A maximum principle for sums of subharmonic functions, and the convexity of level sets (Q1124008)

Let (x,y) denote a typical point of \({\mathbb{R}}^ n\times {\mathbb{R}}^ n\), and let V denote an open neighbourhood of (0,0). The following is proved. Theorem 1.4. Let \(\rho (x,y)=\rho_ 1(x)+\rho_ 2(y)\) on V, where \(\rho_ 1,\rho_ 2\) are harmonic functions, \(\rho_ 1(0)=\rho_ 1(0)=0\), and \((\nabla \rho_ 2)(0)\neq 0\). Also, let \(h(x,y)=h_ 1(x)+h_ 2(y)\) in V, where \(h_ 1,h_ 2\) are \(C^ 2\) functions satisfying \(\Delta h_ i>0\). Then the restriction of h to \(\{\rho =0\}\) does not have a local maximum at (0,0). This ``maximum principle'' is then used to give a new proof of the following known result. Theorem I. Let \(\Omega =W\setminus K\), where W is a bounded convex open subset of \({\mathbb{R}}^ n(n\geq 2)\) and K is a compact convex subset of W. If u is continuous on \({\bar \Omega}\), harmonic on \(\Omega\), valued 0 on \(\partial K\) and 1 on \(\partial W\), then every level set of u in \(\Omega\) is a strictly convex hypersurface. It is also shown that the level sets of u are ``at least as convex'' as are \(\partial W\) and \(\partial K\). Two plausible generalizations of Theorem I (involving Riemannian metrics) are shown not to hold.