Algebraic stability theorem for derived categories of zigzag persistence modules

Yasuaki Hiraoka, Yuichi Ike, Michio Yoshiwaki

TL;DR

This paper provides a unified mathematical framework for the stability of zigzag persistence modules—a flexible tool in Topological Data Analysis (TDA) for tracking features in non-linear data sequences. By leveraging derived categories and "tilting modules" from representation theory, the authors prove an Algebraic Stability Theorem that guarantees these topological "barcodes" are robust against small perturbations in the underlying data.

What Problem Does This Paper Address?

In TDA, we typically use persistent homology to extract "barcodes" that represent the lifespan of topological features (like holes or loops) in a dataset. Standard persistence assumes a linear "filtration" where data only grows (e.g., increasing a radius). However, real-world data—like time-series or moving point clouds—often requires zigzag persistence, where data can be added or removed at each step.

The "holy grail" for any TDA tool is a Stability Theorem: a proof that if your input data changes slightly (noise), your output barcode only changes slightly. While stability was well-proven for ordinary persistence, zigzag persistence is mathematically much more complex. Previous work (by Botnan and Lesnick) proved stability for "purely" zigzag patterns, but a general, computable framework for arbitrary zigzag orientations remained elusive.

Methods

The authors shift the perspective from the modules themselves to their derived categories. This is a high-level abstraction in algebra that treats sequences of modules (complexes) as the primary objects of study.

  1. Derived Equivalence: They prove that the derived category of ordinary persistence modules is equivalent to the derived category of any zigzag persistence module.
  2. Tilting Modules: They use a "classical tilting module" to bridge these two worlds. This allows them to "transport" well-understood stability results from ordinary persistence over to the more difficult zigzag setting.
  3. AR Quivers: They utilize Auslander–Reiten (AR) quivers—graphical representations of the relationships between indecomposable modules—to explicitly calculate distances between barcodes.
  4. Distance Metrics: They define new "derived" versions of the interleaving and bottleneck distances, which are the standard metrics for comparing barcodes.

Datasets

As a theoretical paper in algebraic representation theory, no empirical datasets were used. The work focuses on proving theorems and establishing mathematical relationships. The authors do, however, provide illustrative examples using $A_n$-type quivers (small, finite sequences of vector spaces) to demonstrate their calculations.

Key Insights and Ideas

  • The "Derived" Shortcut: Instead of trying to prove stability for zigzag modules directly (which is hard because they lack a natural "shift" operator), the authors show that zigzag modules are just "rotated" versions of ordinary modules within a larger derived category.
  • Incomparability: A surprising finding is that their new distance metric is incomparable to the "block distance" previously introduced by Botnan and Lesnick. This means that for some data, the new metric is more sensitive, while for others, the block distance is. There is no "one size fits all" distance for zigzag persistence.
  • Sheaf Theory Connection: The paper bridges TDA with microlocal sheaf theory by showing that their derived interleaving distance is essentially a version of the convolution distance used by Kashiwara and Schapira. This connects TDA to deep results in algebraic geometry and topology.

Key Results

  • Algebraic Stability Theorem (AST): The authors formally prove that the derived bottleneck distance between two zigzag barcodes is bounded by the derived interleaving distance ($d_B^D \leq d_I^D$).
  • Isometry Theorem: They go a step further and prove an isometry ($d_B^D = d_I^D$), showing that the bottleneck distance (which is easier to visualize) perfectly captures the algebraic interleaving of the modules.
  • Explicit Computation: Using AR quivers, the authors provide a concrete way to calculate these distances for any zigzag orientation, moving the theory toward practical algorithmic implementation.

Limitations and Future Work

  • Computational Complexity: While the paper provides the mathematical formulas for distances, it does not provide an optimized algorithm. Calculating these distances in the derived category for very large datasets could be computationally expensive.
  • Interpretation of "Derived" Barcodes: The "derived barcodes" include information about the "grading" (or shift) of features. Translating what these graded features mean in terms of physical data (e.g., "this hole exists at time $t$ in the 1st homology group") is more abstract than in standard TDA.
  • Future Direction: A natural next step is to implement these metrics in a software library (like Hera or Gudhi) to see how they perform on real-world time-series data compared to existing zigzag methods.

Summary updated

Abstract

We study distances on zigzag persistence modules from the viewpoint of derived categories and Auslander–Reiten quivers. The derived category of ordinary persistence modules is derived equivalent to that of arbitrary zigzag persistence modules, depending on a classical tilting module. Through this derived equivalence, we define and compute distances on the derived category of arbitrary zigzag persistence modules and prove an algebraic stability theorem. We also compare our distance with the distance for purely zigzag persistence modules introduced by Botnan–Lesnick and the sheaf-theoretic convolution distance due to Kashiwara–Schapira.
BibTeX
@article{hiraoka2023algebraic,
  author = {Yasuaki Hiraoka and Yuichi Ike and Michio Yoshiwaki},
  title = {{Algebraic stability theorem for derived categories of zigzag persistence modules}},
  journal = {Journal of Topology and Analysis},
  year = {2023},
  doi = {10.1142/s1793525323500292},
}

Loading related content...

Figure