Strong maximum principle for mean curvature operators on subRiemannian manifolds (Q1627782)

The following extension of Bony's strong maximum principle (cf. [\textit{E. Battaglia} et al., Ann. Inst. Fourier 66, No. 2, 589--631 (2016; Zbl 1353.35086)]) to a class of quasi elliptic equations \(Q\): \[ Q\phi=a^{ij}(x,D\psi)D_{ij}\phi+b(x,\phi,D\phi) \] on a domain \(\Omega \subset \mathbb{R}^{m+1}\), where \((a^{ij})\) is assumed to be nonnegative, \(a^{ij}=a^{ji}\), \(\frac{\partial b}{\partial z}\leq 0\), and \(b=b(x,z,p)\), is given: \textbf{Theorem A}. Assume the existence of vector fields \(X_1,\dots,X_r\) and \(Y\) of class \(C^\infty\) depending on \(D\phi(x)\) such that \[ a^{ij}(x, D\phi(x))D_{ij}=\sum_{i=1}^rX_i^2+Y. \tag{1} \] Let \(\Gamma\) be an integral curve of a vector field \(Z\in\mathcal{L}(X_1,\dots,X_r)\), the Lie algebra generated by \(X_1,\dots,X_n\). Suppose \(\phi-\psi\) achieves a nonnegative maximum in \(\Omega\) at a point \(\Gamma\), then the maximum is obtained at all points of \(\Gamma\). When \(\Omega\) is equipped with a sub-Riemannian metric, the \(p\)-mean curvature operator and the \(p\)-sub Laplacian belong this to class. This paper devoted to give refinements and extensions of Theorem A for quasi elliptic operators on sub-Riemannian manifolds. As an application, non-existence of horizontal \(p\)-minimal hypersurfaces properly immersed in a domain of \(H_n\), \(n\geq 2\), the Heisenberg group of dimension \(m=2n+1\) is proved (pseudo halfspace theorem, proved in \S7). Other applications are also given. Since this paper is working on sub-Riemannian manifolds, sub-Riemannian geometry is reviewed in \S2 from the viewpoint of differential forms. The inner product of a sub-Riemannian manifold \(M\) is denoted by \(\langle u,v\rangle^\ast\). The sub-Riemannian cotangent bundle \(T^*M\) is introduced by \(\langle v,v\rangle^\ast\geq 0\). The cotangent bundle is denoted by \(\xi\). Then by using the sub-Riemannian area, the generalized mean curvature is introduced (Proposition 2.1). These notions are explicitly exposed when \(M=H_n\), the Heisenberg group of dimension \(m=2n+1\). Theorem A is proved in \S3 adopting the idea in [\textit{D. Gilbarg} and \textit{N. S. Trudinger}, Elliptic partial differential equations of second order. 2nd ed. Berlin: Springer (1983; Zbl 0562.35001)] (Proof of Theorem 10.1). Then it is shown that condition (1) of Theorem A can be replaced by the constant rank condition of \((a^{ij}(x,D\phi(x))\) (Theorem \(\tilde{\mathrm{A}}\) cf. (3.3)). The authors say that Theorem \(\tilde{\mathrm{A}}\) is more practical than Theorem A. By this theorem and from the study of graphs under symmetry (\S4. Theorem C), introducing the horizontal (or \(p\)-) mean curvature \(H_{\overrightarrow{F}}(u)=\operatorname{div}N_{\overrightarrow{F}}(u)\), where \(N_{\overrightarrow{F}}(u)\) is the horizontal normal, the following Theorem is proved (\S5): \textbf{Theorem F}. Suppose \(m\geq 3\), \(H_{\overrightarrow{F}}(u)\geq H_{\overrightarrow{F}}(v)\) in \(U\subset \mathbb{R}^m\), which is a nonsingular domain for both \(u\) and \(v\), and \(v=u\) at \(p_0\in U\). Assume that in \(U\) an orthonormal basis \(N_1^\perp(u)\cdot N_2^\perp(u), \dots,N_{m-1}^\perp(u)\) of the space perpendicular to \(N_{\overrightarrow{F}}(u)\) exists and the rank of \(\mathcal{L}(N_1^\perp(u), \dots,N_{m-1}^\perp(u))\) is constant \(m\). Then \(u=v\) on \(U\). Some alternative forms of Theorem F, which are convenient for the study of Heisenberg groups, are also given in \S5 (Corollaries G and H). \S6 studies the case when \(p_0\in U\) is an isolated singular point of \(u\) and \(v\), and extends Theorem F to this case (Theorems J and J'). via extending Theorem C' in [Zbl 0562.35001] (\S6, Theorem C''). Applications of these results to the geometry of Heisenberg groups are described in \S7, the last section. As sub-Riemannian manifolds, the contact form \(\Theta\) of \(H_n\) is \(dz+\sum_{j=1}^n(x^jdx^{n+j}-x^{n+j}dx^j)\) and the distance function \(\rho\) of \(H_n\) is \([(\sum_{k=1}^{2n}(x^k)^2)^2+4z^2]^{1/4}\) (cf. \S2). Then the following theorems are proved. \textbf{Theorem L}. Let \(\Sigma\) be a closed immersed hypersurface in a Heisenberg cylinder \((H_n\setminus\{0\},\rho^{-2}\Theta)\) with \(n\geq 2\). Suppose either that the horizontal \(p\)-mean curvature satisfies \(H_\Sigma\leq 0\), or \(H_\Sigma\geq 0\) and the interior region of \(\Sigma\) contains the origin of \(H_n\). Then \(\Sigma\) must be a Heisenberg sphere defined by \(\rho^4=c\) for some constant \(c>0\). In particular, \(H_\Sigma=0\). \textbf{Corollary M}. There does not exist a closed immersed hypersurface of positive constant \(p\)-mean curvature in a Heisenberg cylinder with \(n\geq 2\), whose interior region contains the origin. \textbf{Theorem N} (Pseudo-half space theorem). Let \(\Omega\) be a domain of \(H_n\), \(n\geq 2\), defined by either \(z>\phi(\sqrt{x_1^2+\dots+x_n^2})\) or \(x_1>\phi(\sqrt{x_2^2+\dots+x_n^2+z^2})\), where \(\lim_{\tau\to \infty}\phi(\tau)=\infty\). Then there does not exist any horizontal \(p\)-minimal hypersurface properly immersed in \(\Omega\). The authors say that the simplest example of \(\phi\) is \(\phi(x)=cx\), \(c>0\). But the half space theorem does not hold if \(c=0\), since there are catenoid-type horizontal hypersurfaces with finite heights in \(H_n\), \(n\geq 2\) [\textit{M. Ritoré} and \textit{C. Rosales}, J. Geom. Anal. 16, No. 4, 703--720 (2006; Zbl 1129.53041)]. On the other hand, the half space theorem for \(H_1\) is proved in [\textit{J.-H. Cheng} and \textit{J.-F. Hwang}, Bull. Aust. Math. Soc. 70, No. 3, 507--520 (2004; Zbl 1062.35046)].