The Laplace-Beltrami operator in almost-Riemannian geometry (Q2445511)

In the reviewed article, properties of geodesics, Laplacians, area elements and related notions of the so-called \(2\) dimensional almost Riemannian structures are investigated. A two dimensional almost Riemannian structure (2-ARS) is a real rank 2 vector bundle \(E\) over a smooth two dimensional manifold \(M\) which satisfies the following conditions. The vector bundle has to be equipped with a smoothly varying Euclidean structure defined in each of its fibers and also by a bundle morphism \(\mathfrak{f}: E \to TM\). The image of the push-forward \(\mathfrak{f}_\ast:\Gamma(E) \to \Gamma(TM),\) mapping the sections of \(E\) into vector fields on \(M\), have to generate the tangent bundle in each point of the manifold by the commutator of vector fields (Hörmander condition). A regular point on \(M\) is such a point over which the bundle morphism is an isomorphism and a singular point is a point on \(M\) which is not regular. The Euclidean structure on \(E\) can be pushed forward over any point in \(M\). In this way, one obtains a Riemannian metric in each regular points in \(M\). In a similar way, one transfers orthonormal frames in \(E\) over regular points in \(M\) obtaining fields over regular points which are declared to be orthonormal bases in the tangent spaces over these points. Further, one defines a metric structure \(d: M\times M \to \mathbb{R}\) on \(M\) and the energy functional. Then geodesics are defined as curves minimizing the energy functional. In general, they do not coincide with the geodesics for the Riemannian metric obtained over regular points in \(M\). Certain condition among a class of generic conditions, called \(\mathbf{H0}\), is introduced in the text. A classification theorem on types of points on \(2\)-ARS satisfying \(\mathbf{H0}\) is recalled. According to it, there are just three classes of points -- so-called ordinary, Grushin, and tangency type points. Suppose that an \(2\)-ARS over a compact orientable manifold satisfying \(\mathbf{H0}\) has no singular points of tangency type. The authors prove that there is a tubular neighborhood of any connected component of the set of singular points in which the orthonormal bases are described globally by a certain ``normal form'' which equals to the one describing orthonormal bases in local neighborhoods of Grushin points. The proof uses the Pontryagin maximum principle. During the investigation of the Grushin plane with one coordinate compactified, the authors give an explicit example of a geodesic passing through the set of singular points with no singularity. The Laplace-Beltrami operator is defined as the composition of the divergence with the gradient both derived from the Riemannian metric out of the set of singular points. The authors show that the Laplace-Beltrami operator is essentially self-adjoint. Modifying the metric in the Grushin plane, they give a one parameter family of generalized \(2\)-ARS whose Laplace-Beltrami operators are or are not essentially self-adjoint depending on the parameter. Several auxiliary theorems of independent interest are derived which also serve to prove the following theorem in which certain important properties of the Laplace-Beltrami operator are examined. Theorem. Let \(M\) be compact orientable \(2D\) manifold endowed with a \(2\)-ARS such that [\(\mathbf{HA}\)] the set of singular points is an embedded one-dimensional submanifold of \(M\) and [\(\mathbf{HB}\)] each point is an ordinary or a Grushin point Let \(d\omega\) be the corresponding Riemannian area and \(\triangle\) the Laplace-Beltrami operator both defined on the set \(M\setminus \mathcal{Z},\) where \(\mathcal{Z}\) denotes the set of singular points. Then \(\triangle\) with the domain \(C_0^{\infty}(M\setminus \mathcal{Z})\) is essentially self-adjoint on \(L^2(M, d\omega).\) The domain of \(\triangle\) is given by \(D(\triangle) = \{u\in L^2(M,d\omega)|\, \triangle u_{D,g} \in L^2(M,d\omega)\}\) where \(\triangle_{D,g} u\) is \(\triangle u\) seen as the distribution in \(M\setminus \mathcal{Z}\) The resolvent \((-\triangle + 1)^{-1}\) is compact and therefore its spectrum is discrete and consists of eigenvalues with finite multiplicity. In the case of a specific metric on \((\mathbb{R}\setminus 0) \times S^1,\) this theorem is applied to conclude on some qualitative aspects of solutions of the associated Schrödinger equation. Especially, the authors prove an interesting phenomenon namely, that there is no ``wall-crossing'' in contrast to the solutions of the Schrödinger equation corresponding to the ``sum of square'' Laplacian for which it is known that the solution can cross the barrier. The sum of square Laplacian is different from the Laplace-Beltrami operator in general. The authors compares the behaviour of the quantities derived from the Riemannian metric, such as the volume form, with the ones which uses the metric structure \(d,\) as the geodesics. In the appendix, the self-adjointness of the Laplace-Beltrami operator in the Martinet case (a specific sub-Riemannian structure in \(\mathbb{R}\times S^1 \times \mathbb{R})\) is proved.