Monoidal Rips: Stable Multiparameter Filtrations of Directed Networks

Nello Blaser, Morten Brun, Odin Hoff Gardaa, Lars M. Salbu · 2025

Read Monoidal Rips: Stable Multiparameter Filtrations of Directed Networks by Nello Blaser, Morten Brun, Odin Hoff Gardaa, Lars M. Salbu (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-2503-14019.txt.

Introduction

Nello Blaser, Morten Brun, Odin Hoff Gardaa, and Lars Salbu extend a key build in topological data analysis to networks that are directed and weighted. The Vietoris-Rips filtration builds simplicial complexes from metric data with a maximum rule. Their monoidal Rips filtration swaps that rule for a general monoidal product, which widens the networks it can take in.

The team prove interleaving bounds for the persistent homology that comes out, and they get back known stability results for networks with real-number values. With product lattices, the frame turns into multiparameter persistence, and they bound distances to match. A new stability result for the sublevel Rips bifiltration comes next. In tests they beat the Flagser tool on graph regression, and a mix of monoidal products helped sort point clouds.

Catalog record

CREATORNello Blaser, Morten Brun, Odin Hoff Gardaa, Lars M. Salbu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE103 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-14019

Browse the full library · About the Tower of Babel

Full text

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

Monoidal Rips: Stable Multiparameter Filtrations of Directed Networks Nello Blaser, Morten Brun, Odin Hoff Gardaa, Lars M. Salbu Abstract We introduce the monoidal Rips filtration, a filtered simplicial set for weighted directed graphs and other lattice-valued networks. Our construction generalizes the Vietoris-Rips filtration for metric spaces by replacing the maximum operator, determining the filtration values, with a more general monoidal product. We establish interleaving guarantees for the monoidal Rips persistent homology, capturing existing stability results for real-valued networks. When the lattice is a product of totally ordered sets, we are in the setting of multiparameter persistence. Here, the interleaving distance is bounded in terms of a generalized network distance. We use this to prove a novel stability result for the sublevel Rips bifiltration. Our experimental results show that our method performs better than flagser in a graph regression task, and that combining different monoidal products in point cloud classification can improve performance. 2020 Mathematics Subject Classification: 55N31, 62R40, 55U10. Keywords: Topological Data Analysis, Directed Networks, Multiparameter Persistence, Stability, Simplicial Sets 1 Introduction Many methods in topological data analysis (TDA) are based on persistent homology and focus on deriving topological descriptors from point cloud data. Common examples include filtrations derived from the Čech complex, the Vietoris-Rips complex and the alpha complex (see [ 16 ] for a comprehensive overview). A point cloud in a metric space can be represented as a complete weighted graph where the nodes are the points and the edge weights are determined by the metric. In this setting, the graph is symmetric and satisfies the triangle inequality. Other types of data can naturally be represented as (weighted directed) graphs that are not necessarily symmetric or satisfy the triangle inequality. Examples include social networks, citation networks, and biological networks. A (real-valued) network is a function G\colon V\times V\to P with P being a subset of the extended reals, typically \mathbb{R} or [0,\infty] . Methods for computing persistent homology of real-valued networks include directed flag complexes [ 42 , 34 , 11 ] , the directed Rips complex [ 45 , 39 ] and the Rips and Dowker filtrations for networks [ 19 ] . Another direction is path homology, first introduced in [ 29 ] . Path homology has been extended to persistent path homology [ 20 ] , including grounded and weighted versions [ 15 , 33 ] . Computational aspects of persistent path homology are discussed in [ 24 , 23 ] . For a review of homology-based methods for networks and their applications, see [ 2 ] . When working with point cloud data, simplicial complexes are typically used. While simplicial complexes are fundamental in TDA, their simplices—being sets—lack an ordering of vertices. Hence, they are not a natural choice for modeling directed structures, even though such directionality can in principle be encoded indirectly via higher-dimensional simplices. Directed simplicial complexes, which are used for the directed flag complex [ 42 ] , address this by introducing an ordering of the vertices in a simplex. They still disallow tuples with repeated vertices, which limits their expressiveness. For example, the cycle w=(u,v)+(v,u) cannot be the boundary of any 2 -chain, causing it to persist indefinitely. In contrast, simplicial sets allow repeated vertices, enabling w to be the boundary of the 2 -simplex (u,v,u) . This greater flexibility makes simplicial sets more suitable for analyzing directed structures. The \ell_{p} -Vietoris-Rips construction [ 30 ] , Cho’s nerve construction [ 17 ] and the directed Rips complex [ 45 ] are simplicial sets. Stability guarantees are an important part of the theoretical appeal of TDA because they guarantee robustness to noise. The Gromov-Hausdorff distance is a popular choice for measuring the distance between compact metric spaces, including point clouds. A fundamental stability result in TDA tells us that the interleaving distance between Čech (and Vietoris-Rips) persistent homology modules built from two point clouds is bounded in terms of the Gromov-Hausdorff distance between them. See, for example, [ 16 , Section 5.7] . The network distance [ 12 , 13 , 18 , 21 ] generalizes the Gromov-Hausdorff distance to real-valued networks. In [ 21 ] , the authors provide a detailed study of the space of infinite networks, assumed compact and first countable, equipped with the network distance. Their contributions include a characterization of its metric and isomorphism structure, extensions of persistent homology methods to this setting, and convergence results for clustering methods. Many of the previously mentioned methods for real-valued networks have stability guarantees with respect to the network distance [ 45 , 19 , 20 , 39 , 30 ] . Contributions In this paper, we do not restrict ourselves to real-valued networks. Rather, we consider L -graphs, which are functions G\colon V\times V\to L where V is a finite set and L is a complete lattice. Additionally, we require a commutative monoidal structure (L,\otimes,e) that is compatible with the lattice order in the following sense: • The product is functorial: s\otimes t\leq s^{\prime}\otimes t^{\prime} whenever s\leq s^{\prime} and t\leq t^{\prime} in L , and • the neutral element e is minimal in the sense that e\leq t for all t in L . If L=(L,\leq,\otimes,e) satisfies the above conditions, we call L a symmetric monoidal lattice . A recurring example of a symmetric monoidal lattice is [0,\infty] equipped with the usual order and the following monoidal structure: For 1\leq p<\infty , the p -sum +_{p}\colon[0,\infty]\times[0,\infty]\to[0,\infty] is defined as a+_{p}b=(a^{p}+b^{p})^{1/p} , and for p=\infty we define a+_{\infty}b=\sup\{a,b\} . Given an L -graph G\colon V\times V\to L , the monoidal Rips filtration of G is an L -filtered simplicial set R^{\bullet}_{\otimes}(G) , where the n -simplices are the elements in V^{n+1} and their filtration values depend solely on G and the product \otimes\colon L\times L\to L . When the monoidal product is the \infty -sum on [0,\infty] , the monoidal Rips filtration agrees with the directed Rips filtration from [ 45 ] ( Example 3.4 ). Furthermore, we recover network distance stability for the directed Rips filtration ( Example 5.5 ). The \ell_{p} -Vietoris-Rips filtration from [ 30 ] is a filtered simplicial set constructed from a metric space, and is like our construction, inspired by Cho’s work in [ 17 ] . We show that the \ell_{p} -Vietoris-Rips simplicial set is a special case of the monoidal Rips filtration when the L -graph is a metric, and the monoidal product is the p -sum defined above ( Example 3.5 ). We also establish a Gromov-Hausdorff stability result for the \ell_{p} -Vietoris-Rips filtration that is slightly stronger than the one presented in [ 30 ] ( Example 5.6 ). When speaking of stability, we consider symmetric duoidal lattices L=(L,\leq,\otimes,\oplus,e) . Here we have two symmetric monoidal products \otimes and \oplus on the same lattice (L,\leq) sharing the same neutral element e . In addition, these are required to satisfy the following generalized Minkowski inequality: (s\oplus t)\otimes(s^{\prime}\oplus t^{\prime})\leq(s\otimes s^{\prime})\oplus(t\otimes t^{\prime})\text{ for all }s,t,s^{\prime},t^{\prime}\in L. An example of a symmetric duoidal lattice is ([0,\infty],\leq,+_{p},+_{q},0) where \leq is the usual ordering and 1\leq q\leq p\leq\infty . One can also construct a symmetric duoidal lattice from any symmetric monoidal lattice (L,\leq,\oplus,e) by defining the product s\otimes t=\sup\{s,t\} . In this case, we

Read the complete text