Persistence Diagram Universality

• Persistence diagram universality is the study of canonical, model-independent behavior where diagrams are exactly determined by spectral data, optimal metrics, and global topological structures.
• It unifies techniques from Morse theory, random matrix theory, and optimal transport to establish stability and rigidity in both one- and multiparameter persistence settings.
• The framework identifies concrete invariants, such as persistence entropy and geodesic rigidity, while also revealing computational challenges and limitations in extending barcode-like universality beyond classical cases.

Persistence diagram universality denotes a family of results in which persistence diagrams, diagram spaces, or diagram-derived statistics exhibit canonical or model-independent behavior. In the recent literature, universality has several distinct technical meanings: exact determinability of a diagram from underlying spectral data; universal or maximal compatible metrics on diagram spaces; rigidity of geodesics under matching metrics; asymptotic probability laws for random diagrams; and, conversely, sharp failures of barcode-like universality in multiparameter settings (Loftus, 29 Mar 2026, Bubenik et al., 2019, Bauer et al., 2020, Bobrowski et al., 2022).

1. Universality as a family of mathematical principles

The cited work uses the term universality in non-equivalent but structurally related senses. In one line of work, a persistence diagram is universal because it is analytically forced by another invariant: for symmetric matrices, the sublevel-set persistence diagram of a quadratic form on the sphere is exactly determined by the ordered eigenvalues, so random-matrix universality transfers directly to persistence diagrams (Loftus, 29 Mar 2026). In a second line, universality is categorical or metric: persistence diagrams with the $p$-Wasserstein distance form the universal $p$-subadditive commutative monoid generated by a metric pair, and a boundary-sensitive bottleneck distance becomes the universal stable distance for realizable extended persistence diagrams (Bubenik et al., 2019, Bauer et al., 2020).

A third use of the term concerns rigidity of the geometry of diagram space. For several families of matching metrics $d_p[l^q]$, every geodesic is induced by an optimal bijection and pointwise linear interpolation, so the geodesic structure is universal in the sense that all geodesics have the same canonical form; for other metric regimes, this fails through branching and deviant geodesics (Chowdhury, 2019). A fourth use is probabilistic: some works propose or prove ensemble-level laws for random persistence diagrams, ranging from exact solvable distributions for random triangular matrices over finite fields to conjectural universal laws for the noise portion of point-cloud persistence diagrams (Mészáros, 16 Jun 2026, Bobrowski et al., 2022).

These meanings are not interchangeable. Some are exact theorems about algebraic or metric structure, some are asymptotic theorems in stochastic models, and some are explicitly conjectural. This distinction is essential for interpreting the scope of any universality claim.

2. Spectral universality via Morse theory and random matrices

A precise theorem of this type is established for quadratic forms on spheres. Let $M$ be a symmetric matrix with distinct eigenvalues

$\lambda_1<\lambda_2<\cdots<\lambda_n.$

For the restriction of

$$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$

to $S^{n-1}$, the critical points are the eigenvectors $\pm \mathbf e_i$, with critical value $\lambda_i$. The persistence diagram of the sublevel-set filtration has exactly $n-1$ finite bars, with the $p$0-th finite bar equal to

$p$1

length

$p$2

and homological dimension $p$3. The diagram also contains two infinite bars, $p$4 in $p$5 and $p$6 in $p$7. Moreover, for $p$8, the sublevel set is homotopy equivalent to $p$9, producing the ladder

$d_p[l^q]$0

The total persistence telescopes to

$d_p[l^q]$1

This makes the persistence diagram an exact re-encoding of adjacent eigenvalue spacings (Loftus, 29 Mar 2026).

Once bar lengths are identified with spacings, random-matrix universality becomes persistence-diagram universality. For GOE, the paper derives the closed form

$d_p[l^q]$2

for persistence entropy. It also reports that the coefficient of variation of $d_p[l^q]$3 decays roughly as $d_p[l^q]$4, and more specifically gives $d_p[l^q]$5 at $d_p[l^q]$6, $d_p[l^q]$7 at $d_p[l^q]$8, and $d_p[l^q]$9 at $M$0. By contrast, the normalized maximum-bar statistic

$M$1

has coefficient of variation around $M$2–$M$3, consistent with extreme-value behavior. For GUE, the same Morse-theoretic identification applies, but the spacing statistics differ because of stronger level repulsion, $M$4 instead of $M$5. For Wishart matrices, the limiting density is Marchenko–Pastur rather than semicircular. The result is that GOE, GUE, and Wishart ensembles produce distinct universal persistence diagrams, interpreted in the paper as topological fingerprints of random-matrix universality classes (Loftus, 29 Mar 2026).

The same work also develops persistence entropy as a spectral diagnostic. It compares

$M$6

with $M$7, emphasizing that $M$8 is local whereas $M$9 depends on the full spacing distribution. Using $\lambda_1<\lambda_2<\cdots<\lambda_n.$0 samples per class at $\lambda_1<\lambda_2<\cdots<\lambda_n.$1, the reported AUC values for GOE versus GUE discrimination are $\lambda_1<\lambda_2<\cdots<\lambda_n.$2, $\lambda_1<\lambda_2<\cdots<\lambda_n.$3, and $\lambda_1<\lambda_2<\cdots<\lambda_n.$4 for $\lambda_1<\lambda_2<\cdots<\lambda_n.$5, versus $\lambda_1<\lambda_2<\cdots<\lambda_n.$6, $\lambda_1<\lambda_2<\cdots<\lambda_n.$7, and $\lambda_1<\lambda_2<\cdots<\lambda_n.$8 for $\lambda_1<\lambda_2<\cdots<\lambda_n.$9. At $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$0, the bootstrap $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$1 confidence intervals are $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$2 and $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$3, which do not overlap. In the Rosenzweig–Porter model, the paper reports that $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$4 has signal-to-noise ratio $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$5 up to $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$6, whereas $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$7 reaches $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$8 deviation by $$f(\mathbf{x})=\mathbf{x}^\top M\mathbf{x}=\sum_{i=1}^n \lambda_i x_i^2$$9 and spacing variance reaches $S^{n-1}$0 by $S^{n-1}$1, indicating sensitivity to global spectral broadening rather than only local level repulsion (Loftus, 29 Mar 2026).

3. Universal metric constructions for persistence diagrams

A different meaning of universality is categorical. For a metric pair $S^{n-1}$2, where $S^{n-1}$3 is a distinguished subset, the commutative monoid of persistence diagrams is defined as

$S^{n-1}$4

equivalently $S^{n-1}$5, where $S^{n-1}$6 is the free commutative monoid of finite formal sums of elements of $S^{n-1}$7. The $S^{n-1}$8-Wasserstein distance $S^{n-1}$9 on $\pm \mathbf e_i$0 is defined by matching points while allowing unmatched points to be paired with the distinguished subset $\pm \mathbf e_i$1. In the classical case $\pm \mathbf e_i$2, $\pm \mathbf e_i$3 is the bottleneck distance and $\pm \mathbf e_i$4 is the usual $\pm \mathbf e_i$5-Wasserstein distance on finite persistence diagrams (Bubenik et al., 2019).

The universality theorem states that $\pm \mathbf e_i$6 is the universal $\pm \mathbf e_i$7-subadditive commutative metric monoid generated by $\pm \mathbf e_i$8. Equivalently, the forgetful functor from $\pm \mathbf e_i$9-subadditive commutative metric monoids to metric pairs has a left adjoint. Concretely, if $\lambda_i$0 is any $\lambda_i$1-subadditive commutative metric monoid and $\lambda_i$2 is $\lambda_i$3-Lipschitz, then there exists a unique $\lambda_i$4-Lipschitz monoid homomorphism

$\lambda_i$5

factoring $\lambda_i$6 through the canonical inclusion. A further characterization identifies $\lambda_i$7 as the largest $\lambda_i$8-subadditive metric on $\lambda_i$9 compatible with the metric on generators. The same framework applies not only to ordinary persistence diagrams, but also to barcodes and to settings in which multiparameter persistence modules decompose into finite sums of indecomposables (Bubenik et al., 2019).

For $n-1$0, the same paper proves a Kantorovich–Rubinstein duality formula for persistence diagrams. This places $n-1$1 on diagrams in direct analogy with classical optimal transport and strengthens the interpretation of Wasserstein geometry as the canonical metric-monoid completion associated with a metric pair (Bubenik et al., 2019).

4. Universal stable distances and rigid geodesic geometry

Universality also appears as a maximal stability principle. For extended persistence diagrams of piecewise linear functions on finite simplicial complexes, the relevant structure is organized through relative interlevel set homology

$n-1$2

where $n-1$3 is a strip-shaped poset carrying ordinary, relative, and extended persistence data in one functorial object. The paper proves that a more discriminative variant of the bottleneck distance, using boundary-sensitive matchings on admissible upsets of the strip, is universal among stable distances on realizable extended persistence diagrams. More precisely, for any admissible upset $n-1$4 and any two realizable diagrams $n-1$5 on $n-1$6 with finite bottleneck distance, there exist a finite simplicial complex $n-1$7 and PL functions $n-1$8 such that

$n-1$9

and

$p$00

The same work shows that the resulting bottleneck geometry is geodesic through

$p$01

and contrasts this with the interleaving distance of sheaves on $p$02, which is shown to be not intrinsic and therefore not universal; the same non-intrinsic pathology transfers to Reeb graphs (Bauer et al., 2020).

A complementary rigidity theorem concerns geodesics in persistence diagram space under the matching metrics $p$03. For a persistence diagram space $p$04 equipped with these metrics, the canonical geodesic associated with an optimal bijection $p$05 has the form

$p$06

The paper proves that for the metric families $p$07 and $p$08 with $p$09, every geodesic is, up to zero distance, a convex-combination geodesic induced by an optimal bijection. This holds for finite diagrams and for countably infinite diagrams. The same paper also proves that rigidity fails for $p$10, $p$11, and for $p$12, where there exist infinite families of branching geodesics and deviant geodesics. In this sense, geodesic universality is a theorem in some metric regimes and explicitly false in others (Chowdhury, 2019).

5. Probabilistic universality for random persistence diagrams

In stochastic topology, universality may refer to limiting laws for random diagrams or for statistics derived from them. For random point clouds, one paper formulates a sequence of conjectures about the noise portion of persistence diagrams. Writing

$p$13

it focuses on multiplicative persistence

$p$14

rather than additive lifetime. The empirical distribution of $p$15-values is

$p$16

Its weak universality conjecture asserts that for fixed ambient dimension $p$17, filtration type $p$18, and homological degree $p$19, the limit law depends only on $p$20 and not on the specific sampling space or distribution. A stronger conjecture introduces

$p$21

with

$p$22

and proposes that the empirical law of the $p$23-values is universal across $p$24, $p$25, and $p$26. The candidate universal law is the left-skewed Gumbel distribution

$p$27

The paper supports these claims with experiments on iid samples from boxes, balls, annuli, spheres, tori, Klein bottles, projective planes, Henneberg surfaces, the Neptune surface mesh, Beta, normal, and Cauchy distributions, on stratified spaces and linkage configuration spaces, on Brownian motion and the Lorenz system, and on natural image patches and audio delay embeddings. It also reports two notable departures from the conjectured pattern: nearly regular grids with small perturbations and the Ginibre ensemble (Bobrowski et al., 2022).

The same conjectural framework is used for feature-level hypothesis testing. Under the null hypothesis

$p$28

the one-sided p-value is

$p$29

and a Bonferroni correction declares significance when

$p$30

The paper applies this to annuli, cut annuli, figure-8 datasets, natural image patches, and truncated torus filtrations (Bobrowski et al., 2022).

A theorem-level stochastic universality result is available for random infinite lower triangular matrices over a finite field $p$31. Let $p$32 be the verbose persistence diagram of the evolving row-span process. For the finite truncation $p$33, the paper proves an explicit formula:

$p$34

for admissible sets $p$35. It also proves a law of large numbers for lifetimes:

$p$36

where

$p$37

with $p$38 and the $p$39 given by shifted geometric distributions. Fluctuation limits for persistent Betti numbers are then expressed through the same universal corank laws that arise in finite-field and $p$40-adic random matrix theory. This is an exact solvable example in which persistence-diagram statistics are governed by explicit model-specific laws built from broader universal rank-fluctuation inputs (Mészáros, 16 Jun 2026).

6. Generalization, coarse survival, and failure beyond the classical setting

One major direction seeks to extend barcode-like universality beyond ordinary one-parameter homology. The theory of saecular persistence introduces a canonical interval decomposition for chain functors

$p$41

under generic conditions such as $p$42 being well ordered and $p$43 being a category of modules or groups. The main construction is a unique complete lattice homomorphism into subobjects of $p$44, yielding interval factors supported on intervals of $p$45. In finite-dimensional vector spaces, this recovers the classical barcode decomposition, and for constructible abelian-valued functors it relates to generalized persistence diagrams through Jordan–Hölder vectors of saecular factors. The framework also extends persistence ideas from homology to homotopy groups. The paper is explicit, however, that existence and uniqueness depend on structural hypotheses on the index order and on subobject lattices, and that the group case requires normality conditions for quotient-group-valued factors (Ghrist et al., 2021).

A weaker form of universality survives in localized multiparameter settings. For persistence modules over $p$46, a large-scale quotient category is obtained by Serre localization, formalizing the idea that modules are equivalent when they differ only on a negligible region. In the two-parameter case, every object in the localized category decomposes uniquely as a direct sum of vertical strips

$p$47

horizontal strips

$p$48

and quadrants

$p$49

For $p$50, the situation changes: there exist indecomposable torsion-free objects of rank $p$51, and the category has wild representation type. The rank invariant determines the torsion part in the localized setting but does not give a full universal invariant in general. This suggests that barcode-like universality survives only in a coarse or localized form once more than one parameter is present (Frankland et al., 2022).

The strongest negative result concerns the generalized persistence diagram (GPD) for multiparameter persistence. For $p$52, there does not exist $p$53 such that the support size of the $p$54-th GPD of every finite $p$55-parameter filtration with $p$56 simplices is $p$57. The paper constructs filtrations whose GPD support has size at least

$p$58

while the number of simplices remains polynomial in $p$59, and extends the construction from $p$60 to all $p$61 by projection. The same super-polynomial behavior is also shown for degree-Rips and degree-\v{C}ech bifiltrations, and for sublevel-Rips and sublevel-\v{C}ech bifiltrations arising from finite metric spaces. As a computational consequence, the work concludes that the GPD is not a universally compact summary in multiparameter persistence and that exact computation cannot generally be polynomial in the filtration size (Kim et al., 2024).

Taken together, these results delineate the present scope of persistence diagram universality. In one-parameter settings and in several structured extensions, universality can mean exact reconstruction, maximal stability, or canonical geometry. In multiparameter settings, by contrast, the literature identifies both partial survivals of barcode-like behavior and fundamental obstructions, including wild representation type and super-polynomial diagram growth.