Demystifying Latschev's theorem: manifold reconstruction from noisy data (Q6926513)

A theorem of \textit{J. Latschev} [Arch. Math. 77, No. 6, 522--528 (2001; Zbl 1001.53026)] guarantees that for every closed Riemannian manifold \(M\) there is a sufficiently small scale \(\epsilon > 0\) such that the for every \(0 <\beta\leq \epsilon\) there is constant \(\delta>0\) so that if \(S\) is a metric space within Gromov-Hausdorff distance \(\delta\) from \(M\), then the Vietoris-Rips complex \(\mathcal{R}_\beta(S)\) approximating \(S\) at scale \(\beta\) is homotopy equivalent to \(M\). This result is seen as a corner stone in topological data analysis, because \(S\) can be interpreted as a noisy sampling of a manifold \(M\), and so the theorem ensures that (the homotopy type of) \(M\) can always be recognised by using a sufficiently fine sampling. However, the proof of Latschev's Theorem does not provide explicit estimates for the constants involved, which makes it of limited use in concrete applications. The paper under review puts a remedy to this, by proving a version of this result where the constants \(\beta\) and \(\delta\) are explicitly controlled in terms of geometric data, namely the sectional curvatures and convexity radius of \(M\).\N\NThe strategy of proof is roughly as follows. Since \(M\) is homotopy equivalent to its own Vietoris-Rips complex (at any scale) it suffices to find explicit scales \(\beta\) and \(\xi\) so that the \(\mathcal{R}_\xi(M)\) and \(\mathcal{R}_\beta (S)\) are homotopy equivalent. The advantage of looking at Vietoris-Rips complexes is that it is easy to define maps between them: one only needs to define the images of the vertices and then extend by linearity (which can be done if the scales are chosen judiciously).\NIn particular, the author explains how to construct one such map \(\phi\) that yields a homotopy equivalence. The proof that \(\phi\) is a homotopy equivalence relies on Whitehead's Theorem, which states that it is enough to verify that \(\phi\) induces isomorphisms of all homotopy groups. In turn, the main technical difficulty turns out to be in proving that \(\phi_*\) is surjective on the \(n\)-th homotopy group (Lemma 4.5). To do this, the author constructs ``by hand'' for every simplicial map \(\mathbb{S}^n\to \mathcal{R}_\beta(S)\) a preimage for it in the Vietoris-Rips complex of \(M\). This is done by using circumcenters and an extended version of \textit{H. Jung}'s theorem [J. Reine Angew. Math. 123, 241--257 (1901; JFM 32.0296.05)] which allows to control how circumcenters vary. This is a key place where the geometric information is used to produce explicit estimates on the constants that can be used in the reconstruction process.\N\NIn a similar vein, the author also produces estimates on the constant that can be used to reconstruct a submanifold \(M\subset \mathbb{R}^n\) from a subset \(S\subset\mathbb{R}^n\) that is sufficiently close to \(M\) in the Hausdorff distance. Here, instead of sectional curvature and convexity radius one may use the reach of \(M\) (Definition 4.1).\N\NThe paper is well written and mostly self-contained. It is certainly a recommended read for the interested reader.