pith. sign in

arxiv: 2509.11706 · v2 · submitted 2025-09-15 · 💻 cs.SI · cond-mat.stat-mech· physics.soc-ph

Threshold and quasi-stationary distribution for the SIS model on networks

George Cantwell , Cristopher Moore This is my paper

Pith reviewed 2026-05-18 16:59 UTC · model grok-4.3

classification 💻 cs.SI cond-mat.stat-mechphysics.soc-ph
keywords SIS epidemic modelpair approximationquasi-stationary distributionepidemic thresholdnetwork epidemiologystate space expansiondynamic memory
0
0 comments X

The pith

Expanding the state space with a memory of when nodes last became susceptible improves the pair approximation for the SIS epidemic model on networks.

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

The paper develops an improved approximation for the Susceptible-Infectious-Susceptible model on arbitrary networks. It begins with the standard pair approximation, which treats neighboring pairs exactly but approximates the rest of the network, and then dynamically expands the state space so each node carries a memory of when it last became susceptible. This change captures temporal correlations that the basic pair method misses. If the improvement works, it gives a parameter-free way to locate the epidemic threshold and compute the long-term fraction of infected nodes, both on finite real networks and on infinite random graphs.

Core claim

The authors introduce a dynamic state-space expansion that equips nodes with a memory of their last susceptibility time. This refined pair approximation accurately identifies the epidemic threshold and the quasi-stationary fraction of infected individuals for the SIS process on both finite graphs and infinite random graphs.

What carries the argument

Dynamic expansion of the state space that gives each node a memory of when it last became susceptible, thereby refining the pair approximation by accounting for additional temporal correlations.

If this is right

  • The method locates the epidemic threshold more accurately than the ordinary pair approximation on a given network.
  • It supplies reliable estimates of the quasi-stationary infection level once the threshold is crossed.
  • The same procedure works for both small finite graphs and large infinite random graphs.
  • Implementation stays simple and requires no network-specific parameters to be fitted.

Where Pith is reading between the lines

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

  • The same memory device could be added to other compartmental models such as SIR or SIRS to improve their network approximations.
  • On empirical contact networks the refined threshold might better guide the minimal set of interventions needed to prevent sustained spread.
  • Testing the approximation on networks with heterogeneous recovery rates or with slowly changing links would reveal how far the memory trick generalizes.

Load-bearing premise

Adding a memory of last susceptibility time captures the higher-order correlations that the standard pair approximation misses, without needing extra fitting parameters tuned to each network.

What would settle it

Exact solution of the full Markov-chain master equation on a small network whose true threshold and quasi-stationary distribution can be computed directly, then comparison against the approximation's predictions for the same network.

Figures

Figures reproduced from arXiv: 2509.11706 by Cristopher Moore, George Cantwell.

Figure 1
Figure 1. Figure 1: FIG. 1: The one-node Markov process [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Approximations for the SIS model on 3-regular ran [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Fraction of infected nodes at different rates of infec [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Estimates of the survival function for a 4-regular [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Simulations on example networks. Panels (a) and (b) are the results of simulations on a sexual contact network of [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

We study the Susceptible-Infectious-Susceptible (SIS) model on arbitrary networks. The well-established pair approximation treats neighboring pairs of nodes exactly while making a mean field approximation for the rest of the network. We improve the method by expanding the state space dynamically, giving nodes a memory of when they last became susceptible. The resulting approximation is simple to implement and appears to be highly accurate, both in locating the epidemic threshold and in computing the quasi-stationary fraction of infected individuals above the threshold, for both finite graphs and infinite random graphs.

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 paper studies the SIS epidemic model on arbitrary networks and proposes an improvement to the standard pair approximation by dynamically expanding the state space to include a memory variable tracking the time since each susceptible node last became susceptible. The authors claim that the resulting closed equations yield a simple-to-implement approximation that accurately locates the epidemic threshold and computes the quasi-stationary infected fraction, both for finite graphs and for infinite random graphs.

Significance. If the central claim holds without network-specific tuning or post-hoc adjustments, the method would provide a parameter-free extension of pair approximations that better captures higher-order correlations via finite memory, offering a practical advance for analyzing epidemic thresholds and endemic states on general networks.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (derivation of enlarged-state equations): the claim that augmenting each susceptible node with its susceptibility age allows 'exact pair-level equations' for the enlarged space to close without further ad-hoc closures is not fully supported; the probability that a neighbor of a node with susceptibility age a is infected still appears to require a mean-field or pair-level assumption on the remaining graph, which may miss persistent triple correlations.
  2. [§4] §4 (numerical validation): the reported high accuracy on finite graphs and infinite random graphs lacks explicit error bars, direct comparison tables against standard pair approximation and simulation, and tests on a broader range of topologies; without these, it is unclear whether accuracy is general or specific to the tested networks.
minor comments (2)
  1. [§2] Notation for the memory variable and its update rules should be introduced with a clear table or diagram early in the methods section to aid readability.
  2. [§3] The manuscript would benefit from an explicit statement of the closure relation used for the neighbor-infection probability in the enlarged state space.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for the careful and constructive report. We respond point by point to the major comments below, indicating the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (derivation of enlarged-state equations): the claim that augmenting each susceptible node with its susceptibility age allows 'exact pair-level equations' for the enlarged space to close without further ad-hoc closures is not fully supported; the probability that a neighbor of a node with susceptibility age a is infected still appears to require a mean-field or pair-level assumption on the remaining graph, which may miss persistent triple correlations.

    Authors: We thank the referee for this observation. The central technical point of the derivation in §3 is that conditioning on susceptibility age renders the enlarged-state process Markovian, so that the time derivative of each pair probability [S_a X] (where X is S_b or I) can be written exactly using only single-node probabilities with age and other pair probabilities. The infection probability for a neighbor of an S_a node is then obtained directly from the ratio of the appropriate pair probability to the marginal [S_a], without introducing any new closure relation beyond the standard pair-level neglect of correlations outside the focal pair. This is the same structural assumption used in classical pair approximations; the age variable simply refines the state space so that temporal memory is retained at the pair level. We agree that short cycles can still induce persistent triple correlations that the method does not capture, and we will revise the abstract and §3 to state the closure assumption more explicitly and to note this limitation. revision: partial

  2. Referee: [§4] §4 (numerical validation): the reported high accuracy on finite graphs and infinite random graphs lacks explicit error bars, direct comparison tables against standard pair approximation and simulation, and tests on a broader range of topologies; without these, it is unclear whether accuracy is general or specific to the tested networks.

    Authors: We accept this criticism. In the revised manuscript we will add (i) error bars obtained from at least 100 independent Monte Carlo realizations for both the threshold location and the quasi-stationary infected fraction, (ii) explicit tables that report the absolute and relative errors of the standard pair approximation, our dynamic-expansion method, and direct simulation for each network and parameter value, and (iii) additional numerical experiments on Barabási–Albert scale-free networks and Watts–Strogatz small-world networks with varying rewiring probabilities. These additions will make the scope and robustness of the accuracy claims clearer. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation chain is self-contained

full rationale

The paper augments the standard pair approximation for the SIS process by dynamically expanding each node's state with a discrete memory variable tracking time since last becoming susceptible. The governing equations for the joint probabilities over these enlarged states are written directly from the underlying continuous-time Markov chain transition rules, then closed with an explicit mean-field assumption on the remaining network that does not reference the target threshold or quasi-stationary infected fraction. No parameters are fitted to simulation data, no self-citations supply load-bearing uniqueness theorems, and the memory variable is not defined in terms of the quantities it is later used to predict. Consequently the reported accuracy on finite and infinite graphs follows from the structure of the closed system rather than from any definitional or fitting equivalence to the inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approximation rests on the assumption that adding a single memory variable per node closes the moment hierarchy sufficiently for both threshold location and quasi-stationary density; no new particles or forces are postulated.

axioms (2)
  • domain assumption The network is static and the infection and recovery processes are Markovian with constant rates.
    Standard SIS setup invoked throughout the abstract.
  • ad hoc to paper Higher-order correlations beyond pairs can be captured by a finite memory of last susceptibility time.
    This is the key modeling choice that enables the dynamic state expansion.

pith-pipeline@v0.9.0 · 5615 in / 1291 out tokens · 23425 ms · 2026-05-18T16:59:53.467910+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Our SISK pair approximation also allows us to compute this and if it approximatesP ij more accurately, then it should approximate spatial correlations more accurately

    Dynamic correlation The standard pair approximation allows us to approx- imate spatial correlations, i.e.,⟨I iIj⟩=P ij(Ii, Ij). Our SISK pair approximation also allows us to compute this and if it approximatesP ij more accurately, then it should approximate spatial correlations more accurately. How- ever, SISK additionally allows us to improve on our ap- ...

    work page

  2. [2]

    surface- to-volume ratio

    Random regular graphs For the SIS2 model on a random (q+ 1)-regular graph, we again have that every node and every edge is equiv- alent, and soϕ x ij =ϕ x for allijandB ij(ϕx) =qϕ x. Inserting these into to the 9×9 transition matrix, after considerable algebra, we arrive at the the self-consistent equation ϕ1 ϕ2 =   βqϕ 1(β+ϕ2)(βq+q(ϕ 1+ϕ2)+1) qϕ2 1(βq+...

    work page

  3. [3]

    While these equations describe the large-Klimit of our approach, we believe that that they are still not exact, even on random regular graphs

    The factor of (q+1)/2 is due to the fact that a (q+1)- regular graph withnvertices hasn(q+ 1)/2 edges. While these equations describe the large-Klimit of our approach, we believe that that they are still not exact, even on random regular graphs. In particular, we have only incorporated one particular kind of memory—the time since a node was last infectiou...

    work page

  4. [4]

    I. Z. Kiss, J. C. Miller, and P. L. Simon,Math- ematics of Epidemics on Networks: From Exact to Approximate Models, volume 46 ofInterdisciplinary Applied Mathematics. Springer International Publish- ing, Cham (2017), URLhttp://link.springer.com/ 10.1007/978-3-319-50806-1

    work page doi:10.1007/978-3-319-50806-1 2017

  5. [5]

    Pastor-Satorras, C

    R. Pastor-Satorras, C. Castellano, P. Van Mieghem, and A. Vespignani, Epidemic processes in complex networks.Reviews of Modern Physics87(3), 925– 979 (2015), URLhttps://link.aps.org/doi/10.1103/ RevModPhys.87.925

    work page 2015

  6. [6]

    M. J. Keeling and K. T. Eames, Networks and epidemic models.Journal of The Royal Society Interface2(4), 295–307 (2005), URL https://royalsocietypublishing.org/doi/10.1098/ rsif.2005.0051

    work page arXiv 2005

  7. [7]

    P. Grassberger, On the critical behavior of the general epidemic process and dynamical percolation.Math- ematical Biosciences63(2), 157–172 (1983), URL https://linkinghub.elsevier.com/retrieve/pii/ 0025556482900360

    work page 1983

  8. [9]

    Moore and M

    C. Moore and M. E. J. Newman, Epidemics and per- colation in small-world networks.Physical Review E 61(5), 5678–5682 (2000), URLhttps://link.aps.org/ doi/10.1103/PhysRevE.61.5678

    work page doi:10.1103/physreve.61.5678 2000

  9. [10]

    Diekmann, M

    O. Diekmann, M. C. M. De Jong, and J. A. J. Metz, A deterministic epidemic model taking account of re- peated contacts between the same individuals.Jour- nal of Applied Probability35(2), 448–462 (1998), URL http://www.jstor.org/stable/3215698. 10

    work page arXiv 1998

  10. [11]

    suicide coach

    R. Pastor-Satorras and A. Vespignani, Epidemic spread- ing in scale-free networks.Physical Review Letters86, 3200–3203 (2001), URLhttps://link.aps.org/doi/ 10.1103/PhysRevLett.86.3200

    work page doi:10.1103/physrevlett.86.3200 2001

  11. [12]

    T. E. Harris, Contact interactions on a lattice.The An- nals of Probability2(6), 969 – 988 (1974), URLhttps: //doi.org/10.1214/aop/1176996493

    work page doi:10.1214/aop/1176996493 1974

  12. [13]

    G. H. Weiss and M. Dishon, On the asymptotic be- havior of the stochastic and deterministic models of an epidemic.Mathematical Biosciences11(3-4), 261– 265 (1971), URLhttps://linkinghub.elsevier.com/ retrieve/pii/0025556471900873

    work page arXiv 1971

  13. [14]

    Y. Wang, D. Chakrabarti, C. Wang, and C. Faloutsos, Epidemic spreading in real networks: an eigenvalue view- point. In22nd International Symposium on Reliable Dis- tributed Systems, 2003. Proceedings., pp. 25–34 (2003), URLhttps://doi.org/10.1109/RELDIS.2003.1238052

    work page doi:10.1109/reldis.2003.1238052 2003

  14. [15]

    Chakrabarti, Y

    D. Chakrabarti, Y. Wang, C. Wang, J. Leskovec, and C. Faloutsos, Epidemic thresholds in real networks.ACM Transactions on Information and System Security10(4), 1–26 (2008), URLhttps://dl.acm.org/doi/10.1145/ 1284680.1284681

    work page arXiv 2008

  15. [16]

    Van Mieghem, J

    P. Van Mieghem, J. Omic, and R. Kooij, Virus spread in networks.IEEE/ACM Transactions on Networking 17(1), 1–14 (2009), URLhttps://dl.acm.org/doi/10. 1109/TNET.2008.925623

    work page arXiv 2009

  16. [17]

    Van Mieghem, The N-intertwined SIS epidemic net- work model.Computing93(2), 147–169 (2011), URL https://doi.org/10.1007/s00607-011-0155-y

    P. Van Mieghem, The N-intertwined SIS epidemic net- work model.Computing93(2), 147–169 (2011), URL https://doi.org/10.1007/s00607-011-0155-y

    work page doi:10.1007/s00607-011-0155-y 2011

  17. [18]

    M. J. Keeling, The effects of local spatial structure on epi- demiological invasions.Proceedings of the Royal Society of London. Series B: Biological Sciences266(1421), 859– 867 (1999), URLhttps://royalsocietypublishing. org/doi/abs/10.1098/rspb.1999.0716

    work page doi:10.1098/rspb.1999.0716 1999

  18. [19]

    A. S. Mata and S. C. Ferreira, Pair quenched mean- field theory for the susceptible-infected-susceptible model on complex networks.Europhysics Letters103(4), 48003 (2013), URLhttps://iopscience.iop.org/article/ 10.1209/0295-5075/103/48003

    work page doi:10.1209/0295-5075/103/48003 2013

  19. [20]

    Shrestha, S

    M. Shrestha, S. V. Scarpino, and C. Moore, Message- passing approach for recurrent-state epidemic models on networks.Physical Review E92(2), 022821 (2015), URL https://doi.org/10.1103/PhysRevE.92.022821

    work page doi:10.1103/physreve.92.022821 2015

  20. [21]

    Cator and P

    E. Cator and P. Van Mieghem, Second-order mean- field susceptible-infected-susceptible epidemic threshold. Physical Review E85(5), 056111 (2012), URLhttps: //link.aps.org/doi/10.1103/PhysRevE.85.056111

    work page doi:10.1103/physreve.85.056111 2012

  21. [22]

    J. P. Gleeson, High-accuracy approximation of binary- state dynamics on networks.Physical Review Letters 107, 068701 (2011), URLhttps://link.aps.org/doi/ 10.1103/PhysRevLett.107.068701

    work page doi:10.1103/physrevlett.107.068701 2011

  22. [23]

    Karrer and M

    B. Karrer and M. E. J. Newman, Message passing ap- proach for general epidemic models.Physical Review E 82(1), 016101 (2010), URLhttps://doi.org/10.1103/ PhysRevE.82.016101

    work page 2010

  23. [24]

    Karrer, M

    B. Karrer, M. E. J. Newman, and L. Zdeborov´ a, Per- colation on sparse networks.Phys. Rev. Lett.113, 208702 (2014), URLhttps://link.aps.org/doi/10. 1103/PhysRevLett.113.208702

    work page 2014

  24. [25]

    R. S. Ferreira and S. C. Ferreira, Critical behavior of the contact process on small-world networks.The European Physical Journal B86(11), 462 (2013), URLhttp:// link.springer.com/10.1140/epjb/e2013-40534-0

    work page doi:10.1140/epjb/e2013-40534-0 2013

  25. [26]

    J. P. Gleeson, Binary-state dynamics on complex net- works: Pair approximation and beyond.Physical Review X3(2), 021004 (2013), URLhttps://link.aps.org/ doi/10.1103/PhysRevX.3.021004

    work page doi:10.1103/physrevx.3.021004 2013

  26. [27]

    Zhang, Inference of kinetic Ising model on sparse graphs.Journal of Statistical Physics148(3), 502– 512 (2012), URLhttp://link.springer.com/10.1007/ s10955-012-0547-1

    P. Zhang, Inference of kinetic Ising model on sparse graphs.Journal of Statistical Physics148(3), 502– 512 (2012), URLhttp://link.springer.com/10.1007/ s10955-012-0547-1

    work page 2012

  27. [28]

    A. Y. Lokhov, M. M´ ezard, and L. Zdeborov´ a, Dynamic message-passing equations for models with unidirec- tional dynamics.Physical Review E91, 012811 (2015), URLhttps://link.aps.org/doi/10.1103/PhysRevE. 91.012811

    work page doi:10.1103/physreve 2015

  28. [29]

    Calabrese and J

    E. Dom´ ınguez, D. Machado, and R. Mulet, The cavity master equation: average and fixed point of the ferro- magnetic model in random graphs.Journal of Statisti- cal Mechanics: Theory and Experiment2020(7), 073304 (2020), URLhttps://dx.doi.org/10.1088/1742-5468/ ab9eb6

    work page doi:10.1088/1742-5468/ 2020

  29. [30]

    Aurell, D

    E. Aurell, D. Machado Perez, and R. Mulet, A closure for the master equation starting from the dynamic cavity method.Journal of Physics A: Mathematical and Theo- retical56(17), 17LT02 (2023), URLhttps://dx.doi. org/10.1088/1751-8121/acc8a4

    work page doi:10.1088/1751-8121/acc8a4 2023

  30. [31]

    Behrens, B

    F. Behrens, B. Hudcov´ a, and L. Zdeborov´ a, Back- tracking dynamical cavity method.Phys. Rev. X13, 031021 (2023), URLhttps://link.aps.org/doi/10. 1103/PhysRevX.13.031021

    work page 2023

  31. [32]

    Ortega, D

    E. Ortega, D. Machado, and A. Lage-Castellanos, Dy- namics of epidemics from cavity master equations: Susceptible-infectious-susceptible models.Physical Re- view E105, 024308 (2022), URLhttps://link.aps. org/doi/10.1103/PhysRevE.105.024308

    work page doi:10.1103/physreve.105.024308 2022

  32. [33]

    J. R. Norris,Markov Chains. Cambridge Series in Statis- tical and Probabilistic Mathematics, Cambridge Univer- sity Press (1997)

    work page 1997

  33. [34]

    Morris and R

    M. Morris and R. Rothenberg, HIV Transmission Network Metastudy Project: An Archive of Data From Eight Network Studies, 1988–2001: Version 1 (2011), URLhttps://www.icpsr.umich.edu/web/ NAHDAP/studies/22140/versions/V1

    work page 1988

  34. [35]

    ˇSubelj and M

    L. ˇSubelj and M. Bajec, Robust network community de- tection using balanced propagation.The European Physi- cal Journal B81(3), 353–362 (2011), URLhttp://link. springer.com/10.1140/epjb/e2011-10979-2

    work page doi:10.1140/epjb/e2011-10979-2 2011

  35. [36]

    Krzakala, C

    F. Krzakala, C. Moore, E. Mossel, J. Neeman, A. Sly, L. Zdeborov´ a, and P. Zhang, Spectral redemption in clustering sparse networks.Proceedings of the Na- tional Academy of Sciences110(52), 20935–20940 (2013), URLhttps://www.pnas.org/doi/abs/10.1073/ pnas.1312486110. Appendix A: Linear stability for pair-approximations First we derive a result, reminisce...

    work page 2013