Dimensionality reduction for \(k\)-distance applied to persistent homology (Q2063202)

While persistent homology is known to enjoy stability properties with respect to small-scale perturbations of an underlying distance function, it is known to be susceptible to the existence of large-scale `outliers'. In practice, this issue is often addressed by making the distance function more robust, leading to functions such as the \textit{distance to a measure} function or the \textit{\(k\)-distance}~(for discrete sets). Given the high complexity of calculating such distances in high dimensions, it is necessary to study to what extent they are preserved under dimensionality reduction algorithms. A result by \textit{D. R. Sheehy} [in: Proceedings of the 30th annual symposium on computational geometry, SoCG '14, Kyoto, Japan, June 8--11, 2014. New York, NY: Association for Computing Machinery (ACM). 328--334 (2014; Zbl 1397.68207)] shows that, given any \(\epsilon \in [0, 1]\), persistent homology can be approximately preserved under random projections, up to a multiplicative factor of (\(1 \pm \epsilon\)). This opens the door towards analysing the persistent homology of high-dimensional point clouds based on lower-dimensional embeddings. A long-standing question, raised by Sheehy in the aforementioned paper, was whether the \(k\)-distance itself would also be preserved in a similar fashion under random projections. This paper answers the question in the affirmative, showing that Čech filtrations based on the \(k\)-distance are preserved under any linear mapping. In addition to proving this result, the paper also presents additional extensions and implications of it, such as improved bounds for the preservation of distances in the case of low-dimensional submanifolds of a high-dimensional Euclidean space.