Hausdorff stability of persistence spaces (Q285432)

Persistence, in the classical case of a topological space \(X\) filtered in sublevel sets by a continuous function \(f:X \to\mathbb{R}\), can be represented by \textit{Persistent Betti Number} (PBN) functions, which are defined on a half-plane, with nonnegative integer values: For each point \((u,v)\) with \(u <v\) the corresponding \(k\)-PBN is the rank of the homomorphism from the \(k\)-homology module of the sublevel set under \(u\) to the one of the sublevel set under \(v\), induced by inclusion [\textit{H. Edelsbrunner} et al., Discrete Comput. Geom. 28, No. 4, 511--533 (2002; Zbl 1011.68152)]. It is well-known that PBN's can be reconstructed from -- and summarized by -- \textit{persistence diagrams} i.e. sets of points of the same half-plane, proper or at infinity (\textit{cornerpoints} and \textit{cornerlines} in the language of size functions, the forerunners of persistence [\textit{P. Frosini} and \textit{C. Landi}, Appl. Algebra Eng. Commun. Comput. 12, No. 4, 327--349 (2001; Zbl 0974.68229)]). When the range of the filtering function is multi-dimensional (an extension started in [\textit{P. Frosini} and \textit{M. Mulazzani}, Bull. Belg. Math. Soc. - Simon Stevin 6, No. 3, 455--464 (1999; Zbl 0937.55010)]), essential difficulties arise, as clearly explained in [\textit{G. Carlsson} and \textit{A. Zomorodian}, Discrete Comput. Geom. 42, No. 1, 71--93 (2009; Zbl 1187.55004)]. What takes the role of the points (0-dimensional objects in a 2-dimensional space) of persistence diagrams if the range is \(\mathbb{R}^n\)? It is a set of \((2n-2)\)-dimensional objects in a \(2n\)-dimensional space. The present paper defines the Persistence Space as this set -- generalizing the set of points of a persistence diagram -- through multiplicity, which in turn is computed from PBN's. The Multidimensional Representation Theorem (Theorem 3.12) grants that PBN's can be reconstructed from persistence spaces. The relationship between discontinuities of PBN's and homological critical points is then investigated. The main result of the article is the stability of persistence spaces, Theorem 5.1: The Hausdorff distance of persistence spaces relative to two filtering functions \(f, g: X \to \mathbb{R}^n\) is bounded from above by the maximum on \(X\) of the \(L_\infty\)-norm of \((f(x)-g(x))\). The possibility of cutting persistence spaces into persistence diagrams of suitable leaves of a foliation of \(\mathbb{R}^{2n}\) is finally recalled [\textit{S. Biasotti} et al., J. Math. Imaging Vision 32, No. 2, 161--179 (2008; \url{doi:10.1007/s10851-008-0096-z}); \textit{F. Cagliari} et al., Proc. Am. Math. Soc. 138, No. 8, 3003--3017 (2010; Zbl 1214.68424)].