Interleaving by Parts: Join Decompositions of Interleavings and Join-Assemblage of Geodesics

TL;DR

This paper introduces a novel way to decompose interleaving distances in topological data analysis using join representations of poset maps, enabling new insights into geodesics, distances, and metrics.

Contribution

It presents a join-based decomposition of poset maps that simplifies interleaving distances and extends metric concepts to broader topological structures.

Findings

Constructed geodesics between poset maps using join decompositions.

Extended Gromov-Hausdorff distance to simplicial filtrations over arbitrary posets.

Clarified relationships among various metrics in multiparameter hierarchical clustering.

Abstract

Metrics of interest in topological data analysis (TDA) are often explicitly or implicitly in the form of an interleaving distance $d_{\mathrm{I}}$ between poset maps (i.e. order-preserving maps), e.g. the Gromov-Hausdorff distance between metric spaces can be reformulated in this way. We propose a representation of a poset map $\mathbf{F}:\mathcal{P}\to\mathcal{Q}$ as a join (i.e. supremum) $\bigvee_{b\in B} \mathbf{F}_b$ of simpler poset maps $\mathbf{F}_b$ (for a join dense subset $B\subset \mathcal{Q}$) which in turn yields a decomposition of $d_{\mathrm{I}}$ into a product metric. The decomposition of $d_{\mathrm{I}}$ is simple, but its ramifications are manifold: (1) We can construct a geodesic path between any poset maps $\mathbf{F}$ and $\mathbf{G}$ with $d_{\mathrm{I}}(\mathbf{F},\mathbf{G})<\infty$ by assembling geodesics between all $\mathbf{F}_b$s and $\mathbf{G}_b$s via the join operation. This construction generalizes at least three constructions of geodesic paths that have appeared in the literature. (2) We can extend the Gromov-Hausdorff distance to a distance between simplicial filtrations over an arbitrary poset with a flow, preserving its universality and geodesicity. (3) We can clarify equivalence between several known metrics on multiparameter hierarchical clusterings. (4) We can illuminate the relationship between the erosion distance by Patel and the graded rank function by Betthauser, Bubenik, and Edwards, which in turn takes us to an interpretation on the representation $\bigvee_b \mathbf{F}_b$ as a generalization of persistence landscapes and graded rank functions.