Circumcenter extension maps for non-positively curved spaces (Q6120829)

It is a classical fact from hyperbolic geometry that isometries of hyperbolic space induce Möbius transformations (cross-ratio preserving maps) of the boundary at infinity, and that conversely every Möbius transformation extends to an isometry. This rather elementary result (and the notion of cross-ratio) has been generalized in many ways to more general spaces that have some flavor of (non-constant) negative curvature. In the setting of CAT(-1) spaces, the recent work of \textit{K. Biswas} [``Circumcenter extension of Moebius maps to CAT(-1) spaces'', Preprint, \url{arXiv:1709.09110}] proved that Möbius homeomorphisms between ideal boundaries can be extended to quasi-isometries. The present paper attempts to further generalize this to CAT(0)-spaces, which is possible under some restrictive assumptions. For a CAT(0) space \(X\), let \(\partial X\) be its boundary at infinity, consisiting of equivalence classes of endpoints of geodesics. To define a cross ratio on \(\partial X\), the author uses the Gromov product \((\xi\vert\eta)_x\), which roughly measures how long geodesics starting at \(x\) and ending at \(\xi\) resp.\ \(\eta\) stay ``close together''. Letting \(\rho_x=e^{-(\xi\vert\eta)_x}\), he defines the cross ratio \[ cr_x(\xi_1,\xi_2,\xi_3,\xi_4):=\frac{\rho_x(\xi_1,\xi_2)\rho_x(\xi_3,\xi_4)}{\rho_x(\xi_1,\xi_3)\rho_x(\xi_2,\xi_4)}\in\left[0,\infty\right]. \] This is defined for ``admissible quadruples'', that is, quadruples of points in \(\partial X\) with no \(\frac{\infty}{\infty}\) occuring in the above formula, and the author shows that the definition does not depend on the choice of \(x\in X\). A map is called Möbius if it preserves cross ratios and preserves the algebraically visility of pairs, where the latter means finiteness of the Gromov product. (This is implied by visibility, which means that the points are connected by an bi-infinite geodesic.) In this setting, the author generalizes several results known for CAT(-1) spaces to CAT(0) spaces. For the main results, the author has to assume some visiblity properties. One assumption is 4-visibility, which means, that for each quadruple \((\xi_1,\xi_2,\xi_3,\xi_4)\) there is some \(\eta\) with \((\xi_i,\eta)_x\not=\infty\) for \(i=1,2,3,4\). The other assumption is that all \(\xi\) are in a rank 1 hinge, which means, that there exist \(\eta,\sigma\) such that \((\eta\vert\sigma)_x<\infty\) and there are rank 1 geodesics between \((\xi,\sigma)\) and \((\eta,\sigma)\), respectively. Given a Möbius transformation \(f\colon\partial X\to\partial Y\) between ideal boundaries of Hadamard manifolds satisfying the above conditions, assuming in addition that \(f\) and \(f^{-1}\) preserve visibility (not just algebraical visibility) of pairs, the author constructs a ''circumcenter extension'' \(F\colon X\to Y\). He proves local \(\frac{1}{2}\)-Hölder continuity and local Lipschitz continuity of \(F\) on certain dense subsets of \(X\). Better results are obtained for 2-dimensional Hadamard manifolds or if one additionally assumes \(f\) to be equivariant with respect to some cocompact actions on \(X\) and \(Y\). In the latter case, the author proves that \(F\) is a quasi-isometry. In the 2-dimensional case, he shows that \(F\) is a homeomorphism and is differentiable almost everywhere.