Table of Contents
Fetching ...

Universality of persistence diagrams and the bottleneck and Wasserstein distances

Peter Bubenik, Alex Elchesen

TL;DR

This work develops a category-theoretic foundation for persistence diagrams by treating them as free commutative monoids on metric pairs $(X,A)$ endowed with $p$-Wasserstein metrics $W_p$. The authors prove a universal property: persistence diagrams with $W_p$ are the universal $p$-subadditive commutative metric monoid generated by $(X,A)$, yielding a maximal, functorial, and stable distance framework that encompasses barcodes and multiparameter persistence. They establish a precise link between metric and monoid structures through $p$-subadditive monoids internal to symmetric monoidal categories, derive a left adjoint $D_p$ to the forgetful functor, and show that $W_p$ is the largest compatible metric extending the base metric. For $p=1$, Kantorovich-Rubinstein duality holds, connecting persistence diagrams to Lipschitz-function duality. Overall, the results provide a robust, universal, and computable approach to distances between persistence diagrams with broad applicability to various persistence models and stability analyses.

Abstract

We prove that persistence diagrams with the p-Wasserstein distance form the universal p-subadditive commutative monoid on an underlying metric space with a distinguished subset. This result applies to persistence diagrams, barcodes, and to multiparameter persistence modules. In addition, the 1-Wasserstein distance satisfies Kantorovich-Rubinstein duality.

Universality of persistence diagrams and the bottleneck and Wasserstein distances

TL;DR

This work develops a category-theoretic foundation for persistence diagrams by treating them as free commutative monoids on metric pairs endowed with -Wasserstein metrics . The authors prove a universal property: persistence diagrams with are the universal -subadditive commutative metric monoid generated by , yielding a maximal, functorial, and stable distance framework that encompasses barcodes and multiparameter persistence. They establish a precise link between metric and monoid structures through -subadditive monoids internal to symmetric monoidal categories, derive a left adjoint to the forgetful functor, and show that is the largest compatible metric extending the base metric. For , Kantorovich-Rubinstein duality holds, connecting persistence diagrams to Lipschitz-function duality. Overall, the results provide a robust, universal, and computable approach to distances between persistence diagrams with broad applicability to various persistence models and stability analyses.

Abstract

We prove that persistence diagrams with the p-Wasserstein distance form the universal p-subadditive commutative monoid on an underlying metric space with a distinguished subset. This result applies to persistence diagrams, barcodes, and to multiparameter persistence modules. In addition, the 1-Wasserstein distance satisfies Kantorovich-Rubinstein duality.

Paper Structure

This paper contains 20 sections, 34 theorems, 35 equations, 1 figure.

Key Result

Proposition 1.1

Given a metric pair $(X,d,A)$, $(D(X,A),W_p,+,0)$ is a $p$-subadditive commutative metric monoid and the canonical inclusion $i:(X,d,A) \to (D(X,A),W_p,0)$ is $1$-Lipschitz. Furthermore, this construction is functorial.

Figures (1)

  • Figure 1: For $1\leq p\leq q\leq \infty$, the relationships between the categories $\mathbf{Met_*^p}$, $\mathbf{Met_*^q}$, $\mathbf{CMon(Met_*^p)}$, $\mathbf{CMon(Met_*^q)}$, and $\mathbf{Met_{pairs}}$ via the functors $U_r, D_r, \overline{U}_r, \overline{D}_r$, $S_r$ ($r \in \{p,q\}$), and $Q, S_{p,q}$. The solid triangles commute and the dashed triangles commute.

Theorems & Definitions (94)

  • Proposition 1.1: Functorial construction of persistence diagrams with Wasserstein distance
  • Theorem 1.2: Persistence diagrams with the Wasserstein distance as an adjoint
  • Theorem 1.3: Universality of persistence diagrams with the Wasserstein distance
  • Corollary 1.4: Wasserstein distance as largest subadditive distance
  • Theorem 1.5: Symmetric monoidal structures for pointed metric spaces
  • Definition 2.1
  • Definition 2.2
  • Example 2.3
  • Example 2.4
  • Example 2.5
  • ...and 84 more