Nesting inertial manifolds for reaction and diffusion equations with large diffusivity (Q880285)

The authors study the asymptotic behaviour of solutions of a weakly coupled system of reaction diffusion equations, ie. \[ \begin{aligned} & u_t-\text{div}({\mathcal D}^\varepsilon\nabla u)+V^\varepsilon u+\lambda u=f(u)\text{ on }\Omega\subset \mathbb R^N\\ & {\mathcal D}^\varepsilon\partial_nu+B^\varepsilon u=g(x)\text{ on }\Gamma=\partial\Omega,\quad t\in (0,T)\\ & u(0)=u^\varepsilon_0\text{ on }\Omega,\quad u=(u_1,\dots,u_m).\end{aligned}\tag{1} \] Here \(\Omega\) is a smooth bounded domain, \({\mathcal D}^\varepsilon\in\text{diag}(d^\varepsilon_i)\) is a diagonal matrix with \(d^\varepsilon_i\in L^\infty(\Omega)\); moreover \(V^\varepsilon=\text{diag}(V^\varepsilon_i)\), \(B^\varepsilon=\text{diag}(b^\varepsilon_i)\) where \[ V^\varepsilon_i\in L^p(\Omega),\quad b^\varepsilon_i\in L^1(\Gamma) \] with \(p\), \(q\) subject to some conditions depending on the dimension \(N\). The coefficients \(d^\varepsilon_i\), \(V^\varepsilon_i\), \(b^\varepsilon_i\) depend on the parameter \(\varepsilon>0\) and satisfy some asymptotic conditions as \(\varepsilon\to 0\) such as \[ \inf_\Omega (d^\varepsilon_i(x))\to\infty\text{ as }\varepsilon\to 0.\tag{2} \] The aim is to study the asymptotics of solutions of (1) as \(\varepsilon\to 0\) in terms of inertial manifolds. It turns out that this behaviour is related to the asymptotic behaviour of solutions of the system of ODE's: \[ u_t+L_0u=f(u)+|\Gamma|\cdot|\Omega|^{-1} g(u),\quad u(0)=u_0\in\mathbb R^m\tag{3} \] where \(L_0\) is an \(m\times m\) matrix related to the linear part of (1) in a specific way. In order to prove the relevant statements the authors need a number of preparatory steps which permit them to use the machinery of inertial manifolds. Here we content us with some indications. In a first step one considers the linear part of (1) which gives rise to a positive selfadjoint operator \(A_\varepsilon=\text{diag}(A^\varepsilon_i)\), \(\text{dom}(A_\varepsilon)=\Pi \text{dom}(A^\varepsilon_i)\) where \[ \text{dom}(A^\varepsilon_i)=\{u_i\in H^1(\Omega);\;\text{div}(d^\varepsilon_i\nabla u_i)+(V^\varepsilon_i+\lambda)u\in L^2(\Omega)d^\varepsilon_i\partial_nu+b^\varepsilon_i u_i=0\text{ on }\Gamma\}. \] Fractional power spaces are now defined via \(X^\alpha_\varepsilon=\text{dom}(A^\alpha_\varepsilon)\), \(\alpha\in\mathbb R\). By defining linearities and nonlinearities properly one is led to consider (1) and its linear part as abstract evolution equations on the scale of spaces \(X^\alpha_\varepsilon\) which assume the form \[ u_t+L^\varepsilon u=0,\quad u(0)=u^\varepsilon_0\tag{\(4_\ell\)} \] \[ u_t+L^\varepsilon u=h(u), u(0)\equiv u^\varepsilon_0\tag{\(4_{n\ell}\)} \] It is then proved that \(-L^\varepsilon\) generates an analytic semigroup (Thm. 2.1) while Thm. 2.2 guarantees local existence and uniqueness of solutions of \((4)_{n\ell}\). In order to obtain global existence an assumption on the nonlinearities \(f\), \(g\) is needed which in the authors terminology reads as follows: \[ \lim_{|u|\to\infty}\frac{\sup N(u)}{u}<\infty\quad\text{where }N\in\{f,g\}.\tag{5} \] Theorem 3.1 then asserts global existence and also the existence of an absorbing set. The proof, based on (5), uses a suitable Ljapounov function. The existence of a global, compact attractor \({\mathcal A}^\varepsilon\subseteq X^\alpha_\varepsilon\) for \(\alpha<\frac 34\) then follows. Based on these preparations, the authors proceed to the construction of inertial manifolds for (1) resp. \((4)_{n\ell}\). This construction, while along standard lines, is nevertheless quite involved. Theorem 4.7 asserts the existence of an inertial manifold \({\mathcal M}_\varepsilon\) for \((4)_{n\ell}\). In the last part of the paper, the asymptotic behaviour of the attractors and the manifolds \({\mathcal M}_\varepsilon\), \(\varepsilon\to 0\) is discussed and it is shown that they converge in some sense toward the corresponding objects of the ODE (3). We have to omit details.