Persistent Cup-Length Invariants

Updated 3 July 2026

Persistent cup-length is a topological invariant that generalizes classical cup-length by capturing nontrivial multiplicative cohomological interactions across data filtrations.
It refines the analysis of persistent homology using cup modules and partition modules, computed efficiently via persistent cohomology algorithms.
Its strong stability and discriminative power enable the detection of complex structures, such as toroidal components, beyond standard Betti barcodes.

Persistent cup-length is a topological invariant in the persistent setting that generalizes the classical notion of cup-length from cohomology rings to filtered spaces or data filtrations. It is fundamentally grounded in the graded algebra structure imparted by the cup product, providing strictly richer information than classical persistent homology by incorporating multiplicative cohomological interactions. Persistent cup-length invariants, including their multi-parameter and module-theoretic refinements, have emerged as highly stable, computable, and powerful tools for distinguishing and quantifying higher-order topological features in data and filtrations, with demonstrated discriminative capacity beyond the reach of standard persistence barcodes.

1. Algebraic Foundations: Cup Product and Classical Cup-Length

The cup product endows the cohomology $H^*(X)$ of a topological space $X$ with a graded-commutative algebra structure. For a finite simplicial complex (or more generally, a CW complex), with cochains $C^p(X)$ and $C^q(X)$ , the cup product

$\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$

is determined (assuming an ordering on vertices) by

$(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$

Passing to cohomology gives

$\smile\colon H^p(X) \times H^q(X) \to H^{p+q}(X).$

Classical cup-length $\operatorname{cup}(X)$ is defined as the maximal $\ell \geq 1$ for which there exist classes $\eta_1, \dots, \eta_\ell \in H^{>0}(X)$ such that $X$ 0 (Mémoli et al., 2022, Contessoto et al., 2021, Ivshina et al., 15 Jul 2025).

2. Persistent Cup-Length: Definition and Structures

Given a one-parameter filtration $X$ 1, $X$ 2, cohomology yields a persistent diagram of graded rings. For each closed interval $X$ 3, the persistent cup-length is the maximal $X$ 4 for which a nontrivial $X$ 5-fold cup product survives in the image of the ring map $X$ 6: $X$ 7 Equivalently, persistent cup-length may be constructed using explicit cocycle representatives supporting nonvanishing cup products across given intervals, yielding a "persistent cup-length diagram" $X$ 8 that records, for each interval, the largest $X$ 9 of living $C^p(X)$ 0-fold products. A Möbius inversion demonstrates that the function $C^p(X)$ 1 is the maximum $C^p(X)$ 2 for which there exists an interval $C^p(X)$ 3 with $C^p(X)$ 4 (Mémoli et al., 2022, Contessoto et al., 2021).

The persistent setting also admits the notion of zero-divisor-cup-length, wherein one tracks the nilpotency in the ideal of zero-divisors for the multiplicative cohomology structure under filtration, known as persistent zero-divisor-cup-length or persistent cup-length in certain contexts (Mémoli et al., 22 Jun 2025).

The information content of persistent cup-length is refined by encoding the full family of multi-parameter modules generated via powers of the cup product. Let

$C^p(X)$ 5

The functor $C^p(X)$ 6 defines the two-parameter "persistent cup module" structure. For fixed $C^p(X)$ 7, the associated $C^p(X)$ 8-cup module over $C^p(X)$ 9 isolates $C^q(X)$ 0-fold cup product persistence and yields standard barcodes per cohomological degree (Mémoli et al., 2022, Dey et al., 2022). The persistent cup-length can be recovered from these barcodes: $C^q(X)$ 1 Partition modules further decompose these structures, organizing barcode data according to degree patterns in $C^q(X)$ 2-fold products, and provide strictly stronger invariants. These refinements can separate filtrations with indistinguishable ordinary or cup-module barcodes (Dey et al., 2022).

4. Computation and Algorithms

Persistent cup-length and its module-theoretic refinements admit efficient computation. A representative workflow leverages persistent cohomology algorithms (notably Ripser) to generate cocycle barcodes, then systematically computes cup products of representatives to track persistence of nontrivial multiplicative products. For finite filtrations of $C^q(X)$ 3 simplices and maximum dimension $C^q(X)$ 4, persistent $C^q(X)$ 5-cup modules (for all $C^q(X)$ 6) can be determined in $C^q(X)$ 7 time, while partition modules are computable in $C^q(X)$ 8 time for $C^q(X)$ 9 subexponential in $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 0 (Dey et al., 2022, Ivshina et al., 15 Jul 2025, Contessoto et al., 2021). Key computational advances include:

Matrix reduction and sparse linear algebra for coboundary computation;
Batch cup product calculation among selected persistent cocycles;
Efficient memory usage via on-the-fly computation of boundary submatrices.

For large data, landmark-based farthest-point subsampling ensures scalability, and the bottleneck or erosion distance between diagrams for the full and subsampled data is $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 1 if the landmarks are $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 2-dense (Ivshina et al., 15 Jul 2025).

5. Stability and Metric Properties

Persistent cup-length and all associated module/partition invariants exhibit strong stability properties. For persistent filtrations $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 3, $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 4, the cup-length functional is 1-Lipschitz with respect to the interleaving distance $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 5: $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 6 where $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 7 denotes erosion distance on interval-valued invariants. For Vietoris–Rips filtrations of compact metric spaces $\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 8, this further implies

$\smile\colon C^p(X) \times C^q(X) \to C^{p+q}(X)$ 9

with $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 0 the Gromov–Hausdorff distance (Mémoli et al., 2022, Mémoli et al., 22 Jun 2025, Dey et al., 2022). This stability extends to all $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 1-cup modules and to the cup module's rank function, as well as to the persistent LS-category invariant.

Stability is also crucial practically, ensuring robustness to geometric noise and perturbations in the input data or filtration.

6. Discriminative Power and Applications

Persistent cup-length can distinguish spaces and filtrations indistinguishable by standard persistent homology or even persistent cohomology barcodes. For instance, it robustly detects toroidal summands: the existence of a length- $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 2 persistent cup product among independent $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 3 classes certifies the presence of a torus component in each sublevel set (Ivshina et al., 15 Jul 2025). Persistent cup-length separates equilateral triangles from paths with identical $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 4 barcodes, and differentiates $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 5 from wedges of spheres, providing improved lower bounds on $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 6 for metric shape comparisons (Mémoli et al., 22 Jun 2025).

In applied contexts, persistent cup-length enables detection of toroidal structure in neural data, specifically grid cell population activity, providing a test for topologically nontrivial attractors where ordinary Betti counts are ambiguous (Ivshina et al., 15 Jul 2025). Its refined invariants are useful for enhanced data analysis, model selection, and analysis of periodic structures.

The persistent cup-length construction is part of a broader categorical framework. It naturally extends traditional rank invariants (persistent Betti numbers), incorporates Puuska's epi-mono rank invariants, and connects with the persistent LS-category—a higher categorical invariant measuring homotopic complexity of filtrations. The LS-category function $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 7 satisfies

$(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 8

pointwise, and shares similar stability bounds under the interleaving distance (Mémoli et al., 2022).

Other generalizations include persistent analogs of topological complexity, the use of persistent Steenrod modules, and exploration of higher cohomological operations (Massey products, Steenrod squares). These extensions suggest a landscape of persistent invariants capturing increasingly subtle ring-theoretic and higher-structural information (Mémoli et al., 2022, Dey et al., 2022).

Summary Table: Persistent Cup-Length and Refined Invariants

Invariant	Structure	Discriminative Power
Persistent cup-length	Interval function $(\alpha \smile \beta)\bigl([v_0,\dots,v_{p+q}]\bigr) = \alpha([v_0, \dots, v_p]) \cdot \beta([v_p, \dots, v_{p+q}]).$ 9	Distinguishes spaces with identical Betti barcodes
$\smile\colon H^p(X) \times H^q(X) \to H^{p+q}(X).$ 0-cup modules	1D persistence modules for $\smile\colon H^p(X) \times H^q(X) \to H^{p+q}(X).$ 1-fold products	Refines cup-length, tracks $\smile\colon H^p(X) \times H^q(X) \to H^{p+q}(X).$ 2-fold structure
Partition modules	Multigraded submodules indexed by partitions	Separates filtrations with same cup barcodes
Persistent LS-category	Interval function $\smile\colon H^p(X) \times H^q(X) \to H^{p+q}(X).$ 3	Upper bound for cup-length; measures homotopic complexity