Landscape Intervals: A Geometric Framework for Multiparameter Persistence Landscapes
Abstract
Classical persistent homology provides a summary of the multiscale topology of a filtration of a space. These summaries are not naturally valued in a vector space, limiting their direct use in standard machine learning pipelines. To address this issue, Bubenik introduced persistence landscapes, vector-valued invariants of persistence modules, which operate by conveniently encoding the behavior of rank functions of modules along the parameter axis. When passing to multiparameter persistent homology, changes in the rank functions vary by different amounts in different directions. In this paper, we introduce the landscape interval, a subset of the parameter space of a multiparameter persistence module which records changes in the rank function in all directions simultaneously. This geometric perspective leads to two novel invariants which serve as notions of persistence landscapes for multiparameter modules. We show that our invariants generalize existing landscape-type invariants: in particular, they recover Bubenik's landscapes in the one-parameter setting and allow us to obtain the multiparameter landscapes introduced by Vipond. For our main theoretical contributions, we establish two exact identities connecting landscapes with erosion distances, prove that our invariants are stable under perturbations, and refine the interleaving-stability bounds for the existing landscape-type invariants. We also show that on a certain class of multiparameter persistence modules, the full invariants can be recovered from finite information about the module. On the practical side, we introduce a robust numerical framework for extracting our invariants from real data and for incorporating them into standard machine learning pipelines. The framework is illustrated on surface mesh and medical image data.
Related
- COMPLEX: A Closed-Form Certified Embedding of Multiparameter Persistence Modules
- PLDiv: Measuring Dataset Diversity and Semantic Collapse via Persistence Landscapes
- Geometric effects on the persistence of patch structures for 2D Euler equations in simply-connected bounded domains
- The stability landscape in wave-packet scattering: geometric rigidity and sharp Sobolev thresholds
- Graded Automorphism of Multiparameter Quantum Matrix Algebra
BibTeX
@misc{fang2026landscape,
title = {{Landscape Intervals: A Geometric Framework for Multiparameter Persistence Landscapes}},
author = {Pan Fang and Washington Mio and Tom Needham},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.42090},
eprint = {2610.42090},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.42090},
}
RIS
TY - UNPB
TI - Landscape Intervals: A Geometric Framework for Multiparameter Persistence Landscapes
AU - Pan Fang
AU - Washington Mio
AU - Tom Needham
PY - 2026
DA - 2026/10/30
PB - arXiv
DO - 10.48550/arxiv.2610.42090
UR - https://arxiv.org/abs/2610.42090
AB - Classical persistent homology provides a summary of the multiscale topology of a filtration of a space. These summaries are not naturally valued in a vector space, limiting their direct use in standard machine learning pipelines. To address this issue, Bubenik introduced persistence landscapes, vector-valued invariants of persistence modules, which operate by conveniently encoding the behavior of rank functions of modules along the parameter axis. When passing to multiparameter persistent homology, changes in the rank functions vary by different amounts in different directions. In this paper, we introduce the landscape interval, a subset of the parameter space of a multiparameter persistence module which records changes in the rank function in all directions simultaneously. This geometric perspective leads to two novel invariants which serve as notions of persistence landscapes for multiparameter modules. We show that our invariants generalize existing landscape-type invariants: in particular, they recover Bubenik's landscapes in the one-parameter setting and allow us to obtain the multiparameter landscapes introduced by Vipond. For our main theoretical contributions, we establish two exact identities connecting landscapes with erosion distances, prove that our invariants are stable under perturbations, and refine the interleaving-stability bounds for the existing landscape-type invariants. We also show that on a certain class of multiparameter persistence modules, the full invariants can be recovered from finite information about the module. On the practical side, we introduce a robust numerical framework for extracting our invariants from real data and for incorporating them into standard machine learning pipelines. The framework is illustrated on surface mesh and medical image data.
ER -