{"id":"0e4c5649-5658-4505-a6dd-4a774a494ec2","arxiv_id":"1908.09105","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Edge overlap between layers controls whether the AND-rule SIS model on duplex networks has a continuous or a discontinuous, hysteretic epidemic transition.","lead":"This paper studies an SIS epidemic on two-layer networks where infection requires an infected neighbor in each layer, and finds that low overlap between the layers changes the outbreak from a smooth to an abrupt, bistable transition. The result matters for predicting outbreaks in systems where spreading requires reinforcement across different interaction types.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Mean-field Eq. (18) contradicts the unqualified 'O<1 is discontinuous' claim: for O ≥ ⟨k⟩/(⟨k⟩+1) the transition is continuous.","rationale":"The reader's weakest assumption identifies exactly the same issue: the paper's own Eq. (18) only produces a saddle-node bifurcation when O < ⟨k⟩/(⟨k⟩+1), yet the abstract and conclusions claim discontinuous behavior for all O<1. My independent check of the stationary equation confirms this: for sufficiently large overlap relative to the mean degree, F(y) is monotone and the nonzero branch bifurcates continuously. This is a genuine internal inconsistency in the central claim, not merely a disagreement with external consensus. It does not, however, invalidate the core low-overlap result: for the parameters simulated (⟨k⟩=20, O≤0.8), O<20/21, so the reported discontinuous transition and hysteresis are consistent with the mean-field theory. The paper deserves credit for the derivation of the individual-based mean-field equations and for the demonstration that low overlap produces bistability. The required repair is to replace the unqualified 'O<1' statement with the condition O < ⟨k⟩/(⟨k⟩+1) (or its network-specific analogue), and to test a case such as ⟨k⟩=2, O=0.8 to confirm the continuous regime. Because the reader's CONDITIONAL verdict already reflects the need for this correction, my stress-test does not move the verdict.","tokens_in":17616,"tokens_out":9673,"duration_ms":100519,"concrete_test":"Evaluate the saddle-node condition for Eq. (18) analytically: compute F'(y)=1−(⟨k⟩+1)(1−O)y^{⟨k⟩} on y∈[0,1]. If F' has no zero in (0,1), no bistable window exists and the transition is continuous. Apply this to ⟨k⟩=2 and O=0.8: F'(y)=1−0.6y^2>0 everywhere, so the theory itself predicts a continuous transition at βC=1/(2·0.8)=0.625. A supporting simulation on two Poisson layers with N=10^5, ⟨k⟩=2, and O=0.8, initialized from ρ0=0.02 and 0.98, should show no hysteresis if Eq. (18) is correct; observing hysteresis would indicate a mechanism beyond the homogeneous mean-field and would require new theoretical justification.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and conclusions state that every O<1 gives a discontinuous transition with a bistable window. This is not supported by the paper's own homogeneous mean-field equation. With y=1−ρ, stationary nonzero solutions of Eq. (18) satisfy 1 = β⟨k⟩ y [1 − (1−O) y^{⟨k⟩}]. Define F(y)=y[1−(1−O)y^{⟨k⟩}]. A saddle-node bifurcation, hence bistability, requires F to have an interior maximum on (0,1), which occurs only if F'(y*)=0 for some y*<1, i.e. only if O < ⟨k⟩/(⟨k⟩+1). For O ≥ ⟨k⟩/(⟨k⟩+1), F is monotone increasing on [0,1], so the nonzero endemic branch emerges continuously at βC=1/(⟨k⟩O) in a transcritical bifurcation. For example, with ⟨k⟩=2 and O=0.8, since 0.8 > 2/3, Eq. (18) predicts a continuous transition, not the claimed discontinuous one. The simulations in Figs. 2 and 3 use ⟨k⟩=20 and O≤0.8, for which O<20/21, so they do not probe the regime where the universal statement fails. The correct condition for the reported discontinuous transition and hysteresis is O < ⟨k⟩/(⟨k⟩+1), not O<1; the paper should state this caveat explicitly.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a susceptible-infected-susceptible epidemic model on duplex networks in which a susceptible node can be infected only when it has at least one infectious neighbor in each layer. The authors define an overlap parameter O for the fraction of edges common to both layers and claim that the nature of the epidemic phase transition is controlled by O: for O=1 the transition is continuous, while for every O<1 it is discontinuous and accompanied by a bistable region, which widens as O decreases. They derive an individual-based mean-field equation (Eq. 13), its homogeneous-network reduction (Eq. 18), and the epidemic threshold βC = 1/Λmax(Ã) (Eq. 16), and they support the claims with simulations on Erdős-Rényi, Barabási-Albert, Watts-Strogatz, lattice, three-layer, and real multiplex networks.","tokens_in":17916,"tokens_out":5321,"duration_ms":48979,"significance":"The model itself is a simple and natural extension of SIS dynamics to multiplex networks, and the observation that edge overlap can change the order of the transition is potentially interesting. The paper is commendably explicit: the mean-field equations are derived rather than fitted, the epidemic threshold is obtained by linear stability analysis, and the simulation protocol is described in enough detail to be reproduced. If the claims are corrected as described below, the paper would be a useful contribution to the literature on multilayer spreading dynamics. However, as written, the central statement that 'O<1 implies a discontinuous transition' is not supported by the paper's own mean-field equation, and the issue is load-bearing because the abstract, introduction, phase diagram, and conclusions all rest on this statement.","major_comments":[{"comment":"The central claim that every O<1 yields a discontinuous transition is contradicted by the paper's own homogeneous mean-field equation. Setting y=1-ρ in Eq. (18), nonzero stationary solutions satisfy 1 = β⟨k⟩ y [1 - (1-O) y^{⟨k⟩}], i.e. β = 1/(⟨k⟩ F(y)) with F(y)=y[1-(1-O)y^{⟨k⟩}]. A saddle-node bifurcation, and hence bistability, requires F to have an interior maximum on (0,1), which occurs only when O < ⟨k⟩/(⟨k⟩+1). For O ≥ ⟨k⟩/(⟨k⟩+1), F is monotonically increasing, so the endemic branch emerges continuously at βC=1/(⟨k⟩O) via a transcritical bifurcation. The abstract's 'Otherwise, a discontinuous phase transition is observed' is therefore false as a universal statement. For ⟨k⟩=20, the simulations in Figs. 2 and 3 and the mean-field plots in Fig. 7 use O≤0.8 < 20/21, so the regime O∈[20/21,1) is never probed. The text should replace the condition O<1 by O<⟨k⟩/(⟨k⟩+1) (or a corresponding network-specific condition) and should discuss the tricritical point at O=⟨k⟩/(⟨k⟩+1).","section":"Abstract; Section IV.B, Eq. (18); Section VII"},{"comment":"The phase diagram in the β-O plane is misleading as drawn. It appears to show a bistable region for all O<1, but Eq. (18) implies that no closed bistable region exists for O ≥ ⟨k⟩/(⟨k⟩+1). The βF boundary should terminate at the tricritical value O=⟨k⟩/(⟨k⟩+1), and the phase diagram should clearly indicate that the discontinuous transition and hysteresis are confined to small O. The paper's claim that 'the bistable region is enlarged as O decreases' is consistent only within the regime O<⟨k⟩/(⟨k⟩+1), and the phase diagram should reflect this restriction.","section":"Section IV.B and Figs. 6 and 7(d)"},{"comment":"An additional concrete failure of the unqualified claim appears in the square-lattice example (Fig. 9). For a square lattice the degree is k=4, so the homogeneous mean-field criterion gives a tricritical point at O=4/5=0.8. The simulation in Fig. 9(b) uses O=0.8, which is exactly at this boundary, where Eq. (18) predicts no interior saddle-node and hence no bistable window. The paper presents this panel as an example of the discontinuous transition, without noting that the homogeneous theory places it at the transition of the transition-order boundary. This strengthens the need to state the parameter condition explicitly and to test values of O above the tricritical value.","section":"Section VI, Fig. 9 (square lattice)"}],"minor_comments":[{"comment":"The first sentence contains a grammatical error: 'In spite of the study ... has received' should be 'Although the study ... has received' or 'Despite the attention that the study ... has received'.","section":"Abstract"},{"comment":"The figure numbering in the text is inconsistent with the captions: the Watts-Strogatz results are referenced as 'Fig. 7', but Fig. 7 is the homogeneous mean-field figure; the square-lattice results are referenced as 'Fig. 8', but the caption labels them as Fig. 9; and the three-layer results are referenced as 'Fig. 9', but the caption labels them as Fig. 10.","section":"Section VI"},{"comment":"The phrase 'they colloid and annihilate' should be 'they collide and annihilate'.","section":"Section IV.B"},{"comment":"Reference [9] contains a typo, 'Strcuture' for 'Structure', and the text repeatedly spells the real network name 'SACCHCERE' where the standard name is 'Saccharomyces cerevisiae' (the multicode database entry).","section":"References"},{"comment":"The notation O1(2) is under-defined; the reader must infer that O1 and O2 denote the overlap fractions in the two layers. A sentence defining O1 and O2 explicitly before Eq. (17) would improve readability.","section":"Eq. (17)"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing to know: the core phenomenon is real and the theory is mostly sound, but the headline claim is stated too broadly. The paper shows that the AND-rule infection mechanism on a duplex network—a susceptible node needs at least one infected neighbor in each layer—can produce a discontinuous transition with hysteresis when the layers are not fully overlapping. That mechanism is new relative to prior work, which got first-order transitions via coupling to other dynamics. The mean-field derivation (Eqs. 5–13) is careful, and the linearization gives the epidemic threshold as 1/Λmax of the overlap adjacency matrix, with no fitted parameters. For homogeneous networks, Eq. (18) captures the simulation quite well in the tested regime.\n\nNow the soft spots, in proportion. The main problem is the universal claim. The abstract and conclusions say that whenever O < 1, the transition is discontinuous. That is not supported by the paper's own equation. With y = 1 − ρ, the stationary nonzero solutions satisfy 1 = β⟨k⟩ y [1 − (1 − O) y^⟨k⟩]. The function F(y) = y[1 − (1 − O) y^⟨k⟩] has an interior maximum only if O < ⟨k⟩/(⟨k⟩+1). For O ≥ ⟨k⟩/(⟨k⟩+1), F is monotone on [0,1], so the endemic branch emerges continuously at βC = 1/(⟨k⟩O). Example: ⟨k⟩ = 2, O = 0.8 gives a continuous transition. All their simulations use ⟨k⟩ = 20 and O ≤ 0.8, which is inside the discontinuous regime (20/21 ≈ 0.952), so they never hit the counterexample. The correct statement is that the transition is discontinuous for O < ⟨k⟩/(⟨k⟩+1), not for all O < 1.\n\nA second issue is the rewiring scheme. The paper rewires edges in the second layer to tune overlap but does not disclose how this changes the degree distribution. For Poisson networks the effect may be mild, but for BA networks it can matter; the reported BA results may reflect degree distribution changes as much as overlap. Third, the simulation details are thin: no error bars, no number of realizations, no code or data. These are fixable and do not undermine the main mechanism.\n\nWho this is for: people working on multilayer spreading and phase transitions in networks. The AND-rule mechanism is worth knowing, and the threshold result is useful. I would send it to peer review, not desk reject, but with a clear request to correct the O < 1 claim, report the rewiring effects, and add simulation reproducibility details.","headline":"Solid mean-field and simulation study of an AND-rule SIS model on duplex networks, with a real overlap-controlled transition, but the paper overclaims: its own Eq. (18) gives continuous behavior for O ≥ ⟨k⟩/(⟨k⟩+1), not for all O < 1.","tokens_in":18395,"tokens_out":2436,"would_cite":true,"duration_ms":24082,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"On two-layer networks, the AND-rule SIS model crosses a continuous phase transition only when every edge is shared; the moment any edge differs between layers, the outbreak becomes discontinuous and bistable, with a wider hysteresis…","keywords":["multiplex networks","SIS model","AND-rule infection","layer overlap","discontinuous phase transition","bistability","hysteresis","mean-field theory"],"falsifier":"Solve Eq. (18) numerically for a homogeneous network with $\\langle k\\rangle=20$ and $O=0.95$, continuing the stable and unstable branches as functions of $\\beta$. If the endemic branch grows continuously from $\\beta_C=1/(20O)$ with no saddle-node and no hysteresis, the paper's claim of a discontinuous transition for every $O<1$ is refuted.","tokens_in":17410,"feed_emoji":"🦠","tokens_out":8466,"duration_ms":79021,"temperature":0.7,"pith_summary":"The paper proposes a susceptible-infected-susceptible (SIS) model on two-layer networks in which a susceptible node is infected only when it has at least one infected neighbor in each layer. It argues that the fraction $O$ of edges present in both layers controls the character of the epidemic transition. For $O=1$ the dynamics reduces to the standard single-layer SIS model, and the transition is continuous. For every $O<1$, the paper claims, the transition is discontinuous: a disease-free and an endemic state coexist in a bistable window whose width grows as $O$ falls. The practical message is that inter-layer correlation, not just layer structure, can make an outbreak appear abruptly.","feed_headline":"Epidemic onset turns abrupt when network layers partially overlap","feed_subtitle":"In a two-layer SIS model, any missing overlap replaces a smooth transition with a bistable, discontinuous outbreak.","key_machinery":"The machinery is the AND-rule infection rate of Eq. (2), which counts infectious contacts as $(n_i^1+n_i^2)/2$ but multiplies by step functions requiring $n_i^1\\ge1$ and $n_i^2\\ge1$. Averaging this rate over Poisson-binomial neighbor sets gives the individual-based mean-field equation (Eq. (13)), with $R_{\\rm inf}$ written explicitly in Eq. (12). Linearizing near extinction leaves only the common-neighbor term, producing the overlapping adjacency matrix $\\tilde{A}_{ij}=A^1_{ij}A^2_{ij}$ and the threshold $\\beta_C=1/\\Lambda_{\\max}(\\tilde{A})$. In the homogeneous reduction this becomes the scalar equation (Eq. (18)), whose extra term $-(1-O)(1-\\rho)^{\\langle k\\rangle}$ is what creates the saddle-node bifurcation and the bistable region. This product adjacency matrix and the nonlinear term it induces carry the argument.","core_discovery":"The central claim is a dichotomy controlled by edge overlap. When all edges coincide ($O=1$), the AND-rule infection rate collapses to the usual SIS rate, so the endemic state emerges continuously at the threshold $\\beta_C=1/\\Lambda_{\\max}(A)$. For any partial overlap, the homogeneous mean-field equation (Eq. (18)) is claimed to develop a saddle-node pair: below $\\beta_F$ only the healthy state is stable, above $\\beta_C=1/(\\langle k\\rangle O)$ only the endemic state is stable, and between them the two coexist, producing hysteresis. The paper reports this behavior in simulations on Poisson, scale-free, small-world, lattice, and real multiplex networks and finds that the individual-based mean-field theory reproduces the simulations. The threshold is set by the overlapping adjacency matrix $\\tilde{A}_{ij}=A^1_{ij}A^2_{ij}$, so only edges common to both layers contribute to the onset of the outbreak.","pith_inferences":["The paper's own Eq. (18) actually has a saddle-node only for $O<\\langle k\\rangle/(\\langle k\\rangle+1)$; testing $O$ near 1 would reveal whether the 'discontinuous for all $O<1$' statement needs a boundary in $O$ rather than a dichotomy.","Because $\\beta_C$ is governed solely by common edges, measuring overlap is as important as measuring degree; a layered contact network with tiny overlap could suppress a pathogen that would otherwise spread on either layer.","The AND-rule is formally a two-body higher-order interaction, so the same saddle-node mechanism should appear in hypergraph or simplicial contagion models with pairwise-only interactions in each layer.","In finite populations, the observed $\\beta_F$ may be shifted by absorbing-state fluctuations, so a finite-size scaling test would show whether the bistable window survives in the thermodynamic limit."],"forward_implications":["For any two layers that are not identical, an AND-rule SIS outbreak has a hysteresis window: low-seed outbreaks die while high-seed outbreaks persist at the same infection rate.","The epidemic threshold depends only on the overlap matrix, so edges that are not shared between layers do not change $\\beta_C$; only common edges matter for onset.","In homogeneous networks $\\beta_C=1/(\\langle k\\rangle O)$, so the threshold rises as overlap decreases and diverges as $O\\to0$.","At $O=1$ the model reduces to standard SIS, confirming that the discontinuity is generated by the non-additive AND condition rather than by multiplexity alone.","Real multiplex datasets with partial overlap should exhibit the same bistable, discontinuous outbreak because the mechanism requires only common and distinct edges, not a specific network family."],"supporting_citations":[{"why":"Defines the overlap fraction O that parameterizes the two layers throughout the paper.","marker":"[56]"},{"why":"Supplies the edge-rewiring procedure used to realize a chosen value of O in simulations.","marker":"[57]"},{"why":"Gives the standard SIS mean-field equation and threshold baseline that the O=1 limit must recover.","marker":"[5]"},{"why":"Provides the largest-eigenvalue epidemic threshold for single-layer SIS that Eq. (16) generalizes to the overlap matrix.","marker":"[68]"},{"why":"Shows that a simple superposition of two SIS layers lowers the threshold, the baseline the AND-rule deviates from.","marker":"[32]"},{"why":"Derives multiplex SIS/SIR thresholds in a tensorial formalism, providing the continuous-transition reference for multiplex networks.","marker":"[40]"}],"fun_headline_variants":["Overlap in multiplex networks turns epidemic onset discontinuous","Partial layer overlap makes disease outbreaks abrupt","When multiplex layers differ, infection emerges with a jump","Edge overlap dictates smooth vs abrupt epidemic transitions","Multiplex network overlap: continuous to discontinuous outbreak shift"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The dichotomy 'continuous only at $O=1$, discontinuous for every $O<1$' assumes that the homogeneous mean-field equation has three coexisting fixed points for all partial overlap; that condition actually holds only for $O<\\langle k\\rangle/(\\langle k\\rangle+1)$, so near-complete overlap may still show a continuous onset.","fun_headline_variants_meta":{"raw":{"variants":["Overlap in multiplex networks turns epidemic onset discontinuous","Partial layer overlap makes disease outbreaks abrupt","When multiplex layers differ, infection emerges with a jump","Edge overlap dictates smooth vs abrupt epidemic transitions","Multiplex network overlap: continuous to discontinuous outbreak shift"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000146,"raw_usage":{"total_tokens":1164,"prompt_tokens":906,"completion_tokens":258,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":522,"completion_tokens_details":{"reasoning_tokens":187}},"tokens_in":522,"tokens_out":258,"duration_ms":3177,"temperature":1.0,"reasoning_tokens":187,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:22:20.232453+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve Eq. (18) numerically for a homogeneous network with $\\langle k\\rangle=20$ and $O=0.95$, continuing the stable and unstable branches as functions of $\\beta$. If the endemic branch grows continuously from $\\beta_C=1/(20O)$ with no saddle-node and no hysteresis, the paper's claim of a discontinuous transition for every $O<1$ is refuted.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the overlap fraction O that parameterizes the two layers throughout the paper."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the edge-rewiring procedure used to realize a chosen value of O in simulations."},{"cited_title":"Pastor-Satorras, C","cited_arxiv_id":null,"evidence_quote":"Gives the standard SIS mean-field equation and threshold baseline that the O=1 limit must recover."},{"cited_title":"Cozzo, R","cited_arxiv_id":null,"evidence_quote":"Shows that a simple superposition of two SIS layers lowers the threshold, the baseline the AND-rule deviates from."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Derives multiplex SIS/SIR thresholds in a tensorial formalism, providing the continuous-transition reference for multiplex networks."}],"review_version":1}