Characterization of \(f\)-extremal disks (Q2423241)

Given a continuously differentiable function \(f : (0,+\infty) \to \mathbb{R}\), a classical result by \textit{J. Serrin} [Arch. Ration. Mech. Anal. 43, 304--318 (1971; Zbl 0222.31007)] asserts that if there exists a solution \(u\) of the overdetermined problem \[ \begin{cases} \Delta u + f(u) = 0,\ u > 0 &\text{in }\Omega, \\ u = 0 &\text{on }\partial \Omega \\ u_\eta = \text{const.}& \text{on }\partial \Omega \end{cases} \tag{1} \] in a smooth, bounded domain \(\Omega \subset \mathbb{R}^n\), then \(\Omega\) is a ball. Here \(u_\eta\) denotes the directional derivative of \(u\) with respect to the outward normal \(\eta\) to \(\partial\Omega\). Serrin's result followed \textit{A. D. Aleksandrov}'s \textit{soup bubble} theorem (see [Transl., Ser. 2, Am. Math. Soc. 21, 341--354, 354--388 (1962; Zbl 0122.39601)] and [Ann. Mat. Pura Appl., IV. Ser. 58, 303--315 (1962; Zbl 0107.15603)]) asserting that the only closed surface having constant mean curvature in \(\mathbb{R}^n\) is the sphere, and inspired a large number of subsequent contributions. Not to be confused with the Aleksandrov theorem is Hopf's theorem asserting that the only compact immersed surface of constant mean curvature and genus zero (a topological sphere) in \(\mathbb{R}^3\) is a round sphere, see p. 131 in [\textit{H. Hopf}, Differential geometry in the large. Berlin: Springer (1983; Zbl 0669.53001)]. Hopf's theorem deals with immersed surfaces, thus allowing for self-intersections. The statement ceases to hold in dimension \(n \ge 4\): see \textit{W.-Y. Hsiang} [J. Differ. Geom. 17, 337--356 (1982; Zbl 0493.53043)]. The statement fails even in \(\mathbb R^3\) if the assumption that the surface be of genus zero is dropped, see \textit{H. C. Wente} [Pac. J. Math. 121, 193--243 (1986; Zbl 0586.53003)]. In the paper under review, Serrin's result mentioned at the beginning is rephrased by means of the concept of an \textit{\(f\)-extremal domain}: the authors say that \textit{the only bounded \(f\)-extremal domain in\/ \(\mathbb{R}^n\) is the ball}. By taking unbounded domains into consideration, Berestycki, Caffarelli and Nirenberg raised the conjecture (referred to as the BCN conjecture) that ``if \(\Omega \subset \mathbb{R}^n\) is an \(f\)-extremal domain whose complement \(\mathbb{R}^n \setminus \Omega\) is connected, then \(\Omega\) is either a half-space, or a ball, or the complement of a ball, or an infinite cylinder (i.e., a domain isometric to the Cartesian product \(\mathbb{B}_r^k \times \mathbb{R}^{n - k}\), where \(\mathbb{B}_r^k\) is the canonical ball with radius \(r\) in \(\mathbb R^k\))'', see p. 1110 in [\textit{H. Berestycki} et al., Commun. Pure Appl. Math. 50, No. 11, 1089--1111 (1997; Zbl 0906.35035)]. Without any restriction on the function \(f\), there are counterexamples, see [\textit{A. Ros}, \textit{D. Ruiz} and \textit{P. Sicbaldi}, ``Solutions to overdetermined elliptic problems in nontrivial exterior domains'', Preprint, \url{arXiv:1609.03739}] for \(n = 2\) and [\textit{P. Sicbaldi}, Calc. Var. Partial Differ. Equ. 37, No. 3--4, 329--344 (2010; Zbl 1188.35122)] for \(n \ge 3\). The paper under review investigates the validity of the BCN conjecture in the case when the domain \(\Omega\) lies in a Riemannian manifold \(M\). More precisely, the authors prove a rather general result (Theorem A) concerning the construction of a suitable bilinear form \(Q\), which is associated to any solution \((u,\Omega)\) of Problem (1) and enjoys useful properties. In order to construct the bilinear form \(Q\), the authors need to assume the existence of a family of \textit{candidate solutions}: roughly speaking, they assume that Problem (1) has a solution \(v_{q,w,a}\) in a convenient family of domains \(\Omega_{q,w,a}\). Then, for every \(x \in \Omega\), the bilinear form \(Q\) at \(x\) is defined (on p. 2070) as the difference between the Hessian \(\nabla^2 u(x)\) of the solution \(u\) of Problem (1) and the Hessian \(\nabla^2 v_{x,\nabla u(x),u(x)}(x)\) of a particular member \(v_{x,\nabla u(x),u(x)}\) of the family of \textit{candidate} solutions. One of the properties of \(Q\) is that the condition \(Q \equiv 0\) holds if and only if \(u\) belongs to a special subset of the family of candidate solutions. In this sense, the bilinear form \(Q\) plays the role of a \(P\)-function (such functions owe their name to the contributions of L. E. Payne, see \textit{R. P. Sperb} [Maximum principles and their applications. New York etc.: Academic Press (1981; Zbl 0454.35001)]). In order to apply Theorem A, the construction of a family of \textit{candidate solutions} is needed. This is achieved in the special case when the manifold \(M\) is the two-dimensional sphere \(\mathbb{S}^2 \subset \mathbb{R}^3\). As a consequence, under some restrictions on the function \(f\) (see (H) on p. 2064), the authors extend Serrin's theorem to the case when \(\Omega\) is a simply connected domain in \(\mathbb{S}^2\): this case is not included in the results by \textit{G. Ciraolo} and \textit{L. Vezzoni} in [Springer Proceedings in Mathematics \& Statistics 176, 87--96 (2016; Zbl 1356.35150)], where Serrin's problem on manifolds is also considered. Moreover, since the function \(f(u) = \lambda u\) satisfies condition (H), the authors are able to prove \textit{the Schiffer conjecture D for the first eigenvalue} of the Laplacian in \(\mathbb{S}^2\). The two Schiffer conjectures (D and N) are discussed in [\textit{V. E. Shklover}, Rev. Mat. Iberoam. 16, No. 3, 529--569 (2000; Zbl 0976.35015)]. Further results on the Schiffer conjecture N in the Euclidean space are found in [\textit{B. Kawohl} and \textit{M. Lucia}, ``Some results related to Schiffer's problem'', J. Anal. Math. (to appear)].