 2 d s ≤ ∥ U ( τ ) ∥ H 1 2 + c 1 ∫ τ t ∥ U k ( s ) ∥ L 2 2 d s
Inserting this one into (72) , we get
∫ τ t ∥ U k ( s ) ∥ L 2 2 d s ≤ λ − 1 ∥ U k ( τ ) ∥ L 2 2 + C 6 k 2 ∥ U ( τ ) ∥ H 1 2 + C 7 k 2 ∫ τ t ∥ U k ( s ) ∥ L 2 2 d s . \int_\tau^t\lVert U_k(s)\rVert_{L^2}^2\,ds\leq\lambda^{-1}\lVert U_k(\tau)\rVert_{L^2}^2+\frac{C_6}{k^2}\lVert U(\tau)\rVert_{H^1}^2+\frac{C_7}{k^2}\int_\tau^t\lVert U_k(s)\rVert_{L^2}^2\,ds. ∫ τ t ∥ U k ( s ) ∥ L 2 2 d s ≤ λ − 1 ∥ U k ( τ ) ∥ L 2 2 + k 2 C 6 ∥ U ( τ ) ∥ H 1 2 + k 2 C 7 ∫ τ t ∥ U k ( s ) ∥ L 2 2 d s .
By choosing k k k sufficiently large, we obtain the desired estimate (71) . □
Theorem 5.4. For every t ∈ R t\in\mathbb{R} t ∈ R , the pullback attractor A μ ( t ) \mathcal{A}_{\mu}(t) A μ ( t ) has a finite fractal dimension in H 1 ( R ) H^1(\mathbb{R}) H 1 ( R ) , i.e.,
dim H 1 ( A μ ( t ) ) < + ∞ . \dim_{H^1}(\mathcal{A}_{\mu}(t))<+\infty. dim H 1 ( A μ ( t )) < + ∞.
Proof. We apply the Theorem 5.1 with
X = H 1 ( R ) , A ( t ) = A μ ( t ) . X=H^1(\mathbb{R}),\qquad\mathcal{A}(t)=\mathcal{A}_{\mu}(t). X = H 1 ( R ) , A ( t ) = A μ ( t ) .
By Theorem 3.8, the family { A ( t ) } t ∈ R \{\mathcal{A}(t)\}_{t\in\mathbb{R}} { A ( t ) } t ∈ R is compact in H 1 ( R ) H^1(\mathbb{R}) H 1 ( R ) . Let
V ( t , s ) = U μ ( t , s ) \mathcal{V}(t,s)=U_{\mu}(t,s) V ( t , s ) = U μ ( t , s )
be the process generated by (1) . By the invariance of the pullback attractor, for t ≥ s t\geq s t ≥ s
A ( t ) = V ( t , s ) A ( s ) , \mathcal{A}(t)=\mathcal{V}(t,s)\mathcal{A}(s), A ( t ) = V ( t , s ) A ( s ) ,
so the condition ( i ) (i) ( i ) of the Theorem 5.1 holds.
Let u 1 , u 2 ∈ A ( s ) u_1,u_2\in\mathcal{A}(s) u 1 , u 2 ∈ A ( s ) and set
U ( t ) = V ( t , s ) u 1 − V ( t , s ) u 2 . U(t)=\mathcal{V}(t,s)u_1-\mathcal{V}(t,s)u_2. U ( t ) = V ( t , s ) u 1 − V ( t , s ) u 2 .
Then U U U solves (61) . From Lemma 5.2, for k ≥ k 0 k \ge k_0 k ≥ k 0 ,
∥ U ( t ) ∥ H 1 2 ≤ e − λ ( t − s ) ∥ U ( s ) ∥ H 1 2 + c 1 ∫ s t ∫ ∣ x ∣ ≤ k U 2 ( τ , x ) d x d τ . \lVert U(t)\rVert_{H^1}^2 \le e^{-\lambda(t-s)}\lVert U(s)\rVert_{H^1}^2+c_1\int_s^t\int_{\lvert x\rvert\le k}U^2(\tau,x)\,dx\,d\tau. ∥ U ( t ) ∥ H 1 2 ≤ e − λ ( t − s ) ∥ U ( s ) ∥ H 1 2 + c 1 ∫ s t ∫ ∣ x ∣ ≤ k U 2 ( τ , x ) d x d τ .
Since ∫ ∣ x ∣ ≤ k U 2 d x ≤ ∥ U k ∥ L 2 2 \int_{\lvert x\rvert\le k}U^2\,dx\le\lVert U_k\rVert_{L^2}^2 ∫ ∣ x ∣ ≤ k U 2 d x ≤ ∥ U k ∥ L 2 2 , Lemma 5.3 gives
∫ s t ∥ U k ( τ ) ∥ L 2 2 d τ ≤ c 6 ∥ U k ( s ) ∥ L 2 2 + c 7 k 2 ∥ U ( s ) ∥ H 1 2 . \int_s^t\lVert U_k(\tau)\rVert_{L^2}^2\,d\tau\le c_6\lVert U_k(s)\rVert_{L^2}^2+\frac{c_7}{k^2}\lVert U(s)\rVert_{H^1}^2. ∫ s t ∥ U k ( τ ) ∥ L 2 2 d τ ≤ c 6 ∥ U k ( s ) ∥ L 2 2 + k 2 c 7 ∥ U ( s ) ∥ H 1 2 .
Because U k ( s ) = φ k U ( s ) U_k(s)=\varphi_kU(s) U k ( s ) = φ k U ( s ) and φ k = 1 \varphi_k=1 φ k = 1 on ∣ x ∣ ≤ k \lvert x\rvert\le k ∣ x ∣ ≤ k , we have
∥ U k ( s ) ∥ L 2 ≤ ∥ U ( s ) ∥ L 2 ( ∣ x ∣ ≤ 2 k ) ≤ ∥ U ( s ) ∥ H 1 . \lVert U_k(s)\rVert_{L^2}\le\lVert U(s)\rVert_{L^2(\lvert x\rvert\le2k)}\le\lVert U(s)\rVert_{H^1}. ∥ U k ( s ) ∥ L 2 ≤ ∥ U ( s ) ∥ L 2 (∣ x ∣ ≤ 2 k ) ≤ ∥ U ( s ) ∥ H 1 .
Therefore,
∥ U ( t ) ∥ H 1 2 ≤ e − λ ( t − s ) ∥ U ( s ) ∥ H 1 2 + c 1 c 6 ∥ U ( s ) ∥ L 2 ( ∣ x ∣ ≤ 2 k ) 2 + c 1 c 7 k 2 ∥ U ( s ) ∥ H 1 2 . \lVert U(t)\rVert_{H^1}^2\le e^{-\lambda(t-s)}\lVert U(s)\rVert_{H^1}^2+c_1c_6\lVert U(s)\rVert_{L^2(\lvert x\rvert\le2k)}^2+\frac{c_1c_7}{k^2}\lVert U(s)\rVert_{H^1}^2. ∥ U ( t ) ∥ H 1 2 ≤ e − λ ( t − s ) ∥ U ( s ) ∥ H 1 2 + c 1 c 6 ∥ U ( s ) ∥ L 2 (∣ x ∣ ≤ 2 k ) 2 + k 2 c 1 c 7 ∥ U ( s ) ∥ H 1 2 .
Note that the constants c 1 , c 6 , c 7 c_1,c_6,c_7 c 1 , c 6 , c 7 are independent of s s s , t t t , and k k k . Taking square roots and using a + b ≤ a + b \sqrt{a+b}\le\sqrt a+\sqrt b a + b ≤ a + b ,
∥ U ( t ) ∥ H 1 ≤ e − λ ( t − s ) / 2 ∥ U ( s ) ∥ H 1 + ( c 1 c 6 ) 1 / 2 ∥ U ( s ) ∥ L 2 ( ∣ x ∣ ≤ 2 k ) + ( c 1 c 7 k 2 ) 1 / 2 ∥ U ( s ) ∥ H 1 . \lVert U(t)\rVert_{H^1}\le e^{-\lambda(t-s)/2}\lVert U(s)\rVert_{H^1}+(c_1c_6)^{1/2}\lVert U(s)\rVert_{L^2(\lvert x\rvert\le2k)}+\left(\frac{c_1c_7}{k^2}\right)^{1/2}\lVert U(s)\rVert_{H^1}. ∥ U ( t ) ∥ H 1 ≤ e − λ ( t − s ) /2 ∥ U ( s ) ∥ H 1 + ( c 1 c 6 ) 1/2 ∥ U ( s ) ∥ L 2 (∣ x ∣ ≤ 2 k ) + ( k 2 c 1 c 7 ) 1/2 ∥ U ( s ) ∥ H 1 .
Choose k ≥ k 0 k\ge k_0 k ≥ k 0 large enough so that
( c 1 c 7 k 2 ) 1 / 2 ≤ 1 8 . \left(\frac{c_1c_7}{k^2}\right)^{1/2}\le\frac18. ( k 2 c 1 c 7 ) 1/2 ≤ 8 1 .
Since t t t is fixed, for s s s sufficiently negative we also have
e − λ ( t − s ) / 2 ≤ 1 8 . e^{-\lambda(t-s)/2}\le\frac18. e − λ ( t − s ) /2 ≤ 8 1 .
Hence, for all such s s s ,
∥ U ( t ) ∥ H 1 ≤ 1 4 ∥ U ( s ) ∥ H 1 + K ∥ U ( s ) ∥ L 2 ( ∣ x ∣ ≤ 2 k ) , \lVert U(t)\rVert_{H^1}\le\frac14\lVert U(s)\rVert_{H^1}+K\lVert U(s)\rVert_{L^2(\lvert x\rvert\le2k)}, ∥ U ( t ) ∥ H 1 ≤ 4 1 ∥ U ( s ) ∥ H 1 + K ∥ U ( s ) ∥ L 2 (∣ x ∣ ≤ 2 k ) ,
where
K = ( c 1 c 6 ) 1 / 2 . K=(c_1c_6)^{1/2}. K = ( c 1 c 6 ) 1/2 .
This is exactly the condition ( i i i ) (iii) ( iii ) of Theorem 5.1 with
η t , s = 1 4 , n 1 ( v ) = ∥ v ∥ L 2 ( ∣ x ∣ ≤ 2 k ) , n 2 ≡ 0. \eta_{t,s}=\frac14,\qquad n_1(v)=\lVert v\rVert_{L^2(\lvert x\rvert\le2k)},\qquad n_2\equiv0. η t , s = 4 1 , n 1 ( v ) = ∥ v ∥ L 2 (∣ x ∣ ≤ 2 k ) , n 2 ≡ 0.
The semi-norm n 1 n_1 n 1 is compact in H 1 ( R ) H^1(\mathbb{R}) H 1 ( R ) , since the embedding H 1 ( ∣ x ∣ ≤ 2 k ) ↪ L 2 ( ∣ x ∣ ≤ 2 k ) H^1(\lvert x\rvert\le2k)\hookrightarrow L^2(\lvert x\rvert\le2k) H 1 (∣ x ∣ ≤ 2 k ) ↪ L 2 (∣ x ∣ ≤ 2 k ) is compact. Moreover,
lim sup s → − ∞ η t , s = 1 4 < 1 2 , lim sup s → − ∞ K t , s = K < + ∞ . \limsup_{s\to-\infty}\eta_{t,s}=\frac14<\frac12,\qquad\limsup_{s\to-\infty}K_{t,s}=K<+\infty. s → − ∞ lim sup η t , s = 4 1 < 2 1 , s → − ∞ lim sup K t , s = K < + ∞.
For the condition ( i i ) (ii) ( ii ) , the same estimate gives
∥ U ( t ) ∥ H 1 ≤ L t , s ∥ U ( s ) ∥ H 1 , \lVert U(t)\rVert_{H^1}\le L_{t,s}\lVert U(s)\rVert_{H^1}, ∥ U ( t ) ∥ H 1 ≤ L t , s ∥ U ( s ) ∥ H 1 ,
where
L t , s = ( e − λ ( t − s ) + c 1 c 6 + c 1 c 7 k 2 ) 1 / 2 . L_{t,s}=\left(e^{-\lambda(t-s)}+c_1c_6+\frac{c_1c_7}{k^2}\right)^{1/2}. L t , s = ( e − λ ( t − s ) + c 1 c 6 + k 2 c 1 c 7 ) 1/2 .
Thus
lim sup s → − ∞ L t , s = ( c 1 c 6 + c 1 c 7 k 2 ) 1 / 2 < + ∞ . \limsup_{s\to-\infty}L_{t,s}=\left(c_1c_6+\frac{c_1c_7}{k^2}\right)^{1/2}<+\infty. s → − ∞ lim sup L t , s = ( c 1 c 6 + k 2 c 1 c 7 ) 1/2 < + ∞.
Therefore, all conditions of Theorem 5.1 are satisfied. We conclude that the pullback attractor A μ ( t ) \mathcal{A}_{\mu}(t) A μ ( t ) has a finite fractal dimension in H 1 ( R ) H^1(\mathbb{R}) H 1 ( R ) . □
Acknowledgments
This work was carried out while the first author was visiting the Mathematical Institute of the University of Göttingen. He would like to thank the Institute for its hospitality. N.A. and B.T. are supported by NAWI Graz.
Cung The Anh
Faculty of Mathematics and Informatics, Hanoi National University of Education
136 Xuan Thuy, Cau Giay, Hanoi, Vietnam
Email address: anhoctmath@hnue.edu.vn
Nesibe Ayhan
Department of Mathematics and Scientific Computing, University of Graz Heinrichstrasse 36, 8010 Graz, Austria Email address: nesibe.ayhan@uni-graz.at
Bao Quoc Tang
Department of Mathematics and Scientific Computing, University of Graz Heinrichstrasse 36, 8010 Graz, Austria Email address: quoc.tang@uni-graz.at
Ingo Witt
Mathematical Institute, University of Göttingen, Bunsenstraße 3-5, D-37073 Göttingen, Germany Email address: iwitt@uni-math.gwdg.de