Degree–Rips Bifiltrations

• Degree–Rips bifiltration is a density-sensitive extension of the Vietoris–Rips complex that uses both scale and local density parameters to capture topological features.
• It provides a computationally tractable and stable framework with homotopy interleavings that ensure robust topological inference even in noisy, outlier-rich data.
• Algorithmic implementations leverage neighborhood degree computations and incremental Rips complex constructions to efficiently generate multiparameter persistence modules.

The degree–Rips bifiltration is a fundamental construction in multiparameter topological data analysis that enables density-sensitive, stable, and computationally tractable summaries of topological features across varying local density and scale parameters. It extends the classical Vietoris–Rips (Rips) complex by introducing a density or degree threshold, yielding a bifiltration indexed by two parameters. This construction plays a critical role in extracting robust topological information from data sets, especially in the context of noise and outliers.

1. Formal Definition and Constructions

Given a metric space $(X, d)$ equipped either with a Borel probability measure $\mu$ or as a finite point cloud, the degree–Rips bifiltration introduces two parameters:

• A scale parameter $r > 0$ (or $\varepsilon$): controls the geometric proximity threshold for simplex inclusion.
• A density threshold, either as a normalized parameter $\alpha \in [0,1]$ (in the measure-theoretic setting) or as a degree threshold $k\in \mathbb{N}$ (number of neighbors in radius $2r$).

For each point $x \in X$, define the local measure-mass at scale $r$: $$m_r(x) := \mu(B(x, r)), \quad B(x, r) = \{y \in X : d(x, y) < r\}$$ or, for finite sets, the r-neighborhood degree: $\mu$0

A simplex $\mu$1 is included in the degree–Rips complex at parameters $\mu$2 or $\mu$3 if:

• All its vertices have local density $\mu$4, or
• All vertices satisfy $\mu$5,
• The diameter $\mu$6 (or $\mu$7), as in the usual Rips criterion.

This defines the bifiltration as a functor: $\mu$8 or, equivalently for probability measures,

$\mu$9

This forms a bifiltration, with inclusion maps functorially relating complexes as the parameters ($r > 0$0 or $r > 0$1) vary (Rolle, 2022, Blumberg et al., 2020, Jardine, 2020).

2. Homotopy-Theoretic and Stability Properties

The degree–Rips bifiltration admits a poset-based homotopy type model. For each parameter tuple, one defines a poset of simplices closed under inclusion, and the classifying space (nerve) of this poset yields a Kan simplicial set whose geometric realization is homotopy equivalent to the underlying clique complex (Jardine, 2020). The parameter set carries a partial order: increasing the scale or lowering the degree threshold admits more simplices, leading to a functorial diagram.

Stability of the degree–Rips bifiltration is characterized via homotopy interleavings. If $r > 0$2 and $r > 0$3 are metric probability spaces with Gromov–Prohorov distance at most $r > 0$4, then the respective degree–Rips bifiltrations are homotopy-$r > 0$5-interleaved in their two parameters. Explicitly,

$r > 0$6

are simplicial inclusions (up to homotopy equivalence), and vice versa, with the compositions homotopic to the natural shifts (Rolle, 2022, Blumberg et al., 2020, Jardine, 2020).

In the normalized discrete case, for finite metric spaces with uniform measures, the homotopy interleaving distance between normalized degree–Rips bifiltrations is bounded by the Gromov–Prohorov distance. However, the shift factors are sharp: a dilation of the scale parameter by at least 3 is required for interleaving in certain configurations (Blumberg et al., 2020).

3. Multiparameter Persistence Modules and Interval Decomposability

Application of homology functors produces two-parameter persistence modules: $r > 0$7 for fixed homological dimension $r > 0$8 (Alonso et al., 2024). A key result is the probabilistic non-interval-decomposability theorem: for a homogeneous Poisson sample in $r > 0$9, the probability that the degree–Rips module decomposes strictly into intervals decays exponentially in the number of points. Indecomposable blocks (“wild type” modules) appear generically in high dimensions and with large samples.

This non-interval-decomposability is structurally explained: local configurations in the point cloud induce features in the module that cannot align with a totally ordered, interval-type summand structure, and such configurations become overwhelmingly frequent as the sample size increases (Alonso et al., 2024).

4. Algorithmic and Computational Aspects

Degree–Rips bifiltrations are polynomial in size for fixed skeleton dimension, e.g., $\varepsilon$0 for $\varepsilon$1-simplices in an $\varepsilon$2-point cloud, and with grid coarsening this reduces to $\varepsilon$3, matching that of the classical Rips filtration. In contrast, analogous density-sensitive or subdivision-based filtrations such as the full subdivision–Rips bifiltration can be exponentially larger (Blumberg et al., 2020).

Practical computation is tractable in moderate dimensions and data set sizes. Algorithms optimize by precomputing neighborhood degrees (using range search or kNN structures), filtering vertices, and constructing the Rips clique complex incrementally:

• For low-dimensional persistence, "line-sweep" algorithms as implemented in RIVET enable efficient computation for hundreds of data points.
• Detection of non-interval indecomposables in small parameter grids can be accomplished by checking ranks of four boundary-map compositions for a $\varepsilon$4 subposet, with linear complexity in the small complex size (Alonso et al., 2024).

5. Homology Inference and Limit Object Analysis

Homology inference with the degree–Rips bifiltration connects finite-sample topology to the limit topology on probability-measure spaces. As finite samples converge in Prokhorov (measure) and Hausdorff (geometry) distance to the support, the finite bifiltrations converge in homotopy-interleaving sense to the continuous bifiltration on the support (Rolle, 2022).

A paradigmatic example analyzes the bifiltration for a planar annulus with a low-mass disc (representing outliers): the region in parameter space (scale and density) supporting nontrivial first homology is shifted but not destroyed by the low-density region. This is formalized by explicit geometric calculations of level sets and homotopy types, ultimately showing that the effect of outliers is to move density-threshold boundary curves upward in the parameter plane, rather than fundamentally changing the sequence of topological features (e.g., spheres $\varepsilon$5) (Rolle, 2022).

This suggests future directions in characterizing and exploiting such limit objects for robust topology inference and in designing parameter slices for barcodes or clustering that optimize for noise resilience.

6. Applications and Further Directions

Degree–Rips bifiltrations are used for density-aware multiscale clustering, stable topological inference in noisy and outlier-rich settings, and as a vehicle for multiparameter persistence methods in computational topology. Their functorial and stable behavior accommodates homotopy-theoretic operations such as Mayer–Vietoris sequences and controlled equivalences, enabling rigorous analysis of topological features and their robustness under data perturbations (Jardine, 2020).

The explicit computational tractability of degree–Rips bifiltrations, especially for moderate sample sizes, makes them suitable for practical applications in data analysis, though for large data or high dimensions the decomposition and module-theoretic complexity remains a significant challenge.

Ongoing research targets refined stability bounds, efficient multiparameter persistence calculations, and a deeper understanding of limit dynamics and indecomposable module structure in high-dimensional random settings (Alonso et al., 2024).