1. Introduction
This paper considers the 3D Navier–Stokes–Voigt (NSV) equation with damping and double delays as follows:
\begin{equation}
\left\{
\begin{aligned}
&\frac{\partial}{\partial t}(1-\alpha^2\Delta)u-\nu\Delta u+(u(t-\rho(t)) \cdot \nabla)u+\beta u+\gamma|u|^{r-1}u\\
&\quad +\nabla p=f(t,u_t)+g(x,t),\\
& \mbox{div} \ u=0,\ \ \ \ (x,t) \ \ \mbox{in}\ \ \Omega_{\tau} ,\\
& u=0,\ \ \ \ x\in \partial\Omega,\\
&u(\tau,x)=u^{\tau}(x), \ \ x \in \Omega,\\
&u(s+\tau,x)=\phi(s,x),\ \ (x,s)\in \Omega_h.\\
\end{aligned}
\right.
\end{equation}We provide some commentaries for the system (1.1).
(I) (Unbounded domain) Assume that Ω is an unbounded Poincaré domain with a boundary
$\partial\Omega$ (for example in [Reference Wang41],
$\Omega=\mathcal{O}\times\mathbb{R}$ is usually a 3D unbounded channel, where
$\mathcal{O}$ is a smooth bounded domain in
$\mathbb{R}^2$). Then, there is a
$\lambda_1 \gt 0$ such that
\begin{align*}
\int_{\Omega}|v|^2ds\leq\frac{1}{\lambda_1}\int_{\Omega}|\nabla v|^2ds,\ \forall v\in H^1_0(\Omega),
\end{align*}which is crucial for obtaining dissipative solution. For the research results on unbounded Poincaré domains of NS equations, refer to [Reference Anh and Trang3, Reference Garrido-Atienza and Marín-Rubio19, Reference Ju21, Reference Marín-Rubio and Real29, Reference Rosa34, Reference Wang41].
(II) (Non-autonomous and delay external force term) Memory processes and no-instant transmission are often found in practical systems. Therefore, it is necessary to consider delay systems. Caraballo in [Reference Caraballo, Marín-Rubio and Valero8, Reference Caraballo, Marín-Rubio and Valero9] has already conducted extensive theoretical research on the existence and stability of solutions and attractors for delay differential equations and has applied these results to the 2D Navier–Stokes (NS) equation with delay in[Reference Caraballo and Real10, Reference Caraballo and Real11, Reference Liu and Caraballo23, Reference Liu, Caraballo and Marín-Rubio24, Reference Marín-Rubio, Márquez-Durán and Real28]. We consider the non-autonomous external forcing term
$g(t,x)$ and external forcing term containing some memory effects during a fixed interval of time of length h
$ \gt $ 0, i.e.,
$f(t,u_t)=f(t,u(t-\rho(t)))$, where
$\rho\in C^1(\mathbb{R};[0,h])$,
$u^{\tau}(x)$ is the initial velocity field,
and ϕ represents the initial value in the time interval
$(-h,0)$ and and
$\Omega_{\tau}=\Omega\times (\tau,\infty)$,
$\Omega_h=\Omega\times (-h,0)$. The case
$f(t,u_t)$ is considered including situations such as distributed delay, variable delay and delay in gradient, or second-order derivative terms in [Reference Caraballo, Marín-Rubio and Valero9, Reference Caraballo and Real10, Reference Marín-Rubio and Real30].(III) (Convective term with delay) Considering a time-delay function in the convective term
$(u(t-\rho(t)) \cdot \nabla)u$ is also meaningful, that is, when the trajectory appears, it may occur that the delay of fluid particles follows the fluid. The motivation for introducing this delay has been well explored in the introduction of [Reference Liu25] for Burgers equation. Later on, the research of 2D NS equation with double delays can be found in [Reference Bessaih, Garrido-Atienza and Schmalfuß6, Reference Garcia-Luengo, Marín-Rubio and Planas18, Reference Shi, Cui, Li and Yang35, Reference Yang, Chen and Caraballo44].(IV) (Voigt-term) The system (1.1) includes Voigt term
$-\alpha^2\frac{\partial}{\partial t}\triangle u \ (\alpha \gt 0)$, which was introduced by Oskolkov in [Reference Oskolkov32] as a model of motion of linear, viscoelastic fluids. Afterwards, the NSV equation in [Reference Cao, Lunasin and Titi7] with a small value of α was also proposed as a regularization for simulating periodic and no-slip Dirichlet boundary conditions in 3D NS equations for direct numerical calculations.(V) (Damping term) The damping term is considered as
$\beta u+\gamma|u|^{r-1}u,$ where
$\gamma \gt 0,\ \beta\in\mathbb{R}$ are constants, and
$r\in[1,\infty)$. It should be pointed out that the damping term originates from the resistance to the motion of flow in [Reference Anh and Trang4], which describes various physical situations, including porous media flow, drag, or friction effects, and other dissipative mechanisms.
The study of system (1.1) in this paper can be extended to some systems by investigating limit behaviours of invariant measures with parameters and delay, for example,
(a) (The case of
$\rho(t)=0$)
In the absence of delay, i.e.,
$\rho(t)=0$, the system (1.1) is
(1.2)
\begin{equation}
\left\{
\begin{aligned}
&\frac{\partial}{\partial t}(1-\alpha^2\Delta)u-\nu\Delta u+(u(t) \cdot \nabla)u+\beta u+\gamma|u|^{r-1}u\\
&\quad +\nabla p=f(t,u)+g(x,t),\\
& \mbox{div} \ u=0,\ \ \ \ (x,t) \ \ \mbox{in}\ \ \Omega_{\tau}.\\
\end{aligned}
\right.
\end{equation}The study of well-posedness and attractors for (1.2) in some unbounded domains can be found in [Reference Anh and Trang3, Reference Anh and Trang4]. However, the research result of the invariant measure of system (1.2) is unknown. When
$\rho(t)\rightarrow0$, can the invariant measure of the system (1.2) be characterized by the limit behaviour of the invariant measure of system (1.1)? The answer is yes, and refer to Section 6.1 for details.(b) (The case of α = 0) When the parameter α = 0, the system (1.1) is
(1.3)
\begin{equation}
\left\{
\begin{aligned}
&\frac{\partial u}{\partial t}-\nu\Delta u+(u(t-\rho(t)) \cdot \nabla)u+\beta u+\gamma|u|^{r-1}u+\nabla p=f(t,u_t)+g(x,t),\\
& \mbox{div} \ u=0,\ \ \ \ (x,t) \ \ \mbox{in}\ \ \Omega_{\tau}.\\
\end{aligned}
\right.
\end{equation}The research results on the well-posedness, regularity, and attractors of system (1.3) in space-periodic domain can be found in [Reference Shi, Cui, Li and Yang35]. There are also no research results of invariant measure on unbounded domains for the system (1.3). Specifically, it is also highly significant to consider the limit behavior of the invariant measure of system (1.1) approaching the system (1.3) as
$\alpha\rightarrow 0$ and see Section 6.2 for details.(c) (The case of
$\beta=\gamma=0$) If
$\beta=\gamma=0$, the system (1.1) is
(1.4)
\begin{equation}
\left\{
\begin{aligned}
&\frac{\partial}{\partial t}(1-\alpha^2\Delta)u-\nu\Delta u+(u(t-\rho(t)) \cdot \nabla)u+\nabla p=f(t,u_t)+g(t,x),\\
& \mbox {div}\ u=0,\ \ \ \ (x,t) \ \ \mbox{in}\ \ \Omega_{\tau}.\\
\end{aligned}
\right.
\end{equation}The pullback attractors of 3D NSV equations in some unbounded domains can be found in [Reference Anh and Trang3, Reference García-Luengo, Marín-Rubio and Real17]. The research results on long-time behaviour of solutions for 3D NSV equations with delay on bounded damain can be found in [Reference Qin and Su33, Reference Toi39]. Currently, there are no research results for the system (1.4) on unbounded domains, especially concerning invariant measures. However, the results of system (1.1) in this paper can be extended to system (1.4).
In this paper, our main goal is to consider the limit behaviours of invariant measures of system (1.1) with respect to parameters and delay, particularly in the case of the parameters
$\alpha \ \mbox{or} \ \gamma\rightarrow 0$. We first prove the well-posedness and pullback attractor of system (1.1), and then we construct an invariant Borel probability measure supported on the pullback attractor. Finally, the limit behaviours of invariant measures are considered. It is highly significant to investigate the properties of solutions to the classical 3D NS equation in the future, by exploring the limit behaviour of the invariant measure of system (1.1) as the parameters
$\alpha, \beta, \gamma, \rho(t)$ all approach 0. Before obtaining the above results, let us first introduce the concept of invariant measure.
Invariant measures: Studying invariant measures in fluid systems is crucial [Reference Foias, Manley, Rosa and Temam16], as measurements of various aspects of turbulence are essentially measurements of time-averaged quantities. In particular, the existence of invariant measures for autonomous dissipative dynamical systems is obtained by applying generalized Banach limits in [Reference Lukaszewicz, Real and Robinson26]. Subsequently, the proofs can be simplified in [Reference Chekroun and Glatt-Holtz12] and can be extended to a broad class of autonomous dissipative systems. In [Reference Lukaszewicz and Robinson27], further research was conducted on the invariant measure of non-autonomous dissipative dynamical systems, and the abstract results were applied to the 2D NS equations. For additional insights into the study of invariant measures and their application to the NS equations with delays, refer to [Reference Wang, Zhao and Caraballo42, Reference Yang, Chen and Caraballo44, Reference Zhao and Yang46]. Based on the above motivation, the main results and features of this present paper can be summarized as follows.
The first goal of this paper is to extend the above results of long-time behaviours of solutions to the NSV equation with double delays and damping on unbounded Poincaré domain. The unbounded domain case was introduced by Abergel [Reference Abergel1] and Babin [Reference Babin5]. Another condition required in the works of Abergel [Reference Abergel1, Reference Abergel2] and Babin [Reference Babin5] is the smoothness of boundary of the domain in order to obtain the appropriate regularity of the domain of the Stokes operator. This regularity is no longer required in [Reference Rosa34], and the domain can be an arbitrary open set as long as the Poincaré inequality holds. This includes the channel flow, the multi-channel flow, and other variations. The existence of attractors in dissipative evolution equations has always depended on the compactness of the semigroup generated by these equations. Usually, this compactness is obtained through the compact embedding of relevant Sobolev spaces by combining some regularization properties of these equations. This method is only applicable to bounded domains; otherwise, Sobolev embedding is no longer compact. However, by using the energy equation method, Rosa [Reference Rosa34] obtained the so-called asymptotic compactness of the semigroup generated by the 2D NS equations on unbounded Poincaré domains, which was still valid in proving the existence of attractors. Later on, Marín-Rubio [Reference Garrido-Atienza and Marín-Rubio19, Reference Marín-Rubio and Real29] investigated the well-posedness and pullback attractor of NS equations with delay. The results of the NSV equations on unbounded domains can be found in [Reference Anh and Trang3, Reference Ju21], and NSV equations with damping on unbounded domains can be found in [Reference Anh and Trang4]. In this paper, for the long-time behaviour of solution for system (1.1), since Ω is an unbounded domain without regularity assumptions at the boundary, the main difficulty is that compact embedding of
$V\hookrightarrow H$ is not valid on whole Ω, which causes difficulty in taking the limit of the approximation equation. In addition, we investigate the existence and limit behaviours of invariant measures that were not considered in [Reference Anh and Trang3, Reference Anh and Trang4]. Here, we also introduce double delays
$(u(t-\rho(t))\cdot\nabla)u$ and
$f(t,u_t)$. Due to the presence of variable delay in the convective term, the estimates of trilinear operator
$b(u(t-\rho(t)),u,v)$ and sublinear delay operator
$f(t,u(t-\rho(t))$ are more complex to achieve well-posedness and asymptotic compactness property of pullback attractor for system (1.1) and can refer to
$\S$ 3 and
$\S$ 4 for details.
The second goal of this paper is to prove the existence and periodicity of invariant measures associated with the process
$\{U(t,\tau)\}_{t\geq\tau}$ for system (1.1). The key point is to establish the continuity of
$U(t,\tau)\phi$ with respect to initial time τ and initial value ϕ. Then, by applying Theorem 3.1 in [Reference Lukaszewicz and Robinson27], we can obtain an invariant measure. Furthermore, if the external forcing terms of system (1.1) satisfy periodicity with respect to time t, a periodic dynamical system can be obtained. It is meaningful to consider the periodic behaviour of solution in [Reference Chen and Yang13, Reference Da Prato and Debussche15, Reference Taniuchi36, Reference Yang, Caraballo and Chen43]. Compared to [Reference Yang, Caraballo and Chen43], we need to consider not only the unbounded domain Ω but also the periodicity of variable delay
$\rho(t)$. By the periodicity of dynamical system, a periodic invariant measure is attained and see
$\S$ 5 for details.
Finally, the limit behaviours of invariant measures with parameters and delay are considered for system (1.1). We prove that when the limits of invariant measures with respect to delay approaches zero and parameter α tend to a number, any limit of the invariant measure sequences for this class of equations must be the invariant measure of the corresponding limit equations. Compared to [Reference Yang, Caraballo and Chen43], the presence of variable delay in the convective term makes the analysis of
$B(u(t-\rho(t)),u)$ more challenging as
$\rho(t)\rightarrow 0$. When the parameter
$\alpha\rightarrow\alpha_0,\ 0 \lt \alpha_0 \lt \infty$, the limiting behaviour of the invariant measures in the V space is obtained. Specifically, it should be noted that when α = 0, system (1.1) is reduced into the 3D incompressible NS equation with double delays and damping, and the limiting behaviour of the invariant measures in the CH space needs to be considered. The limit behaviour of the invariant measure for parameter
$\gamma\rightarrow\gamma_0 \ (0\leq\gamma_0 \lt \infty)$ is similar to α. The results are novel and significant, and see
$\S$ 6 for details.
This paper is organized as follows. In
$\S$ 2, the functional spaces and some useful inequalities are provided. In
$\S$ 3, the well-posedness of equation (1.1) can be obtained in an unbounded domain. In
$\S$ 4, the existence of minimal pullback attractors for the system (1.1) about weak solution can be obtained. In
$\S$ 5, we construct a family of invariant Borel probability measures (respectively, periodic invariant measure), which are supported on the pullback attractors. In
$\S$ 6, we will show the limits of invariant measures with respect to delay approaches 0 and parameters
$\alpha,\gamma$ tend to some numbers.
2. Preliminaries
In this section, we will first present some functional spaces and operations and then give some assumptions on the time delay and the external forcing in (1.1), which are needed for the well-posedness of problem (1.1) on unbounded Poincaré domains.
2.1. Functional spaces
Consider the following usual abstract spaces:
\begin{align*}
&\mathcal{V}:=\{u\in (C_0^{\infty}(\Omega))^3: \nabla\cdot u=0\},\nonumber\\
&H:=\mbox{the closure of} \ \mathcal{V} \ \mbox{in the Lebesgue space} \ (L^2(\Omega))^3,\nonumber\\
&V:=\mbox{the closure of} \ \mathcal{V} \ \mbox{in the Sobolev space} \ (H_0^1(\Omega))^3,
\end{align*}with inner products and norms as
\begin{align*}
&(u,v)_H=\sum_{i=1}^3\int_{\Omega}u_i(x)\cdot v_i(x)dx\;\;\mbox{and}\;\;|u|=(u,u)^{\frac{1}{2}}=\left(\int_{\Omega}|u(x)|^2dx\right)^{\frac{1}{2}},\nonumber\\
&(u,v)_V=((u,v))=\sum_{i=1}^3(D_iu,D_iv \big)_{H} \;\;\mbox{and}\;\;\|u\|^2=|\nabla u|^2.
\end{align*} Define
$\mathbb{L}^p=(L^p(\Omega))^3 \ (1\leq p\leq \infty)$. Furthermore, we define Banach spaces with delay by
\begin{equation*}||u||_{C_H}=\sup\limits_{s\in[-h,0]}|u(s)|,\ ||u||_{C_V}=\sup\limits_{s\in[-h,0]}||u(s)||.\end{equation*} Specifically,
$L_X^p=L^p(-h,0;X).$ Define
$\widetilde{V}=V\cap (L^{r+1}(\Omega))^3$.
Define a bilinear form
$a: V\times V\rightarrow \mathbb{R}$ as
Clearly, since the bilinear form a coincides with the scalar product
$((\cdot,\cdot))$ in V, it is continuous on V, meaning that it satisfies
$|a(u,u)|\leq C_1||u||^2$ for some
$C_1 \gt 0$ and all
$u\in V$. From Riesz Lemma, it is easy to obtain that there exists a unique linear operator
$\widetilde{A}:V\rightarrow V'$; here, Vʹ is the dual of V, such that
$a(u,v)=\langle \widetilde{A}u, v\rangle$, for
$u,v\in V$. Additionally, the V-coercive property of a implies that
$a(u,u)\geq C_2||u||^2$ for some
$C_2 \gt 0$ and all
$u\in V$. Thus, by the Lax-Milgram theorem, the operator
$\widetilde{A}:V\rightarrow V'$ is an isomorphism. By Chapter II in [Reference Foias, Manley, Rosa and Temam16], the following relation
holds, where the injections are continuous and the
$V\subset H$ is dense.
Define an unbounded operator A in H by
\begin{equation*}
\left\{
\begin{aligned}
&D(A):=\{u\in V; \widetilde{A}u\in H \},\\
&Au:=\widetilde{A}u,u\in D(A).
\end{aligned}
\right.
\end{equation*} It has been well proven that under some additional assumptions related to the regularity of domain Ω, space D(A) can be represented by Sobolev space. By Chapter II in [Reference Foias, Manley, Rosa and Temam16], if
$\Omega\subset\mathbb{R}^3$ is open, bounded, and connected, with C 2 boundary
$\partial\Omega$, it follows from the orthogonal projection of
$P_L:(L^2(\Omega))^3\rightarrow H$ that
\begin{equation*}
\left\{
\begin{aligned}
&D(A):=V\cap H^2,\\
&Au:=P_L(-\Delta u),~u\in D(A).
\end{aligned}
\right.
\end{equation*}A is also known as the Stokes operator that is a non-negative self-adjoint operator in H.
2.2. Properties of linear operators
$\bullet$ The operator A :
(
$\mathbf{A}_\mathbf{1}$) If Ω is a bounded domain, then the operator A is invertible and A −1 is bounded, self-adjoint and compact in H. Hence the spectrum of A generates eigenvalues
with
$\displaystyle{\lim_{m\rightarrow\infty}}\lambda_m=+\infty$ and corresponding eigenvectors
$\{w_m\}_{m=1}^\infty$ which satisfy
in H; see Chapter II in [Reference Foias, Manley, Rosa and Temam16].
(
$\mathbf{A}_\mathbf{2}$) If Ω is an unbounded Poincaré domain, then the operator A is invertible, the inverse A −1 is bounded, and
where
$D(A)=\{u\in V, \widetilde{A}u\in H\}$. Moreover, the norm on D(A) is equivalent to the
$|A\cdot|$-norm by
$\bullet$ The trilinear operator:
Define the trilinear operator
$b(u,v,w)$ on Ω, as well as
$B(u,v)$, by
\begin{align*}
b(u,v,w)=(B(u,v),w)=\sum_{i,j=1}^{3}\int_{\Omega}u_i\frac{\partial v_j}{\partial x_i}w_jdx
\end{align*} for
$u,v,w\in V$. It satisfies
\begin{equation}
\left\{
\begin{aligned}
&b(u,v,w)=-b(u,w,v),\ \mbox{for all}\ u,v,w\in V,\\
&b(u,v,v)=0,\ \mbox{for all}\ u,v\in V,\\
&b(u,u,Au)=0,\ \mbox{for all}\ u\in D(A) \ \mbox{only}\ Au=-\Delta u.
\end{aligned}
\right.
\end{equation}2.3. Some hypotheses and inequalities
$\bullet$ Hypotheses of the term with time delay:
(H 0)
$\rho:\mathbb{R}\rightarrow [0,h]$ and satisfies
$|\frac{d\rho}{dt}|\leq \rho^{\ast} \lt 1$.(H 1)
$f(\cdot,u):[\tau,+\infty)\rightarrow H$ is measurable for all
$u\in H$, and
$f(t,\cdot):H\rightarrow H$ is continuous for all
$t\geq \tau$.(H 2) There exist
$a\in L^{\infty}_{loc}(\mathbb{R};[0,\infty))$ and
$b\in L^1_{loc}(\mathbb{R};[0,\infty))$ such that
\begin{equation*}|f(t,u)|^2\leq a(t)|u|^2+b(t), \ \ \mbox {for any} \ u\in H.\end{equation*}(H 3) For any R
$ \gt $ 0, there exists a constant
$L(R) \gt 0$ such that
\begin{equation*}|f(t,u)-f(t,v)|\leq L(R)\Lambda^{\frac{1}{2}}(t)|u-v|,~\text{for any}\ |u|,|v|\leq R,\end{equation*}where
$\Lambda\in L^{\infty}_{loc}(\mathbb{R};[0,+\infty)).$
$\bullet$ Some lemmas:
Lemma 2.1. (Theorem 13.3, [5.6])
Let X and Y be two (not necessarily reflexive) Banach spaces with compact injection
$Y\subset X$. Given
$1\leq q \lt p\leq +\infty$, and for any
$T \gt \tau$, let
$\mathcal{H}$ be a bounded set in
$L^1(\tau,T;Y)\cap L^p(\tau,T;X)$ such that
\begin{align}
\int_{\tau}^{T-l}|| u(l+s)-u(s)||^q_Xds\rightarrow 0, \ \mbox{as}\ l\rightarrow 0,
\end{align} uniformly for
$u\in\mathcal{H}$. Then
$\mathcal{H}$ is relatively compact in
$L^q(\tau,T;X)$.
Lemma 2.2. (Gagliardo–Nirenberg inequality, [Reference Nirenberg31])
Assume
$1\leq q,r\leq\infty$, and
$j,m\in\mathbb{N}$ satisfying
$0\leq j \lt m$. For any
$u\in C_0^{\infty}(\mathbb{R}^n)$, there is a constant C depending on
$n,m,j,q,r,\theta$ such that
\begin{align*}
||D^ju||_{L^p(\mathbb{R}^n)} \leq C||D^mu||^{\theta}_{L^r(\mathbb{R}^n)}||u||^{1-\theta}_{L^q(\mathbb{R}^n)},
\end{align*} where
$\frac{1}{p}-\frac{j}{n}=\theta(\frac{1}{r}-\frac{m}{n})+(1-\theta)\frac{1}{q},\ \frac{j}{m}\leq\theta\leq 1.$ The above inequality holds for
$\frac{j}{m}\leq\theta \lt 1$ when
$m-j-\frac{n}{r}$ is a non-negative integer.
Based on Lemma 2.2, we provide the following interpolation inequality, which is very useful in the subsequent derivation process.
Lemma 2.3. (Interpolating inequality, [Reference Foias, Manley, Rosa and Temam16, Reference Nirenberg31])
Assume that
$2 \lt r\leq 5$, if
$u\in L^{\infty}(\tau,T;H)$ and
$u\in L^{r+1}(\tau,T; (L^{r+1}(\Omega))^3)$, then we can get
\begin{align}
||u||_{L^q(\tau,T;(L^p(\Omega))^3)}\leq C||u||^{\theta}_{L^{\infty}(\tau,T;H)}||u||^{1-\theta}_{L^{r+1}(\tau,T;(L^{r+1}(\Omega))^3)},
\end{align} where
$p=\frac{6(r+1)}{11-r}$,
$q=\frac{3(r-1)(r+1)}{4(r-2)}$, and
$\theta=\frac{5-r}{3(r-1)}$.
Lemma 2.4. [Reference Constantin and Foias14, Reference Foias, Manley, Rosa and Temam16]
In 3D case, Ladyzhenskaya, Agmon’s, and Sobolev inqualities are given as follows:
\begin{equation}
\left\{
\begin{aligned}
&||u||_{L^4(\Omega)}\leq c_1|u|^{\frac{1}{4}}|\nabla u|^{\frac{3}{4}}
, \ \mbox{for}\ \ \mbox{all}\ \ u\in V,\\
&||u||_{L^{\infty}(\Omega)}\leq c_2||u||^{\frac{1}{2}}||u||^{\frac{1}{2}}_{D(A)}
, \ \mbox{for}\ \ \mbox{all}\ \ u\in D(A),\\
&||u||_{L^6(\Omega)} \leq c_3||u||, \ \ u\in V.\\
\end{aligned}
\right.
\end{equation}3. Well-posedness of system (1.1)
In this section, we will investigate the existence and uniqueness of solutions to (1.1). To this end, we first consider the autonomous equation
$u+\alpha^2Au=g$. Using the Lax–Milgram lemma, we can find that for any given function
$g\in V'$, there is a unique solution
$u_g\in V$ such that
The mapping
$\mathcal{L}:u\in V\rightarrow u+\alpha^2Au\in V'$ is linear and bijective with
$\mathcal{L}^{-1}g=u_g$. By (3.1), we have
$|u_g|^2+\alpha^2||u_g||^2\leq ||g||_{V'}||u_g||$ and, in particular,
$||u_g||\leq\alpha^{-2}||g||_{V'}$, that is,
which is crucial to obtain the existence of solutions for system (1.1).
Theorem 3.1. Suppose that
$ g\in L_{loc}^2(\mathbb{R};V'), \ \phi\in C_V$, and
$(H_0)$-
$(H_2)$ hold, and assume
$2 \lt r\leq 5,\ \alpha,\gamma \gt 0,~\beta\in\mathbb{R}$. Then for all
$T\geq\tau$, the system (1.1) has a weak solution u such that
\begin{equation*}u\in C([\tau-h,T];V)\cap L^{r+1}(\tau,T;\mathbb{L}^{r+1}),\ \frac{du}{dt}\in L^{\frac{r+1}{r}}(\tau,T;V)\end{equation*}and satisfies
\begin{align}
\frac{d}{dt}(u+\alpha^2 Au,w)+ \nu a(u,w)&+b(u(t-\rho(t)),u,w)+(\gamma|u|^{r-1}u,w)+(\beta u,w)\nonumber\\
&\quad=(f(t,u(t-\rho(t)))
+g(t),w),
\end{align} for any
$w\in V$ in weak sense. Furthermore, if
$f(t,u(t-\rho(t)))$ also satisfies the assumption (H 3), then the solution is unique.
Proof. We will divide it into two parts to prove the theorem.
$\bullet$ Part I. Existence of solutions. To derive our results, we apply the Faedo–Galerkin method and compactness method, which are divided into the following four steps.
Step 1: Existence of local approximated solutions
Noting that Ω is an unbounded Poincaré domain, it is not valid to take a special basis here since compactness is lost by using the Stokes operator. However,
$\mathcal{V}$ is dense in
$\widetilde{V}$ and H, and
$\widetilde{V}$ is separable because it is a subspace of
$(H_0^1(\Omega))^3\cap (L^{r+1}(\Omega))^3$. Therefore, one can take a sequence
$\{v_i\}_{i\geq1}\subset \widetilde{V}$ and is dense in
$\widetilde{V}$. Then there exists a sequence
$\{\nu_n^i\}_{n\geq 1}\subset \mathcal{V}$ accumulating to vi for any
$i\geq 1$, such that the linear combination of
$\{\nu_n^i\}_{i,n\geq 1}$ are dense in
$\widetilde{V}$ and H. Thus, by the Gram–Schmidt orthonormalization process, we can get an orthonormal
$\{w_j\}_{j=1}^{\infty} \subset \mathcal{V}$ of H such that the linear combinations of these elements is dense in
$\widetilde{V}$.
Define
$V_m=\{w_1,\cdots,w_m\}$ and let the orthogonal projection
$P_m:H\rightarrow V_m$ be given as
$u_m(x,t):=P_mu(x,t)=\displaystyle{\sum_{i=1}^m}g_{i}(t)\omega_i(x)$ (here
$g_{i}(t)=(u,w_i),\ i=1,\ldots,m$ and
$g_{i}\in C^1[0,t_m], \ 0\leq t_m \lt T$). We construct the approximated solution
$u_m=\displaystyle{\sum_{i=1}^m}g_{i}(t)\omega_i(x)$ satisfying the following finite-dimensional problem:
\begin{equation}
\left\{
\begin{aligned}
&(u_m(t)+\alpha^2 Au_m(t),w_m)+\nu\int_{\tau}^{t}(Au_m(s)+B(u_m(s-\rho(s)),u_m(s))\\
&\quad +\gamma|u_m|^{r-1}u_m+\beta u_m,w_m)ds\\
&=\int_{\tau}^{t}(P_mf(s,u_m(s-\rho(s))
+P_mg(s,x),w_m)ds+(u^{\tau}_m(x),w_m)\\
&\quad +\alpha^2((u_m^{\tau}(x),w_m)),\\
&u_m(\tau,x)=u^{\tau}_m(x), \ \ x \in \Omega,\\
&u_m(s+\tau,x)=P_m\phi(s,x),\ \ s\in(-h,0),\ \ x\in \Omega.\\
\end{aligned}
\right.
\end{equation} By the theory of functional differential equations in [Reference Hale and Verduyn-Lunel20], the system (3.4) has a local solution on interval
$[0,t_m]$.
Step 2: Uniform estimates of
$\{u_m\}$ and
$\{\frac{du_m}{dt}\}$
Multiplying (3.4) by
$g_{i}(t)$ in differential form and then summing from i = 1 to m, using Hölder’s and Young’s inequalities, we can obtain the estimate as follows:
\begin{align}
&\quad\frac{1}{2}\frac{d}{dt}\big(|u_m(t)|^2+\alpha^2||u_m(t)||^2\big)+\frac{\nu}{2}||u_m(t)||^2+\gamma||u_m(t)||^{r+1}_{\mathbb{L}^{r+1}}\nonumber\\
&\leq |\beta||u_m(t)|^2+\frac{1}{2\nu}||g(t)||^2_{V'}+\big(a^{\frac{1}{2}}(t)|u_m(t-\rho(t))|+b^{\frac{1}{2}}(t) \big)|u_m(t)|\\
&\leq |\beta||u_m(t)|^2+\frac{1}{2\nu}||g(t)||^2_{V'}+\frac{1}{2}b(t)
+\frac{1}{2}|u_m(t)|^2 +\frac{1-\rho^*}{2}|u_m(t-\rho(t))|^2 \nonumber\\
&\quad +\frac{a(t)}{2(1-\rho^*)}|u_m(t)|^2.\nonumber
\end{align} By
$(H_0)$, we get
\begin{align}
(1-\rho^*)\int_{\tau}^{t}|u_m(s-\rho(s))|^2ds
\leq & \int_{\tau}^{t}|u_m(s)|^2ds+||\phi||^2_{L^2_H}.
\end{align}Now, it follows from (3.5) and (3.6) that
\begin{align*}
&\quad|u_m(t)|^2+\alpha^2||u_m(t)||^2+\nu\int_{\tau}^{t}||u_m(s)||^2ds
+2\gamma\int_{\tau}^{t}||u_m(s)||^{r+1}_{{\mathbb{L}^{r+1}}}ds\\
&\leq \int_{\tau}^{t}(2|\beta|+2+\frac{a(s)}{1-\rho^*})|u_m(s)|^2ds
+\frac{1}{\nu}||g||^2_{L^2(\tau,T;V')}+||\phi||^2_{L_H^2}\\
&\quad +||b||_{L^1(\tau,T)}+|u_m^{\tau}|^2+\alpha^2||u_m^{\tau}||^2. \nonumber
\end{align*}By Gronwall’s inequality, it can be concluded that
\begin{align}
|u_m(t)|^2
&\leq \left(||b||_{L^1(\tau,T)}+(\frac{2}{\lambda_1}+\alpha^2)||\phi||^2_{C_V} \right. \nonumber \\
&\quad \left.+\frac{1}{\nu}||g||^2_{L^2(\tau,T;V')}\right)e^{\int_{\tau}^{T}(2|\beta|+2+\frac{a(s)}{1-\rho^*})ds}:=M,
\end{align}from which, we have
\begin{align}
&\quad\alpha^2||u_m(t)||^2+\nu\int_{\tau}^{t}||u_m(s)||^2ds+2\gamma\int_{\tau}^{t}||u_m(s)||^{r+1}_{{\mathbb{L}^{r+1}}}ds\nonumber\\
&\leq M\int_{\tau}^{T}(2|\beta|+2+\frac{a(s)}{1-\rho^*})ds+\frac{1}{\nu}||g||^2_{L^2(\tau,T;V')}\nonumber\\
&\quad+||b||_{L^1(\tau,T)}+(\frac{2}{\lambda_1}+\alpha^2)||\phi||^2_{C_V}+M:=N.
\end{align} From (3.7) and (3.8), it follows that
$t_m=T$ and
Next our aim is to derive the uniform boundedness of
$\{\frac{du_m}{dt}\}$. Let
$v_m=u_m+\alpha^2Au_m$. It follows from (3.4) that
\begin{align}
\frac{dv_m}{dt}&=-\nu Au_m-B(u_m(t-\rho(t)),u_m)-\beta u_m\nonumber\\
&\quad-\gamma |u_m|^{r-1}u_m+g(x,t)+P_mf(t,u_m(t-\rho(t))).
\end{align} By
$u_m\in L^{\infty}(\tau,T;V)\cap L^{r+1}(\tau,T;{\mathbb{L}^{r+1}})$, we obtain
\begin{align}
||B(u_m(t-\rho(t)),u_m)||_{L^2(\tau,T;V')}
&=\int_{\tau}^{T}\sup\limits_{||w_j||=1}| \lt (u_m(s-\rho(s))\cdot\nabla)u_m(s), \nonumber\\
&\quad w_j \gt |^2ds\nonumber\\
&\leq\int_{\tau}^{T}||u_m(s)||\ ||u_m(s-\rho(s))||ds\leq C.
\end{align} Similarly, it can be concluded that
$-\nu A u_m\in L^2(\tau,T;V')$,
$|u_m|^{r-1}u_m\in L^{\frac{r+1}{r}}(\tau,T;\mathbb{L}^{\frac{r+1}{r}})$.
In addition, it follows from (H 2) that
\begin{align}
&\quad\int_{\tau}^{T}|f(t,u_m(t-\rho(t)))|^2dt\nonumber\\
&\leq ||a||_{L^{\infty}(\tau,T)}\int_{\tau}^{T}|u_m(t-\rho(t))|^2dt+||b||_{L^1(\tau,T)}\nonumber\\
&\leq \frac{||a||_{L^{\infty}(\tau,T)}}{1-\rho^{\ast}}||\phi||^2_{L_H^{2}}
+\frac{||a||_{L^{\infty}(\tau,T)}}{1-\rho^{\ast}}
||u_m||^2_{L^{2}(\tau,T;H)}
+||b||_{L^1(\tau,T)},
\end{align} which implies that
$f(t,u_m(t-\rho(t)))$ belongs to
$L^2(\tau,T;H)$ for any
$T\geq\tau$. Combining (3.11)-(3.12), we can obtain
\begin{equation*}\frac{dv_m}{dt}\in L^{\frac{r+1}{r}}(\tau,T;\mathbb{L}^{\frac{r+1}{r}})+L^{2}(\tau,T;V').\end{equation*} Noting
$2 \lt r\leq5$, we can obtain that
$V\hookrightarrow \mathbb{L}^{r+1}$. Then using (3.1)–(3.2), one has
\begin{align}
\frac{du_m}{dt}\in L^{\frac{r+1}{r}}(\tau,T;V).
\end{align} Due to the presence of nonlinear terms in (1.1), we need to obtain a strongly convergent sequence of
$\{u_m\}$ in a specific Sobolev space.
Step 3: Compactness argument
Since Ω is unbounded and there is no assumption of regularity on the boundary
$\partial\Omega$, thus
$V \hookrightarrow\hookrightarrow H$ is not valid. Next, we will use Lemma 2.1 to prove the result of local compactness. From uniform estimates of (3.7) and (3.8), it is sufficient to show that condition (2.2) holds, that is,
\begin{align*}
\lim\limits_{l\rightarrow0}\sup\limits_{m\in\mathbb{N}}
||u_m(t+l)-u_m(t)||_{L^2(\tau,T-l;(L^2(\mathcal{O}))^3)}=0,
~~\forall~\tau\leq t\leq t+l\leq T,
\end{align*} for any bounded subset
$\mathcal{O}$. It follows from (3.4) that
\begin{align}
&\int_{\Omega}(u_m(t+l)-u_m(t))w_idx+\alpha^2\int_{\Omega}(\nabla u_m(t+l)-\nabla u_m(t))\cdot \nabla w_idx \nonumber\\
&\quad +\nu\int_{t}^{t+l} \lt \nabla u_m(s), \nabla w_i \gt ds\nonumber\\ &\quad+\int_{t}^{t+l}b(u_m(s-\rho(s)),u_m(s),w_i)ds+\beta\int_{t}^{t+l}\int_{\Omega}u_m(s)w_idxds \nonumber\\
&\quad +\gamma\int_{t}^{t+l}\int_{\Omega}|u_m|^{r-1}u_m\cdot w_idxds\nonumber\\
&=\int_{t}^{t+l} \lt g(s),w_i \gt ds+\int_{t}^{t+l} \lt f(s,u_m(s-\rho(s))),w_i \gt ds.
\end{align} Multiplying (3.14) by
$g_{i}(t+l)-g_{i}(t)$ and summing about i from 1 to m, we obtain
\begin{align*}
&\int_{\Omega}|u_m(t+l)-u_m(t)|^2dx+\int_{t}^{t+l}b(u_m(s-\rho(s)),u_m(s),u_m(t+l)-u_m(t))ds\nonumber\\
&\quad+\alpha^2\int_{\Omega}|\nabla u_m(t+l)-\nabla u_m(t)|^2dx\\
&\quad +\nu\int_{t}^{t+l}\int_{\Omega}\nabla u_m(s)\cdot \nabla(u_m(t+l)-u_m(t))dxds\nonumber\\
&\quad+\gamma\int_{t}^{t+l}\int_{\Omega}|u_m(s)|^{r-1}u_m(s)(u_m(t+l)-u_m(t))dxds\\
&\quad +\beta\int_{t}^{t+l}\int_{\Omega}u_m(s)(u_m(t+l)-u_m(t))dxds\nonumber\\
&=\int_{t}^{t+l}(f(s,u_m(s-\rho(s))),u_m(t+l)-u_m(t))ds\\
&\quad +\int_{t}^{t+l}(g(s),u_m(t+l)-u_m(t))ds,
\end{align*} which, along with Hölder’s inequality and
$(H_2)$, shows that
\begin{align}
&\quad|u_m(t+l)-u_m(t)|^2+\alpha^2|\nabla u_m(t+l)-\nabla u_m(t)|^2\nonumber\\
&\leq
||u_m(t+l)-u_m(t)||\int_{t}^{t+l}G_m(s)ds +\gamma||u_m(t+l)-u_m(t)||_{{\mathbb{L}^{r+1}}}\nonumber\\
&\quad \int_{t}^{t+l}||u_m(s)||^r_{{\mathbb{L}^{r+1}}}ds,
\end{align}where
\begin{align*}
G_m(s)
&=||u_m(s-\rho(s))|| \ ||u_m(s)||+|\beta|\lambda_1^{-\frac{1}{2}}|u_m(s)|\nonumber\\
&\quad+\lambda_1^{-\frac{1}{2}}(a^{\frac{1}{2}}(s)|u_m(s-\rho(s))|+b^{\frac{1}{2}}(s))
+\nu|\nabla u_m(s)|+||g(s)||_{V'}.
\end{align*} Integrating the inequality (3.15) with respect to t over
$(\tau,T-l)$, we have
\begin{align}
&\qquad
\int_{\tau}^{T-l} |u_m(t+l)-u_m(t)|^2 dt
+ \alpha^2 \int_{\tau}^{T-l} |\nabla u_m(t+l)-\nabla u_m(t)|^2 dt \nonumber\\
&\leq
\int_{\tau}^{T-l}||u_m(t+l)-u_m(t)||\int_{t}^{t+l}G_m(s)dsdt\nonumber\\
&\quad+\gamma\int_{\tau}^{T-l}||u_m(t+l)-u_m(t)||_{{\mathbb{L}^{r+1}}}\int_{t}^{t+l}||u_m(s)||^r_{{\mathbb{L}^{r+1}}}dsdt.
\end{align} Define
$\overline{s}$ as follows:
\begin{equation*}
\overline{s}=\left\{
\begin{aligned}
&\tau \ \ \ \mbox{if}\ \ s\leq\tau,\\
& s ,\ \ \ \mbox{if} \ \ \tau \lt s\leq T-l ,\\
&T-l, \ \ \mbox{if}\ \ s \gt T-l.\\
\end{aligned}
\right.
\end{equation*}For the right-hand side of (3.16), by the Fubini theorem, we obtain
\begin{align*}
&\quad\int_{\tau}^{T-l}||u_m(t+l)-u_m(t)||\int_{t}^{t+l}G_m(s)dsdt\nonumber\\
&\qquad+\gamma\int_{\tau}^{T-l}||u_m(t+l)-u_m(t)||_{{\mathbb{L}^{r+1}}}\int_{t}^{t+l}||u_m(s)||^r_{{\mathbb{L}^{r+1}}}dsdt\nonumber\\
&\leq
\int_{\tau}^{T}G_m(s)\int_{\overline{s-l}}^{\overline{s}}
||u_m(t+l)-u_m(t)||dtds\nonumber\\
&\quad+\gamma\int_{\tau}^{T}||u_m(s)||^r_{{\mathbb{L}^{r+1}}}
\int_{\overline{s-l}}^{\overline{s}}||u_m(t+l)-u_m(t)||_{{\mathbb{L}^{r+1}}}dtds
=: I_1+I_2,
\end{align*} where
$0\leq \overline{s}-\overline{s-l}\leq l$.
By
$(H_0)$, using Hölder’s and Young’s inequalities, the estimate of I 1 can be obtained as follows:
\begin{align*}
I_1&\leq
l^{\frac{1}{2}} \big(\int_{\overline{s-l}}^{\overline{s}}||u_m(t+l)-u_m(t)||^2dt\big)^{\frac{1}{2}}\int_{\tau}^{T}G_m(s)ds\nonumber\\
&\leq
l^{\frac{1}{2}}||u_m(t+l)-u_m(t)||_{L^2(\overline{s-l},\overline{s};V)}\bigg[\int_{\tau}^{T}\bigg((1-\rho^*)||u(s-\rho(s))||^2\nonumber\\
&\quad+\frac{1}{2(1-\rho^*)}||u_m(s)||^2\bigg)ds+\nu\int_{\tau}^{T}|\nabla u_m|ds+|\beta|\lambda_1^{-\frac{1}{2}}\int_{\tau}^{T}|u_m|ds\nonumber\\
&\quad+\int_{\tau}^{T}\frac{a(s)}{2(1-\rho^*)\lambda_1^2}ds+\frac{(T-\tau)^{\frac{1}{2}}}{\lambda_1}||b||^{\frac{1}{2}}_{L^1(\tau,T)}+\int_{\tau}^{T}||g(s)||_{V'}ds\bigg].
\end{align*}It then follows from (3.7) and (3.8) that
\begin{align}
I_1&\leq \frac{2l^{\frac{1}{2}}N^{\frac{1}{2}}}{\nu^{\frac{1}{2}}}\bigg[\int_{\tau}^{T}(\frac{1}{2(1-\rho^*)}+1)||u_m||^2ds+||\phi||^2_{L^2_V} \nonumber\\
&\quad +|\beta|\lambda_1^{-\frac{1}{2}}\int_{\tau}^{T}|u_m|ds+\nu\int_{\tau}^{T}|\nabla u_m|ds\nonumber\\ &\quad+\frac{||a||_{L^{\infty}(\tau,T)}}{2(1-\rho^*)\lambda_1^2}(T-\tau)+\frac{(T-\tau)^{\frac{1}{2}}}{\lambda_1}||b||^{\frac{1}{2}}_{L^1(\tau,T)}+(T-\tau)^{\frac{1}{2}}||g||_{L^2(\tau,T;V')}\bigg]\nonumber\\
&\leq \frac{2l^{\frac{1}{2}}N^{\frac{1}{2}}}{\nu^{\frac{1}{2}}}\bigg[(\frac{1}{2\nu(1-\rho^*)}+\frac{1}{\nu})N+\nu^{\frac{1}{2}}N^{\frac{1}{2}}(T-\tau)^{\frac{1}{2}}+||\phi||^2_{L^2_V}+|\beta|(T-\tau)\lambda_1^{-\frac{1}{2}}M^{{\frac{1}{2}}}\nonumber\\ &\quad+\frac{||a||_{L^{\infty}(\tau,T)}}{2(1-\rho^*)\lambda_1^2}(T-\tau)+\frac{(T-\tau)^{\frac{1}{2}}}{\lambda_1}||b||^{\frac{1}{2}}_{L^1(\tau,T)}+(T-\tau)^{\frac{1}{2}}||g||_{L^2(\tau,T;V')}\bigg].
\end{align}Similarly, it can be concluded that
\begin{align}
I_2 &\leq l^{\frac{1}{r+1}}(T-\tau)^{\frac{r+1}{r}}N.
\end{align} Letting
$l\rightarrow0$ in (3.16), by (3.17) and (3.18), we obtain
\begin{align}
&\lim\limits_{l\rightarrow0}\sup\limits_{m\in\mathbb{N}}
\left(||u_m(t+l)-u_m(t)\right.\nonumber\\
&\quad\left.||_{L^2(\tau,T-l;H)}+\alpha^2||\nabla u_m(t+l)-\nabla u_m(t)||_{L^2(\tau,T-l;H}\right)=0.
\end{align} Consider a cut-off function
$\psi\in C^1(\mathbb{R}^+)$ and
$0\leq\psi(s)\leq 1$ such that
\begin{equation*}
\psi(s)=\left\{
\begin{aligned}
&1, \ \ \ \mbox{if}\ \ 0\leq s\leq 1,\\
&0, \ \ \ \mbox{if} \ \ s\geq 2 .\\
\end{aligned}
\right.
\end{equation*} For each R
$ \gt $ 0, define
$u_{m,R}=\psi(\frac{|x|^2}{R^2})u_m(x)$, then from (3.9) and (3.19), we obtain
\begin{equation*}\{u_{m,R}\}\ \mbox{is bounded in}\ L^{\infty} (\tau,T;(L^2(\mathcal{O}_R))^3)\cap L^2 (\tau,T;(H_0^1(\mathcal{O}_R))^3)\end{equation*}and
\begin{equation*}
\lim\limits_{l\rightarrow0}\sup\limits_{m\in\mathbb{N}}
||u_{m,R}(t+l)-u_{m,R}(t)||_{L^2(\tau,T-l;(L^2(\mathcal{O}_R))^3)}=0,
\end{equation*} where
$\mathcal{O}_R=B(0,R)\cap\Omega$. In this case, the compact injection
$(H_0^1(\mathcal{O}_R))^3\subset (L^2(\mathcal{O}_R))^3$ is true. Then by Lemma 2.1, we have
\begin{equation*}
\{u_{m,R}\} \ \mbox{is relatively compact in} \ L^2(\tau,T;(L^2(\mathcal{O}_R))^3).
\end{equation*} By applying the standard diagonal procedure, one can find that there exists a subsequence
$\{u_m\}$ (still denoted
$\{u_m\}$ as the whole sequence), such that
Step 4: Taking the limit
Combining the previous uniform estimates, there exist a subsequence
$\{u_m\}$ (still defined as
$\{u_m\}$) and
$u\in L^{\infty}(\tau,T;V)$ with
$\frac{du}{dt}\in L^{\frac{r+1}{r}}(\tau,T;V)$ such that
\begin{equation}
\left\{
\begin{aligned}
&u_m\rightharpoonup u \ \mbox{weakly} \ast \ \mbox{in} \ L^{\infty}(\tau,T;V),\\
&u_m\rightharpoonup u \ \mbox{weakly} \ \mbox{in} \ L^{r+1}(\tau,T;{\mathbb{L}^{r+1}}),\\
&u_m\rightarrow u \ \mbox{strongly in} \ L^2(\tau,T;(L_{loc}^2(\Omega))^3),\\
&\frac{dv_m}{dt}=\mathcal{L}\frac{du_m}{dt}\rightharpoonup \mathcal{L}\frac{du}{dt} \ \mbox{weakly} \ \mbox{in} \ L^{\frac{r+1}{r}}(\tau,T;V'),\\
&\frac{du_m}{dt}\rightharpoonup \frac{du}{dt} \ \mbox{weakly} \ \mbox{in} \ L^{\frac{r+1}{r}}(\tau,T;V),\\
&|u_m|^{r+1}u_m\rightharpoonup \chi \ \mbox{weakly}\ \mbox{in} \ L^{\frac{r+1}{r}}(\tau,T;\mathbb{L}^{\frac{r+1}{r}}).\\
\end{aligned}
\right.
\end{equation} It then follows from Lions lemma that
$u\in C([\tau,T];V)$ (revise a zero measurement set), which implies
\begin{equation*}\frac{1}{2}\frac{d}{dt}(|u(t)|^2+\alpha^2||u(t)||^2)
=(\frac{d}{dt}u(t),u(t))+\alpha^2((\frac{d}{dt}u(t),u(t)))\end{equation*} and makes the initial value u τ meaningful. From [Reference Lions22] and (3.20), there is a subsequence of
$\{u_m\}$ such that the limits in (3.21) satisfy
By
$P_m\phi\rightarrow\phi$ in CV and
$u_m\rightarrow u$ in
$L^2(\tau,T;(L^2(\mathcal{O}))^3)$, we can obtain
Then combining with the continuity of f, we have
which, combining with
$(H_2)$ and Mazur’s theorem, shows that
Taking a diagonal subsequence, denoted again um, that satisfies (3.20), (3.22), and (3.23) for a sequence of regular bounded open sets
$\mathcal{O}_j\in\Omega$ that contain all supports of functions wj of the basis (
$\{w_1,w_2,\cdots\}$ is density in V). Then by (3.22) and (3.23), we can deduce that
\begin{align}
& \int_{\tau}^{T}\alpha \langle|u_m(s)|^{r-1}u_m(s)-|u(s)|^{r-1}u(s),w_j\rangle ds
\nonumber\\
&\quad\leq \int_{\tau}^{T}|||u|^{r-1}+|u_m|^{r-1}||_{\mathbb{L}^{\frac{r+1}{r-1}}}||u_m-u||_{\mathbb{L}^{\frac{6(r+1)}{11-r}}}||w_j||_{\mathbb{L}^6}ds\nonumber\\
&\quad\leq C\big[||u_m||_{L^{r+1}(\tau,T;\mathbb{L}^{r+1})}+||u||_{L^{r+1}(\tau,T;\mathbb{L}^{r+1})}\big]^{r-1}\nonumber\\
&\qquad\cdot||u_m-u||^{1-\theta}_{L^{r+1}(\tau,T;\mathbb{L}^{r+1})}||u_m-u||^{\theta}_{L^{\infty}(\tau,T;(L^2(\mathcal{O}))^3)}\nonumber\\
&\qquad\rightarrow 0, \ \mbox{as} \ \ m\rightarrow\infty,
\end{align} where
$\frac{r-1}{r+1}+\frac{11-r}{6(r+1)}+\frac{1}{6}=1$ and
\begin{align}
\int_{\tau}^{T} \langle f(s,u_m(s-\rho(s)))
- f(s,u(s-\rho(s))) ,w_j \rangle ds\rightarrow 0, \ \mbox{as} \ \ m\rightarrow\infty.
\end{align}By (3.21), we can find
\begin{align} &\qquad\int_{\tau}^{T}|b(u_m(t-\rho(t)),u_m(t),w_j)-b(u(t-\rho(t)),u(t),w_j)|dt\nonumber\\
&\leq
\int_{\tau}^{T}|b(u_m(t-\rho(t)),u_m-u,w_j)|dt+\int_{\tau}^{T}|b(u_m(t-\rho(t))-u(t-\rho(t)),u,w_j)|dt \nonumber\\
&\leq C_T
||u_m-u||_{L^2(\tau,T;(L^2(\mathcal{O}))^3)}
+C_T
||u_m-u||_{L^2(\tau-h,T;(L^2(\mathcal{O}))^3)}\nonumber\\
& \rightarrow 0, \ \mbox{as} \ m\rightarrow\infty.
\end{align}Therefore, by (3.21), (3.23), and (3.24)-(3.26), one can obtain the limit on equation (3.4) that satisfies (3.3).
$\bullet$ Part II. Uniqueness of solution. Let
$u_1(t),u_2(t)$ be two weak solutions of (1.1) with initial data
$\phi_1,\phi_2$. Define
$w(t)=u_1(t)-u_2(t)$ and w(t) satisfies
\begin{align*}
\frac{1}{2}\frac{d}{dt}&(|w(t)|^2+\alpha^2||w(t)||^2)+\nu|\nabla w(t)|^2+b(w(t-\rho(t)),u_2,w)\nonumber\\
&\qquad+\beta|w(t)|^2+\gamma(|u_1|^{r-1}u_1-|u_2|^{r-1}u_2,w)\nonumber\\
&=(f(t,u_1(t-\rho(t)))-f(t,u_2(t-\rho(t))),w)\nonumber\\
&\leq L(R)\Lambda^{\frac{1}{2}}(t)|w(t-\rho(t))||w(t)|,
\end{align*} with initial data
$w^{\tau}=\phi_1-\phi_2$. It follows from [Reference Yang and Yang45] that
\begin{align}
\int_{\Omega}(|u_1|^{r-1}u_1-|u_2|^{r-1}u_2)\cdot(u_1-u_2)dx\geq 0.
\end{align} By (3.27), Hölder’s and Young’s inequalities imply that for any
$t\in[\tau,T]$,
\begin{align}
&\frac{d}{dt}(|w(t)|^2+\alpha^2||w(t)||^2)+2\nu||w(t)||^2\nonumber\\
&\quad\leq (2|\beta|+\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1(1-\rho^*)})|w(t)|^2+2||w(t-\rho(t))|| ||u_2|| ||w||\nonumber\\
&\qquad+(1-\rho^*)||w(t-\rho(t))||^2.
\end{align} Integrating (3.28) over
$[\tau,t]$ with
$t\in[\tau,T]$, by (3.8), we obtain
\begin{align}
&\quad|w(t)|^2+\alpha^2||w(t)||^2+2\nu\int_{\tau}^{t}||w(s)||^2ds\nonumber\\
&\leq \int_{\tau}^{t}(2|\beta|+\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1(1-\rho^*)})|w(s)|^2ds+\int_{\tau}^{t}(2+\frac{N}{(1-\rho^*)\alpha^2})||w(s)||^2ds\nonumber\\
&\quad+2||\phi_1-\phi_2||^2_{L^2_V}+|w^{\tau}|^2+\alpha^2||w^{\tau}||^2.
\end{align}Using Gronwall’s inequality, we get
\begin{align}
\quad|w(t)|^2+\alpha^2||w(t)||^2
\leq \bigg[ 2+\frac{1}{\lambda_1}+\alpha^2\bigg]||\phi_1-\phi_2||^2_{C_V}\cdot e^{\widetilde{H}(t-\tau)},
\end{align} where
$\widetilde{H}=\max\{(2+\frac{N}{(1-\rho^*)\alpha^2})\frac{1}{\alpha^2}, 2|\beta|+\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1(1-\rho^*)} \}$. Since
$\phi_1=\phi_2$, the uniqueness of solutions of (1.1) can be obtained from (3.30), as desired.
4. Existence of pullback attractors for the system (1.1)
In this section, we will show the existence of pullback attractor of (1.1) in CV. To this end, we further assume that
(H 4)
$a\in L^{\infty}(\mathbb{R})$ satisfies
\begin{equation*}
||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}
+2|\beta|+\frac{||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}}{1-\rho^*} \lt \nu\lambda_1,\end{equation*} which implies that there exists σ
$ \gt $ 0 such that
\begin{equation*}\sigma(1+\alpha^2\lambda_1)+||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}e^{\sigma h}+2|\beta|+\frac{||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}}{1-\rho^*}\leq\nu\lambda_1.\end{equation*}(H 5) Suppose that g and b satisfy
\begin{equation*}\int_{-\infty}^{t}e^{\sigma s}||g(s)||^2_{V'}ds \lt +\infty,\ \ \int_{-\infty}^{t}e^{\sigma s}b(s)ds \lt +\infty, \
\mbox{for any} \ t\in\mathbb{R}.\end{equation*} Based on the well-posedness of (1.1), we are able to define a continuous nonautonomous dynamical system. Define a mapping
$U(t,\tau): C_V \rightarrow C_V$ by
where u is the solution of (1.1) with initial value
$\phi\in C_V$.
Let
$\mathcal{P}(C_V)$ be the class of all families of nonempty subsets in CV.
$\mathcal{D}_{\sigma}$ is the class of all families
$ D =\{D(t): t\in\mathbb{R}\}\subset \mathcal{P}(C_V)$
such that
\begin{equation*}\lim\limits_{t\rightarrow -\infty} e^{\sigma t}
\| D(t) \|^2_{C_V} =0,\end{equation*} where
$\| D(t) \|_{C_V} = \sup\limits_{\phi \in D(t)} \| \phi \|_{C_V}$. Such a D is said to be tempered in CV. Next, we establish the existence of
$\mathcal{D}_{\sigma}$-pullback absorbing set for the system U. For simplicity, define the Hilbert space
$\mathcal{M}:=H\times V$ with the norm
\begin{equation*}||u||^2_{\mathcal{M}}=|u|^2+\alpha^2||u||^2.\end{equation*} Obviously,
$||u||^2_{\mathcal{M}}$ is equivalent to
$||u||^2$ by Poincaré’s inequality.
Lemma 4.1. Suppose that
$ g\in L_{loc}^2(\mathbb{R};V'), \ \phi\in C_V$, and
$(H_0)$-
$(H_5)$ hold, and assume
$2 \lt r\leq 5,\ \alpha,\gamma \gt 0,~\beta\in\mathbb{R}$. Then the solution u of (1.1) satisfies
\begin{align*}
&2\gamma \int_{t-h}^{t}||u(s)||^{r+1}_{{\mathbb{L}^{r+1}}}ds\vee ||u_t||^2_{C_{\mathcal{M}}}\nonumber\\
&\quad\leq e^{\sigma h}e^{-\sigma(t-\tau)}\big(e^{\sigma h}||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}||\phi||^2_{L_H^2}+||\phi||^2_{C_{\mathcal{M}}}\big)\nonumber\\
&\qquad +\frac{4e^{-\sigma (t-h)}}{\nu}\int_{\tau}^{t}e^{\sigma s}||g(s)||_{V'}^2ds
+\frac{4e^{-\sigma (t-h)}}{\nu\lambda_1}\int_{\tau}^{t}e^{\sigma s}b(s)ds,
\end{align*} where
$||u_t||^2_{C_{\mathcal{M}}}=||u_t||^2_{C_H}+\alpha^2||u_t||^2_{C_V}$.
Proof. Taking the inner product of (1.1) with
$e^{\sigma t}u(t)$, we deduce
\begin{align}
&\frac{1}{2}\frac{d}{dt}\big[e^{\sigma t}(|u(t)|^2+\alpha^2||u(t)||^2)\big]+\frac{\nu}{2}e^{\sigma t}||u(t)||^2+\beta e^{\sigma t}|u(t)|^2+\gamma e^{\sigma t}||u(t)||^{r+1}_{{\mathbb{L}^{r+1}}}\nonumber\\
&\quad\leq e^{\sigma t} ||g(t)||_{V'}||u(t)||+\frac{1}{2}\sigma e^{\sigma t}(|u|^2+\alpha^2||u||^2)\nonumber\\
&\qquad+e^{\sigma t}|f(t,u(t-\rho(t)))||u(t)|-\frac{\nu}{2}e^{\sigma t}||u(t)||^2.
\end{align}It follows from (4.2), Hölder’s and Young’s inequalities that
\begin{align*}
\frac{1}{2}\frac{d}{dt}&\big[e^{\sigma t}(|u(t)|^2+\alpha^2||u(t)||^2)\big]+\frac{\nu}{4}e^{\sigma t}||u(t)||^2+\gamma e^{\sigma t}||u(t)||^{r+1}_{{\mathbb{L}^{r+1}}}\nonumber\\
&\leq
\frac{2e^{\sigma t}}{\nu}||g(t)||^2_{V'}
+\frac{1}{2}\sigma e^{\sigma t}(|u(t)|^2+\alpha^2||u(t)||^2)
+\frac{(1-\rho^*)a^{\frac{1}{2}}(t)}{2}e^{\sigma t}|u(t-\rho(t))|^2\nonumber\\
&\quad+e^{\sigma t}\frac{a^{\frac{1}{2}}(t)|u(t)|^2}{2(1-\rho^*)}
+|\beta|e^{\sigma t}|u(t)|^2+\frac{2}{\nu\lambda_1}e^{\sigma t}b(t)-\frac{\nu}{2}e^{\sigma t}||u(t)||^2.
\end{align*}Then, we can obtain
\begin{align*}
& e^{\sigma t}(|u(t)|^2+\alpha^2||u(t)||^2)+\frac{\nu}{2}\int_{\tau}^{t}e^{\sigma s}||u(s)||^2ds+2\gamma \int_{\tau}^{t}e^{\sigma s}||u(s)||^{r+1}_{{\mathbb{L}^{r+1}}}ds\nonumber\\
&\quad\leq \frac{4}{\nu}\int_{\tau}^{t}e^{\sigma s}||g(s)||^2_{V'}ds+\biggr[\sigma(\frac{1}{\lambda_1}+\alpha^2)+\frac{||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}e^{\sigma h}+2|\beta|}{\lambda_1}+\frac{||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}}{(1-\rho^*)\lambda_1}-\nu \biggr]\nonumber\\
&\qquad\times\,\int_{\tau}^{t}e^{\sigma s}||u(s)||^2ds+\frac{4}{\nu\lambda_1}\int_{\tau}^{t}e^{\sigma s}b(s)ds\nonumber\\
&\qquad+e^{\sigma (h+\tau)}||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}||\phi||^2_{L_H^2}+e^{\sigma \tau}(|u^{\tau}|^2+\alpha^2||u^{\tau}||^2),
\end{align*} which, along with
$(H_4)$, shows
\begin{align*}
&||u(t)||^2_{\mathcal{M}}+\frac{\nu}{2}e^{-\sigma t}\int_{\tau}^{t}e^{\sigma s}||u(s)||^2ds+2\gamma e^{-\sigma t}\int_{\tau}^{t}e^{\sigma s}||u(s)||^{r+1}_{{\mathbb{L}^{r+1}}}ds\nonumber\\
&\quad\leq\frac{4}{\nu}e^{-\sigma t}
\int_{\tau}^{t}e^{\sigma s}||g(s)||^2_{V'}ds+\frac{4}{\nu\lambda_1}
e^{-\sigma t}\int_{\tau}^{t}e^{\sigma s}b(s)ds+e^{-\sigma (t-\tau)}\nonumber\\
&\qquad\times\bigg(e^{\sigma h}||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}||\phi||^2_{L_H^2}+||\phi||_{C_{\mathcal{M}}}^2\bigg).
\end{align*} Thus, we have for any
$s'\in [-h,0]$
\begin{align*}
||u(t+s')||_{\mathcal{M}}^2\leq ||u_t||^2_{C_{\mathcal{M}}}&\leq e^{\sigma h}e^{-\sigma(t-\tau)}\bigg(e^{\sigma h}||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}||\phi||^2_{L_H^2}+||\phi||^2_{C_{\mathcal{M}}}\bigg)\nonumber\\
&\quad+\frac{4}{\nu} e^{\sigma h}e^{-\sigma t}\int_{\tau}^{t}e^{\sigma s}||g(s)||_{V'}^2ds+\frac{4}{\nu\lambda_1}e^{\sigma h}e^{-\sigma t}\nonumber\\
&\qquad\times\int_{\tau}^{t}e^{\sigma s}b(s)ds ,
\end{align*} and for all
$\tau\leq t-h$,
\begin{align*}
2\gamma \int_{t-h}^{t}||u(s)||^{r+1}_{{\mathbb{L}^{r+1}}}ds
&\leq 2\gamma e^{\sigma h}\int_{\tau}^{t}e^{\sigma (s-t)}||u(s)||^{r+1}_{{\mathbb{L}^{r+1}}}ds\nonumber\\
&\leq e^{\sigma h}e^{-\sigma(t-\tau)}\bigg(e^{\sigma h}||a||^{\frac{1}{2}}_{L^{\infty}(\mathbb{R})}||\phi||^2_{L_H^2}+||\phi||^2_{C_{\mathcal{M}}}\bigg)\nonumber\\
&\quad+\frac{4}{\nu} e^{\sigma h}e^{-\sigma t}\int_{\tau}^{t}e^{\sigma s}||g(s)||_{V'}^2ds+\frac{4}{\nu\lambda_1}e^{\sigma h}e^{-\sigma t}\int_{\tau}^{t}e^{\sigma s}b(s)ds.
\end{align*}The proof is complete.
Lemma 4.2. Suppose that
$(H_0)$–
$(H_5)$ hold. Then the nonautonomous dynamical system U has a closed
$\mathcal{D}_\sigma$-pullback absorbing set
$K = \{K(t) : t\in \mathbb{R} \} \in \mathcal{D}_\sigma$ given as follows:
\begin{align}
K(t) = \left\{\phi \in C_V : \| \phi \|^2_{C_V} \leq \mathcal{R}_{C_V}^2(t) \right\},
\end{align} where
$\mathcal{R}_{C_V}^2(t)
= 1+\frac{4 e^{\sigma h}
e^{-\sigma t}}{\nu\alpha^2}
\int_{-\infty}^{t}e^{\sigma s}||g(s)||_{V'}^2ds
+\frac{4 e^{\sigma h}
e^{-\sigma t}}{\nu\alpha^2\lambda_1}\int_{-\infty}^{t}
e^{\sigma s}b(s)ds$.
Proof. Given
$t \in \mathbb{R}$ and
$D\in \mathcal{D}$, it follows from Lemma 4.1 that there exists
$\mathfrak{T} = \mathfrak{T}(t,D) \lt t$ such that for all
$\tau \lt \mathfrak{T}$,
which shows that K pullback attracts all elements in
$\mathcal{D}$. In addition, by
$(H_5)$, we can obtain that K is tempered, as desired.
Next, we discuss the asymptotic compactness of the nonautonomous dynamical system U associated with problem (1.1).
Lemma 4.3. Suppose that
$(H_0)$–
$(H_5)$ hold. Then the process U is
$\mathcal{D}_{\sigma}$-pullback asymptotically compact in CV.
Proof. Given a sequence
$\{\tau_n\}$ with
$\tau_n\rightarrow-\infty$, and
$D = \{D(t) : t\in \mathbb{R} \} \in \mathcal{D}$. We will prove that for any
$\phi^n\in D(\tau_n)$, the sequence
is relatively compact in CV. The proof will be split into the following two steps.
Step 1. We will prove that
$\{u(t,\tau_n,\phi^n)\}$ is relatively in V.
For any integer
$k\geq 0$, there is a time
$\tau_{D}(k)\leq t-k$, and it follows from Lemma 4.2 that
Then we can find a subsequence
$\{\tau_{n'}\}\subset \{\tau_{n}\}$,
$\{\phi^{n'}\}\subset\{\phi^{n}\}$ and
$w^k\in V$ such that
\begin{align}
&u(t-k,\tau_{n'},\phi^{n'})
\rightharpoonup w^k \ \mbox{weakly} \ \mbox{in} \ V,\ \mbox{for all} \ \tau_{n'}\leq t-k.
\end{align}From (4.4) with k = 0 and the lower semi-continuity of the norm, it can be concluded that
\begin{align}
||w^0||\leq\liminf\limits_{n'\rightarrow \infty} ||u(t,\tau_{n'},\phi^{n'})||.
\end{align}To prove that the limit is strong topology in V, we only show that
\begin{align*}
||w^0||\geq\limsup\limits_{n'\rightarrow \infty} ||u(t,\tau_{n'},\phi^{n'})||.
\end{align*}From (1.1), we obtain
\begin{align}
&\quad ||u(t)||^2_{\mathcal{M}}+2\int_{\tau}^{t}e^{-\sigma (t-s)}\big(\nu||u(s)||^2+ \gamma||u(s)||^{r+1}_{{\mathbb{L}^{r+1}}}+\beta|u(s)|^2\big)ds\nonumber\\
&=
\sigma\int_{\tau}^{t}e^{-\sigma (t-s)}||u(s)||^2_{\mathcal{M}}ds+e^{-\sigma (t-\tau)}||u^{\tau}||^2_{\mathcal{M}}\nonumber\\
&\quad+2\int_{\tau}^{t}e^{-\sigma (t-s)}\big((f(s,u(s-\rho(s))),u)+ \lt g(s),u \gt \big)ds.
\end{align} Then for any fixed
$k\geq 0$ and
$\tau_{n'}\leq t-k,$ there exists
\begin{align}
||u^{n'}(t)||^2_{\mathcal{M}}&=-2\int_{t-k}^{t}e^{-\sigma (t-s)}\big(\nu||u^{n'}(s)||^2+ \gamma||u^{n'}(s)||^{r+1}_{{\mathbb{L}^{r+1}}}+\beta|u^{n'}(s)|^2\big)ds\nonumber\\
&\quad+\sigma\int_{t-k}^{t}e^{-\sigma (t-s)}||u^{n'}(s)||^2_{\mathcal{M}}ds+e^{-\sigma k}||u^{n'}(t-k)||^2_{\mathcal{M}}\nonumber\\
&\quad+2\int_{t-k}^{t}e^{-\sigma (t-s)}(f(s,u^{n'}(s-\rho(s))),u^{n'})+ \lt g(s),u^{n'} \gt )ds.
\end{align} By the arguments of Theorem 3.1, we have that for any bounded open subset
$\mathcal{O}\in \Omega$,
\begin{align}
&u(\cdot,t-k, \varphi^{n} )\rightharpoonup u(\cdot,t-k,\varphi) \ \mbox{weakly} \ \mbox{in} \ L^{r+1}(t-k,t;{\mathbb{L}^{r+1}}),\nonumber\\
&u(\cdot,t-k, \varphi^{n} ) \rightharpoonup u(\cdot,t-k, \varphi ) \ \mbox{weakly} \ \mbox{in} \ L^2(t-h-k,t;V),\nonumber\\
&u(\cdot,t-k, \varphi^{n} ) \rightarrow u(\cdot,t-k, \varphi ) \ \mbox{strongly} \ \mbox{in} \ L^2(t-h-k,t;(L^2(\mathcal{O}))^3),
\end{align} where
$\{\varphi^{n}\}$ weakly converges to φ in CV.
It follows from (4.7) that
\begin{align*}
||u^{n'}(t)||^2_{\mathcal{M}}&\leq\int_{t-k}^{t}e^{-\sigma (t-s)}\big(-2\nu||u^{n'}(s)||^2-2\gamma||u^{n'}(s)||^{r+1}_{{\mathbb{L}^{r+1}}}\nonumber\\
&\quad-2\beta|u^{n'}(s)|^2+\sigma||u^{n'}(s)||^2_{\mathcal{M}}\big)ds+(\frac{1}{\lambda_1}+\alpha^2)e^{-\sigma k}K(t-k)\nonumber\\
&\quad+2\int_{t-k}^{t}e^{-\sigma (t-s)}(f(s,u^{n'}(s-\rho(s))),u^{n'})+ \lt g(s),u^{n'} \gt )ds.
\end{align*} It follows from
$(H_4)$ that
$2\nu\lambda_1-\big(2|\beta|+(1+\alpha^2\lambda_1)\sigma\big)\geq 0$, and thus, we can define a new norm as follows:
\begin{equation*}[u^{n'}]^2=(2\nu-\sigma\alpha^2)||u^{n'}||^2+(2\beta-\sigma)|u^{n'}|^2,\end{equation*} which is equivalent to the norm in
$\mathcal{M}$. Then
$(\int_{t-k}^{t}e^{-\sigma (t-s)}[u^{n'}]^2ds)^{\frac{1}{2}}$ is equivalent to the norm in
$L^2(t-k,t;\mathcal{M})$. Therefore by (4.8), one has
\begin{align}
&\int_{t-k}^{t}
e^{-\sigma (t-s)}[u(s,t-k,w^k)]^2ds\nonumber\\
&\quad\leq\liminf\limits_{n'\rightarrow\infty}
\int_{t-k}^{t}e^{-\sigma (t-s)}[ u^{n'} (s)]^2 ds,
\end{align}
\begin{align}
&\gamma\int_{t-k}^{t}
e^{-\sigma (t-s)}||u(s,t-k,w^k)||^{r+1}_{{\mathbb{L}^{r+1}}}ds\nonumber\\
&\quad\leq\gamma\liminf\limits_{n'\rightarrow\infty}
\int_{t-k}^{t}e^{-\sigma (t-s)}||u^{n'}(s)||^{r+1}_{{\mathbb{L}^{r+1}}}ds,
\end{align}and
\begin{align}
&\lim\limits_{n\rightarrow\infty}\int_{t-k}^{t}e^{-\sigma (t-s)} \lt g(s),u^{n'}(s) \gt ds\nonumber\\
&\quad=\int_{t-k}^{t}e^{-\sigma (t-s)} \lt g(s), u(s,t-k,w^k) \gt ds.
\end{align} Since
$u^{n'}\rightarrow u$ strongly in
$L^2(t-h-k,t;(L^2(\mathcal{O}))^3)$ for all bounded subset
$\mathcal{O}\subset \Omega$, similar to (3.23), we have
\begin{align}
f(s,u^{n'}(s-\rho(s)))\rightharpoonup f(s,u(s-\rho(s)))\ \mbox{weakly in} \ L^2(t-k,t;(L^2(\mathcal{O}))^3).
\end{align} Then there exists a function
$\widetilde{\chi}$ such that supp
$\widetilde{\chi}\subset B(0,R)\cap \Omega=\mathcal{O}$ and
$c(\cdot)\in C^1([t-k,t])$ satisfies
\begin{align}
&\lim\limits_{n'\rightarrow\infty}\int_{t-k}^{t}e^{-\sigma (t-s)}(f(s,u^{n'}(s-\rho(s))),c(s)\widetilde{\chi})ds\nonumber\\
&\quad=\int_{t-k}^{t}e^{-\sigma (t-s)}(f(s,u(s-\rho(s))),c(s)\widetilde{\chi})ds,
\end{align} for all
$c(\cdot)\widetilde{\chi}\in L^2(t-h-k,t;(L^2(\mathcal{O}))^3)$. Then by the standard diagonal process, we get
\begin{align}
&\lim\limits_{n'\rightarrow\infty}\int_{t-k}^{t}e^{-\sigma (t-s)}(f(s,u(s-\rho(s),t-k,u(t-k,\tau_{n'},\phi^{n'}))),u^{n'})ds\nonumber\\
&\qquad-\int_{t-k}^{t}e^{-\sigma (t-s)}(f(s,u(s-\rho(s),t-k,w^k)),u(s,t-k,w^k))ds\nonumber\\
&=\lim\limits_{n'\rightarrow\infty}\int_{t-k}^{t}e^{-\sigma (t-s)}(f(s,u(s-\rho(s),t-k,u(t-k,\tau_{n'},\phi^{n'})))\nonumber\\
&\qquad-f(s,u(s-\rho(s),t-k,w^k)),u^{n'})ds\nonumber\\
&\qquad+\lim\limits_{n'\rightarrow\infty}\int_{t-k}^{t}e^{-\sigma (t-s)}(f(s,u(s-\rho(s),t-k,w^k)),u^{n'}-u(s,t-k,w^k))= 0.
\end{align}It then follows from (4.9)–(4.11) and (4.14) that
\begin{align}
\alpha^2\limsup\limits_{n'\rightarrow\infty}||u^{n'}(t)||^2
&\leq \int_{t-k}^{t}e^{-\sigma (t-s)}\big((-2\nu+\sigma\alpha^2) ||u(s,t-k,w^k)||^2\nonumber\\
&\quad +(-2\beta+\sigma) |u(s,t-k,w^k)|^2 \big)ds+(\frac{1}{\lambda_1}+\alpha^2)e^{-\sigma k} K(t-k)\nonumber\\
&\quad -2\gamma\int_{t-k}^{t}e^{-\sigma (t-s)} ||u(s,t-k,w^k)||^{r+1}_{{\mathbb{L}^{r+1}}}ds\nonumber\\
&\quad +2\int_{t-k}^{t}e^{-\sigma (t-s)}\big((f(s,u(s-\rho(s),t-k,w^k)), u(s,t-k,w^k) )\nonumber\\
&\quad + \lt g(s), u(s,t-k,w^k) \gt \big)ds.
\end{align}By virtue of (4.5) and (4.8), then together with (4.6), one has
\begin{align}
\alpha^2||w^0||^2 &\leq \int_{t-k}^{t}e^{-\sigma (t-s)}\big((-2\nu+\sigma\alpha^2)||u(s,t-k,w^k)||^2 \nonumber\\
&\quad+(-2\beta+\sigma)|u(s,t-k,w^k)|^2\big)ds+(\frac{1}{\lambda_1}+\alpha^2)e^{-\sigma k}||w^k||^2\nonumber\\
&\quad+2\int_{t-k}^{t}e^{-\sigma (t-s)}\big((f(s,u(s-\rho(s),t-k,w^k)),u(s,t-k,w^k))\nonumber\\
&\quad+ \lt g(s),u(s,t-k,w^k) \gt \big)ds-2\gamma\int_{t-k}^{t}e^{-\sigma (t-s)}||u(s,t-k,w^k)||^{r+1}_{{\mathbb{L}^{r+1}}}ds.
\end{align}Subtracting (4.16) from (4.15), we verify that
\begin{align*}
\limsup\limits_{n'\rightarrow\infty}||u^{n'}(t)||^2&\leq (\frac{1}{\lambda_1\alpha^2}+1)
e^{-\sigma k}K(t-k)-(\frac{1}{\lambda_1\alpha^2}+1)e^{-\sigma k}||w^k||^2+||w^0||\nonumber\\
&\rightarrow ||w^0||, \ \mbox{as} \ k\rightarrow\infty.
\end{align*}Now we get
\begin{align*}
\lim\limits_{n'\rightarrow\infty}||u^{n'}-w^0||\rightarrow 0.
\end{align*} Then we obtain the compactness of
$\{u(t,\tau_n,\phi^n)\}$ in V.
Step 2. We will prove that
$\{u_t(\cdot,\tau_n,\phi^n)\}$ is relatively in CV.
From Step 1, using a diagonal argument, there exists a subsequence
$\phi^{n'}$ with
$\tau_{n'}\leq t-h$, for all rational numbers
$\theta_j\in [-h,0]$; there is a
$\widetilde{w}^{j}\in V$ such that
\begin{align}
\lim\limits_{n'\rightarrow\infty}||u^{n'}(t+\theta_j)-\widetilde{w}^{j}||\rightarrow 0,
\end{align} which implies that
$\{u^{n'}(t+\theta_j)\}$ is relatively compact in V.
Next, we show that for all irrational numbers
$\theta\in [-h,0]$, and
$\{u^{n'}(t+\theta)\}$ is relatively compact in V.
Firstly, from (3.13), for all
$t-h\leq t_1 \lt t_2\leq t$, it is easy to conclude that
\begin{align}
||u^{n'}(t_1)-u^{n'}(t_2)||
&\leq\int_{t_1}^{t_2}||\frac{du^{n'}(s)}{ds}||ds\nonumber\\
&\leq(t_2-t_1)^{\frac{1}{r+1}}||\frac{du^{n'}(s)}{ds}||_{L^{\frac{r+1}{r}}(t-h,t;V)}\nonumber\\
&\leq C(t_2-t_1)^{\frac{1}{r+1}},
\end{align} which implies that
$\{u^{n'}\}$ is equicontinuous in
$C([t-h,t],V)$.
Then for any
$n',m'\in \mathbb{N}$, we get
\begin{align}
||u^{n'}(t+\theta)-u^{m'}(t+\theta)||&\leq||u^{n'}(t+\theta)-u^{n'}(t+\theta_j)||+||u^{n'}(t+\theta_j)\nonumber\\
&-u^{m'}(t+\theta_j)||+||u^{m'}(t+\theta_j)-u^{m'}(t+\theta)||,
\end{align} where θj ia a rational number. From (4.18), one can choose an appropriate number
$\theta_j\in[-h,0]$ such that
\begin{equation*}||u^{n'}(t+\theta)-u^{n'}(t+\theta_j)||+||u^{m'}(t+\theta_j)-u^{m'}(t+\theta)|| \lt \frac{\varepsilon}{2},\end{equation*} for any ɛ > 0. In addition, noting that
$\{u^{n'}(t+\theta_j)\}$ is relatively compact in V, we can find that there exists
$N_0\in\mathbb{N}$ such that for any
$n',m'\geq N_0$,
$||u^{n'}(t+\theta_j)-u^{m'}(t+\theta_j)|| \lt \frac{\varepsilon}{2}.$ Then by (4.19), we obtain that
$u^{n'}(t+\theta)$ is a Cauchy sequence in V, which shows that
\begin{align}
\{u^{n'}(t+\theta)\}~\text{is relatively compact in}~V.
\end{align} Finally, from (4.17) and (4.20), it follows that
$\{u^n(s)\}$ is relatively compact in V for all
$s\in(t-h,t)$. Then together with (4.18) and by using the Ascoli–Arzel
$\grave{a}$ theorem, there exist a
$\psi^0\in C_V$ and a subsequence of
$\{u^n\}$ (still defined as
$\{u^n\}$) such that
which shows the asymptotically compactness of
$\{u_t^n\}$ in CV, as desired.
We are now ready to prove the existence and uniqueness of pullback attractors for the system (1.1).
Theorem 4.4. Assume that
$(H_0)$–
$(H_5)$ hold, then the system
$U(t,\tau)$ generated by the solutions of (1.1) has a unique pullback
$\mathcal{D}$-attractor
$\mathcal{A} = \{\mathcal{A}(\tau):\tau\in \mathbb{R}\}$ in CV.
Proof. From Lemma 4.2, it follows that U has a pullback absorbing set and is pullback asymptotically compact by Lemma 4.3. Hence, the existence and uniqueness of the pullback attractors in CV can be immediately established from [Reference Wang40, Proposition 3.6].
5. Invariant measures supported on the pullback attractors
5.1. The existence of invariant measures
In this subsection, we will prove the existence of invariant Borel probability measures for
$\{U(t,\tau)\}_{t\geq\tau}$ on space CV. To this end, we first establish the following continuity of the CV-valued function
$(\tau,\phi)\mapsto U(t,\tau)\phi$.
Lemma 5.1. Assume that
$(H_0)$–
$(H_5)$ hold, then for any
$\phi\in C_V$ and
$t\in\mathbb{R}$, the CV-valued function
$(\tau,\phi)\mapsto U(t,\tau)\phi$ is continuous on
$(-\infty,t]\times C_V$.
Proof. Setting
$\tau_*\in(-\infty,t]$ and
$\phi_*\in C_V$, we will show that for any ɛ
$ \gt $ 0, there is a δ
$ \gt $ 0 such that for any
$|\tau-\tau_*|\vee ||\phi_*-\phi||_{C_V} \lt \delta$ satisfying
It follows from (5.1) that
\begin{align}
&||U(t,\tau)\phi-U(t,\tau_*)\phi_*||_{C_V}\leq||U(t,\tau)\phi-U(t,\tau)\phi_*||_{C_V}\nonumber\\
&\quad+||U(t,\tau)\phi_*-U(t,\tau_*)\phi_*||_{C_V}.
\end{align} For the first term on the right-hand side of (5.2), by the uniqueness of Theorem 3.1, there is a
$\delta_1 \gt 0$ such that for any
$|\tau-\tau_*|\vee ||\phi_*-\phi||_{C_V} \lt \delta_1$,
Next, our goal is to prove that
\begin{align}
\sup\limits_{-h\leq s'\leq 0}||u(t+s';\tau,\phi_*)-u(t+s';\tau_*,\phi_*)|| \lt \varepsilon,
\end{align} where
$u(t;\tau,\phi_*)$ and
$u(t;\tau_*,\phi_*)$ are the solutions of (1.1) with initial value
$\phi_*$ at the initial times τ and
$\tau_*$, respectively.
To obtain (5.4), we will divide the proof into the following steps.
Step 1. We will prove that for any
$\varepsilon_1 \gt 0$, there is a
$\delta_2=\delta_2(\varepsilon_1,\tau_*,\phi_*) \gt 0$ such that for any
$\tau\in(\tau_*-\delta_2,\tau_*)$,
\begin{align}
\sup\limits_{\tau\leq \eta\leq \tau_*}||u(\eta;\tau,\phi_*)-\phi_*(0)|| \lt \varepsilon_1.
\end{align} For any
$\eta\in [\tau,\tau_*]$, note that
\begin{align}
||u(\eta;\tau,\phi_*)-\phi_*(0)||^2
&\leq||u(\eta;\tau,\phi_*)-\phi_*(0)||^2_{\mathcal{M}}\nonumber\\
&=||u(\eta;\tau,\phi_*)||^2_{\mathcal{M}}-||\phi_*(0)||^2_{\mathcal{M}}
-2\alpha^2((u(\eta;\tau,\phi_*)\nonumber\\
&\quad-\phi_*(0),\phi_*(0)))-2(u(\eta;\tau,\phi_*)-\phi_*(0),\phi_*(0)).
\end{align}From (3.5) and (5.6), it follows that
\begin{align*}
& \big|||u(\eta;\tau,\phi_*)||^2_{\mathcal{M}}-||\phi_*(0)||^2_{\mathcal{M}}\big|
=\big|\int_{\tau}^{\eta}\frac{d||u(s;\tau,\phi_*)||^2_{\mathcal{M}}}{ds}ds\big|\nonumber\\
&\leq \frac{1}{\nu}\int_{\tau}^{\tau_*}||g(s)||^2_{V'}ds
+ (1-\rho^*)\int_{\tau-\rho(s)}^{\tau_*-\rho(s)}|u(s)|^2ds\nonumber\\
&\quad+ \int_{\tau}^{\tau_*}(1+\frac{a(s)}{1-\rho^*}+2|\beta|)|u(s)|^2ds
+ \int_{\tau}^{\tau_*}b(s)ds,
\end{align*} which implies that there exists
$\delta'_2 \gt 0$ such that for any s satisfying
$\tau-\delta'_2 \lt \tau \lt s \lt \tau^*$,
\begin{align}
\big|||u(\eta;\tau,\phi_*)||^2_{\mathcal{M}}-||\phi_*(0)||^2_{\mathcal{M}}\big| \lt \frac{\varepsilon_1}{2}.
\end{align}For the last two terms on the right-hand side of (5.6), we have the estimate as follows:
\begin{align*}
& 2\alpha^2((u(\eta;\tau,\phi_*)-\phi_*(0),\phi_*(0)))
+2(u(\eta;\tau,\phi_*)-\phi_*(0),\phi_*(0))\nonumber\\
&=2\left(\int_{\tau}^{\eta}\big(\frac{du(s;\tau,\phi_*)}{ds}
+\alpha^2\frac{dAu(s;\tau,\phi_*)}{ds}\big)ds,\phi_*(0) \right)\nonumber\\
&\leq
2\int_{\tau}^{\eta}
\left\|\frac{dv(s;\tau,\phi_*)}{ds}\right\|_{V'}ds
\|\phi_*(0)\|\nonumber\\
&\leq
2\|\phi_*(0)\| \bigg[\sqrt{\tau_*-\tau}\big(\nu\|u\|_{L^2(\tau,\tau_*;V)}
+\|B(u(t-\rho(t)),u)\|_{L^2(\tau,\tau_*;V')}\nonumber\\
&\quad
+|\beta| \|u\|_{L^2(\tau,\tau_*;V')}+\|g\|_{L^2(\tau,\tau_*;V')}
+\lambda_1^{-\frac{1}{2}}\|f(t,u(t-\rho(t)))\|_{L^2(\tau,\tau_*;H)}\big)\nonumber\\
&\quad +(\tau_*-\tau)^{\frac{1}{r+1}}
\| |u|^{r-1}u \|_{L^{\frac{r+1}{r}}(\tau,\tau_*;V')}\bigg].
\end{align*} By virtue of (3.10)–(3.12), then there exists
$\delta''_2 \gt 0$ such that for any s satisfying
$\tau-\delta''_2 \lt \tau \lt s \lt \tau^*$,
\begin{align}
\big|2\alpha^2((u(\eta;\tau,\phi_*)-\phi_*(0),\phi_*(0)))
+2(u(\eta;\tau,\phi_*)-\phi_*(0),\phi_*(0))\big| \lt \frac{\varepsilon_1}{2},
\end{align}which, together with (5.7) and (5.8), shows that (5.5) holds.
Step 2. Next, we will show that for any
$\varepsilon_2 \gt 0$, there is a
$\delta_3=\delta_3(\varepsilon_2,\tau_*,\phi_*)$ such that for any
$\tau\in(\tau_*-\delta_3,\tau_*)$ satisfying
\begin{align}
\sup\limits_{\tau_*\leq\eta\leq t}||u(\eta;\tau_*,\phi_*)-u(\eta;\tau,\phi_*)||^2 \lt \varepsilon_2.
\end{align} By Theorem 3.1, we can deduce that
$u(\cdot;\tau_*,\phi^*)\in C([\tau_*-h,t];V)$. Since
$t,\tau_*$ are fixed and from (3.30), it can be concluded that
\begin{align}
&\quad\sup\limits_{\tau_*\leq\eta\leq t}||u(\eta;\tau_*,\phi_*)-u(\eta;\tau,\phi_*)||^2\nonumber\\
&\leq\sup\limits_{\tau_*\leq\eta\leq t}||u(\eta;\tau_*,\phi_*)-u(\eta;\tau,\phi_*)||^2_{\mathcal{M}}\nonumber\\
&\leq \big( 2+\frac{1}{\lambda_1}+\alpha^2\big)\cdot e^{\widetilde{H}(t-\tau)}\sup\limits_{-h\leq s\leq 0}||u(\tau_*+s;\tau,\phi_*)-\phi_*(s)||^2\nonumber\\
&\leq \widetilde{C}\sup\limits_{\tau-\tau_*\leq s\leq 0}||u(\tau_*+s;\tau,\phi_*)-\phi_*(0)||^2+\widetilde{C}\sup\limits_{\tau-\tau_*\leq s\leq 0}||\phi_*(s)-\phi_*(0)||^2\nonumber\\
&\quad+\widetilde{C}\sup\limits_{-h\leq s\leq \tau-\tau_*}||\phi_*(\tau_*-\tau+s)-\phi_*(s)||^2,
\end{align} where
$\widetilde{C}=\big( 2+\frac{1}{\lambda_1}+\alpha^2\big) e^{\widetilde{H}(t-\tau)}$. Then by (5.5) and the fact that
$\phi_*(s)\in C_V$, one can obtain (5.9) holds.
Step 3. Now we will show that for any
$\varepsilon_3 \gt 0$, there is a
$\delta_4=\delta_4(\varepsilon_3,\tau_*,\phi_*) \gt 0$ such that for any
$\tau\in(\tau_*-\delta_4,\tau_*)$,
\begin{align}
\sup\limits_{\tau_*-h\leq\eta\leq \tau_*}||u(\eta;\tau_*,\phi_*)-u(\eta;\tau,\phi_*)||^2 \lt \varepsilon_3.
\end{align} For any
$\eta\in[\tau,\tau_*]$, it can be concluded that
\begin{align*}
&\sup\limits_{\tau_*-h\leq\eta\leq \tau_*}||u(\eta;\tau_*,\phi_*)-u(\eta;\tau,\phi_*)||^2\nonumber\\
&\quad\leq \sup\limits_{\tau_*-h\leq\eta\leq \tau}||\phi_*(\eta-\tau_*)-\phi_*(\eta-\tau)||^2+\sup\limits_{\tau\leq\eta\leq \tau_*}||\phi_*(\eta-\tau_*)-\phi_*(0)||^2\nonumber\\
&\qquad+\sup\limits_{\tau\leq\eta\leq \tau_*}||\phi_*(0)-u(\eta;\tau,\phi_*)||^2.
\end{align*} Then from (5.5) and
$\phi_*(s)\in C_V$, it follows that there is a
$\delta_3=\delta_3(\varepsilon_2,\tau_*,\phi_*) \gt 0$ such that (5.11) holds.
From Step 1–Step 3, it follows that (5.4) holds. Then combining with (5.3), we can obtain the desired result.
Now let us recall the definition of generalized Banach limits, which is useful for showing the existence of invariant measure of the process
$\{U(t,\tau)\}_{t\geq\tau}$ (see [Reference Foias, Manley, Rosa and Temam16] for more details).
Definition 5.2. A generalised Banach limit is any linear functional denoted by
$\mathop{\rm LIM}\limits_{t\rightarrow\infty}$, defined on the space of all bounded real-valued functions on
$[0,\infty)$ that satisfies
(1)
$\mathop{\rm LIM}\limits_{t\rightarrow\infty}f(t)\geq0$ for non-negative functions f.
(2)
$\mathop{\rm LIM}\limits_{t\rightarrow\infty}f(t)=\lim\limits_{t\rightarrow\infty}f(t)$ if the usual limit
$\lim\limits_{t\rightarrow\infty}f(t)$ exists.
Theorem 5.3. Assume that
$(H_0)$–
$(H_5)$ hold. Fix a generalized Banach limit
$\mathop{\rm LIM}_{t\rightarrow\infty}$ and
$\xi_{\cdot}:\mathbb{R}\rightarrow C_V$ which is a continuous map such that
$\{\xi_{t}\}_{t\in \mathbb{R}}\in\mathcal{D}$. Then there is a unique family of Borel probability measures
$\{\mu_t\}_{t\in\mathbb{R}}$ in CV such that the support of the measure µt is contained in
$\mathcal{A}(t)$ and
\begin{equation*}\mathop{\rm LIM}_{t\rightarrow\infty}\frac{1}{t-\tau}\int_{\tau}^{t}\Psi(U(t,s)\xi_s)ds=\int_{\mathcal{A}(t)}\Psi(u)d_{\mu_t}(u),\end{equation*}for every real-valued continuous functional Ψ on CV. In addition, µt is invariant in the sense that
\begin{equation*}\int_{\mathcal{A}(t)}\Psi(u)d_{\mu_t}(u)=\int_{\mathcal{A}(\tau)}\Psi(U(t,\tau)u)d_{\mu_{\tau}}(u),\ \forall~t\geq\tau.\end{equation*}Proof. Theorem 4.4 shows that
$\{U(t,\tau)\}_{t\geq\tau}$ has a pullback attractor
$\mathcal{A} = \{\mathcal{A}(t) :t\in \mathbb{R}\}$ in CV. It follows from Lemma 5.1 that
$U(t,\tau)\phi$ is continuous in τ and ϕ. Then by [Reference Lukaszewicz and Robinson27, Theorem 3.1], the result can be immediately obtained.
5.2. The periodicity of invariant measures
Based on Theorem 5.3, we obtain that there exists a family of Borel probability measures
$\{\mu_t \}_{t\in \mathbb{R}}$ such that µt is an invariant measure for dynamical system U and is supported on
$\mathcal{A}(t)$. Now, we give the definition of periodic invariant measure (see Theorem 3.1 in [Reference Yang, Caraballo and Chen43]).
Definition 5.4. For any
$\mathcal{T} \gt 0$, it is called a
$\mathcal{T}$-periodic invariant measure, if there exists an invariant measure µt such that
Next, we will prove the existence of periodicity of invariant measures.
Theorem 5.5. Under the conditions of Theorem 5.3, and assume that
$f(t,\cdot),\ g(t), \ \rho(t)$ are
$\mathcal{T}$-periodic with respect to t and ξt is a
$\mathcal{T}$-periodic continuous mapping in t, for every real-valued continuous functional Ψ on CV, then we obtain
\begin{align*}
\int_{C_V}\Psi(u)d_{\mu_t}(u)=\int_{C_V}\Psi(u)d_{\mu_{t+\mathcal{T}}}(u)=\int_{\mathcal{A}(t)}\Psi(u)d_{\mu_t}(u).
\end{align*}Proof. Let
$\overline{v}(t)=u(t+\mathcal{T},\tau+\mathcal{T},\phi)$. Since
$f(t,\cdot),\ g(t),\ \rho(t)$ are
$\mathcal{T}$-periodic with respect to t, then for any
$w\in V$, we have
\begin{align*}
&(\overline{v}(t),w)+\alpha^2((\overline{v}(t),w))=(\phi(0),w)+\alpha^2((\phi(0),w))\nonumber\\
&\quad-\nu\int_{\tau+\mathcal{T}}^{t+\mathcal{T}}(-\Delta \overline{v}(s-\mathcal{T}),w)ds
+ \int_{\tau+\mathcal{T}}^{t+\mathcal{T}}(g(s),w)ds\nonumber\\
&\qquad-\int_{\tau+\mathcal{T}}^{t+\mathcal{T}}b(\overline{v}(s-\mathcal{T}-\rho(s-\mathcal{T})),\overline{v}(s-\mathcal{T}),w)ds-\beta\int_{\tau+\mathcal{T}}^{t+\mathcal{T}}(\overline{v}(s-\mathcal{T}),w)ds\nonumber\\
&\qquad-\gamma\int_{\tau+\mathcal{T}}^{t+\mathcal{T}}(|\overline{v}(s-\mathcal{T})|^{r-1}\overline{v}(s-\mathcal{T}),w)ds\nonumber\\
&\qquad+\int_{\tau+\mathcal{T}}^{t+\mathcal{T}}(f(s,\overline{v}(s-\mathcal{T}-\rho(s-\mathcal{T}))),w)ds\nonumber\\
\end{align*}
\begin{align*}
&=(\phi(0),w)+\alpha^2((\phi(0),w))-\nu\int_{\tau}^{t}((\overline{v}(s),w))ds
+\int_{\tau}^{t}(g(s),w)ds\nonumber\\
&\quad-\int_{\tau}^{t}b(\overline{v}(s-\rho(s)),\overline{v}(s),w)ds
-\beta\int_{\tau}^{t}(\overline{v}(s),w)ds\nonumber\\
&\quad
-\gamma\int_{\tau}^{t}(|\overline{v}(s)|^{r-1}\overline{v}(s),w)ds+\int_{\tau}^{t}(f(s,\overline{v}(s-\rho(s )),w)ds,
\end{align*} which shows that
$\overline{v}(t)$ is also a solution of (1.1). By the uniqueness of solution, we obtain
which implies that
$U(t,\tau)\phi$ is a periodic dynamical system with period
$\mathcal{T}$. By (5.12), Theorem 5.3, and the periodicity of ξt in t, we can obtain that
\begin{align}
\int_{C_V}\Psi(u)d_{\mu_{t+\mathcal{T}}}(u)
&=\mathop{\rm LIM}_{\tau\rightarrow-\infty}\frac{1}{t-\tau+\mathcal{T}}\int_{\tau}^{t+\mathcal{T}}\Psi(U(t+\mathcal{T},s)\xi_s)ds\nonumber\\ &=\mathop{\rm LIM}_{\tau\rightarrow-\infty}\frac{1}{t-\tau+\mathcal{T}}\int_{\tau-\mathcal{T}}^{t}\Psi(U(t+\mathcal{T},s+\mathcal{T})\xi_{s+\mathcal{T}})ds\nonumber\\ &=\mathop{\rm LIM}_{\tau\rightarrow-\infty}\frac{1}{t-\tau+\mathcal{T}}\int_{\tau-\mathcal{T}}^{t}\Psi(U(t+\mathcal{T},s+\mathcal{T})\xi_s)ds\nonumber\\
&=\mathop{\rm LIM}_{\tau\rightarrow-\infty}\frac{1}{t-\tau+\mathcal{T}}\int_{\tau}^{t}\Psi(U(t,s)\xi_s)ds.
\end{align} Due to the boundedness of Ψ, we have that for all
$t\geq\tau$,
\begin{equation*}\mathop{\rm LIM}_{\tau\rightarrow-\infty}\frac{1}{t-\tau+\mathcal{T}}\int_{\tau-\mathcal{T}}^{t}\Psi(U(t,s)\xi_s)ds=\mathop{\rm LIM}_{\tau\rightarrow-\infty}\frac{1}{t-\tau}\int_{\tau}^{t}\Psi(U(t,s)\xi_s)ds.\end{equation*}Combining with (5.13), we prove the periodicity of the invariant measure µt. Now the proof is completed.
6. Limiting behaviours of invariant measures
In this section, we study the limiting behaviours of the invariant measures for system (1.1) as the delay approaches zero and parameter α tends to a non-negative number α 0, respectively.
6.1. Limits of invariant measures with respect to delay
Firstly, we study the limiting behaviour of the invariant measures for system (1.1) as the delay goes to zero. For simplicity, let
$\widehat{\rho} \equiv \rho(t)$. For all
$\phi\in C([-\widehat{\rho},0];V)$, define
$\widehat{\phi}$ as
\begin{equation*}
\widehat{\phi}(t)=
\left\{
\begin{aligned}
&\phi(t), \ \mbox{if} \ t\in[-\widehat{\rho},0], \\
&\phi(-\widehat{\rho}), \ \mbox{if} \ t\in[-1,-\widehat{\rho}]. \\
\end{aligned}
\right.
\end{equation*} Therefore, we can extend ϕ from
$C([-\widehat{\rho},0];V)$ to
$C([-1,0];V)$. For any
$\widehat{\rho}\in(0,1]$ and
$\phi \in C([-\widehat{\rho},0];V)$, define
$\mathbb{T}_{\widehat{\rho}}$ as
$\mathbb{T}_{\widehat{\rho}} \phi=\phi(0)$. For any
$\widehat{\rho}\in (0,1]$ and
$\phi\in C([-1,0];V)$, let
$\mathcal{T}_{\widehat{\rho}}(\phi)=\phi|_{(-\widehat{\rho},0)}$, where
$\phi|_{(-\widehat{\rho},0)}$ is the restriction of ϕ in
$(-\widehat{\rho},0)$.
Given
$X\subseteq C([-\widehat{\rho},0];V)$, we give
$\widehat{X}=\{\widehat{\phi}:\phi\in X\}$ and
$\mathbb{T}_{\widehat{\rho}} X=\{\mathbb{T}_{\widehat{\rho}} \phi: \phi\in X\}$. Similarly, define
$Y\subseteq C([-\widehat{\rho},0];V)$ and
$\mathcal{T}_{\widehat{\rho}} Y =\{\mathcal{T}_{\widehat{\rho}}\phi: \phi\in Y\}$. It is easy to conclude that if X and Y are compact, then
$\mathbb{T}_{\widehat{\rho}} X$ and
$\mathcal{T}_{\widehat{\rho}} Y$ are also compact.
It follows from Theorem 4.4 that system (1.1) exists a dynamical system and has a
$\mathcal{D}$-pullback attractor
$\mathcal{A}^{\widehat{\rho}}\in \mathcal{D}$ in
$C([-\widehat{\rho},0];V)$. For any
$\widehat{\rho}\in [0,h]$, the solution and dynamical system of (1.1) can be written as
$u^{\widehat{\rho}}$ and
$U^{\widehat{\rho}}$. If
$\widehat{\rho}=0$, system (1.1) is reduced to the system (1.2) without delay [Reference Anh and Trang4]. Similarly, we define u 0 and U 0 as the solution and the dynamical system of (1.2).
Next, we prove the convergence between the solutions of (1.1) and (1.2) with respect to delay.
Lemma 6.1. Assume that
$(H_0)$–
$(H_5)$ hold. Then for any
$t\in[\tau,T]$ and compact set
$K\subseteq C([-h,0];V)$ such that
\begin{align*}
\lim\limits_{\widehat{\rho}\rightarrow0}\sup\limits_{\phi\in\mathcal{T}_{\widehat{\rho}_n}K}
||u^{\widehat{\rho}}(t,\tau,\phi)-u^0(t,\tau,\phi(0))||=0,
\end{align*} where
$u^{\widehat{\rho}}$ and u 0 are the solutions of (1.1) and (1.2), respectively.
Proof. By Theorem 3.1 and Definition 3.1 in [Reference Anh and Trang4], we can find that there exists a positive constant S such that
$||u^{\widehat{\rho}}(t,\tau,\phi)||^2\vee||u^0(t,\tau,\phi(0))||^2\leq S$. Let
$v(t)=u^{\widehat{\rho}}(t)-u^0(t)$, then by (1.1) and (1.2), we have that
\begin{align}
&\frac{1}{2}(|v(t)|^2+\alpha^2||v(t)||^2)+\nu||v(t)||^2+(B(u^{\widehat{\rho}}(t-\widehat{\rho}),u^{\widehat{\rho}}(t))-B(u^0(t),u^0(t)),v(t))\nonumber\\
&\qquad+\beta v(t)+\gamma(|u^{\widehat{\rho}}(t)|^{r-1}u^{\widehat{\rho}}(t)-|u^0(t)|^{r-1}u^0(t),v(t))\nonumber\\
&=(f(s,u^{\widehat{\rho}}(s-\widehat{\rho}))-f(s,u^0(s)),v(s)).
\end{align}It follows from (2.1) that
\begin{align}
&\quad(B(u^{\widehat{\rho}}(t-\widehat{\rho}),u^{\widehat{\rho}}(t))-B(u^0(t),u^0(t)),v(t))\nonumber\\
&=b(u^{\widehat{\rho}}(t-\widehat{\rho})-u^0(t-\widehat{\rho}),u^0(t),v(t)) +b(u^0(t-\widehat{\rho})-u^0(t),u^0(t),v(t))\nonumber\\
&\leq \frac{S}{\nu}||u^{\widehat{\rho}}(t-\widehat{\rho})-u^0(t-\widehat{\rho})||^2+\frac{S}{\nu}||u^0(t-\widehat{\rho})-u^0(t)||^2+\frac{\nu}{2}||v(t)||^2.
\end{align} By (6.1)–(6.2),
$(H_0)$ and
$(H_3)$, we obtain
\begin{align}
&|v(t)|^2+\alpha^2||v(t)||^2+\nu\int_{\tau}^t||v(s)||^2ds+2\gamma\int_{\tau}^t(|u^{\widehat{\rho}}|^{r-1}u^{\widehat{\rho}}-|u^0|^{r-1}u^0,v(s))ds\nonumber\\
&\leq (2|\beta|+1)\int_{\tau}^t|v(s)|^2ds
+ \frac{2S}{\nu(1-\rho^*)}\int_{\tau}^{t}||v(s)||^2ds\nonumber\\
&\quad+L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}\int_{\tau}^t|u^{\widehat{\rho}}(s-\widehat{\rho})-u^0(s)|^2ds\nonumber\\
&\quad+\frac{2S}{\nu(1-\rho^*)}\int_{\tau-\widehat{\rho}}^{\tau}||u^{\widehat{\rho}}(s)-u^0(s)||^2ds
+\frac{2S}{\nu}\int_{\tau}^{t}||u^0(s-\widehat{\rho})-u^0(s)||^2ds.
\end{align}By (3.27), it can be inferred from (6.3) that
\begin{align*}
&\quad|v(t)|^2+\alpha^2||v(t)||^2\nonumber\\
&\leq (2|\beta|+1)\int_{\tau}^t|v(s)|^2ds+L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}\int_{\tau}^t|u^{\widehat{\rho}}(s-\widehat{\rho})-u^0(s-\widehat{\rho})|^2ds\nonumber\\
&\quad+\frac{2S}{\nu(1-\rho^*)}\int_{\tau-\widehat{\rho}}^{\tau}||u^{\widehat{\rho}}(s)-u^0(s)||^2ds+\frac{2S}{\nu}\int_{\tau}^{t}||u^0(s-\widehat{\rho})-u^0(s)||^2ds\nonumber\\
&\quad+\frac{2S}{\nu(1-\rho^*)}\int_{\tau}^{t}||v(s)||^2ds+L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}\int_{\tau}^t|u^0(s-\widehat{\rho})-u^0(s)|^2ds.
\end{align*}The Poincaré inequality implies
\begin{align}
&\quad|v(t)|^2+\alpha^2||v(t)||^2\nonumber\\
&\leq
\left(2|\beta|+1+\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{1-\rho^*}\right)
\int_{\tau}^t|v(s)|^2ds+\frac{2S}{\nu(1-\rho^*)}\int_{\tau}^{t}||v(s)||^2ds\nonumber\\
&\quad+\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu}\right)
\int_{\tau}^{t}||u^0(s-\widehat{\rho})-u^0(s)||^2ds
\nonumber\\
&\quad
+ 2S\widehat{\rho}\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{(1-\rho^*)\lambda_1}
+\frac{2S}{\nu(1-\rho^*)}\right).
\end{align} Due to the compactness of K in CV, for any ɛ > 0, there exists a finite open covering sphere
$\{B(\phi_i;\varepsilon)\}_{i=1}^{m}\subseteq C_V$ such that for each
$\phi\in K,\phi\in\{B(\phi_i;\varepsilon)\}_{i=1}^{m}$, then there exists every
$\phi\in K$ such that
For the third term on the right-hand side of (6.4), we have
\begin{align}
&\quad \left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu} \right)
\int_{\tau}^{t}||u^0(s-\widehat{\rho})-u^0(s)||^2ds\nonumber\\
&\leq 6\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu}\right)\int_{\tau}^{t}||u^0(s-\widehat{\rho};\phi(0))-u^0(s-\widehat{\rho};\phi_i(0))||^2ds\nonumber\\
&\quad+ 6\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu}\right)\int_{\tau}^{t}||u^0(s-\widehat{\rho};\phi_i(0))-u^0(s;\phi_i(0))||^2ds\nonumber\\
&\quad+ 6\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu}\right)\int_{\tau}^{t}||u^0(s;\phi_i(0))-u^0(s;\phi(0))||^2ds.
\end{align} By Definition 3.1 in [Reference Anh and Trang4], it can be concluded that for any
$i\in \{1,...,m\}, u^0(s,\phi_i(0))\in C([\tau,T];V)$. Then, there is a
$\delta=\delta(\varepsilon,T,K)$ such that for any
$\widehat{\rho}\in [0,\delta)$,
\begin{align}
6\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu}\right)\int_{\tau}^{t}||u^0(s-\widehat{\rho};\phi_i(0))-u^0(s;\phi_i(0))||^2ds \lt \frac{\varepsilon}{3}
\end{align}holds. Furthermore, it follows from Theorem 3.1 in [Reference Anh and Trang4] that
\begin{align}
&\quad 6\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu}\right)\int_{\tau}^{t}||u^0(s;\phi_i(0))-u^0(s;\phi(0))||^2ds\nonumber\\
&\leq 6(T-\tau)\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu}\right)||\phi(0)-\phi_i(0)||^2e^{-\widetilde{K}(T-\tau)}\nonumber\\
& \lt \frac{\varepsilon}{3},
\end{align} where
$\widetilde{K}$ is a constant independent of x and t. Similar to (6.7), the first term on the right-hand side of (6.5) can be estimated as follows:
\begin{align}
&\quad 6\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{\lambda_1}
+\frac{2S}{\nu}\right)\int_{\tau}^{t}||u^0(s-\widehat{\rho};\phi(0))-u^0(s-\widehat{\rho};\phi_i(0))||^2ds \lt \frac{\varepsilon}{3}.
\end{align}From (6.3) and (6.6)–(6.8), it follows that
\begin{align}
&\quad|v(t)|^2+\alpha^2\|v(t)\|^2\nonumber\\
&\leq \left(2|\beta|+1+\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{1-\rho^*}\right)\int_{\tau}^{t}|v(s)|^2ds\nonumber\\
&\quad+\frac{2S}{\nu(1-\rho^*)}
\int_{\tau}^{t}||v(s)||^2ds+2S\widehat{\rho}\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{(1-\rho^*)\lambda_1}
+\frac{2S}{\nu(1-\rho^*)}\right)+\varepsilon.
\end{align}By using Gronwall’s inequality, we have
\begin{align}
|v(t)|^2+\alpha^2\|v(t)\|^2
&\leq \left[2S\widehat{\rho}\left(\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{(1-\rho^*)\lambda_1}
+\frac{2S}{\nu(1-\rho^*)}\right)
+\varepsilon \right]e^{\widetilde{N}(T-\tau)}\nonumber\\
&\quad\rightarrow 0,~\mbox{as} \ \widehat{\rho}\rightarrow 0,
\end{align} where
$\widetilde{N}=\max \left\{2|\beta|+1+\frac{L^2(R)||\Lambda||_{L^{\infty}(\tau,T)}}{1-\rho^*},
\frac{2S}{\nu(1-\rho^*)\alpha^2} \right\}$, as desired.
Let
$\mathcal{H}_{\widehat{\rho}}$ be the set of all invariant probability measures of (1.1) and satisfy that the support of such a measure is contained in the pullback attractor.
Theorem 6.2. For all
$n\in \mathbb{N}$, let
$\widehat{\rho}_n\in (0,1]$ satisfying
$\widehat{\rho}_n\rightarrow 0$ as
$n\rightarrow\infty$. If
$\mu_t^{\widehat{\rho}_n}\in \mathcal{H}_{\widehat{\rho}}$, then there is a subsequence
$\{\mu_t^{\rho_{n_k}}\circ \mathbb{T}^{-1}_{\rho_{n_k}}\}_{k=1}^{\infty}$ of
$\{\mu_t^{\rho_n}\circ \mathbb{T}^{-1}_{\rho_n}\}_{n=1}^{\infty}$ and an invariant measure µt of U 0 such that
$\mu_t^{\rho_{n_k}}\circ \mathbb{T}^{-1}_{\rho_{n_k}}\rightarrow \mu_t$ weakly as
$k\rightarrow\infty$.
In addition, assume that
$f(t,\cdot),\ g(t)$, and
$\rho(t)$ are
$\mathcal{T}$-periodic with respect to t, then there exist a subsequence
$\{\mu_t^{\rho_{n_k}}\circ \mathbb{T}^{-1}_{\rho_{n_k}}\}_{k=1}^{\infty}$ of
$\{\mu_t^{\rho_n}\circ \mathbb{T}^{-1}_{\rho_n}\}_{n=1}^{\infty}$ and an
$\mathcal{T}$-periodic invariant measure µt of U 0 such that
$\mu_t^{\rho_{n_k}}\circ \mathbb{T}^{-1}_{\rho_{n_k}}\rightarrow \mu_t$ weakly as
$k\rightarrow\infty$.
Proof. Assume that
$(H_0)$–
$(H_5)$ hold. For any
$\tau\in\mathbb{R}$, from Theorem 4.4 and similar proof to Lemma 5.1 in [Reference Yang, Caraballo and Chen43], the union
$\bigcup\limits_{0 \lt \widehat{\rho}\leq 1}\widetilde{\mathcal{A}}^{\widehat{\rho}}$ is precompact in
$C([-1,0];V)$. Based on the abstract theory of [Reference Yang, Caraballo and Chen43, Theorem 5.1] and Lemma 6.1, combined with the compactness of
$\bigcup\limits_{0 \lt \widehat{\rho}\leq 1}\widetilde{\mathcal{A}}^{\widehat{\rho}}$, the results of Theorem 6.2 can be obtained.
6.2. Limits of invariant measures for parameter
$\alpha\rightarrow\alpha_0$
In this section, we will investigate the limit behaviour of the (periodic) invariant measure of (1.1) when
$\alpha\rightarrow\alpha_0$, where
$0\leq\alpha_0 \lt \infty$.
However, for system (1.3), the unbounded domain is a special case of this paper, which will not be proved in detail here, refer to Theorem 3.1 and [Reference Shi, Cui, Li and Yang35, Theorem 3.1].
Now we consider
$0\leq\alpha \lt \infty$ and denote the solution of (1.1) and the dynamical system as u α and U α. From
$\S$ 4, U α has a
$\mathcal{D}$-pullback attractor
$\mathcal{A}^{\alpha}\in \mathcal{D}$ in
$C([-h,0];V)$, especially,
$\mathcal{A}^0\in \mathcal{D}$ in
$C([-h,0];H)$ as α = 0.
We first prove the compactness of the set of all attractors of (1.1) for all
$\alpha\in [0,\infty)$.
Lemma 6.3. Assume that
$(H_0)$–
$(H_5)$ hold. For any
$\tau\in\mathbb{R}$, the union
$\bigcup\limits_{0 \lt \alpha \lt \infty}\mathcal{A}^{\alpha}$ is precompact in
$C([-h,0];V)$. Specifically,
$\mathcal{A}^0$ is precompact in
$C([-h,0];H)$ as α = 0.
Proof. By (4.3), U α has a pullback-
$\mathcal{D}$ attractor
$\mathcal{A}^{\alpha}$ in CV and exists a pullback-
$\mathcal{D}$ absorbing set K α given as
\begin{equation*}K^{\alpha}(\tau)=\{\psi\in C([-h,0];V): ||\psi||^2_{C_V}\leq \mathcal{R}_V^2(\tau)\},\end{equation*} where
$\mathcal{R}_V^2(\tau)$ is given by (4.3). Then combining the invariance of
$\mathcal{A}^{\alpha}$, we can derive
\begin{align}
\bigcup\limits_{0 \lt \alpha \lt \infty}\mathcal{A}^{\alpha}(\tau)\subset\bigcup\limits_{0 \lt \alpha \lt \infty}K^{\alpha}(\tau)=E(\tau),
\end{align} where
$E(\tau)=\{\phi\in C([-h,0];V): ||\phi||^2_{C_V}\leq \mathcal{R}_V^2(\tau)\}$. Taken
$\{u^n\}^{\infty}_{n=1}\subset\bigcup\limits_{0 \lt \alpha \lt \infty}\mathcal{A}^{\alpha}(\tau)$, for any n, there exists
$\alpha_n\in (0,\infty)$ such that
$u^n\in \mathcal{A}^{\alpha_n}(\tau)$. Taken a sequence
$\{t_n\}^{\infty}_{n=1}$ satisfies
$t_n\rightarrow\infty$ as
$n\rightarrow\infty$. By the invariance of
$\mathcal{A}^{\alpha_n}$, there is a
$u^n_{\tau-t_n}\in\mathcal{A}^{\alpha_n}(\tau-t_n)$ such that
From (6.11), we obtain
$u^n_{\tau-t_n}\in E(\tau-t_n)\in\mathcal{D}$. Similar to the discussion of Theorem 4.3, it can be concluded that
$U^{\alpha_n}(\tau,\tau-t_n,u^n_{\tau-t_n})$ is precompact in CV. Similar to the discussion of Theorem 5.3 in [Reference Shi, Cui, Li and Yang35], we find that
$\mathcal{A}^0$ is precompact in
$C([-h,0];H)$.
Next, we prove the convergence between the solutions of (1.1) and (1.3) with respect to parameter α.
Lemma 6.4. Assume that
$(H_0)$–
$(H_5)$ hold. Then for every compact set
$K\subseteq C([-h,0];V)$ (
$K\subseteq C([-h,0];H)$ as α = 0) and
$0 \lt \alpha_0 \lt \infty$, we obtain
\begin{align*}
\lim\limits_{\alpha\rightarrow\alpha_0}\sup\limits_{\phi\in K}
||u_t^{\alpha}(\cdot,\tau,\phi)-u_t^{\alpha_0}(\cdot,\tau,\phi)||_{C_V}=0,
\end{align*}especially,
\begin{align*}
\lim\limits_{\alpha\rightarrow 0}\sup\limits_{\phi\in K}
||u_t^{\alpha}(\cdot,\tau,\phi)-u_t^0(\cdot,\tau,\phi)||_{C_H}=0, \ \mbox{for any} \ \ t\in[\tau,T].
\end{align*}Proof. The proof will be considered with the following two cases.
$\bullet$ Case I:
$\alpha\rightarrow\alpha_0 \ \mbox{with} \ 0 \lt \alpha_0 \lt \infty$.
Let
$v(t)=u^{\alpha}(t)-u^{\alpha_0}(t)$, it follows from (1.1) that
\begin{align}
&\frac{d}{dt}(1-\alpha_0^2\triangle)v-\nu\triangle v+B(v(t-\rho(t)),u^{\alpha})+B(u^{\alpha_0}(t-\rho(t)),v)\nonumber\\
&\qquad+\beta v+\gamma(|u^{\alpha}|^{r-1}u^{\alpha}-|u^{\alpha_0}|^{r-1}u^{\alpha_0})
\nonumber\\
&=f(t,u^{\alpha}(t-\rho(t)))-f(t,u^{\alpha_0}(t-\rho(t)))+(\alpha^2-\alpha_0^2)\triangle u^{\alpha}.
\end{align} where
$-\alpha^2\triangle u^{\alpha}+\alpha^2_0\triangle u^{\alpha_0}
= (\alpha^2-\alpha^2_0)(-\triangle u^{\alpha})-\alpha^2_0\triangle v$ and
$B(u^{\alpha}(t-\rho(t)),u^{\alpha})-B(u^{\alpha_0}(t-\rho(t)),u^{\alpha_0})
=B(v(t-\rho(t)),u^{\alpha})+B(u^{\alpha_0}(t-\rho(t)),v)$.
Taking the inner product with (6.12) by v(t), then from
$(H_3)$, using Hölder’s and Young’s inequalities, we obtain
\begin{align*}
&\frac{1}{2}\frac{d}{dt}(|v(t)|^2+\alpha^2_0||v(t)||^2)+\nu||v(t)||^2\nonumber\\
&\leq|b(v(t-\rho(t)),u^{\alpha},v)|+(|\beta|+\frac{L^2(R)\Lambda(t)}{2(1-\rho^*)})|v(t)|^2\nonumber\\
&\quad+\frac{1-\rho^*}{2}|v(t-\rho(t))|^2+\frac{|\alpha^2-\alpha_0^2|^2}{2\nu}||u^{\alpha}(t)||^2+\frac{\nu}{2}||v(t)||^2.
\end{align*}Then by (3.8), we have
\begin{align*}
|v(t)|^2+\alpha_0^2||v(t)||^2&\leq \int_{\tau}^{T}\left(\frac{N}{(1-\rho^*)\alpha^2}+1\right)||v(s)||^2ds\nonumber\\
&\quad+\int_{\tau}^{T}\left(2|\beta|+\frac{L^2(R)\Lambda(s)}{1-\rho^*}+1\right)|v(s)|^2ds
+\frac{N}{\nu^2}|\alpha^2-\alpha_0^2|^2.
\end{align*}Gronwall’s inequality implies that
\begin{align}
|v(t)|^2+\alpha^2_0||v(t)||^2
\leq \frac{N}{\nu^2}|\alpha^2-\alpha_0^2|^2\cdot e^{\int_{\tau}^{T}\widetilde{T}dt}
\rightarrow 0, \ \mbox{as} \ \alpha\rightarrow\alpha_0,
\end{align} where
$\widetilde{T}=\max \left\{\frac{N}{(1-\rho^*)\alpha^2\alpha_0^2}
+\frac{1}{\alpha_0^2},2|\beta|+\frac{L^2(R)\Lambda(t)}{1-\rho^*}+1\right\}$.
$\bullet$ Case II:
$\alpha\rightarrow\alpha_0 \ \mbox{with} \ \alpha_0=0.$
Similar to the discussion in (6.12), we can conclude that
\begin{align*}
&\frac{dv}{dt}-\nu\triangle v+B(v(t-\rho(t)),u^{\alpha})+B(u^0(t-\rho(t)),v)\nonumber\\
&\qquad+\beta v+\gamma(|u^{\alpha}|^{r-1}u^{\alpha}-|u^0|^{r-1}u^0)
\nonumber\\
&=f(t,u^{\alpha}(t-\rho(t)))-f(t,u^0(t-\rho(t)))+\alpha^2\triangle\frac{du^{\alpha}}{dt}.
\end{align*}Then we have
\begin{align}
&\quad\frac{1}{2}\frac{d}{dt}|v(t)|^2+\nu||v(t)||^2\nonumber\\
&\leq |b(v(t-\rho(t)),u^{\alpha},v)|+(|\beta|+\frac{L^2(R)\Lambda(s)}{2(1-\rho^*)})|v(t)|^2+\frac{1-\rho^*}{2}|v(t-\rho(t))|^2\nonumber\\
&\quad+\frac{\alpha^4}{\nu}||\frac{du^{\alpha}}{dt}||^2+\frac{\nu}{4}||v(t)||^2.
\end{align}From Lemma 2.5 and Theorem 3.1 in [Reference Shi, Cui, Li and Yang35], it implies
\begin{align}
&\quad\int_{\tau}^{T}|b(v(s-\rho(s)),u^{\alpha},v)|ds\nonumber\\
&\leq C \int_{\tau}^{T} ||u^{\alpha}(s)||_{(L^p(\Omega))^3}||v(s)|||v(s-\rho(s))|^{\frac{2}{q(1-\kappa)}}||v(s-\rho(s))||^{\frac{3}{p(1-\kappa)}}ds,
\end{align} where
$\frac{2}{q}+\frac{3}{p}+\kappa=1$ (
$0\leq\kappa \lt 1$) and the definitions of p and q are in Lemma 2.3.
By repeatedly applying Young’s inequality to (6.15) (here we directly use the result of estimate (18) in Theorem 3.2 of [Reference Shi, Cui, Li and Yang35]), we obtain
\begin{align}
&\int_{\tau}^{T}\left|b\left( v(s-\rho(s)),u^{\alpha}(s),v(s) \right)\right|ds\\
&\quad\leq
\frac{1}{q}\int_{\tau}^{T}
||u^{\alpha}(s)||^q_{\mathbb{L}^p}
\sup\limits_{\sigma\in(s-h,s)}|v(\sigma)|^2 ds
+\frac{\nu}{4}\int_{\tau}^{t} \sup\limits_{\sigma\in(s-h,s)}||v(\sigma)||^2ds\nonumber\\
&\quad+C(\nu)\int_{\tau}^{t} \sup\limits_{\sigma\in(s-h,s)}|v(\sigma)|^2ds.\nonumber
\end{align}Integrating (6.14) in time, combining with (6.16), we have
\begin{align}
\sup\limits_{\sigma\in(t-h,t)}|v(\sigma)|^2
&\leq\int_{\tau}^{T}(2|\beta|+1+\frac{L^2(R)\Lambda(s)}{(1-\rho^*)}
+\frac{2}{q}||u^{\alpha}(s)||^q_{\mathbb{L}^p }) \sup\limits_{\sigma\in(s-h,s)}|v(\sigma)|^2ds\nonumber\\
&\quad+\frac{\alpha^4}{\nu}\int_{\tau}^{T}||\frac{du^{\alpha}(s)}{ds}||^2ds.
\end{align}It follows from (2.3), (3.13), and Gronwall’s inequality that
\begin{align}
&\sup\limits_{\sigma\in(t-h,t)}|v(\sigma)|^2\nonumber\\
&\quad\leq \frac{C_L\alpha^4}{\nu}\cdot e^{\int_{\tau}^{T}(2|\beta|+1+\frac{L^2(R)\Lambda(s)}{(1-\rho^*)}+\frac{2}{q}||u^{\alpha}(s)||^q_{\mathbb{L}^p})ds}\nonumber\\
&\quad\leq \frac{C_L\alpha^4}{\nu}\cdot e^{\int_{\tau}^{T}(2|\beta|+1+\frac{L^2(R)\Lambda(s)}{(1-\rho^*)})ds
+C(\theta,q)(||u||_{L^{\infty}(\tau,T;H)}+||u||_{L^{r+1}(\tau,T;\mathbb{L}^{r+1})})}\nonumber\\
&\quad\rightarrow 0, \ \mbox{as} \ \alpha\rightarrow 0,
\end{align} where CL is a constant independent of
$s,x$. Now, we have completed the proof from (6.13) and (6.18).
Given that
$\mathcal{H}_{\alpha} $ is the set of all invariant probability measures of (1.1) with parameter α and satisfies that the support of such a measure is contained in the pullback attractor. From the abstract result of [Reference Yang, Caraballo and Chen43, Theorem 4.1], we can obtain the following result about limit behavior of the invariant measure for the parameter α.
Theorem 6.5. For all
$n\in \mathbb{N}$, let
$\alpha_n,\alpha_0\in [0,\infty)$ satisfy
$\alpha_n\rightarrow \alpha_0$ as
$n\rightarrow\infty$. If
$\mu_t^{\alpha_n}\in \mathcal{H}_{\alpha}$, then there is a subsequence
$\{\mu_t^{\alpha_{n_k}}\}_{k=1}^{\infty}$ of
$\{\mu_t^{\alpha_n}\}_{n=1}^{\infty}$ and an invariant measure µt of
$U^{\alpha_0}$ such that
$\mu_t^{\alpha_{n_k}}\rightarrow \mu_t$ weakly as
$k\rightarrow\infty$.
In addition, assume that
$f(t,\cdot),\ g(t)$, and
$\rho(t)$ are
$\mathcal{T}$-periodic with respect to t, then there exist a subsequence
$\{\mu_t^{\alpha_{n_k}}\}_{k=1}^{\infty}$ of
$\{\mu_t^{\alpha_n}\}_{n=1}^{\infty}$ and a
$\mathcal{T}$-periodic invariant measure µt of
$U^{\alpha_0}$ such that
$\mu_t^{\alpha_{n_k}}\rightarrow \mu_t$ weakly as
$k\rightarrow\infty$.
Proof. Based on [Reference Yang, Caraballo and Chen43, Theorem 4.1], Lemma 6.4, and the compactness of
$\bigcup\limits_{0\leq\alpha \lt \infty}\mathcal{A}^{\alpha}(\tau)$, the results of Theorem 6.5 can be obtained immediately.
Remark 6.6. Denote the solution of (1.1) and the dynamical system as u γ and U γ. We can also consider the limiting behaviour of the invariant measures as the parameter
$\gamma\rightarrow\gamma_0 \ (0\leq\gamma_0 \lt \infty)$. The key is to handle
$\gamma |u^{\gamma}|^{r-1}u^{\gamma}-\gamma_0 |u^{\gamma_0}|^{r-1}u^{\gamma_0}=\gamma(|u^{\gamma}|^{r-1}u^{\gamma}-|u^{\gamma_0}|^{r-1}u^{\gamma_0})+(\gamma-\gamma_0)|u^{\gamma_0}|^{r-1}u^{\gamma_0}$. By applying the abstract result of [Reference Yang, Caraballo and Chen43, Theorem 4.1], proceeding similar to Lemma 6.3 and 6.4, we can obtain the result.
Remark 6.7. If
$\alpha=\gamma=\beta=0$, equation (1.1) is reduced to the 3D NS equation with double delays as follows:
\begin{equation}
\left\{
\begin{aligned}
&\frac{\partial u}{\partial t}-\nu\Delta u+(u(t-\rho(t)) \cdot \nabla)u+\nabla p=f(t,u_t)+g(t,x),\\
& \mbox {div} \ u=0.\\
\end{aligned}
\right.
\end{equation} The long time dynamics of 2D NS equations are also important for understanding the turbulence of fluid flow [Reference Temam37]. The well-posedness, attractors of 2D NS with delay on the unbounded domain were studied in [Reference Garrido-Atienza and Marín-Rubio19, Reference Marín-Rubio and Real29]. For system (6.19), there are only some results on bounded domain [Reference Garcia-Luengo, Marín-Rubio and Planas18]. As is well known, the regularity of solution for the 3D NS equation is still an open problem. In the case of bounded domains, can the invariant measure of the system (6.19) be characterized by the limit behaviour of the invariant measure of system (1.1) when
$\alpha,\gamma,\beta\rightarrow0$? This is a problem to be investigated in the future.
Acknowledgements
This work was supported by the National Natural Science Foundation of China (Nos.12471468 and 12371499) and Tianjin Natural Science Foundation of China (No. 23JCZDJC00470).







