{"id":"5c59be21-c796-4fae-b8d3-116ffd68f20a","arxiv_id":"1909.00397","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A heterogeneous edge-based compartmental theory predicts the outbreak threshold and final size of misinformation spreading on correlated two-layer networks, with good agreement to simulations.","lead":"This paper models how misinformation spreads through two social platforms at once, with each person represented on both. It finds that higher average connections, more unequal connection counts, and positive overlap between platforms all make misinformation reach more people and break out faster.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Jacobian threshold criterion yields a β_c that depends on λ and γ separately, not on the effective ratio β=λ/γ alone as the paper's central claim asserts.","rationale":"The reader identified the threshold as semi-numerical and the mean-field closure as the weakest assumption, but did not notice that the Jacobian's eigenvalue problem couples λ and γ separately through the (λ−1) factor in Eq. (17). This is more load-bearing than the general clustering limitation because it directly undermines the paper's headline assertion that β_c is a well-defined network constant depending only on structure. The derivation of ω = λ μ + γ(λ−1) is straightforward from Eq. (21), so the concern is not speculative; it is an internal mathematical consequence of the authors' own equations. The paper's simulation plots of R(∞) versus β for different γ may still appear to collapse because the γ-dependence is weak at small λ, but a quantitative check is needed. The qualitative trends about average degree, heterogeneity, and correlation may survive, which is why I do not recommend outright rejection; the threshold method must be re-derived (e.g., via the next-generation matrix) and the γ-independence confirmed before the central claim can be accepted. Credit is due for the careful construction of the EBCM equations and the agreement of final-size predictions, which are independent of the threshold extraction issue.","tokens_in":15804,"tokens_out":31869,"duration_ms":296464,"concrete_test":"For the ER-ER multiplex network of Fig. 3(a) (⟨k⟩1=10, ⟨k⟩2=5), compute the leading eigenvalue of the Jacobian in Eq. (21) at (λ=0.1, γ=0.1) and at (λ=0.9, γ=0.9), both with β=1. If the eigenvalue zero-crossing (or the point where ω departs from zero) differs, β_c is not a function of β alone. As a complementary stochastic check, simulate the original process at γ=0.1 and γ=0.9, plot R(∞) against β, and test whether the curves collapse within the reported variability; if they separate by more than the error bars, the central threshold claim fails quantitatively.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The outbreak threshold is identified from the leading eigenvalue ω of the Jacobian in Eq. (21) at the trivial fixed point. This Jacobian has a block structure J = [[λM, λI],[γ(λ−1)M, γ(λ−1)I]] with M = [[X1−I, Y12],[Y21, X2−I]]. For any eigenvalue μ of M, the corresponding eigenvalue of J is ω = λ μ + γ(λ−1), as can be verified by substituting an eigenvector (u,v) and using v = (ω/λ)u − M u. Setting ω = 0 gives β_c = (1−λ)/μ, which depends on λ (equivalently on γ for fixed β). The factor γ(λ−1) originates from the discrete-time ordering correction in Eq. (17), g_a = (1−λ^{-1}) μ_a f_a. Thus the leading eigenvalue is not a function of β alone; the threshold value inferred from Fig. 2 should shift with γ even for identical network structure. The abstract and Section IV instead state that β_c is reduced by average degree, degree heterogeneity, and inter-layer correlation, implying a single γ-independent threshold. If the numerical eigenvalue computation is run at different γ, the onset point will move, so the claimed 'analytical method' does not actually define a network-intrinsic β_c. The paper does not report the theoretical threshold curve in the (λ,γ) plane, so this dependence remains hidden, but it is a direct consequence of Eq. (21).","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies misinformation spreading on two-layer multiplex networks with tunable inter-layer degree correlations. The authors propose an ignorant-spreader-recovered (ISR) model in which a spreader's recovery probability grows with the number of its spreader/recovered neighbors, develop a heterogeneous edge-based compartmental theory for the final outbreak size R(∞), and derive a threshold from a Jacobian stability analysis of the mean-field equations. They then compare theory with stochastic simulations on Erdős–Rényi and scale-free multiplex networks. The central claims are that R(∞) grows continuously with the effective transmission probability β = λ/γ above an outbreak threshold β_c, and that larger average degree, stronger degree heterogeneity, and more positive inter-layer correlation reduce β_c, while heterogeneity and correlation enlarge (reduce) R(∞) for small (large) β.","tokens_in":16099,"tokens_out":18084,"duration_ms":179342,"significance":"If the threshold analysis is corrected, the paper would offer a useful extension of edge-based compartmental theory to multiplex networks with a nonlinear, neighbor-dependent recovery rule. Strengths include a self-contained derivation that is compared with stochastic simulations without any fitted parameters, and coverage of both homogeneous and heterogeneous multiplex networks with positive and negative inter-layer correlations. The circularity burden is low because the equations are derived from the model and the threshold is not tuned to the simulation data. However, the current definition of β_c is technically flawed: the eigenvalue condition in Section III.B yields a threshold that depends on λ and γ separately, not on the ratio β alone, which undermines the paper's main quantitative claim as stated.","major_comments":[{"comment":"The outbreak threshold obtained from Eq. (21) is not a function of β = λ/γ alone. Writing M = [[X1−I, Y12], [Y21, X2−I]], the Jacobian has the block form J = [[λM, λI], [γ(λ−1)M, γ(λ−1)I]]. For an eigenvector (u,v) with M u = μ u, the eigenvalues of J are 0 and λμ + γ(λ−1). The nontrivial branch crosses zero at λ_c = γ/(μ_max+γ), i.e., β_c = 1/(μ_max+γ), which depends explicitly on γ. Therefore two parameter pairs (λ,γ) with the same ratio β can lie on opposite sides of the outbreak threshold, contradicting the abstract's and Section IV's statements that a single network-structure-dependent β_c governs spreading. The manuscript should present the threshold as a curve in the (λ,γ) plane and qualify the structural-effect statements to fixed γ.","section":"Section III.B, Eq. (21)"},{"comment":"The criterion that the leading Jacobian eigenvalue ω 'deviates from zero' at β_c is not a well-defined stability threshold and is inconsistent with Eq. (21). The vector (u,−Mu)^T is in the kernel of J for every eigenvector u of M, so J has an exact zero eigenvalue for all λ and γ. Below the threshold the leading eigenvalue is therefore exactly zero, not 'a small number very close to zero' as stated in the text. The threshold should be defined by the sign change of the nonzero branch λμ_max + γ(λ−1). In addition, the caption of Fig. 2 does not state the value of γ used for each curve, which is essential under the corrected criterion.","section":"Section III.B"},{"comment":"Because β_c depends on γ, the comparisons of R(∞) versus β at different γ values (Figs. 3, 5, and 6) and the threshold estimates from the peak of Δ (Figs. 3, 5, and 7) need a quantitative report of the theoretical and simulated threshold values for each γ and network structure. The manuscript currently marks the theoretical thresholds only as dotted lines and gives no numeric values, so the claimed 'good agreement' cannot be fully checked from the information provided. A table reporting β_c (or λ_c) from Eq. (21), from the Δ-peak method, and from simulations would resolve this.","section":"Section IV, Figs. 3-7"}],"minor_comments":[{"comment":"The factor '1 − λ −1' should be typeset as 1 − λ^{-1}; the missing superscript makes the equation ambiguous and should be corrected throughout.","section":"Eq. (17) and surrounding text"},{"comment":"The notation for the probabilities r_a^a and r_b^a is garbled in the printed text (they appear as 'ra' and 'rb'), and the subscripts and superscripts in Eqs. (8)-(9) are inconsistent. Please rewrite these equations with unambiguous labels, such as r_{a→a} and r_{b→a}.","section":"Eqs. (7)-(9)"},{"comment":"The manuscript contains several OCR or encoding artifacts, including the stray characters '澳' and 'ť' in the text and garbled axis labels in Figs. 2-7; these should be corrected in the production version.","section":"General production quality"},{"comment":"The cavity and mean-field assumptions underlying Eqs. (4)-(6) and (12) are not stated as explicitly as they should be; a short paragraph noting that the theory targets tree-like configuration-model networks and may be less accurate for clustered or strongly correlated empirical networks would make the scope clear.","section":"Section III.A"},{"comment":"The statement that 'no systematical theoretical study has been performed to date' is stronger than necessary; earlier edge-based and mean-field analyses of multiplex spreading should be cited in this context to avoid overclaiming novelty.","section":"Abstract and Section I"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the main obstacle is the threshold definition in Section III.B. The γ-dependence of the eigenvalue condition is a technical issue in the presentation of the central claim, but it appears fixable by reporting the corrected threshold expression, specifying γ in Fig. 2, and revising the conclusions to state structural trends for fixed γ. The novelty claim is somewhat overstated, but this is not the main barrier to publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is a competent extension of edge-based compartmental theory to a new misinformation-spreading model on correlated multiplex networks. The model—an ISR process with a recovery probability that increases with the number of spreader/recovered neighbors—is new, and the derivation is careful. The simulation work is solid: many independent runs, multiple network types, and the theory matches the simulations for configuration-model networks. The qualitative results about average degree, degree heterogeneity, and inter-layer correlation are plausible and supported. Credit is due for a clean derivation and honest theory-simulation comparison.\n\nThe soft spot is the threshold. The paper claims an 'analytical method based on stability analysis' that yields a network-intrinsic beta_c. But the Jacobian in Eq. (21) has eigenvalues omega = lambda mu + gamma(lambda - 1), where mu is an eigenvalue of the network matrix. The condition omega = 0 gives beta_c = (1 - lambda)/mu_max, which depends explicitly on lambda (or gamma for fixed beta). So there is no single gamma-independent beta_c for a given network. The abstract's statement that beta_c is reduced by degree, heterogeneity, and correlation holds only at fixed gamma. The authors do plot a threshold curve in the (lambda, gamma) plane in Fig. 4, so they know this, but the text overstates it. Also, computing the leading eigenvalue of a large matrix and reading where it leaves zero is semi-numerical, not an analytical derivation.\n\nOne more caveat: the theory inherits the standard cavity/mean-field limitations. It assumes tree-like configuration-model networks and statistically independent neighbor states. That's fine for the simulations here, but real clustered or time-varying multiplex networks are likely outside its quantitative reach. The paper doesn't discuss this.\n\nNo code or data, but the simulation details are sufficient to reproduce. This is an incremental but real contribution to the multiplex spreading literature, mainly useful for network scientists working on social contagions. With a revision that reframes the threshold as a function of (lambda, gamma) and tones down the 'analytical' claim, I'd be satisfied.\n\nI'd send it to peer review—it deserves careful referee time. I would not cite it in its current form because of the threshold overclaim, but after correction I might.\n\nBest,","headline":"Solid incremental multiplex-spreading theory with an oversold threshold: the Jacobian condition gives a beta_c that depends on lambda and gamma separately, not a single network-intrinsic value.","tokens_in":16601,"tokens_out":16265,"would_cite":false,"duration_ms":132086,"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":"The paper claims that a heterogeneous edge-based compartmental theory accurately predicts misinformation outbreak sizes and thresholds on correlated multiplex networks, with positive inter-layer correlation lowering the threshold.","keywords":["misinformation spreading","multiplex networks","inter-layer correlation","edge-based compartmental theory","outbreak threshold","ignorant-spreader-recovered model","degree heterogeneity","stability analysis"],"falsifier":"Take a duplex configuration-model network with the same joint degree distribution used in the paper but with many triangles added inside each layer, run the ignorant-spreader-recovered process, and compare $R(\\infty)$ versus $\\beta$ to Eqs. (12)–(14). A systematic deviation beyond simulation error would show that the neighbour-independence closure, not just parameter error, is the failing step.","tokens_in":15588,"feed_emoji":"📰","tokens_out":9294,"duration_ms":89597,"temperature":0.7,"pith_summary":"This paper tries to predict how misinformation spreads when each person is active on two online platforms at once and the two platform networks are correlated in who is well connected on each. It proposes an ignorant-spreader-recovered model on a two-layer multiplex network and derives a heterogeneous edge-based compartmental theory for the final outbreak size $R(\\infty)$. The theory also yields an outbreak threshold $\\beta_c$ from a stability analysis of the trivial fixed point. The central result is that a larger average degree, stronger degree heterogeneity, or a more positive inter-layer correlation lowers $\\beta_c$, so misinformation breaks out more easily, and these same factors enlarge $R(\\infty)$ for weak transmission but shrink it for strong transmission. The theoretical predictions agree with stochastic simulations on configuration-model duplex networks.","feed_headline":"Correlated platforms accelerate misinformation outbreaks","feed_subtitle":"Theory and simulation agree that hub-heavy, positively correlated layers lower the outbreak threshold.","key_machinery":"The central object is the heterogeneous edge-based compartmental theory, an extension of edge-based compartmental methods that tracks, separately for each degree pair $(k_a,k_b)$, the probability $\\theta_a(k_a,k_b,t)$ that a node in layer $a$ has not transmitted misinformation along a given edge by time $t$. The theory closes the dynamics with the cavity assumption: the recipient of a transmission is treated as unable to transmit, so the remaining neighbors of the sender are independent. This yields the ignorance probability $I(\\vec{k},t)=\\theta_1(t)^{k_1}\\theta_2(t)^{k_2}$, the outbreak size $R(\\infty)=1-\\sum_{\\vec{k}}p(\\vec{k})I(\\vec{k},\\infty)$, and a Jacobian matrix whose leading eigenvalue locates $\\beta_c$.","core_discovery":"On a duplex network in which each individual is represented by one replica node in each layer, the probability that a node with degree pair $(k_1,k_2)$ is still ignorant at time $t$ is written as $I(\\vec{k},t)=\\theta_1(t)^{k_1}\\theta_2(t)^{k_2}$, where $\\theta_a(t)$ is the average probability that a randomly chosen edge in layer $a$ has not yet transmitted the misinformation. The evolution of the degree-dependent transmission probabilities $\\theta_a(k_a,k_b,t)$ is closed by assuming the neighbor reached along an edge is in a cavity state, which makes the states of the other neighbors statistically independent. The outbreak threshold is located by linearizing the resulting dynamical system around the fixed point $\\theta_a=1$, $\\xi_a^R=0$ and finding where the leading eigenvalue $\\omega$ of the Jacobian leaves zero. The paper's claim is that this theory reproduces, for random and scale-free duplex networks, the simulation results for both the final outbreak size $R(\\infty)$ and the threshold $\\beta_c$.","pith_inferences":["A consequence the authors leave implicit is that hub-pairs, users who are central on both platforms, are the natural targets for intervention: reducing their activity on either layer should push the leading eigenvalue back below zero.","The reinforced recovery rule is doing important work in the theory; re-running the same analysis with constant recovery probability would show whether the reported crossover in $R(\\infty)$ is a property of reinforcement or of the multiplex structure itself.","On empirical multi-platform data, the paper's crossover prediction is testable directly: measure the rank correlation of degree rankings across platforms and check whether platforms with more positive correlation show smaller rumor cascades at high exposure but larger ones at low exposure.","If the model is extended to more than two layers, the same formulas generalize by replacing the product $\\theta_1^{k_1}\\theta_2^{k_2}$ with a product over layers, which would make the predicted threshold shift even more sensitive to positive inter-layer correlation."],"forward_implications":["Misinformation outbreaks in this model are continuous: $R(\\infty)$ rises smoothly from zero once $\\beta$ passes $\\beta_c$, so there is no hysteresis or bistability.","Because larger average degree lowers $\\beta_c$, any intervention that reduces connection density in either platform raises the amount of transmission needed for an outbreak.","Positive inter-layer correlation makes outbreaks harder to prevent: when the same individuals are hubs on both platforms, the leading eigenvalue of the Jacobian crosses zero at a smaller $\\beta$.","At large effective transmission rates the picture reverses: stronger heterogeneity and positive correlation leave more low-degree margin nodes ignorant, so the final outbreak size is smaller than in homogeneous or negatively correlated networks."],"supporting_citations":[{"why":"Supplies the generalized configuration model used to generate the correlated duplex networks on which the model is simulated.","marker":"[38]"},{"why":"Provides the edge-based compartmental formalism that the heterogeneous theory extends to degree-pair heterogeneity.","marker":"[43]"},{"why":"Introduces the probability that an edge has not yet transmitted, the central object $\\theta_a$ used in the evolution equations.","marker":"[44]"},{"why":"Gives the reduction of edge-based SIR dynamics to a small set of differential equations that motivates the $\\theta$-equations.","marker":"[45]"},{"why":"Edge-based compartmental theory for dynamical correlations on networks, the direct template for the multiplex extension.","marker":"[49]"},{"why":"Supplies the rank-based correlation measure used for inter-layer correlation and the earlier finding that inter-layer degree correlations can inhibit spreading.","marker":"[36]"},{"why":"Supplies the variability-peak method used in simulations to locate the outbreak threshold for comparison with theory.","marker":"[50]"}],"fun_headline_variants":["Correlation lowers misinformation outbreak threshold","Positive inter-layer correlation speeds misinformation spread","Hub-heavy correlated platforms accelerate misinformation","Multiplex theory and simulation agree on misinformation threshold","When platforms correlate, misinformation outbreaks come faster"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The main load-bearing assumption is the mean-field closure: all nodes with the same degree pair behave identically, and once a neighbor is placed in the cavity state, the states of the remaining neighbors are treated as statistically independent.","fun_headline_variants_meta":{"raw":{"variants":["Correlation lowers misinformation outbreak threshold","Positive inter-layer correlation speeds misinformation spread","Hub-heavy correlated platforms accelerate misinformation","Multiplex theory and simulation agree on misinformation threshold","When platforms correlate, misinformation outbreaks come faster"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000231,"raw_usage":{"total_tokens":1526,"prompt_tokens":1024,"completion_tokens":502,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":640,"completion_tokens_details":{"reasoning_tokens":439}},"tokens_in":640,"tokens_out":502,"duration_ms":10715,"temperature":1.0,"reasoning_tokens":439,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:54:17.556309+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a duplex configuration-model network with the same joint degree distribution used in the paper but with many triangles added inside each layer, run the ignorant-spreader-recovered process, and compare $R(\\infty)$ versus $\\beta$ to Eqs. (12)–(14). A systematic deviation beyond simulation error would show that the neighbour-independence closure, not just parameter error, is the failing step.","supporting_citations":[{"cited_title":"Wang , author M","cited_arxiv_id":null,"evidence_quote":"Supplies the generalized configuration model used to generate the correlated duplex networks on which the model is simulated."},{"cited_title":"Guo , author D","cited_arxiv_id":null,"evidence_quote":"Provides the edge-based compartmental formalism that the heterogeneous theory extends to degree-pair heterogeneity."},{"cited_title":"\\ Lee , author J","cited_arxiv_id":null,"evidence_quote":"Introduces the probability that an edge has not yet transmitted, the central object $\\theta_a$ used in the evolution equations."},{"cited_title":"Correlated couplings and robustness of coupled networks","cited_arxiv_id":"1010.4971","evidence_quote":"Gives the reduction of edge-based SIR dynamics to a small set of differential equations that motivates the $\\theta$-equations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Edge-based compartmental theory for dynamical correlations on networks, the direct template for the multiplex extension."},{"cited_title":"Dickison , author S","cited_arxiv_id":null,"evidence_quote":"Supplies the rank-based correlation measure used for inter-layer correlation and the earlier finding that inter-layer degree correlations can inhibit spreading."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the variability-peak method used in simulations to locate the outbreak threshold for comparison with theory."}],"review_version":1}