Is magnitude ‘generically continuous’ for finite metric spaces?

Hirokazu Katsumasa, Emily Roff, Masahiko Yoshinaga · 2025

Read Is magnitude ‘generically continuous’ for finite metric spaces? by Hirokazu Katsumasa, Emily Roff, Masahiko Yoshinaga (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-08745.txt.

Introduction

Hirokazu Katsumasa, Emily Roff, and Masahiko Yoshinaga ask if magnitude is generically continuous in two thousand twenty-five. Magnitude is a number tied to a metric space. For finite spaces it reads as an effective count of points at a scale. Data analysis uses it, so its stability matters in practice.

They check continuity under the Gromov-Hausdorff view of space shape. Magnitude turns out to be nowhere continuous on that space. Yet signs hint it may be continuous in a generic sense. Limits drawn from most spaces seem to be kept by magnitude. The paper argues generic stability is what counts for use.

Catalog record

CREATORHirokazu Katsumasa, Emily Roff, Masahiko Yoshinaga
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-08745

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Is magnitude ‘generically continuous’ for finite metric spaces? Hirokazu Katsumasa, Emily Roff, Masahiko Yoshinaga Abstract . Magnitude is a real-valued invariant of metric spaces which, in the finite setting, can be understood as recording the ‘effective number of points’ in a space as the scale of the metric varies. Motivated by applications in topological data analysis, this paper investigates the stability of magnitude: its continuity properties with respect to the Gromov–Hausdorff topology. We show that magnitude is nowhere continuous on the Gromov–Hausdorff space of finite metric spaces. Yet, we find evidence to suggest that it may be ‘generically continuous’, in the sense that generic Gromov–Hausdorff limits are preserved by magnitude. We make the case that, in fact, ‘generic stability’ is what matters for applicability. 1. Introduction Magnitude is a real-valued isometric invariant of metric spaces, with wide-ranging connections to topics in geometry, analysis, combinatorics and category theory. It was first introduced by Leinster in [ 18 ] , where it was defined for finite metric spaces as follows. Given a space X , let Z_{X} denote the X\times X matrix with entries Z_{X}(x,y)=e^{-d(x,y)} ; this is called the similarity matrix of X . A weighting for X is a vector \bm{w}\in\mathbb{R}^{X} with the property that (Z_{X}\bm{w})(x)=1 for every x\in X . If a weighting \bm{w} exists, then the magnitude |X| is defined to be the sum of the entries in \bm{w} . (In the case that X carries more than one weighting, this value is independent of the choice of weighting used to compute it [ 18 , Lem. 1.1.2] .) For almost every finite metric space, a weighting does exist—indeed, in the generic case \det(Z_{X})\neq 0 , so there is a unique weighting on X given by \bm{w}(x)=\sum_{y\in X}Z_{X}^{-1}(x,y) . Thus, for a generic finite metric space X , we have (1.1) |X|=\sum_{x,y\in X}Z_{X}^{-1}(x,y). The magnitude of a metric space is analogous, in a precise sense, to the Euler characteristic of a topological space [ 17 ] . Unlike Euler characteristic, however, the magnitude of a space depends on the scale of the metric. Thus, given a metric space X=(X,d) , it is informative to consider the magnitudes of the scaled spaces tX:=(X,td) for each real number t>0 . The magnitude function of X is the partially defined function (1.2) \displaystyle\begin{split}(0,\infty)&\longrightarrow\mathbb{R}\\ t&\longmapsto|tX|.\end{split} Under favourable conditions—for instance, if X is a subset of Euclidean space—this function is in fact continuous, and satisfies \lim_{t\to 0}|tX|=1 and \lim_{t\to\infty}|tX|=\#X . In such cases, the magnitude function can be interpreted as recording the ‘effective number of points’ in X as the scale of the metric varies, and its instantaneous rate of growth as recording the ‘effective dimension’. For illustrative examples, see [ 19 , Ex. 6.4.6] and [ 31 , §4] . The definition of magnitude can be extended from finite metric spaces to a large class of compact metric spaces [ 23 , §2] , and in that setting it turns out to be remarkably rich in geometric content. The large-scale asymptotics of the magnitude function are of particular interest when interpreted for domains in Euclidean space [ 5 , 9 , 32 , 25 ] or manifolds with boundary [ 11 , 10 ] , capturing information about ‘size-related’ attributes such as volume, surface area, and integrals of curvatures. There is also an associated homology theory, magnitude homology , which has been related in valuable ways to graph-theoretic invariants arising in the context of discrete homotopy theory [ 14 , 4 , 13 , 16 ] . Though its geometric features are displayed most vividly in the world of compact metric spaces, there has been a recent resurgence of interest in the magnitude of finite spaces, both from the perspective of ‘pure’ metric geometry [ 27 , 26 , 12 ] and with a view to applications. Magnitude is increasingly being put to work as an invariant of data sets, in the context of topological data analysis (e.g. [ 1 , 28 ] ) and in machine learning (e.g. [ 2 , 15 , 22 ] ), with new techniques being developed to facilitate its efficient computation or approximation [ 3 ] . Applications such as these bring new theoretical questions into the foreground. In particular, for applicability in data analysis, the stability of geometric or topological invariants is important: small perturbations of the input data should lead to small changes, at most, in the value of an invariant [ 8 , Chap. VIII] . In other words, a good invariant should depend continuously on the input data. Indeed, within the topological data analysis literature, proving a ‘stability theorem’ for a given invariant is usually seen as a prerequisite for its adoption in practice. A stability theorem typically says that the invariant in question is Lipschitz continuous, or at least continuous, with respect to the Hausdorff or Gromov–Hausdorff metric on the relevant space of data (e.g. [ 7 , 6 , 30 ] ). Despite its increasingly widespread application, there is, to date, no stability theorem for magnitude. To describe what such a theorem should look like, let \operatorname{FMet} denote the set of isometry classes of finite metric spaces, equipped with the Gromov–Hausdorff metric, and let \operatorname{FMet}^{\circ}\subset\operatorname{FMet} be the subspace consisting of finite metric spaces for which magnitude is defined. A conventional stability theorem for magnitude might tell us that the mapping (1.3) \displaystyle\begin{split}\operatorname{FMet}^{\circ}&\longrightarrow\mathbb{R}\\ X&\longmapsto|X|\end{split} is continuous (or, better, Lipschitz continuous). Unfortunately, magnitude is not continuous in this sense: examples of discontinuities can be found in [ 18 , Ex. 2.2.7 and Ex. 2.2.8] and [ 29 , Ex. 3.4] . Yet, such examples seem to be rare—so rare, in fact, that intuition suggests one is vanishingly unlikely to encounter instances of magnitude’s instability in ‘the wild’ of real-world data. In this paper we take the point of view that, for the purposes of applicability, the generic behaviour of invariants is what one needs to understand, not their vanishingly rare pathologies. This seems to reflect actual practice in the domains where magnitude is being applied: for instance, in [ 22 , §3.2] , the assumption that \lim_{t\to 0}|tX|=1 is justified by the fact, proved in [ 29 , Thm. 2.3] , that this property holds generically—though not always—for finite metric spaces. We therefore pose the question: is magnitude generically stable as an invariant of finite metric spaces? In other words, is the function X\mapsto|X| generically continuous on \operatorname{FMet}^{\circ} ? Of course, the answer will depend on how one chooses to formulate the notion of ‘generic continuity’. For one natural formulation, we will prove definitively that magnitude is not generically continuous. For another natural formulation, we will provide strong evidence to suggest that it may be. Summary of results In the first formulation, a function is generically continuous if it is continuous at a generic point of its domain: Formulation A . Let A and B be topological spaces. A function f\colon A\to B is generically continuous if there exists a dense open subset U\subseteq A such that f is continuous at every point of U . Equivalently, f is generically continuous if there exists a nowhere-dense subset V\subset A such that f|_{A\backslash V}\colon A\backslash V\to B is continuous. In Section 2 we

Read the complete text