Failure of curvature-dimension conditions on sub-Riemannian manifolds via tangent isometries (Q6175749)

Curvature-dimension inequalities give a synthetic notion of Ricci curvature, suitable for extending the classical notion of Ricci curvature bounds to more general metric-measure spaces. However, they are inadequate for the case of sub-Riemannian manifolds, since sub-Riemannian manifolds do not satisfy curvature-dimension conditions. In particular, various results in this direction have been proven over the years, by different methods, showing that certain classes of sub-Riemannian manifolds cannot satisfy curvature-dimension conditions for various ranges of the parameters. The main goal of this paper is to give a definite version of this fact, by unifying and extending the previous work to essentially full generality. Take a complete, possibly rank-varying, sub-Riemannian manifold \(M\) with \(g\) the metric on the distribution, and let \(\nabla\) denote the corresponding horizontal gradient. In addition, let \(\mathsf{m}\) be a smooth, positive measure on \(M\), and let \(\Delta\) be the corresponding sub-Laplacian, namely \(\Delta u = \mathrm{div}_{\mathsf{m}}\left( \nabla u\right)\) for any smooth function \(u\), where \(\mathrm{div}_{\mathsf{m}}\) denotes the divergence of a vector field with respect to \(\mathsf{m}\). For such a structure, consider the Bakry-Émery inequality \(\mathsf{BE}(K,\infty)\), which holds for some \(K\in \mathbb{R}\) if, for all smooth functions \(u\), \[ \frac{1}{2} \Delta\left( \|\nabla u\|^2 \right) \geq g\left( \nabla u,\nabla \Delta u\right) +K\| \nabla u\|^2 . \] It was recently shown, and the present paper gives a self-contained proof in the cases under consideration, that a sub-Riemannian manifold \(M\) as above is infinitesimally Hilbertian. Further, for any \(K\in \mathbb{R}\), the inequality \(\mathsf{BE}(K,\infty)\) is implied by \(\mathsf{CD}(K,N)\) for any \(N\in[1,\infty]\) on any infinitesimally Hilbertian metric-measure space. Here \(\mathsf{CD}(K,N)\) is the curvature-dimension condition giving a synthetic version of a lower bound of \(K\) for the Ricci curvature and an upper bound of \(N\) on the dimension, with the precise definition being given in terms of optimal transport and Wasserstein spaces. The main result of the paper is that if \(M\) satisfies \(\mathsf{BE}(K,\infty)\) for any \(K\in \mathbb{R}\), then \(M\) must be Riemannian. This result is proven by showing that the \(\mathsf{BE}(K,\infty)\) condition implies that the nilpotent approximation (which agrees with the metric tangent cone in the sense of Gromov) at every point admits a large enough space of isometries to force the tangent cone to be Euclidean, so that \(M\) is in fact Riemannian. In view of the infinitesimal Hilbertianity of \(M\), it further follows that any sub-Riemannian manifold (of the general type described above) that satisfies any \(\mathsf{CD}(K,N)\) condition is in fact Riemannian. This fundamental result is complemented by examples showing that similar results need not hold if the measure \(\mathsf{m}\) is allowed to vanish. In particular, putting a one-parameter family of measures on the Grushin plane, where the measure is allowed to vanish at the non-Riemannian points, provides examples of structures that satisfy the \(\mathsf{BE}(0,\infty)\) inequality almost everywhere. Taking half of such a weighted Grushin plane, namely a weighted Grushin half-plane, gives sub-Riemannian manifolds with boundary that satisfy the \(\mathsf{CD}(0,N)\) condition for large enough \(N\). Finally, these Grushin examples provide a counterexample to certain gluing results for \(\mathsf{RCD}\) metric-measure spaces.