Reduced limits for nonlinear equations with measures (Q2269685)

The authors investigate the convergence of solutions of the equation \[ -\Delta u + g(u) = \mu \quad\text{in}\quad \Omega, \leqno(1) \] where \(\Omega\subset{\mathbb R}^N\), \(N\geq 2\), is a smooth bounded domain, \(g:{\mathbb R}\rightarrow {\mathbb R}\) is a continuous nondecreasing function with \(g(0)=0\), and \(\mu\) is a finite measure on \(\Omega\). By solution we mean a function \(u\in L^1_{loc}(\Omega)\) such that \(g(u)\in L^1_{loc}(\Omega)\) and (1) holds in the sense of distributions. Equation (1) is not solvable for every measure \(\mu\). Let \({\mathcal G}(g)\) denote the set of finite measures for which a solution of (1) exists, and \({\mathcal G}_0(g)\) the set of finite measures for which a solution of \[ -\Delta u + g(u) = \mu \quad\text{in}\quad \Omega, \qquad u=0 \quad\text{on}\quad \partial\Omega, \leqno(2) \] exists. Clearly \({\mathcal G}_0(g) \subset {\mathcal G}(g)\), but the authors show that in fact \({\mathcal G}_0(g)={\mathcal G}(g)\). It is known from the work of \textit{H.~Brezis} and the authors [Nonlinear elliptic equations with measures revisited. Mathematical aspects of nonlinear dispersive equations. Lectures of the CMI/IAS workshop on mathematical aspects of nonlinear PDEs, Princeton, NJ, USA, 2004. Princeton, NJ: Princeton University Press. Annals of Mathematics Studies 163, 55--109 (2007; Zbl 1151.35034)] that if \(\{\mu_k\}\subset{\mathcal G}_0(g)\) is a bounded sequence of measures in \(\Omega\) converging strongly to \(\mu\), then the sequence of solutions \(\{u_k\}\) of (2) with data \(\mu_k\) converges strongly in \(L^1(\Omega)\). This conclusion is also true if \(g(t)=|t|^{q-1}t\) with \(1<q<N/(N-2)\) and \(\mu_k \buildrel * \over \rightharpoonup \mu\). But this is false for \(q\geq N/(N-2)\). The authors investigate what happens if \(\mu_k \buildrel * \over \rightharpoonup \mu\), and what additional minimal assumptions guarantee convergence of \(\{u_k\}\) to a solution of (1). They prove the following results. In each theorem it is assumed that \(\{\mu_k\}\subset {\mathcal G}(g)\) with \(\mu_k \buildrel * \over \rightharpoonup \mu\) and \(\{u_k\}\) are corresponding solutions of (1) or (2). Theorem 1. If \(u_k\rightarrow u^\#\) in \(L^1(\Omega)\), then \(g(u^\#)\in L^1(\Omega)\) and there exists a finite measure \(\mu^\#\) in \(\Omega\) such that \(u^\#\) solves (2) with \(u,\mu\) replaced by \(u^\#,\mu^\#\). Theorem 2. The measure \(\mu^\#\) does not depend on the boundary condition, in the sense that if \(v_k\) are solutions of (1) with data \(\mu_k\) and \(v_k\rightarrow v^\#\) in \(L^1(\Omega)\), then \(v^\#\) solves (1) with \(u,\mu\) replaced by \(v^\#,\mu^\#\), with the same measure \(\mu^\#\) as in Theorem 1. \(\mu^\#\) is called the reduced limit of \(\{\mu_k\}\) if there is a sequence of solutions \(\{v_k\}\) of (1) converging to a solution \(v^\#\) in \(L^1(\Omega)\), as in Theorem 2. Theorem 3. If \(\{\mu_k\}\subset{\mathcal G}(g)\) has reduced limit \(\mu^\#\) and \(\mu_k \geq 0\) for all \(k\), then \(\mu^\#\geq 0\). In the special case \[ -\Delta u + |u|^{q-1}u = \mu \quad\text{in}\quad \Omega \leqno(3) \] in the supercritical range \(q\geq N/(N-2)\), it is known that (3) has a solution if and only if \(\mu\in L^1(\Omega)+ W^{-2,q}(\Omega)\). Furthermore, if \(\{\mu_k\}\) is a bounded sequence converging strongly in \(W^{-2,q}(\Omega)\), then \(\mu^\#=\mu\). If \(\{\mu_k\}\) is just bounded in \(W^{-2,q}(\Omega)\), it can happen that \(\mu^\#=0\) even if \(\mu_k \buildrel * \over \rightharpoonup \mu \neq 0\). But this cannot happen for \(g(t)=|t|^{q-1}t\). Theorem 4. If \(\{\mu_k\}\) is a nonnegative sequence with \(\mu_k \buildrel * \over \rightharpoonup \mu\) and reduced limit \(\mu^\#\), and \(\{\mu_k\}\) is bounded in \(W^{-2,q}(\Omega)\), then \(\mu^\#=0\) if and only if \(\mu=0\). Under the assumptions of Theorem 4, equation (3) has a solution for all \(\mu\), but surprisingly, the authors provide a counterexample showing that it is not always true that \(\mu=\mu^\#\).