pith. sign in

arxiv: 2606.26750 · v1 · pith:5Q3OKGDNnew · submitted 2026-06-25 · 🧮 math.CA · math.AP

Evolving edge weights via local entropy flow and cohesion flow on graphs

Juan Zhao , Jicheng Ma , Yunyan Yang , Liang Zhao This is my paper

Pith reviewed 2026-06-26 02:54 UTC · model grok-4.3

classification 🧮 math.CA math.AP
keywords local entropy flowcohesion flowedge weightsgraph flowscommunity detectionnode classificationasymptotic behaviorparabolic flows on graphs
0
0 comments X

The pith

Local entropy and cohesion flows evolve edge weights on graphs with global existence and uniqueness of solutions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper defines local entropy and cohesion quantities on graphs and constructs two corresponding parabolic flows that evolve the edge weights. It proves global existence and uniqueness for solutions of both flows and analyzes their asymptotic behavior, including limits that tend to positive infinity. These flows are then applied to network tasks such as community detection and node classification, where sequential use of the cohesion flow followed by the local entropy flow yields results competitive with Ollivier Ricci flow and Lin-Lu-Yau Ricci flow while lowering computational cost.

Core claim

By introducing local entropy and cohesion on graphs, the authors define the local entropy flow and the cohesion flow as evolution equations for edge weights. They establish global existence and uniqueness of solutions for both flows and study their long-term asymptotic behaviors, including the case in which the limit tends to positive infinity. The flows are applied to fundamental network analysis tasks including community detection and node classification, with empirical results competitive with established Ricci flows on graphs.

What carries the argument

The local entropy flow and cohesion flow, which are parabolic-type evolution equations for edge weights driven by the newly defined local entropy and cohesion quantities.

If this is right

  • Both flows admit global unique solutions whose asymptotic behavior can be tracked, including divergence to positive infinity.
  • Sequential application of the cohesion flow followed by the local entropy flow improves performance on community detection and node classification.
  • The resulting weighted graphs achieve accuracy competitive with Ollivier Ricci flow and Lin-Lu-Yau Ricci flow on benchmark tasks.
  • The flows are computationally efficient and reduce overall cost relative to the comparison methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The two flows together may offer a practical discrete substitute for continuous curvature flows when only graph data is available.
  • The asymptotic analysis could be used to predict when the evolved weights stabilize into community structure without additional clustering steps.
  • Extending the same quantities to directed or weighted multi-graphs might preserve the global existence result.

Load-bearing premise

The local entropy and cohesion quantities are well-defined and sufficiently regular on the graphs so that the resulting flows admit global smooth solutions.

What would settle it

A concrete graph on which either the local entropy flow or the cohesion flow develops a singularity or loses uniqueness in finite time would disprove the global existence and uniqueness claim.

Figures

Figures reproduced from arXiv: 2606.26750 by Jicheng Ma, Juan Zhao, Liang Zhao, Yunyan Yang.

Figure 1
Figure 1. Figure 1: Regular hexagon graph We next present an example of the two discrete flows. Example 3.2. We consider a graph consisting of two squares connected by four edges, as shown in [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Two squares connected by four edges We then demonstrate community detection via the discrete local entropy flow and the cohe￾sion flow in the following example. Example 3.3. Let G = (V, E, w0) be a graph composed of two triangles connected by the edge x3 x4, as shown in [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Community detection via discrete local entropy flow [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Three squares connected by a triangle 4. Proofs of the main theorems In this section, we present the proofs of the existence, uniqueness, and convergence of so￾lutions to the continuous local entropy flow (2.7) and the discrete local entropy flow (2.8). All proofs are established based on classical ordinary differential equation (ODE) theory. We also provide the corresponding proofs for the cohesion flow (… view at source ↗
Figure 5
Figure 5. Figure 5: Visualization of community detection results [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation study results on test accuracy for LEF [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Impact of the number of LEF iterations on test accuracy [PITH_FULL_IMAGE:figures/full_fig_p024_7.png] view at source ↗
read the original abstract

In this paper, we first propose two different quantities on graphs, namely local entropy and cohesion, then design two corresponding flows for edge weights: the local entropy flow and the cohesion flow. We establish the global existence and uniqueness of solutions for both flows and investigate their asymptotic behaviors, including the case that the limit goes to positive infinity. Moreover, they can be applied to fundamental network analysis tasks, including community detection and node classification. Empirical evaluations demonstrate that our method achieves performance competitive with Ollivier Ricci flow and Lin-Lu-Yau Ricci flow on benchmark network analysis tasks. In experimental scenarios, we first apply the cohesion flow to evolve the edge weights of the graph, and then apply the local entropy flow to further update the resulting weighted graph. Both flows are computationally efficient, leading to a significant reduction in overall computational cost and improved scalability.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces local entropy and cohesion quantities on graphs and the associated local entropy flow and cohesion flow on edge weights. It claims to prove global existence and uniqueness of solutions to both flows, to analyze their long-time asymptotics (including cases where the limit is positive infinity), and to apply the evolved weights to community detection and node classification, reporting performance competitive with Ollivier-Ricci and Lin-Lu-Yau Ricci flows. The method is applied sequentially (cohesion flow followed by local entropy flow) and is asserted to be computationally efficient.

Significance. If the global-existence statements are supported by explicit a priori bounds that rule out finite-time singularities while preserving positivity, the work would supply new, explicitly constructible dynamical systems on weighted graphs with potential advantages in scalability over curvature-based flows. The empirical competitiveness on standard benchmarks would then constitute a concrete, falsifiable contribution to network-analysis methodology.

major comments (2)
  1. [Abstract] Abstract and the global-existence statements: on a finite graph both flows reduce to autonomous ODE systems on the vector of edge weights. Local existence follows from Picard-Lindelöf, but the claimed global smooth solutions for arbitrary positive initial data require uniform a priori control preventing loss of positivity or ||w(t)|| → ∞ in finite time. The explicit mention of asymptotic regimes “including the case that the limit goes to positive infinity” indicates that unbounded solutions are contemplated; without a maximum principle, Lyapunov functional, or comparison argument that supplies the necessary bounds, the global-existence claim is not secured by standard ODE theory alone.
  2. [Applications section] The sequential application (cohesion flow followed by local entropy flow) is presented as the practical algorithm, yet the interaction between the two flows and the precise stopping criteria or termination conditions are not shown to preserve the global-existence guarantees established for each flow separately.
minor comments (2)
  1. Notation for the local entropy and cohesion functionals should be introduced with explicit formulas before the flows are defined; the current presentation leaves their precise dependence on the edge weights and vertex degrees implicit.
  2. The experimental section should include a brief description of the benchmark graphs, the precise community-detection and node-classification pipelines that use the evolved weights, and the number of independent runs used to report performance statistics.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and insightful comments on the global-existence claims and the sequential application of the flows. Below we address each major comment directly, indicating where revisions will be made to strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract] Abstract and the global-existence statements: on a finite graph both flows reduce to autonomous ODE systems on the vector of edge weights. Local existence follows from Picard-Lindelöf, but the claimed global smooth solutions for arbitrary positive initial data require uniform a priori control preventing loss of positivity or ||w(t)|| → ∞ in finite time. The explicit mention of asymptotic regimes “including the case that the limit goes to positive infinity” indicates that unbounded solutions are contemplated; without a maximum principle, Lyapunov functional, or comparison argument that supplies the necessary bounds, the global-existence claim is not secured by standard ODE theory alone.

    Authors: We agree that standard ODE theory on finite-dimensional systems requires explicit a priori bounds to guarantee global existence. In the manuscript (Theorems 3.3 and 4.2), global existence is established by exhibiting a strictly decreasing Lyapunov functional for each flow that yields uniform upper and lower bounds on all edge weights for every finite time interval, thereby ruling out both loss of positivity and finite-time blow-up. The asymptotic regimes in which weights tend to +∞ are analyzed only in the long-time limit t o∞; the Lyapunov control prevents any singularity at finite t. We will add a brief paragraph in the abstract and introduction explicitly referencing these Lyapunov estimates to make the argument self-contained. revision: partial

  2. Referee: [Applications section] The sequential application (cohesion flow followed by local entropy flow) is presented as the practical algorithm, yet the interaction between the two flows and the precise stopping criteria or termination conditions are not shown to preserve the global-existence guarantees established for each flow separately.

    Authors: The referee correctly notes that the composition of the two flows requires justification. Because each flow separately admits a unique global solution for any positive initial data, and because the output of the cohesion flow at any finite time (or at its equilibrium) remains strictly positive, it furnishes admissible initial data for the local-entropy flow; global existence for the second flow then follows directly from the already-proven result. The manuscript currently runs each flow for a fixed, finite number of iterations chosen empirically. We will revise the applications section to state this explicitly, to note that any finite-time truncation preserves positivity, and to add a short remark that the sequential procedure therefore inherits the global-existence guarantees of the individual flows. revision: yes

Circularity Check

0 steps flagged

No circularity: definitions and existence proofs are independent of the claimed results.

full rationale

The paper introduces local entropy and cohesion as new quantities on graphs, defines the corresponding parabolic flows on edge weights, and states that global existence/uniqueness and asymptotics are established by mathematical analysis. No equations or steps reduce a claimed prediction or uniqueness result to a fitted parameter or self-citation by construction. Empirical comparisons to Ollivier and Lin-Lu-Yau flows are external benchmarks. The derivation chain is self-contained against standard ODE/PDE theory on finite graphs and does not rely on renaming known results or smuggling ansatzes via self-citation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be identified from the given text.

pith-pipeline@v0.9.1-grok · 5673 in / 1049 out tokens · 74880 ms · 2026-06-26T02:54:05.810585+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

65 extracted references · 3 linked inside Pith

  1. [1]

    U. Alon, E. Yahav, On the bottleneck of graph neural networks and its practical implica- tions, in: International Conference on Learning Representations, 2021

    2021

  2. [2]

    Anand, G

    K. Anand, G. Bianconi, Entropy measures for networks: Toward an information theory of complex topologies, Phys. Rev. E 80 (4) (2009) 045102

    2009

  3. [3]

    S. Bai, B. Hua, Y . Lin, S. Liu, On the Ricci flow on trees, arXiv: 2509.22140, 2025

    arXiv 2025

  4. [4]

    S. Bai, A. Huang, L. Lu, S. T. Yau, On the sum of Ricci-curvatures for weighted graphs, Pure Appl. Math. Q. 17 (2021) 1599-1617

    2021

  5. [5]

    S. Bai, R. Li, S. Liu, X. Lai, Ricci flow on weighted digraphs with balancing factor, arXiv:2509.19989, 2025

    arXiv 2025

  6. [6]

    S. Bai, Y . Lin, L. Lu, Z. Wang, S. Yau, Ollivier Ricci-flow on weighted graphs, Amer. J. Math. 146 (2024) 1723-1747

    2024

  7. [7]

    S. Bai, S. Liu, X. Lai, The weighted Forman and Lin-Lu-Yau Ricci flow on graphs, arXiv:2601.02673, 2026

    arXiv 2026

  8. [8]

    Bauer, J

    F. Bauer, J. Jost, S. P. Liu, Ollivier-Ricci curvature and the spectrum of the normalized graph Laplace operator. Math. Res. Lett. 19 (2012) 1185-1205

    2012

  9. [9]

    Bhowmick, B

    S. Bhowmick, B. Seah, Clustering and summarizing protein-protein interaction networks: a survey, IEEE Trans. Knowl. Data Eng. 28 (2015) 638-658

    2015

  10. [10]

    Chien, J

    E. Chien, J. Peng, P. Li, O. Milenkovic, Adaptive universal generalized pagerank graph neural network, in: Int. Conf. Learning Representations (ICLR), 2021

    2021

  11. [11]

    B. Chow, F. Luo, Combinatorial Ricci flows on surfaces, J. Differential Geometry 63 (2008) 97-129

    2008

  12. [12]

    Clauset, M

    A. Clauset, M. Newman, C. Moore, Finding community structure in very large networks, Phys. Rev. E 70 (2004) 066111. 26

    2004

  13. [13]

    Cordasco, L

    G. Cordasco, L. Gargano, Community detection via semi-synchronous label propagation algorithms, 2010 IEEE International Workshop on: Business Applications of Social Net- work Analysis (BASNA), 1-8

    2010

  14. [14]

    T. M. Cover, J. A. Thomas, Elements of Information Theory, Wiley, 2nd edn., 2005

    2005

  15. [15]

    Craven, D

    M. Craven, D. DiPasquo, D. Freitag, A. McCallum, T. Mitchell, K. Nigam, S. Slattery, Learning to extract symbolic knowledge from the World Wide Web, in: Proceedings of the Fifteenth National/Tenth Conference on Artificial Intelligence/Innovative Applications of Artificial Intelligence, AAAI’98/IAAI’98, pp. 509-516, 1998

    1998

  16. [16]

    Danon, A

    L. Danon, A. Díaz-Guilera, J. Duch, A. Arenas, Comparing community structure identifi- cation, J. Stat. Mech. Theory Exp. (2005) P09008

    2005

  17. [17]

    Dehmer, Information processing in complex networks: Graph entropy and information functionals, Appl

    M. Dehmer, Information processing in complex networks: Graph entropy and information functionals, Appl. Math. Comput. 201 (2012) 82-94

    2012

  18. [18]

    Fortunato, Community detection in graphs, Phys

    S. Fortunato, Community detection in graphs, Phys. Rep. 486 (2010) 75-174

    2010

  19. [19]

    Gasteiger, A

    J. Gasteiger, A. Bojchevski, S. Günnemann, Predict then propagate: Graph neural networks meet personalized PageRank, in: Int. Conf. Learning Representations (ICLR), 2019

    2019

  20. [20]

    Gasteiger, S

    J. Gasteiger, S. Weißenberger, S. Günnemann, Diffusion improves graph learning, in: Pro- ceedings of the 33rd International Conference on Neural Information Processing Systems, pp. 13366-13378, Red Hook, NY , USA, 2019

    2019

  21. [21]

    Girvan, M

    M. Girvan, M. E. J. Newman, Community structure in social and biological networks, Proc. Natl. Acad. Sci. 99 (2002) 7821-7826

    2002

  22. [22]

    Hamilton, Three-manifolds with positive ricci curvature, J

    R. Hamilton, Three-manifolds with positive ricci curvature, J. Differ. Geom. 17 (1982) 255-306

    1982

  23. [23]

    W. L. Hamilton, R. Ying, J. Leskovec, Inductive representation learning on large graphs, in: Proc. 31st Int. Conf. Neural Inf. Process. Syst. (NeurIPS), pp. 1025-1035, 2017

    2017

  24. [24]

    Hubert, P

    L. Hubert, P. Arabie, Comparing partitions, J. Classif. 2 (1985) 193-218

    1985

  25. [25]

    Jeffreys, Theory of Probability, Oxford University Press, 1998

    H. Jeffreys, Theory of Probability, Oxford University Press, 1998

    1998

  26. [26]

    J. Jost, S. P. Liu, Ollivier’s Ricci Curvature, Local Clustering and Curvature-Dimension Inequalities on Graphs. Discrete Comput. Geom. 51 (2014) 300-322

    2014

  27. [27]

    T. N. Kipf, M. Welling, Semi-supervised classification with graph convolutional networks, in: Int. Conf. Learning Representations (ICLR), 2017

    2017

  28. [28]

    Kullback, R

    S. Kullback, R. A. Leibler, On information and sufficiency, Ann. Math. Statist. 22 (1951) 79-86

    1951

  29. [29]

    X. Lai, S. Bai, Y . Lin, Normalized discrete Ricci flow used in community detection, Phys. A 597 (2022) 127251

    2022

  30. [30]

    Leskovec, SNAP datasets: Stanford large network dataset collection,http://snap

    J. Leskovec, SNAP datasets: Stanford large network dataset collection,http://snap. stanford.edu/data, 2014

    2014

  31. [31]

    R. Li, F. Münch, The convergence and uniqueness of a discrete-time nonlinear Markov chain, J. Funct. Anal. 290 (2026) 111367

    2026

  32. [32]

    Y . Lin, S. Liu, The Ricci flow with prescribed curvature on graphs, arXiv:2603.10479, 2026

    Pith/arXiv arXiv 2026

  33. [33]

    Y . Lin, L. Lu, S. T. Yau, Ricci curvature of graphs, Tohoku Math. J. 63 (2011) 605-627

    2011

  34. [34]

    J. Ma, Y . Yang, A modified Ricci flow on arbitrary weighted graph, J. Geom. Anal. 35 (2025) 332

    2025

  35. [35]

    J. Ma, Y . Yang, Evolution of weights on a connected finite graph, arXiv:2411.06393, 2024

    arXiv 2024

  36. [36]

    J. Ma, Y . Yang, Piecewise-linear Ricci curvature flows on weighted graphs, arXiv:2505.15395, 2025

    arXiv 2025

  37. [37]

    J. Ma, Y . Yang, J. Zhao, L. Zhao, Geometric evolution graph convolutional networks: En- 27 hancing graph representation learning via Ricci flow, arXiv:2603.26178, 2026

    Pith/arXiv arXiv 2026

  38. [38]

    McAuley, C

    J. McAuley, C. Targett, Q. Shi, A. van den Hengel, Image-based recommendations on styles and substitutes, in: Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR’15, pp. 43-52, 2015

    2015

  39. [39]

    Monti, D

    F. Monti, D. Boscaini, J. Masci, E. Rodola, J. Svoboda, M.M. Bronstein, Geometric deep learning on graphs and manifolds using mixture model CNNs, in: Proc. IEEE Conf. Com- put. Vis. Pattern Recognit. (CVPR), pp. 5115-5124, 2017

    2017

  40. [40]

    Newman, Modularity and community structure in networks, Proc

    M. Newman, Modularity and community structure in networks, Proc. Natl. Acad. Sci. 103 (2006) 8577-8582

    2006

  41. [41]

    Newman,Networks, Oxford Univ

    M. Newman,Networks, Oxford Univ. Press (2018)

    2018

  42. [42]

    C. C. Ni, Y . Y . Lin, F. Luo, J. Gao, Community detection on networks with Ricci flow, Sci. Rep. 9 (2019) 9984

    2019

  43. [43]

    Ollivier, Ricci curvature of metric spaces, C

    Y . Ollivier, Ricci curvature of metric spaces, C. R. Math. 345 (2007) 643-646

    2007

  44. [44]

    Ollivier, Ricci curvature of markov chains on metric spaces, J

    Y . Ollivier, Ricci curvature of markov chains on metric spaces, J. Funct. Anal. 256 (2009) 810-864

    2009

  45. [45]

    Perelman, The entropy formula for the Ricci flow and its geometric applications, arXiv:0211159, 2002

    G. Perelman, The entropy formula for the Ricci flow and its geometric applications, arXiv:0211159, 2002

    2002

  46. [46]

    Reichardt, S

    J. Reichardt, S. Bornholdt, Statistical mechanics of community detection, Phys. Rev. E 74 (2006) 016110

    2006

  47. [47]

    Rozemberczki, C

    B. Rozemberczki, C. Allen, R. Sarkar, Multi-scale attributed node embedding, J. Complex Networks 9(2) (2021) cnab014

    2021

  48. [48]

    P. Sen, G. Namata, M. Bilgic, L. Getoor, B. Gallagher, T. Eliassi-Rad, Collective classifi- cation in network data, AI Magazine, 29(3) (2008) 93-106

    2008

  49. [49]

    Sengupta, N

    P. Sengupta, N. Azarhooshang, R. Albert, B. DaGupta, Finding influential cores via nor- malized Ricci flows in directed and undirected hypergraphs with applications, Phys. Rev. E 111 (2025) 044316

    2025

  50. [50]

    C. E. Shannon, A mathematical theory of communication, Bell System Tech. J. 27 (1948) 379-423

    1948

  51. [51]

    D. Shi, Y . Guo, Z. Shao, J. Gao, How curvature enhances the adaptation power of framelet GCNs, arXiv:2307.09768, 2023

    arXiv 2023

  52. [52]

    J. Tang, J. Sun, C. Wang, Z. Yang, Social influence analysis in large-scale networks, in: Proc. 15th ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining (KDD’09), New York, NY , USA, 2009, pp. 807-816

    2009

  53. [53]

    Tauro, C

    S. Tauro, C. Palmer, G. Siganos, M. Faloutsos, A simple conceptual model for the internet topology, GLOBE COM’01 IEEE Global Telecommun. Conf. 3 (2001) 1667-1671

    2001

  54. [54]

    Y . Tian, J. Ma, Y . Yang, L. Zhao, Community detection of undirected hypergraphs by Ricci flow, Phys. Rev. E 112 (2025) 044311

    2025

  55. [55]

    Topping, F

    J. Topping, F. DiGiovanni, B. P. Chamberlain, X. Dong, M. M. Bronstein, Understanding over-squashing and bottlenecks on graphs, in: Int. Conf. Learning Representations (ICLR), 2022

    2022

  56. [56]

    Veli ˇckovi´c, G

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Liò, Y . Bengio, Graph attention networks, in: Int. Conf. Learning Representations (ICLR), 2018

    2018

  57. [57]

    Weber, J

    M. Weber, J. Jost, E. Saucan, Forman-Ricci flow for change detection in large dynamic data sets, Axioms 5 (2016) 26

    2016

  58. [58]

    Weber, E

    M. Weber, E. Saucan, J. Jost, Characterizing complex networks with Forman-Ricci curva- ture and associated geometric flows, J. Complex Networks 5 (2017) 527-550

    2017

  59. [59]

    K. Xu, C. Li, Y . Tian, T. Sonobe, K. Kawarabayashi, S. Jegelka, Representation learning on 28 graphs with jumping knowledge networks, in: Proc. Int. Conf. Machine Learning (ICML), PMLR, pp. 5453-5462, 2018

    2018

  60. [60]

    Z. Ye, K. S. Liu, T. Ma, J. Gao, C. Chen, Curvature graph network, in: Int. Conf. Learn. Represent. (ICLR), 2020

    2020

  61. [61]

    Zachary, An information flow model for conflict and fission in small groups, J

    W. Zachary, An information flow model for conflict and fission in small groups, J. Anthro- pol. Res. 33 (1977) 452-473

    1977

  62. [62]

    Zheng, B

    X. Zheng, B. Zhou, J. Gao, Y . Wang, P. Lió, M. Li, G. Montufar, How framelets enhance graph neural networks, in: Int. Conf. Mach. Learn. (ICML), PMLR, pp. 12761-12771, 2021

    2021

  63. [63]

    J. Zhao, J. Ma, Y . Yang, L. Zhao, Core detection via Ricci curvature flows on weighted graphs, Physica A 692 (2026) 131525

    2026

  64. [64]

    J. Zhao, J. Ma, Y . Yang, L. Zhao, Finding core subgraphs of directed graphs via discrete Ricci curvature flow, arXiv:2512.07899, 2025

    arXiv 2025

  65. [65]

    J. Zhao, J. Ma, Y . Yang, L. Zhao, An efficient entropy flow on weighted graphs: theory and applications, arXiv:2604.08144, 2026. 29

    Pith/arXiv arXiv 2026