Pith. sign in

REVIEW 4 major objections 5 minor 52 references

Ensemble-level loopy message passing with generalized-edge closure for percolation

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read An ensemble-level message-passing scheme based on generalized edges—overlaps of neighborhoods—yields a second-order recursion whose percolation thresholds match simulations on clustered networks, and the newly defined generalized-edge…

desk verdict Genuinely new ensemble-level loopy message-passing framework with a plausible diagnostic, but the central factorization is asserted, not proven, and the derivation sits in an unavailable supplement. read the letter →

arxiv 2608.09397 v1 pith:VXSAAORC submitted 2026-08-10 cond-mat.stat-mech

L.-H. Wang , Y.-M. Du This is my paper · ORCID
classification cond-mat.stat-mech MSC 82B4305C80 PACS 64.60.aq89.75.Hc
keywords percolationthresholdhighlyclusterednetworksrandomgraphensemblesmessagepassinggeneralizededgegeneralized-edgeclosurecoefficientloop-inducedcorrelationssite
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

On clustered networks, short loops break the independence assumption that makes standard message passing tractable, so percolation thresholds are hard to predict from local statistics alone. This paper develops an ensemble-level message-passing framework in which messages travel along generalized edges—overlaps of local subgraphs around pairs of nodes—rather than single edges. The second-order approximation, Eq. (9), retains degree, distance, shared-neighbor count, and residual size, and it produces threshold predictions that track Monte Carlo simulations where tree-like and first-order equations fail. The paper also introduces the generalized-edge closure coefficient (GECC), a structural statistic computed entirely from local information, which indicates in advance whether a given approximation order is reliable. The payoff would be an a priori certificate: from local statistics alone, one can tell when low-order predictions are trustworthy and when higher orders are needed.

What carries the argument

The central object is the nth-order generalized edge E_{kd}^{(n)} = $Γ_k^{{(n)}}$ ∩ $Γ_d^{{(n)}}$, the overlap of the order-n subgraphs around two nodes; the message is carried by the subgraph pair ($Γ_k^{{(n)}}$, $Γ_d^{{(n)}}$), split into the generalized edge and the residual structure R_{d→k}^{(n)}. A single structural function $A^{{(n)}}$ absorbs all correlations induced by overlapping generalized edges and factorizes under the assumed independence of shell-1-neighbor messages. The tractable engine is Eq. (9), the second-order recursion ϵ_k = q Σ_{d,y,w} m_{kdyw}^{(2)} g(q,y,w) ϵ_d P(d,y,w|k), where g(q,y,w)=1-(1-$q^{{y-1}}$)^w is the connection probability through at least one occupied shared neighbor and $m^{{(2)}}$ is the mean residual size. The diagnostic $GECC^{{(n)}}$ averages the fraction of internal node pairs that fail to re-induce the same generalized edge; low GECC certifies closure at that order.

What would settle it

Run site percolation on a random graph ensemble with strong shell-1-neighbor correlations—for example, where each node's direct neighbors form dense shared neighborhoods—while keeping GECC(2) small; if Eq. (9)'s predicted threshold deviates substantially from Monte Carlo, the independence assumption underpinning the recursion is false and closure alone is not sufficient.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that loop-induced correlations in random clustered graph ensembles can be propagated at the ensemble level by replacing single-edge transmission with generalized edges, defined as intersections of order-n subgraphs around pairs of nodes. The message-passing recursion (Eq. 4) is exact at any order, and linearizing it around the percolation transition yields Eq. (6). Retaining second-order structural features—the distance y within the generalized edge, the number w of shared shell-1 neighbors, and the mean residual size m—produces the tractable recursion Eq. (9), whose threshold prediction beats the zeroth-order (tree-like) and first-order (triangle-aware) equations on synthetic and real networks. The paper further claims that the accuracy of a given order is governed by generalized-edge closure, quantified by GECC: when internal node pairs re-induce the original generalized edge, the order is reliable; when they do not, residual threshold error grows, as demonstrated by a controlled synthetic sweep and by four real networks.

Load-bearing premise

The derivation assumes that messages from the direct neighbors of a node are mutually independent, even though those neighbors can share shell-2 nodes or be directly linked in a clustered graph.

Editorial extensions

If this is right

  • The zeroth- and first-order equations reappear as strict limits of the same generalized-edge framework, so the new scheme is a generalization rather than a separate theory.
  • On networks with good second-order closure, q_c^{(2)} follows simulated thresholds closely, so structural statistics alone can replace guesswork about low-order reliability.
  • On networks with GECC(2) close to one, even the second-order prediction remains far from simulation, and GECC identifies these networks before any percolation simulation is run.
  • GECC is computed purely from local statistics and involves no percolation dynamics, so it can warn in advance that a given truncation order is inadequate and that higher-order or alternative treatments are needed.
  • Because the framework operates at the ensemble level and does not require a realized adjacency matrix, it applies to large clustered random-graph ensembles described by local statistics.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If GECC remains small at every tested order, the argument suggests the threshold estimate is converging; a natural testable extension is to compute GECC(n) progressively and stop when it decays, a procedure the authors flag for future work.
  • Because GECC depends only on local structure, it could be adapted to other message-passing-based quantities on clustered networks, such as epidemic thresholds or influence maximization; the paper names these as future directions.
  • The synthetic sweep varies only the conversion parameter φ, so the near-linear relation between GECC(2) and the residual threshold gap is evidence for closure as the driver; a complementary test would hold GECC fixed while changing loop density to see whether closure, not loop density, predicts error.
  • The independence assumption in Eq. (5) is the structural weak point; if shell-1-neighbor messages are correlated, one could try to rescue the framework by conditioning on shared shell-2 neighborhoods, which would be a direct refinement of A^{(n)}.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript develops an ensemble-level loopy message-passing framework for site percolation on random clustered graphs. The authors define nth-order generalized edges as intersections of nth-order local subgraphs, propose an exact recursion (Eq. 4) that is then linearized and reduced to a second-order equation (Eq. 9) depending on degree, distance, shared-neighbor count, and mean residual size. They introduce the generalized-edge closure coefficient (GECC) as an a priori diagnostic of whether a given approximation order is reliable, and test the framework on synthetic networks generated by a controlled unit-replacement procedure and on four real networks from the SNAP collection. The paper reports that GECC(2) tracks the residual second-order threshold error in the synthetic sweep and classifies the real networks into regimes of adequate and inadequate second-order closure.

Significance. If the central recursions are sound, the paper offers a useful step beyond prescribed-motif and single-realization treatments of loopy percolation, with a tractable second-order equation and a structural diagnostic that is computed without percolation simulations. The lower-order limits correctly reduce to known tree-like (Eq. 12) and triangle-aware (Eq. 11) equations, and the synthetic conversion sweep is a clean controlled test showing co-response between GECC(2) and residual threshold error. The validation is not circular: thresholds are computed from independent Monte Carlo simulations, and GECC is computed from local structure alone. The machine-checkable aspects are limited, however, because the derivation of the central recursions and the algorithmic details are deferred to a Supplemental Material that is not included in the manuscript.

major comments (4)
  1. [Sec. II B 2, Eq. (5)] The factorization of A^(n) in Eq. (5) is load-bearing but unsupported. The text claims that message contributions from shell-1 neighbors of the degree-d node 'remain mutually independent because they are direct neighbors,' but in a loopy graph two shell-1 neighbors can be connected by an edge or through a common shell-2 node after the focal node is deleted, so their residual cluster sizes can be correlated. Since Eq. (6) is obtained by linearizing Eq. (4) after this factorization, and Eq. (9) inherits the same assumption, the central second-order result currently rests on an unproved independence claim. The manuscript should either prove the factorization under a stated ensemble condition or test it explicitly, for example by comparing the factorized expression with exact enumeration on small loopy ensembles.
  2. [Sec. II C 2, Eq. (8)] Replacing the residual-size distribution P(m|d,l,y,w) by its mean m^(2)_dlyw in Eq. (8) is another unexamined reduction. The percolation threshold is set by the spectral radius of the message-passing matrix, and replacing random coefficients by their averages inside a spectral-radius condition is not generally exact. The manuscript gives no argument that the threshold is insensitive to fluctuations of m or that the mean-field replacement is controlled. This matters because Eq. (9), the main second-order prediction, uses exactly this averaged quantity.
  3. [Supplemental Material and Sec. II] The derivations of Eq. (4), Eq. (5), Eq. (6), the detailed form of the correlation structure A^(2), and the computational procedure for GECC are all deferred to a Supplemental Material that is not included with the manuscript. A referee cannot verify the central recursions, the claimed recovery of lower-order limits, or the GECC algorithm from the main text alone. The authors should provide the Supplement or move the essential derivations into the main text before the technical claims can be assessed.
  4. [Sec. II D 2 and Sec. II E 3] The claim that GECC serves as an 'a priori validity certificate' is stronger than what the evidence supports. GECC measures one particular consistency property, namely whether internal node pairs re-induce the same generalized edge, but it does not measure the message independence asserted in Eq. (5), nor can it rule out longer-range correlations that only appear at higher orders, as the Discussion itself concedes. The empirical support is a single synthetic sweep showing an approximately linear relation between GECC(2) and the residual threshold error, plus four real networks. This is suggestive but not a certificate; the wording in Sec. II E 3 ('elevate GECC from a descriptive correlate ... into an a priori validity certificate') should be softened or supported by additional controlled tests, such as multiple synthetic families with different loop structures and a demonstration that small GECC(n) reliably implies small threshold error.
minor comments (5)
  1. [Sec. II C 2, Eq. (7)] The notation '−→' in Eq. (7) is nonstandard and unclear; the intended reduction from graph-dependent variables to ensemble-level features should be stated in words or with a more conventional arrow notation.
  2. [Sec. II D 1, Eq. (13)] The symbol R^(n)_G for the set of generalized-edge labels could be confused with the residual structure R^(n)_d→k defined in Eq. (2); a different symbol would improve readability.
  3. [Sec. II E 2 and Fig. 2] The text says the pale-blue shading 'provides a visual guide' for GECC(2), but the shading is not explained in the caption; it should either be removed or described precisely.
  4. [Methods and Data Availability] The data-availability statement says the code and processed data 'will be made publicly available upon publication'; for reproducibility, the authors should deposit the code and data at the time of submission, especially since the Supplemental Material is currently unavailable.
  5. [Sec. I, Introduction] The phrase 'To our knowledge, this is the first ensemble-level formulation' is a novelty claim that would benefit from a more precise comparison with existing ensemble methods, for example clarifying what distinguishes the generalized-edge construction from the motif-based ensembles of Refs. [21,22,31-37].

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: thresholds and GECC are derived from local structural statistics and validated against independent Monte Carlo simulations.

full rationale

The paper's derivation chain is self-contained. The second-order threshold q_c^(2) is obtained by linearizing the message-passing recursion and reducing it to Eq. (9), whose inputs are local ensemble statistics (degree, distance, shared-neighbor count, and mean residual size) estimated directly from the network under study; no parameter is fitted to the simulated threshold q_c^sim, and no percolation simulation output enters the definition of GECC(n) in Eqs. (16)-(17). The recovery of the tree-like and first-order equations (Eqs. (11)-(12)) is an explicit consistency check rather than a circular reuse of the target result. The framework's central approximation, Eq. (5), factorizes A(n) by declaring shell-1 neighbor messages independent 'because they are direct neighbors'; this is an unproved and potentially false assumption in loopy graphs, and the paper itself acknowledges validation only up to second order and notes GECC's blind spot for longer-range correlations. However, an unsupported approximation, even if load-bearing, is a correctness risk rather than a circular step: it does not make the prediction equivalent to its inputs by construction, and the numerical comparison against Monte Carlo is external. The only self-citation ([8], Wang & Du 2025) appears in a general list of percolation references and is not load-bearing. Therefore no circularity is identified.

Assumptions & free parameters 0 free parameters · 4 assumptions · 2 invented entities

The theoretical derivation introduces no fitted constants; its inputs are measured local statistics. The main burden is carried by three uncontrolled modeling assumptions: shell-1 message independence, replacement of the residual-size distribution by its mean, and the identification of closure with approximation validity. Synthetic construction parameters such as d0, m_min, m_max, and phi are validation choices, not part of the theoretical claim, so they are not listed as free parameters.

assumptions (4)
  • domain assumption The ensemble can be described by degree-conditioned local subgraph types Gamma_k^(n) that capture all relevant loop-induced correlations.
    Section II.A replaces the instance-specific subgraphs of Ref. [39] with degree-conditioned types, which assumes statistical homogeneity within degree classes and that measured local statistics suffice to close the equations.
  • ad hoc to paper Message contributions from shell-1 neighbors of a node are mutually independent in loopy ensembles.
    Used to factorize A(n) in Eq. (5); asserted without proof and questionable when shell-1 neighbors share outer neighbors or are directly connected.
  • ad hoc to paper The distribution of residual size m can be replaced by its average m^(2)_dlyw without changing the percolation threshold.
    Eq. (8) replaces P(m|d,l,y,w) by its mean for tractability, with no error estimate for this reduction.
  • ad hoc to paper Validity of a finite-order approximation is captured by generalized-edge closure, meaning that internal node pairs re-induce the same generalized edge.
    GECC measures this property, but the mapping from closure to threshold error is not proven; it is supported only by the empirical correlation in Figs. 2 and 3.
invented entities (2)
  • nth-order generalized edge E_kd^(n)
    purpose: Message-passing unit defined as the intersection of two nodes' nth-order subgraphs; intended to absorb loop-induced correlations into one effective structure.
    A mathematical construction internal to the framework; it has no direct falsifiable prediction outside the message-passing equations in which it is used.
  • GECC(n) generalized-edge closure coefficient independent evidence
    purpose: Quantifies the fraction of internal node pairs that fail to re-induce the same generalized edge; used as an a priori diagnostic of approximation reliability.
    Computed from local statistics alone, and its predictive relation to threshold error is tested against Monte Carlo simulations on synthetic and real networks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Ensemble-level loopy message passing with generalized-edge closure for percolation." pith.science (2026). https://pith.science/paper/VXSAAORC

@misc{pith2026260809397,
  author       = {Pith},
  title        = {Pith review of: Ensemble-level loopy message passing with generalized-edge closure for percolation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VXSAAORC}},
  note         = {Machine review of arXiv:2608.09397}
}
read the original abstract

Predicting the percolation threshold of highly clustered networks from local statistics remains difficult, because short loops break the independence assumption underlying tree-like message passing. Existing remedies address loopy connectivity either through prescribed local motifs in random-graph ensembles or through a single network's realized topology, leaving an ensemble-level treatment of arbitrary connectivity patterns absent. Here, we develop a loopy message-passing framework for random clustered graph ensembles based on generalized-edge statistics, which characterize overlap patterns among the neighborhoods of different nodes. This yields a progressively refined approximation scheme based on neighborhoods of increasing size around each node. The low-order approximations recover previous equations for random network ensembles, and the new result that yields refined threshold prediction is developed by the second-order approximation. We show that the effectiveness of this framework depends not only on short-cycle density but also on the internal consistency of generalized edges. To diagnose this effectiveness, we introduce the generalized-edge closure coefficient (GECC) to quantify this consistency. Because GECC is computed entirely from local statistics and does not rely on any percolation calculation, it serves as an a priori diagnostic for the reliability of the approximation. Using synthetic and real networks, the threshold is evaluated via the second-order and lower-order approximations. Comparisons with Monte Carlo simulations show that GECC captures key structural features that strongly affect the percolation threshold. These results establish ensemble-based loopy message passing as an efficient route for predicting the percolation threshold in large clustered networks.

Figures

Figures reproduced from arXiv: 2608.09397 by the authors.

Figure 1
Figure 1. FIG. 1. Schematic illustration of second-order generalized edges, their closure property, and associated structural features. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 48 canonical work pages

  1. [1]

    For subgraphs ofnth-order, consider two sub- graphs Γ (n) k and Γ (n) d associated with two nodes of de- greeskandd, respectively

    Generalized-edge messages The progressively refined formulation replaces single- edge transmission with message passing on generalized edges. For subgraphs ofnth-order, consider two sub- graphs Γ (n) k and Γ (n) d associated with two nodes of de- greeskandd, respectively. The generalized edge formed by these two nodes is defined as E(n) kd = Γ(n) k ∩Γ (n)...

    work page

  2. [2]

    (4), that holds for arbitrary random networks at arbitrary approximation ordern

    Loopy message-passing formulation on random networks With the generalized edges and the associated proba- bility quantities defined above, we now derive the central technical result of this framework: an exact, ensemble- level message-passing recursion, Eq. (4), that holds for arbitrary random networks at arbitrary approximation ordern. All correlations i...

    work page

  3. [3]

    Extraction of typical structural features For annth-order generalized edge associated with sub- graphs Γ(n) d and Γ(n) l , we extract a minimal set of struc- tural features needed by the reduced message-passing equation: the degree classes involved in the transition; the effective connectivity, extracted from the general- ized edgeE (n) dl ; and the size ...

    work page

  4. [4]

    , sm Γ(n) d ,Γ (n) l1 ,

    Second-order message passing equation For a second-order generalized edge associated with subgraphs Γ (2) d and Γ (2) l , the corresponding message- 4 ˜π s|Γ (n) k ,Γ (n) d g q,Γ (n) k ∩Γ (n) d +δ s,0 h 1−g q,Γ (n) k ∩Γ (n) d i =δ s,0 h 1−g q,Γ (n) k ∩Γ (n) d i +q g q,Γ (n) k ∩Γ (n) d X {Γ(n) li }m i=1 X s1,...,sm A(n) s1, . . . , sm Γ(n) d ,Γ (n) l1 , . ...

    work page

  5. [5]

    At both first and zeroth orders, only shell-1 neighbors are retained, and hence the second-order connectivity variablesyandw do not enter the reduced description

    Recovery of lower-order approximations We next show how lower-order approximations are re- covered from lower-order subgraphs. At both first and zeroth orders, only shell-1 neighbors are retained, and hence the second-order connectivity variablesyandw do not enter the reduced description. As a result, the corresponding connection factor is set tog(q, y, w...

    work page

  6. [6]

    LetGdenote the whole graph

    Statistics of Generalized Edges and the Induced Generalized Degree Distribution We therefore first characterize the generalized degree distribution, defined as the distribution of the number of generalized edges rooted at each node. LetGdenote the whole graph. For a nodeu∈G, let Γ (n) u denote thenth-order subgraph of nodeu. For each nodev̸=u contained in...

    work page

  7. [7]

    (9) is accurate only when a generalized edge behaves as a single effective structure

    Generalized-edge closure coefficient The threshold predicted by the recursion in Eq. (9) is accurate only when a generalized edge behaves as a single effective structure. Concretely, consider a gener- alized edge induced by an initial node pair. Because a message may enter this structure through any node pair internal to it, each such internal pair can eq...

    work page

  8. [8]

    Threshold prediction diagnostics We evaluate the predictive accuracy of approximations of different orders for site percolation. For each net- work, we compare the simulation-based thresholdq sim c with three theoretical estimates: the locally tree-like pre- dictionq tree c , the first-order threshold predictionq (1) c , and the second-order threshold pre...

    work page

Show all 52 references
  1. [9]

    The networks are gen- erated by a two-step unit replacement procedure on a random backbone

    Validation using synthetic networks We first test whether GECC (n) tracks the reliability of the approximations on synthetic networks with con- trolled generalized-edge closure. The networks are gen- erated by a two-step unit replacement procedure on a random backbone. Each ba...

  2. [10]

    For visual clarity, Fig

    Real-world network analysis Having established the closure-departure mechanism on controlled synthetic benchmarks, we next apply the same threshold-diagnostic protocol to real-world net- works from the Stanford Large Network Dataset Collec- tion [42], to test whether the same ...

  3. [11]

    M. E. J. Newman, The structure and function of complex networks, SIAM Rev45, 167 (2003)

    2003

  4. [12]

    D. S. Callaway, M. E. J. Newman, S. H. Strogatz, and D. J. Watts, Network robustness and fragility: Percola- tion on random graphs, Phys Rev Lett85, 5468 (2000)

    2000

  5. [13]

    Cohen, K

    R. Cohen, K. Erez, D. Ben-Avraham, and S. Havlin, Re- silience of the internet to random breakdowns, Phys Rev Lett85, 4626 (2000)

    2000

  6. [14]

    Cohen, K

    R. Cohen, K. Erez, D. Ben-Avraham, and S. Havlin, Breakdown of the internet under intentional attack, Phys Rev Lett86, 3682 (2001)

    2001

  7. [15]

    Albert, H

    R. Albert, H. Jeong, and A.-L. Barab´ asi, Error and at- tack tolerance of complex networks, Nature406, 378 (2000)

    2000

  8. [16]

    Moore and M

    C. Moore and M. E. J. Newman, Epidemics and per- colation in small-world networks, Phys Rev E61, 5678 (2000)

    2000

  9. [17]

    Artime and M

    O. Artime and M. De Domenico, Percolation on feature- enriched interconnected systems, Nat Commun12, 2478 (2021)

    2021

  10. [18]

    L. H. Wang and Y. M. Du, Percolating critical window for correlated scale-free networks, Physica A664, 130441 (2025)

    2025

  11. [19]

    Karrer and M

    B. Karrer and M. E. J. Newman, Message passing ap- proach for general epidemic models, Phys Rev E82, 016101 (2010)

    2010

  12. [20]

    Karrer, M

    B. Karrer, M. E. J. Newman, and L. Zdeborov´ a, Perco- lation on sparse networks, Phys Rev Lett113, 208702 (2014)

    2014

  13. [21]

    Morone and H

    F. Morone and H. A. Makse, Influence maximization in complex networks through optimal percolation, Nature 524, 65 (2015)

    2015

  14. [22]

    Qian, D.-D

    C. Qian, D.-D. Zhao, M. Zhong, B. Zhang, H. Peng, and W. Wang, Message-passing approach for percolation on the networked system: A mini-review, Europhys Lett 147, 11002 (2024)

    2024

  15. [23]

    Pearl,Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference(Morgan Kaufmann, San Francisco, CA, 1988)

    J. Pearl,Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference(Morgan Kaufmann, San Francisco, CA, 1988)

    1988

  16. [24]

    J. S. Yedidia, W. T. Freeman, and Y. Weiss, Constructing free-energy approximations and generalized belief prop- agation algorithms, IEEE Trans Inf Theory51, 2282 (2005)

    2005

  17. [25]

    M´ ezard and A

    M. M´ ezard and A. Montanari,Information, Physics, and Computation(Oxford University Press, Oxford, 2009)

    2009

  18. [26]

    M. E. J. Newman, S. H. Strogatz, and D. J. Watts, Ran- dom graphs with arbitrary degree distributions and their applications, Phys Rev E64, 026118 (2001)

    2001

  19. [27]

    S. N. Dorogovtsev, A. V. Goltsev, and J. F. F. Mendes, Critical phenomena in complex networks, Rev Mod Phys 80, 1275 (2008)

    2008

  20. [28]

    Pastor-Satorras and A

    R. Pastor-Satorras and A. Vespignani, Epidemic spread- ing in scale-free networks, Phys Rev Lett86, 3200 (2001)

    2001

  21. [29]

    Kenah and J

    E. Kenah and J. M. Robins, Second look at the spread of epidemics on networks, Phys Rev E76, 036113 (2007)

    2007

  22. [30]

    D. J. Watts and S. H. Strogatz, Collective dynamics of small-world networks, Nature393, 440 (1998)

    1998

  23. [31]

    M. E. J. Newman, Random graphs with clustering, Phys Rev Lett103, 058701 (2009)

    2009

  24. [32]

    J. C. Miller, Percolation and epidemics in random clus- tered networks, Phys Rev E80, 020901 (2009)

    2009

  25. [33]

    Barrat, M

    A. Barrat, M. Barth´ elemy, and A. Vespignani,Dynamical Processes on Complex Networks(Cambridge University Press, Cambridge, 2008)

    2008

  26. [34]

    M. E. J. Newman,Networks: An Introduction(Oxford University Press, Oxford, 2010)

    2010

  27. [35]

    Cohen and S

    R. Cohen and S. Havlin,Complex Networks: Structure, Robustness and Function(Cambridge University Press, Cambridge, 2010)

    2010

  28. [36]

    Radicchi and C

    F. Radicchi and C. Castellano, Breaking of the site-bond percolation universality in networks, Nat Commun6, 10196 (2015)

    2015

  29. [37]

    Radicchi and G

    F. Radicchi and G. Bianconi, Redundant interdependen- cies boost the robustness of multiplex networks, Phys Rev X7, 011013 (2017). 11

    2017

  30. [38]

    M. ´A. Serrano and M. Bogu˜ n´ a, Clustering in complex networks. i. general formalism, Phys Rev E74, 056114 (2006)

    2006

  31. [39]

    M. ´A. Serrano and M. Bogu˜ n´ a, Clustering in complex networks. ii. percolation properties, Phys Rev E74, 056115 (2006)

    2006

  32. [40]

    M. ´A. Serrano and M. Bogu˜ n´ a, Percolation and epi- demic thresholds in clustered networks, Phys Rev Lett 97, 088701 (2006)

    2006

  33. [41]

    J. C. Miller, Spread of infectious disease through clus- tered populations, J R Soc Interface6, 1121 (2009)

    2009

  34. [42]

    M. E. J. Newman, Properties of highly clustered net- works, Phys Rev E68, 026121 (2003)

    2003

  35. [43]

    Britton, M

    T. Britton, M. Deijfen, A. N. Lager ˚ as, and M. Lindholm, Epidemics on random graphs with tunable clustering, J Appl Probab45, 743 (2008)

    2008

  36. [44]

    J. P. Gleeson, Bond percolation on a class of clustered random networks, Phys Rev E80, 036107 (2009)

    2009

  37. [45]

    Bollob´ as, S

    B. Bollob´ as, S. Janson, and O. Riordan, Sparse random graphs with clustering, Random Struct Algorithms38, 269 (2011)

    2011

  38. [46]

    Karrer and M

    B. Karrer and M. E. J. Newman, Random graphs con- taining arbitrary distributions of subgraphs, Phys Rev E 82, 066118 (2010)

    2010

  39. [47]

    Allard, L

    A. Allard, L. H´ ebert-Dufresne, P.-A. No¨ el, V. Marceau, and L. J. Dub´ e, Bond percolation on a class of correlated and clustered random graphs, J Phys A Math Theor45, 405005 (2012)

    2012

  40. [48]

    J. M. Mooij and H. J. Kappen, Loop corrections for ap- proximate inference on factor graphs, J Mach Learn Res 8, 1113 (2007)

    2007

  41. [49]

    G. T. Cantwell and M. E. J. Newman, Message passing on networks with loops, Proc Natl Acad Sci USA116, 23398 (2019)

    2019

  42. [50]

    Mann and S

    P. Mann and S. Dobson, Belief propagation on net- works with cliques and chordless cycles, Phys Rev E107, 054303 (2023)

    2023

  43. [51]

    R. Milo, S. Shen-Orr, S. Itzkovitz, N. Kashtan, D. Chklovskii, and U. Alon, Network motifs: Simple building blocks of complex networks, Science298, 824 (2002)

    2002

  44. [52]

    Leskovec and A

    J. Leskovec and A. Krevl, SNAP Datasets: Stanford large network dataset collection,http://snap.stanford.edu/ data(2014)

    2014

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.