Local integral manifolds for a nonautonomous parabolic equation (Q1280077)

The authors construct integral manifolds for the nonautonomous equation \[ \partial_t u=\Delta u-f(u)+g(x)+ \varepsilon \bigl(g_1 (x,t)- f_1(u,t)),\;x\in\Omega \subseteq\mathbb{R}^n. \tag{1} \] An integral manifold is to (1) what an inertial manifold is to an autonomous equation. Here, \(\Omega\) is a smooth bounded domain, \(u\in L^2(\Omega)\), with \(u=0\) on \(\partial\Omega\) and \(u(0)=u_r(x)\). The data \(f,g,f_1,g_1\) are subject to a series of growth conditions, denoted simply by (g.c) below. One is then given an equilibrium solution \(z\) of (1) for \(\varepsilon=0\). In order to study (1) in a neighbourhood of \(z\) one sets \(u=z+v\) in (1) so as to get \[ \partial_t v+Av= B_0(v)+ B_1(t,v), \tag{2} \] where \(A=-\Delta+ f'(z)\), \(B_1(t,v)= g_1(x,t)-f_1 (z+v,t)\) and \(B_0(v)= -(f(z+v)- f(v)-f'(z+v))\). By a standard cutoff procedure one replaces \(B_0,B_1\) by \(B_0^*,B_1^*\) respectively which conicide with \(B_0,B_1\) for \(v\) small and all \(t\in\mathbb{R}\). Equation (2) is then replaced by \[ \partial_tv+ Av=B(v,t),\quad B(v,t)=B_0^*(v)+ \varepsilon B_1^*(v,t). \tag{3} \] One has the estimates: \[ \bigl\| B(v_1,t)-B(v_2,t) \bigr\| \leq\bigl(L_\rho +| \varepsilon| L_2\bigr) \| v_1-v_2\|,\quad \bigl\| B(v,t)\bigr \|\leq L_\rho\rho+ |\varepsilon| L_1 \tag{4} \] for all \(v_1,v_2\) and \(t\in\mathbb{R}\). As a consequence of (g.c) one has \(L_\rho\to 0\) as \(\rho\to 0\), where \(\rho\) is a small parameter describing the cutoff. Spectral considerations give rise to projection operators \(P\), \(Q=1-P\) where \(\text{rg}(P)\) has finite dimension. \(P,Q\) satisfy \[ \| e^{At} P\|\leq Ke^{(\lambda- \delta)t},\;\| e^{-AT} Q\|\leq K(t^{-1} +1)^{1/2}e^{-(\lambda+ \delta)t},\;t>0; \tag{5} \] here \(\lambda \geq 0\) and \(\delta>0\) are parameters going via spectral splitting into the construction of \(P,Q\). Application of \(P,Q\) to (3) yields \[ \partial_t p+Ap= PB(p+q,t),\;\partial_tq+ Aq=QB(p+q,t),\;p=Pv,\;q=Qv. \tag{6} \] To (6) an integral manifold construction is applied, too involved to be discussed here. The crucial point is that no spectral gap condition for \(\sigma(A)\) is needed; instead heavy use is made of the fact that \(L_\rho\) and \(|\varepsilon| L_j\) in (4) can be made arbitrarily small by proper choice of \(\varepsilon,\rho\).