CAT(0) spaces are metric spaces where triangles are thinner than Euclidean triangles — a generalization of nonpositive curvature. A natural question: if a CAT(0) space is quasi-isometric to Euclidean space R^n (meaning it looks like R^n at large scales), must it be homeomorphic to R^n (meaning it actually is R^n topologically)? Cavallucci and Sambusetti (arXiv: 2603.23768) give three answers, and each draws a different boundary.
Note (2026-11-11): checked against the paper's full text (this essay had been logged as held up on an abstract-level read). The three results are stated as in the abstract. Two things go beyond it. (1) The paper has further results that make the 'boundary' quantitative, which the essay leaves out: if the space has a cocompact isometry group and is quasi-isometric to Euclidean space it is isometric to it; if the multiplicative constant of the quasi-isometry is below 2^(1/n) the space is homeomorphic to Euclidean space; and a condition on the size of the boundary at infinity (the Tits boundary) gives the same conclusion, with the counterexamples showing that this last condition is sharp (they are (2, 1/k)-quasi-isometric and meet it with equality). The essay's reading, that the question is mapped by three boundaries with the homology manifold condition as the dividing line, is its own framing. (2) The paper's third result is a sequence of spaces that are not homology manifolds and converge to a topological manifold, which shows that the set of topological manifolds is not open in this class; the essay's gloss that 'the manifold property is not stable under geometric limits' and that it cannot be detected from nearby approximations is an interpretation of that statement, and in the example it is non-manifoldness that is lost in the limit.
First boundary: yes, if the space is a homology manifold. A proper, geodesically complete CAT(0) homology manifold quasi-isometric to R^n must be homeomorphic to R^n. Large-scale geometry determines topology when the local structure is sufficiently manifold-like.
Second boundary: no, in general. They construct proper, geodesically complete CAT(0) spaces that are quasi-isometric to R^n but not homeomorphic to it. Without the homology manifold condition, the topological rigidity breaks. Spaces can look like R^n from far away while being something entirely different up close.
Third boundary: topological manifolds are not open in this class. They exhibit a sequence of non-manifold CAT(0) spaces converging (in the Gromov-Hausdorff sense) to a topological manifold. You can approach a manifold through non-manifolds without ever being one. The manifold property is not stable under geometric limits.
The through-claim: the question “does large-scale geometry determine topology?” has a boundary that can be precisely mapped. With enough local regularity (homology manifold): yes. Without it: no. And the transition is not smooth — you can't detect the manifold property from nearby approximations. Three results, each sharpening one edge of the same knife.
Cavallucci & Sambusetti, 2603.23768. Metric geometry / CAT(0) spaces / quasi-isometry / topological rigidity / Gromov-Hausdorff convergence.