Forman–Ricci Curvature on Contact-Sequence Temporal Networks via Spatiotemporal Prism Complexes

Taiki Yamada · 2026

Read Forman–Ricci Curvature on Contact-Sequence Temporal Networks via Spatiotemporal Prism Complexes by Taiki Yamada (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-15685.txt.

Introduction

Taiki Yamada builds a geometric frame for temporal contact networks. Past work flattens time into snapshots. Yamada lifts the network into a simplicial complex via a prism build. On that complex he develops Forman-Ricci curvature and contrasts it with an augmented variant. The two agree under uniform weights, with a closed-form gap otherwise. Tests on synthetic models and a contact data set confirm the theory. The variants disagree on 56 to 67 percent of the 1-simplices, mostly temporal and diagonal ones, yet stay strongly correlated.

Catalog record

CREATORTaiki Yamada
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75512
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.15685

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Forman–Ricci Curvature on Contact-Sequence Temporal Networks via Spatiotemporal Prism Complexes Taiki Yamada Abstract Temporal networks—sequences of time-stamped contacts among nodes—constitute the finest-grained representation of dynamic interaction data; however, geometric and topological analyses of such networks have remained largely confined to time-aggregated or snapshot-based approximations. Such reductions destroy the temporal ordering and interevent statistics essential for understanding spreading dynamics, synchronization, and information flow. This study proposes a geometric framework that lifts a contact-sequence temporal network into a genuine simplicial complex through a prism construction adapted from algebraic topology. On this spatiotemporal prism complex, we develop the Forman–Ricci curvature in its original CW-complex form and contrast it with an augmented variant widely used in network science. We prove that the two variants coincide under uniform weights, derive a closed-form expression for their pointwise discrepancy in the general case, and identify the precise conditions under which they diverge—conditions generically satisfied in temporal networks because temporal edges carry interval-dependent weights. Numerical experiments on three synthetic contact-network models (Erdős–Rényi, activity-driven, and bursty) and on the SocioPatterns Hypertext 2009 face-to-face contact dataset quantitatively confirm the theoretical predictions: the two Forman variants disagree on 56 56 – 67 % 67\% of the 1 1 -simplices—predominantly the temporal and diagonal simplices—while remaining strongly correlated according to the Pearson coefficient. The proposed framework provides a principled, parameter-free method for assigning discrete Ricci curvature to each contact event, thereby opening a new geometric avenue for temporal data analysis. I Introduction Most empirical network data are intrinsically temporal 5 ; 11 . Records of human face-to-face contacts, neuronal spike trains, financial transactions, online communications, and contact-tracing data naturally occur as contact sequences : ordered triples (i,j,t_{k}) , specifying that nodes i and j interacted at time t_{k} . This representation constitutes the most refined description available. Any coarser representation, including the familiar snapshot sequence and time-aggregated graph, can be recovered from the contact sequence through a (necessarily lossy) projection. Despite its conceptual simplicity, the analytical framework for contact-sequence temporal networks remains comparatively underdeveloped. Most existing geometric and topological tools have been designed for static graphs and are therefore applied either to aggregated graphs or, at best, to sequences of static snapshots 9 ; 15 ; 18 . Accordingly, researchers have conflated two qualitatively distinct sources of information about a temporal network: (1). the spatial structure of interactions at a given moment, and (2). the temporal ordering and interevent statistics of contacts, which govern causal reachability and burstiness 8 ; 10 . Snapshot representations preserve (1) at a chosen temporal resolution but obliterate (2) within each observation window, whereas aggregated representations obliterate (2) entirely. Consequently, geometric quantities such as discrete curvature, when computed on snapshots, depend uncontrollably on temporal binning and cannot meaningfully reflect the causally ordered temporal structure. In particular, discrete Ricci curvature has emerged as a powerful descriptor of network geometry because it provides local indicators of bottleneck structure, community separation, and robustness 13 ; 17 ; 19 ; 16 . Two principal variants are currently in use: the optimal-transport-based Ollivier–Ricci curvature 13 and the combinatorial Forman–Ricci curvature 3 , with the latter originally formulated on weighted CW complexes. In network science literature, an augmented variant of Forman’s curvature 19 ; 16 is considerably more popular than the original formula because it admits a particularly simple closed-form expression involving only node degrees and triangle counts. Whether this convenient simplification preserves the geometric content of Forman’s original definition, and whether the answer changes when the underlying complex is intrinsically temporal, has not, to the best of our knowledge, been systematically investigated. This study aims to overcome existing limitations. Specifically, we propose a mathematical framework in which contact-sequence temporal networks are represented as genuine simplicial complexes through the prism construction of algebraic topology 4 , thereby yielding the spatiotemporal prism complex . \mathcal{K}_{\mathrm{ST}} . On this complex, we compute discrete curvatures and analyze how these values reflect both spatial and temporal structure. The main contributions of this study are as follows. (i) A definition of \mathcal{K}_{\mathrm{ST}} that is rigorous, coordinate-free, and naturally inherits the time ordering of the underlying contact sequence. In particular, \mathcal{K}_{\mathrm{ST}} is a bona fide simplicial complex (not merely a clique-rule shorthand) because its construction through prism subdivision automatically generates all faces of every dimension, including the diagonal 1 -simplices that mediate spatial connectivity and temporal persistence in the data. (ii) A precise formulation of two Forman-type curvatures on \mathcal{K}_{\mathrm{ST}} —the original CW formula 3 and the augmented form 19 —along with the proof that they coincide under uniform weights (Lemma 2 ) and a closed-form expression for their pointwise discrepancy in the general case (Proposition 3 ). As temporal edges carry weights that depend on the inter-contact interval, the two variants are generically inequivalent in temporal networks. We make this dependence explicit: spatial edges always satisfy \mathbf{F}=\mathbf{F}^{\mathrm{aug}} on \mathcal{K}_{\mathrm{ST}} , whereas temporal and diagonal edges differ by an amount that is linear in 1-g(\Delta t) to leading order, where g denotes the temporal-edge weight function and \Delta t denotes the time gap. (ii ′ ) Two structural results exploit the prism subdivision itself beyond the \mathbf{F} versus \mathbf{F}^{\mathrm{aug}} comparison: a monotonicity statement asserting that persistent spatial simplices acquire non-negative curvature shifts from prism cofaces (Proposition 8 ), thereby formalizing why long-lived simplices appear less negatively curved than transient simplices in our experiments; and a discrete Gauss–Bonnet identity expressing the total Forman curvature of \mathcal{K}_{\mathrm{ST}} as a sum of snapshot Euler characteristics corrected by persistent-simplex overlaps (Proposition 9 ). The proof of Proposition 8 relies on a curvature-decomposition lemma (Lemma 7 ) that decomposes the Forman–Ricci curvature of any spatial edge into a static intra-snapshot component and a controlled prism correction; this lemma is of independent interest because it isolates the geometric content of prism construction. (iii) An empirical study of three synthetic contact-network models and the SocioPatterns Hypertext 2009 face-to-face contact dataset. We find that the two Forman variants differ numerically on 56 – 67\% of 1 -simplices—almost entirely on temporal and diagonal edges, as the theory predicts—while remaining strongly correlated according to the Pearson coefficient ( \rho\geq 0.83 in all four networks). This result indicates that the augmented form can serve as a fast proxy for rank-based analyses, but the original form is required whenever curvature magnitudes are compared across edges with different temporal extents. The remainder of this paper is organized as follows. Section II establishes the terminology, reviews the background on simplicial complexes and Forman–Ricci curvature, defines the spatiotemporal prism complex, and presents the two curvature formulas that we compare. Section III establishes the equivalence and inequivalence regimes between the two curvature variants, and proves the supporting lemmas. Section IV empirically analyzes the synthetic and real-contact sequence datasets. Section V discusses the methodological implications, positions our framework within existing temporal network geometry literature, and outlines future research directions. II Preliminaries In this section, we present definitions of contact-sequence temporal networks, simplicial complexes, and Forman curvature. II.1 Contact-sequence temporal networks Definition 1 (Contact sequence) . A contact-sequence temporal network on a node set V is represented by the following set of triples. \mathcal{C}=\bigl\{(i_{k},j_{k},t_{k}):k=1,\dots,N\bigr\}\subset V\times V\times\mathbb{R}, (1) where i_{k}\neq j_{k} , and the times t_{k} are not assumed to be distinct. We denote by T(\mathcal{C})=\{t:(i,j,t)\in\mathcal{C}\text{ for some

Read the complete text