Boundary value problems for some fully nonlinear elliptic equations (Q2641910)

The article considers a Yamabe-type problem on locally flat compact manifolds with boundary. The main technique used is the derivation of boundary \(C^{2}\) estimates directly from boundary \(C^{0}\) estimates by controlling the third derivatives on the boundary instead of constructing a barrier function. Let \((M,g)\) be a Riemannian manifold of dimension \(n\geq 3\) with boundary \(\partial M\). We denote the Ricci curvature by Ric, the scalar curvature by \(R\), the boundary mean curvature by \(h\) and the second fundamental form of the embedding \(\partial M\hookrightarrow M\) by \(L_{\alpha\beta}\). The Yamabe problem for manifolds with boundary is to find a conformal metric \(\widehat{g}=e^{-2u}g\) such that the scalar curvature is constant and the mean curvature is zero. The boundary is called umbilic if there exists \(\mu_{g}\) such that the second fundamental form \(L_{\alpha\beta}=\mu_{g}g_{\alpha\beta}\). For example, a totally geodesic boundary is umbilic with zero principal curvatures. In [J. Differ. Geom. 35, No. 1, 21--84 (1992; Zbl 0771.53017)] it is proved by \textit{J. F. Escobar} that for locally flat compact manifolds with umbilic boundary (and some other cases) the Yamabe problem is solvable. As for the non-linear version of the Yamabe problem, it is considered the Schouten tensor defined as \[ A_{g}=\frac{1}{n-2}\left(\text{Ric}-\frac{R}{2(n-1)}g\right). \] The Schouten tensor comes from the curvature decomposition \(\text{Riem}= W+A\odot g\), where the Weyl tensor \(W\) is conformally invariant and \(\odot\) stands for the Kulkarni-Nomizu tensor. In dimension four, the following formula works for closed manifolds: \[ 32\pi^{2}\chi(M)=\int_{M}| W|^{2} + 4\int_{M}\sigma_{2}(A_{g}), \tag{1} \] where \(\chi\) is the Euler characteristic and \(\sigma_{2}(A_{g})\) is the second elementary symmetric function of the eigenvalues of \(A_{g}\). It follows from the formula ~(1) that the term \(\int_{M}\sigma_{2}(A_{g})\) is a conformal invariant. Moreover, \(\text{tr}(A_{g})=\frac{1}{2(n-1)}R\). For closed 4-manifolds, it has been proved by \textit{S.-Y. Chang, M. J. Gursky} and \textit{P. Yang} [J. Anal. Math. 87, 151--186 (2002; Zbl 1067.58028) and Ann. Math. 155, No. 3, 709--787 (2002; Zbl 1031.53062)] that if the Yamabe constant and \(\int_{M}\sigma_{2}(A_{\widehat{g}})\) are both positive, then we can find a conformal metric \(\widehat{g}\) such that \(\sigma_{2}(A_{g})\) is constant. The same sort of result has been proved by \textit{A. Li} and \textit{Y. Y. Li} in [Commun. Pure Appl. Math. 56, No. 10, 1416--1464 (2003; Zbl 1155.35353)] and \textit{P. Guang} and \textit{G. Wang} in [J. Reine Angew. Math. 557, 219--238 (2003; Zbl 1033.53058)] for locally conformally flat closed manifolds under the hypothesis that \(\sigma_{i}(A_{g})>0\) if \(1\leq i\leq k\) and \(k\geq 2\). For closed manifolds which are not conformally flat, Gursky-Viaclovsky proved the result for \(2k>n\). The authors studied the non-linear version of the Yamabe problem for manifolds with boundary. Let \(W\) be a matrix with eigenvalues \(\lambda_{1},\dots, \lambda_{n}\). Then the sum \(\sigma_{k}(W)=\sum_{i_{1}<\dots i_{k}}\lambda_{i_{1}}\lambda_{i_{2}}\dots \lambda_{i_{k}}\), for \(k\leq n\) is called the \(k^{th}\) elementary symmetric function of the eigenvalues of \(W\). Denote \(\sigma_{0}=1\). Suppose the boundary is umbilic. The goal of the paper to is to find a conformal metric \(\widehat{g}\) such that \(\sigma_{k}(A_{g})\) is constant and the boundary is totally geodesic. In order to define a class of locally flat compact manifolds of dimension \(n\geq 3\) with boundary, for which an affirmative answer to the question is achieved, the author introduces the following technical definition: The set \(\Gamma^{+}_{k}=\{\text{the connected component of}\sigma_{k}(\lambda)>0\text{ which contains the identity}\}\) is called the positive \(k\)-cone. Equivalently, it is shown that \(\Gamma^{+}_{k}=\{\lambda\mid \sigma_{i}(\lambda)>0,\;1\leq i\leq k\}\) is an open convex cone with vertex at the origin, e.g., \(\gamma^{+}_{1}=\{\lambda\mid \lambda_{1}+\dots +\lambda_{n}>0\}\) and \(\Gamma^{+}_{n}=\{\lambda\mid \lambda_{i}>0,\;1\leq i\leq n\}\). The following is the nested relation \[ \Gamma^{+}_{1}\supset \Gamma^{+}_{2}\supset \dots \supset\Gamma^{+}_{n}. \] Denote \(W\in \Gamma^{+}_{k}\) if the eigenvalues \(\lambda(W)\in \Gamma^{+}_{k}\). Under the conformal change of metric \(\widehat{g}=e^{-2u}g\), the new tensors carry a \textit{hat} (\(\widehat{A}\), \(\widehat{L}\) and \(\widehat{\mu}\)). The Schouten tensor becomes \[ \widehat{A}=\nabla^{2}u + du\oplus du - \frac{1}{2}| \nabla u|^{2}g +A_{g}, \] where the derivatives are covariant derivatives with respect to the metric \(g\). The second fundamental form satisfies \[ \widehat{L}e^{u}=\frac{\partial u}{\partial n}g+L_{g}, \] where \(n\) is the orthogonal unitary vector field along \(\partial M\). Since umbilicity is conformally invariant, it is natural to consider the class of manifolds with umbilical boundary. When the boundary is umbilic, the above formula becomes \[ \widehat{\mu}e^{-u}=\frac{\partial u}{\partial n} +\mu_{g}. \] Thus, the problem becomes the following equation: \[ \sigma^{1/k}_{k}\left(\nabla^{2}u + du\oplus du - \frac{1}{2}| \nabla u|^{2}g +A_{g}\right)=e^{-2u}, \;\text{in \(M\)}\tag{2} \] \[ \frac{\partial u}{\partial n}+\mu_{g}=0, \;\text{in \(\partial M\)}.\tag{3} \] Theorem: Suppose \((M,g)\) is a locally conformally flat compact manifold of dimension \(n\geq 3\) with umbilic boundary. If \(A_{g}\in \Gamma^{+}_{k}\) fr \(k\geq 2\) and \(\mu\geq 0\), then there exists a smooth solution \(u\) satisfying equations (2)--(3). In other words, there is a conformal metric \(\widehat{g}=e^{-2u}g\) such that \(\sigma_{k}(\widehat{A})=1\) and the boundary is totally geodesic. The paper extends the theorem to the case below \[ F\left(\nabla^{2}u + du\oplus du - \frac{1}{2}| \nabla u|^{2}g +A_{g}\right)= e^{-2u}, \;\text{in \(M\)}\tag{4} \] \[ \frac{\partial u}{\partial n}+\mu_{g}=0, \;\text{in \(\partial M\)}.\tag{5} \] where \(F\) satisfies some structure conditions as listed next: Let \(\Gamma \) be a open convex cone in \(\mathbb R^{n}\) with vertex at the origin satisfying \(\Gamma^{+}_{n}\subset \Gamma\subset\Gamma^{+}_{1}\). Suppose that \(F(\lambda)=F(\sigma_{1},\dots,\sigma_{n}(\lambda))\in C^{\infty}(\Gamma\cap C^{0}(\overline{\Gamma})\) is a homogeneous symmetric function of degree one normalized with \(F(E)=F(1,\dots,1)\). Assume that \(F=0\) on \(\partial\Gamma\) and \(F\) satisfies the following in \(\Gamma\) (S0) \(F\) is positive; (S1) \(F\) is concave (i.e., \(\frac{\partial^{2} u}{\partial\lambda_{i}\partial\lambda_{j}}\) is negative semi-definite); (S2) \(F\) is monotone (i.e., \(\frac{\partial F}{\partial \lambda_{i}}\) is positive); (S3) \(\frac{\partial F}{\partial \lambda_{i}}\geq \varepsilon \frac{\partial u}{\partial \sigma_{1}}\), for some constant \(\varepsilon >0\), for all \(i\). In some cases, it is used the following additional condition: {(A)} \(\sum_{j\neq i}\frac{\partial F}{\partial \lambda_{j}}\leq \rho\frac{\partial F}{\partial \lambda_{i}}\), for some \(\rho>0\), for all \(\lambda\in\Gamma\) with \(\lambda_{i}\leq 0\). As an example, the authors give \(f=\frac{1}{n}\left(\lambda_{1}+\dots +\lambda_{n}\right)\) with \(\Gamma=\{\lambda\mid\;\lambda_{1}+\dots +\lambda_{n}>0\). Suppose \((M,g)\) is a locally conformally flat compact manifold of dimension \(n\geq 3\) with umbilical boundary. Let \(F\) satisfy the structure conditions (S0)--(S3) in a corresponding cone \(\Gamma\). If \(A_{g}\in\Gamma\) and \(\mu_{g}\geq 0\), then there exists a somooth solution \(u\) of the equations (4)--(5).