Local Characterization of Multi Parameter Persistence Modules

26 concepts8 papers

In topological data analysis (TDA), persistent homology tracks how the geometric features of a dataset evolve across different scales. When we vary a single parameter—such as a distance threshold—the resulting mathematical structure is a 1D persistence module. These modules are well-behaved: they decompose uniquely into a collection of intervals, which we can easily visualize as a 1D "barcode."

However, real-world data often depends on multiple parameters simultaneously, such as looking at distance thresholds and local density filters. When we move to multi-parameter persistence, the underlying algebraic structures become notoriously "wild." They generally do not decompose into simple, interpretable intervals, making global analysis computationally intractable and visually opaque.

To overcome this complexity, a powerful research direction has emerged: restricting our focus to well-behaved classes of modules that can be completely understood through local characterizations. By checking simple algebraic conditions on tiny, localized regions of the parameter space, we can guarantee nice global decompositions.

Taming Wildness via Rectangle Decomposability

To make multi-parameter persistence tractable, we can focus on modules built from simple geometric shapes. The most natural extension of a 1D interval to higher dimensions is a multi-dimensional rectangle (a product of intervals). A module is rectangle-decomposable if it can be expressed as a direct sum of indicator functions over these rectangular regions.

As noted by Sergio Cabello et al. in the Guest Editors' Foreword, restricting analysis to rectangle-decomposable modules provides a computationally tractable pathway for analyzing high-dimensional data because it simplifies the calculation of rank invariants.

Crucially, Magnus Bakke Botnan et al. demonstrated in On Rectangle-Decomposable 2-Parameter Persistence Modules that for this specific class, the rank invariant is complete. This means that computing the ranks of the maps within the module is sufficient to uniquely and fully characterize its structure, perfectly recovering the simplicity of 1D barcodes. Under certain structural constraints—such as the existence of a "conqueror function" explored by Chen Cai et al. in Elder-rule-staircodes for Augmented Metric Spaces—multi-parameter modules naturally satisfy the ordering conditions required to guarantee an exact, interval-decomposable representation.

The Local-to-Global Principle

How do we determine if a massive, complex multi-parameter module is rectangle-decomposable without executing an expensive global decomposition algorithm? The core insight of this research direction is that global decomposability is entirely determined by local behavior.

In two dimensions, this manifests as an algebraic property called exactness. As proven by Jérémy Cochoy et al. in Decomposition of Exact pfd Persistence Bimodules, a 2D persistence module can be uniquely decomposed into simple geometric blocks (like quadrants and bands) if and only if it satisfies exactness locally. Local exactness means that the horizontal and vertical maps between vector spaces at the vertices of any small rectangle in the grid form an exact sequence. This algebraic coupling forces the global information flow to align into clean, discrete block summands. For instance, persistence modules derived from interlevel sets naturally satisfy this due to the Mayer-Vietoris theorem.

To make this even more practical, Magnus Bakke Botnan et al. established a strict local-to-global check in Local Characterizations for Decomposability of 2-Parameter Persistence Modules. They proved that a 2-parameter module decomposes into rectangles if and only if its restriction to every tiny $2 \times 2$ grid within the domain is itself decomposable.

Local Check (2x2 Grid)             Global Structure
     V_01 ------> V_11               +---------------+
      ^            ^                 |               |
      |            |       ===>      |   Rectangle   |
     V_00 ------> V_10               |   Component   |
                                     +---------------+

Remarkably, the authors proved a maximality result: rectangles are the largest possible class of shapes for which this $2 \times 2$ local check works. If you expand the allowed building blocks to include slightly more complex shapes, like L-shaped hooks, the local-to-global principle breaks down entirely.

Scaling to Higher Dimensions

Does this local magic scale when we have $n$ parameters instead of 2? Generalizing this local characterization was a significant open problem, because the geometry of an $n$-dimensional grid introduces far more complex interactions than a simple 2D square.

This gap was bridged by Vadim Lebovici et al. in Local characterization of block-decomposability for multiparameter persistence modules. The authors generalized 2D exactness by introducing $k$-middle exactness, an algebraic condition evaluated on small $k$-dimensional hypercubes within the parameter space.

The authors proved that an $n$-dimensional persistence module is block-decomposable if and only if it satisfies $k$-middle exactness for all relevant dimensions $k$. This means that even in highly complex, high-dimensional parameter spaces, we do not need to grasp the global topology all at once; we can completely verify the module's global structural tractability by auditing its behavior on small, lower-dimensional sub-structures.

From Theory to Stable Visualizations

Once we use local characterizations to verify that a module is well-behaved, we can confidently extract visual summaries and invariants for data analysis.

For example, Magnus Bakke Botnan et al. introduced the concept of a "signed barcode" in Signed Barcodes for Multi-Parameter Persistence via Rank Decompositions. By decomposing the rank invariant of a grid-indexed module into the linear combinations of rectangular building blocks, they created a generalized barcode that retains the intuitive visual power of 1D TDA.

However, for these representations to be useful in practice, they must be stable: a tiny amount of noise in the input data should not cause a massive jump in the resulting barcode. In On the bottleneck stability of rank decompositions of multi-parameter persistence modules, Magnus Bakke Botnan et al. discovered that while minimal rank decompositions using pure rectangles are actually unstable, a slight shift in perspective restores stability. By utilizing a rank exact decomposition—which swaps out rectangles for hook-shaped modules as its fundamental building blocks—the authors achieved bottleneck stability. This ensures that the local-to-global framework is not just an elegant algebraic property, but a reliable foundation for multi-parameter data pipelines.

Key Foundational Work

On Rectangle-Decomposable 2-Parameter Persistence Modules (2020) — contributed 7 concept elements to this direction.
Decomposition of Exact pfd Persistence Bimodules (2020) — contributed 3 concept elements to this direction.
Elder-rule-staircodes for Augmented Metric Spaces (2020) — contributed 2 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_generalization_of_local_characterization_c2af1144,
  title = {generalization of local characterization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/generalization-of-local-characterization/rdr_c2af1144476b4cc18c1745a36a2569fb},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_generalization_of_local_characterization_c2af1144_v1,
  title = {generalization of local characterization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/generalization-of-local-characterization/rdr_c2af1144476b4cc18c1745a36a2569fb/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}

See something wrong? Report a mistake