Computation of smooth manifolds via rigorous multi-parameter continuation in infinite dimensions (Q285445)

The article deals with the computation of the two-dimensional smooth manifold \(\mathcal M\) implicitly defined by the formula \[ \mathcal M=\{x \in {\mathbb R}^2 \times B_1:\;f(x) = 0\},\tag{1} \] where \(f:\;{\mathbb R}^2 \times B_1 \to B_2\) is a compact and twice Fréchet differentiable operator whose derivative \(Df(x):\mathbb R^2 \times B_1 \to B_2\) has a bounded inverse at each \(x \in\mathcal M\). The authors describe the general scheme of their arguments: Problem (1) is reduced to a finite-dimensional one using a Galerkin-like method; the multi-parameter technique is applied by a finite-dimensional approximation to obtain an approximate simplicial triangulation \(S\) of the manifold; for each simplex \(\sigma \in S\) the existence of a genuine smooth local chart \(U_\sigma\) of the corresponding part of \(\mathcal M\) is constructed; the smoothness of this chart is proved; the charts \(U_\sigma\) (\(\sigma \in S\)) are united in a global smooth atlas of \(\mathcal M\): \[ \mathcal M := \bigcup_{\sigma \in S} U_\sigma; \] it is proved that this formula determines a smooth two-dimensional manifold embedded in the Banach space \({\mathbb R}^2 \times B_1\). This scheme is consequently realized. As an example, the known Cahn-Hilliard equation \[ \begin{cases} u_t = -(\varepsilon^2\Delta u + u - u^3)_{yy} & \text{ in } [0,1], \\ u_y = u_{yyy} = 0 & \text{ for } y = 0, 1, \end{cases} \] is considered. Some of the authors' statements are vague. The assumption about the invertibility of \(Df(x)\) implies that \(\mathcal M\) consists of isolated points. However, the authors think that \(\mathcal M\) is two-dimensional, and this is correct if the derivative \(Df(x)\) is changed onto the partial derivative of \(f\) along with \(B_1\). Further, the assumption about compactness of \(f\) seems to be strange; really, in the infinite-dimensional case, it means that the problem under consideration is ill-posed; but in this case all arguments for the transfer from the infinite-dimensional problem to a finite-dimensional one are false. One can find some other inaccuracies.