A general study of nonlinear problems with three solutions in Hilbert spaces (Q1083681)

This paper intends to demonstrate that a general criterion can be found for deciding whether a certain nonlinear equation has exactly three solutions in a Hilbert space H. Formally, the equation under study reads \[ u-Lu+\Gamma (u)=0, \] with \(L\in {\mathcal L}(H)\) compact and selfadjoint and \(\Gamma\) the gradient of some functional J verifying \(\Gamma (0)=0\) (so that \(u=0\) is a solution). The above equation is then shown to have exactly two other solutions under (essentially) two separate assumptions on the linear part I-L and the nonlinear part \(\Gamma\). The condition on I-L is geometrical in nature: it is assumed that I-L has exactly one negative eigenvalue, simple, all its other eigenvalues being positive. This hypothesis has a bearing on the geometry of the sets \(\{v\in H,((I- L)v| v)<r\}\) where \(r>0\) is fixed. In contrast, the condition on \(\Gamma\) is related to the convexity of the functional J. More precisely, J must be ''more than quadratic'', which is expressed by an ad hoc definition (0-hyperconvexity) purposely introduced. Typically, this approach can be used to prove existence of two nonzero solutions (and no other) in problems where the same result is usually obtained through the maximum principle. An academic example is \[ -\Delta u+\lambda u+f(u)=0;\quad u\in H^ 1_ 0(\Omega), \] for \(\Omega\) bounded open set of \({\mathbb{R}}^ n\), provided that \(\lambda\) lies between the first two eigenvalues of -\(\Delta\) and suitable growth conditions are required of f(u)/u. As no maximum principle is involved here, \(\Delta\) could as well be replaced by a higher order differential operator. Also, the nonlinearity need not be local (e.g. von Kàrmàn equations). By comparison with other studies along the same lines, the notion of 0- hyperconvexity introduced here allows one to eliminate the quite stringent assumption of homogeneity (and oddness) of \(\Gamma\), so that the nontrivial solutions need not be \(\pm u\).