Ricci Flow on Weighted Digraphs with Balancing Factor
Shuliang Bai, Rui Li, Shuang Liu, Xin Lai · 2025
Read Ricci Flow on Weighted Digraphs with Balancing Factor by Shuliang Bai, Rui Li, Shuang Liu, Xin Lai (2025) 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-2509-19989.txt.
Introduction
Directed graphs, where edges run one way, resist the curvature tools built for symmetric networks. Shuliang Bai and three co-authors define Ricci flow for weighted directed graphs. The flow evolves edge weights while preserving distances, and solutions exist and are unique, they prove. A node-wise balancing factor tunes the tug between inflow and outflow, capturing asymmetry directly.
A discrete version of the flow runs on real machines. Numerical studies test the machinery. Ricci flow on undirected graphs already serves community detection and representation learning. Extending it to directed settings covers citation webs, traffic, and follower networks as they are. The balancing factor is the design's signature: direction is information, and the flow now respects it.
Catalog record
| CREATOR | Shuliang Bai, Rui Li, Shuang Liu, Xin Lai |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-19989
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Ricci Flow on Weighted Digraphs with Balancing Factor Shuliang Bai, Rui Li, Shuang Liu, Xin Lai Abstract Ricci curvature and Ricci flow have proven to be powerful tools for analyzing the geometry of discrete structures, particularly on undirected graphs, where they have been applied to tasks ranging from community detection to graph representation learning. However, their development on directed graphs remains limited, with Ricci flow being especially underexplored. In this work, we introduce a rigorous formulation of Ricci flow on directed weighted graphs, which evolves edge weights while preserving distances, and establish both the existence and uniqueness of its solutions. To capture the essence of asymmetry in directed networks and to enhance the capability of modeling more flexible structures, we incorporate a node-wise balancing factor that regulates between outflow and inflow. Building on the continuous Ricci flow evolution framework, we propose a discrete Ricci flow algorithm that is applicable to numerical computing. Numerical studies on various directed graph examples demonstrate the capacity of the proposed flow to reveal structural asymmetry and dynamic evolutions. 1 Introduction Graphs are pervasive in representing data structures in areas such as Internet of Things (IoT) [ 1 , 2 ] , biology [ 3 ] , social network [ 4 ] , supply chain analysis [ 5 ] , etc. Directed graphs in particular capture causal/preferential/directional relationships between entities. In recent years, discrete Ricci curvature has emerged as a powerful tool to analyze network structure. Ricci curvature and Ricci flow are fundamental notions in Riemannian geometry, playing a significant role in understanding manifold geometry. Introduced by Hamilton [ 6 ] , the Ricci flow evolves a Riemannian metric g(t) using Ricci curvature \mathrm{Ric}(g) according to \frac{\partial g}{\partial t}=-2\,\mathrm{Ric}(g), progressively smoothing metric irregularities and driving the manifold toward canonical geometries. This technique led to significant breakthroughs, most notably Perelman’s resolution of the Poincarè conjecture and Thurston’s geometrization conjecture [ 7 , 8 ] . Following these landmark results, Ricci flow has become a central tool in the study geometric evolution. Motivated by these developments, researchers have sought discrete analogues of Ricci curvature and Ricci flow on graphs. Several notions of discrete Ricci curvature have been developed for graphs. Among the most widely studied are Ollivier’s Ricci curvature [ 9 ] on metric spaces using optimal transportation, Lin–Lu–Yau (LLY) curvature [ 10 ] on graphs using a limit form of Ollivier’s Ricci curvature that is better adapted to graphs, and Forman Ricci curvature [ 11 , 12 ] on graphs using combinatorial approaches. Each of them has solid theoretical foundations and computational properties. Ollivier and LLY curvatures maintain a strong connection to the continuous theory but require computing Wasserstein distances, which can be expensive for large networks. Forman curvature offers simplicity and scalability but lacks some geometric richness. Building on these definitions of Ricci curvature, discrete Ricci flows have been proposed for theoretical studies and applications in network analysis, including network robustness, community detection, bottleneck identification [ 13 , 14 , 15 ] . However, most existing work has focused on undirected graphs , where symmetric interactions simplify the underlying structure. The study of Ricci flow on directed graphs , which inherently exhibit asymmetry and directional dependence, remains largely unexplored. This gap motivates this work to adapt the Ricci flow models to directed and weighted networks, capable of capturing their anisotropic geometry. Directed weighted graphs are ubiquitous in modeling real-world systems. They exhibit inherent asymmetry that significantly influences network structure and dynamics. Recent surveys on machine learning for directed graphs highlight that geometric approaches remain far less developed than their counterparts for undirected graphs [ 16 ] . This scarcity hinders the practical effectiveness for inherently directed data, such as road networks for traffic forecasting and control [ 17 , 18 ] , gene regulatory networks for directional biochemical reaction control [ 19 ] , and social network for information spreading analysis [ 20 ] . Simply adapting directed data to algorithms which are designed for undirected data is critical because many complex systems cannot be accurately represented without accounting for directional dependencies and imbalanced interactions. Incorporating directionality into Ricci curvature and Ricci flow algorithms is therefore essential for understanding anisotropic geometry, improving network robustness, and guiding structural optimization. Furthermore, since the LLY’s Ricci curvature on graphs is based on the optimal transportation between probability distributions, to capture directional heterogeneity, we introduce a node-wise balancing factor \beta , which regulates the ratio of outflow to inflow at each node. This mechanism provides a richer representation of flow asymmetry and facilitates the directed Ricci flow adapting to both directionality and local structural imbalance. Another notable limitation of most existing Ricci flow formulations on graphs is that they do not distinguish between edge weight and the distance metric of the network. For instance, in the traditional Ollivier Ricci flow [ 21 ] , edge weights are updated by formulas such as \frac{d}{dt}w_{e}(t)=-\kappa_{e}(t)\,w_{e}(t),\quad\text{or}\quad w_{xy}^{(i+1)}=\bigl(1-\epsilon\,\kappa_{xy}^{(i)}\bigr)\,d^{i}(x,y), and other variants [ 22 , 23 , 24 ] , implying that distances evolve alongside weights, coupling the network’s distance-related geometric structure with its weight-related dynamic properties. Although such coupling has proven effective for tasks like community detection, it becomes problematic in scenarios where physical distances, such as road lengths, are inherently fixed, while weights (such as road width or capacity) evolve along Ricci flow process. These formulations conflate the dual role of edge weights, i.e., the connection strength and metric distance, obscuring the distinct functions of them. In this work, we introduce a Ricci flow that decouples edge weight from the metric distance, allowing the weights to evolve while maintaining the static distance metric. This design better reflects operational dynamics in real-world systems, such as traffic networks [ 25 ] , communication systems where latency remains constant while bandwidth evolves, and transport or logistics networks where geography is immutable but flow conditions change. In a nutshell, this separation will enable more applicable scenarios for further studies. The main contributions of this work are threefold. Firstly, we extend the concept of Ricci flow to weighted directed graphs and propose a new Ricci flow formulation that preserves the underlying distance metric while evolving edge weights , decoupling distance-related geometry from weight-related dynamics. Secondly, we introduce a node-wise balancing factor \beta , which flexibly regulates the bias between outflow- and inflow-based transport distributions, enabling the model to capture local directional properties and flow asymmetry. Lastly, we prove the existence and uniqueness of the solution to the proposed Ricci flow framework for weighted directed graphs, providing a rigorous theoretical foundation for Ricci flow evolution in weighted directed networks. The remainder of this paper is structured as follows. Section 2 reviews definitions and properties of directed graphs. Section 3 introduces the directed LLY’s Ricci curvature with a balancing factor. Section 4 formulates the directed Ricci flow and proves the existence and uniqueness of its solution. Section 5 gives the discrete time version of our Ricci flow formulation, making it applicable to computational network analysis. Section 6 presents illustrative examples and simulations. Finally, Section 7 concludes and outlines future directions. 2 Preliminaries A directed graph (digraph) {\mathcal{G}}=({\mathcal{V}},{\mathcal{E}}) consists of a finite vertex set {\mathcal{V}} and a set of directed edges {\mathcal{E}}\subseteq{\mathcal{V}}\times{\mathcal{V}} , and n=|{\mathcal{V}}| , m=|{\mathcal{E}}| . Different from the