{"id":"f93ee56a-0464-4592-b400-1437f42815b5","arxiv_id":"2411.18709","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":10,"one_line_summary":"Peer influence creates bistability in a social media addiction model, allowing endemic addiction to persist when R0 is below the classic threshold.","lead":"Using a compartmental epidemic model, this paper shows that peer influence, modeled as a nonlinear relapse term, can make social media addiction persist even when the basic reproduction number is below 1. The authors also simulate spatial clustering of addicted and inert populations, suggesting that peer factors must be considered when designing anti-addiction interventions.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"PDE model in Eq. (11) does not conserve local population: only A and I_t diffuse, so S+A+I_t+I_p cannot stay at 1, invalidating the spatial clustering claim.","rationale":"The reader's weakest_assumption listed two candidate flaws: the mass-action form gamma2*A*I_t and the non-conservative diffusion. I focus on the diffusion flaw because it is an internal contradiction rather than a modeling choice. The mass-action relapse term is a legitimate (if debatable) incidence function; the backward bifurcation analysis, despite a likely sign typo in the 'q>0' condition in Section 4.1, can be repaired (the correct condition for two positive endemic roots is q<0, which is consistent with their gamma2_th threshold), so the core bistability result is not demonstrably false. The global stability overclaim in the abstract is a missing proof, not a demonstrated falsehood. The network narrative inconsistency is a presentation error that does not invalidate the modeling. By contrast, the reaction-diffusion system (11) violates the paper's own normalization: because S and I_p do not diffuse, the local total N is not conserved, even though (12) fixes N=1 initially. This makes the spatiotemporal simulations in Figs. 5-7 physically meaningless as written, and the 'long transient of social clustering and segmentation' is a headline claim of the paper. I therefore select this as the single most load-bearing concern. The concrete test above will settle it: if the max local population stays 1, the concern fails and the paper's spatial results may be salvageable; if it drifts, the result is an artifact.","tokens_in":13333,"tokens_out":16376,"duration_ms":136449,"concrete_test":"Numerically integrate the PDE system (11) with the parameters of Fig. 5 Set I (D_A=0.05, D_I=0.0015) and the initial condition (12); record max_{x,y}(S+A+I_t+I_p) at each time step. If this maximum exceeds 1 by more than 1e−8 (or falls below 1 by a similar tolerance) away from transient rounding, the model violates the claimed normalization N=1 and the spatial patterns in Figs. 5-7 are not population-conserving. A complementary run with D_A=D_I=0 should be used to verify that the ODE reaction terms conserve N=1, isolating the diffusion terms as the cause.","verdict_should_be":"REJECT","load_bearing_attack":"The spatiotemporal section is the paper's advertised 'social segmentation' result, and it rests on the reaction-diffusion system (11). Summing the four equations in (11) gives ∂tN = μ(1−N) + D_A∇²A + D_I∇²I_t, where N = S+A+I_t+I_p. The authors state that the population is normalized to 1 and set the initial condition (12) with N(x,y,0)=1. But since S and I_p have no diffusion while A and I_t do, the Laplacian term is generically nonzero wherever the initial A distribution is nonuniform (D_A=0.05, D_I=0.0015 in Fig. 5). Thus N evolves away from 1 immediately, so at a grid point the compartment fractions can sum to more than 1 (or less). Variables that are explicitly normalized fractions become unphysical densities with local sources/sinks. This is not a cosmetic issue: the cluster formation in Figs. 5-7 is generated by this non-conservative PDE, so the 'evolving clusters of endemic and disease-free population' cannot be interpreted as spatial segregation of a fixed population. Note that a simple repair – giving all compartments the same diffusion coefficient – removes differential diffusion and with it the pattern-forming mechanism, so the advertised phenomenon is not robust under the minimal conservation fix.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a four-compartment SAItIp ODE model for social media addiction, with relapse driven both by individual craving (γ1 I_t) and by peer influence (γ2 A I_t). The authors derive the basic reproduction number via the next-generation matrix, analyze equilibria and bifurcations, perform network simulations, and study a reaction-diffusion version on a 2D lattice. The main advertised results are (i) that the addiction-free equilibrium is globally stable when the reproduction number is below unity in the absence of peer influence, (ii) that peer influence creates a backward bifurcation and a bistable window in which addiction persists below the epidemic threshold, and (iii) that diffusion produces long-transient social clustering and segmentation.","tokens_in":13667,"tokens_out":19092,"duration_ms":161180,"significance":"If established, the paper would provide a concrete mechanism by which peer-induced relapse sustains social media addiction below the classic epidemic threshold and produces spatial segregation of addicted and inert populations. The model formulation and the next-generation matrix computation are standard and clearly presented, and the connected-component analysis of cluster size is a useful quantitative probe. However, the manuscript contains several load-bearing technical errors: the global stability claim is not proved, the stated bistability condition in Section 4.1 is inconsistent with the polynomial as written, the network simulations are run in a parameter regime where the claimed bistability is absent, and the reaction-diffusion system in Section 6 does not conserve the local population. These problems undermine the paper's central claims.","major_comments":[{"comment":"The abstract states that the addiction-free equilibrium is 'globally stable' whenever the addictive-generation number is less than unity, but Section 3 proves only local asymptotic stability via the next-generation matrix. No Lyapunov function or other global argument is supplied, and the text itself at the end of Section 3 refers only to the basin of attraction of E0. The global claim must either be proved or removed from the abstract.","section":"Abstract and Section 3"},{"comment":"The condition stated for bistability is incorrect. With p = αβσγ2 + μβγ2 > 0 as defined in Eq. (7), two positive roots of p A^2 + q A + r = 0 require q < 0 and r > 0 (together with q^2 - 4 p r > 0), not q > 0 as claimed. Using the values in Fig. 3 (μ=0.01, β=0.0542, σ=0.2, α=0.01, γ1=0.015, γ2=0.85) gives p>0 and q≈−3.4×10^-4, i.e. the opposite sign. The subsequent derivation of γ2_th from q=0 is thus presented with an internally inconsistent sign convention, and the bistability region in Fig. 2(c) needs to be re-derived.","section":"Section 4.1, Eq. (7) and Eq. (8)"},{"comment":"The network simulations are performed in a parameter regime that is not bistable. For μ=0.01, β=0.0542, σ=0.2, α=0.01, γ1=0.015, the reproduction number from Eq. (5) is R ≈ 3.57, far above 1, so E0 is unstable and the system has a single endemic equilibrium. The statement that 'even though in deterministic analysis when there is no peer effect, the steady state value related to A population was zero' is contradicted by the authors' own model: for this β, R>1, so a positive endemic equilibrium exists even with γ2=0. The network results therefore do not demonstrate bistability or sub-threshold persistence.","section":"Section 5, Figs. 3 and 4"},{"comment":"The reaction-diffusion system does not conserve the local population. Summing the four equations in Eq. (11) gives ∂t(S+A+I_t+I_p) = μ(1−N) + D_A ∇²A + D_I ∇²I_t, where N = S+A+I_t+I_p. With the initial condition in Eq. (12) and spatially non-uniform A, the Laplacian terms are generically nonzero, so N immediately evolves away from 1. The compartments are therefore no longer normalized fractions of a fixed population, and the cluster patterns in Figs. 5–7 are produced by a non-conservative PDE. This directly affects the advertised 'social segmentation' result and requires a corrected formulation.","section":"Section 6.1, Eq. (11)"}],"minor_comments":[{"comment":"The initial conditions are inconsistent: Eq. (12) sets I_t=0 and I_p=1−S−A−I_t, while the Fig. 6 caption sets I_t=1−S−A−I_p and I_p=0. Please harmonize the notation.","section":"Section 6.1, Eq. (12) and Fig. 6"},{"comment":"The description of the phase diagram says the green region is enclosed by Eq. (8), Eq. (10), and R=1, but the text describing the yellow and green regions is confusing. Please clarify which region corresponds to which combination of conditions.","section":"Section 4, Fig. 2(c)"},{"comment":"Eq. (9) has unbalanced parentheses, and Eq. (10) is missing a closing parenthesis. Please correct the typography.","section":"Eq. (9) and Eq. (10)"},{"comment":"The sentence 'the system poses three solutions, out of which two are physically achievable' is imprecise about the number of equilibria. Please state explicitly how many endemic equilibria coexist with E0 in the bistable window.","section":"Section 4"},{"comment":"The paper does not discuss the sensitivity of the main conclusions to the mass-action form γ2 A I_t. Since a saturating or frequency-dependent peer-influence term could eliminate the backward bifurcation, a brief discussion of this modeling choice would strengthen the paper.","section":"General"}],"recommendation":"reject","confidential_remarks":"The manuscript has multiple load-bearing technical errors, including an unsupported global stability claim, an incorrect sign condition for the bistability threshold, network simulations run outside the bistable parameter region, and a reaction-diffusion model that violates local population conservation. The PDE issue is not a local fix; repairing the spatial model would require reformulating the diffusion terms. I recommend rejection rather than a major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core idea here is reasonable: add a peer-influence relapse term to an SIR-type addiction model, and you get a backward bifurcation and hysteresis, so addiction can persist even when R0<1. That is a well-known mechanism in epidemic models, but it is applied here to social media addiction with a clear compartmental setup (susceptible, addictive, temporary inert, permanent inert) and dual relapse from self-craving and peers. The reproduction number is derived cleanly, and the phase diagram in γ2–R space is a helpful way to package the bistability region. This is a genuine new combination of ingredients, and the network simulations and pattern-formation section are a natural extension.\n\nBut the manuscript as written has several load-bearing problems. First, the abstract claims the addiction-free equilibrium is globally stable when R0<1, but the paper only proves local stability. There is no Lyapunov function or other global argument anywhere, so that claim should be downgraded. Second, Section 4.1 states that bistability requires q>0 in the quadratic for the endemic equilibrium. For two positive roots, the sum of roots must be positive, which requires q<0 when p>0. Their own parameter set gives q<0, so the stated condition is a sign error. It is fixable, but it is in a central condition. Third, Eq. (9) for β_c is garbled and appears without derivation; I could not verify it, and a referee should ask for a clean derivation or a corrected formula. Fourth, the reaction-diffusion system is not locally conservative: only A and I_t diffuse, so the sum S+A+I_t+I_p deviates from 1 wherever those Laplacians are nonzero. The stress-test note is right. The cluster formation in Figs. 5–7 is therefore an artifact of a non-physical local source/sink, not a robust property of the model. Giving all compartments the same diffusion coefficient would fix conservation but remove the pattern mechanism, so the advertised social-segmentation result is fragile. Finally, the network section says that with no peer effect the deterministic steady state for A is zero, but the parameters used (γ1=0.015, β=0.0542, μ=0.01, α=0.01, σ=0.2) give R0>1, so the disease-free state is not stable. That inconsistency should be resolved.\n\nNone of this kills the central qualitative point. A corrected model—fixed sign condition, verified formulas, and a conserving spatial formulation—would likely still show peer-driven bistability. The errors are the kind a careful revision can fix. The manuscript deserves a serious referee: the model is well-motivated, the analysis is mostly transparent, and the topic matters. My recommendation is to send it to peer review, with explicit instructions to address the global-stability overclaim, the q sign error, the derivation of Eq. (9), and the conservation flaw in the PDE system.","headline":"A plausible social-media-addiction model with peer-induced relapse and a backward bifurcation, but the current manuscript overreaches: a sign error in the bistability condition, an unproven global-stability claim, and a non-conservative reaction-diffusion system that undermines the pattern-formation result.","tokens_in":14181,"tokens_out":4418,"would_cite":false,"duration_ms":117110,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92D30","37G10","35K57"],"pacs":["89.65.-s","87.23.Ge"],"model":"deepseek-v4-flash","headline":"The paper claims that peer influence, modeled as a nonlinear relapse rate proportional to the product of addicted and temporarily inert populations, creates a backward bifurcation and a bistable regime, allowing social media addiction to…","keywords":["social media addiction","epidemic compartmental model","backward bifurcation","bistability","peer influence","relapse","reaction-diffusion pattern formation","complex networks"],"falsifier":"A direct test would be to fit the model to longitudinal social-media-use data with and without a saturating peer-relapse function and check whether the best-fitting interaction is linear in the product A I_t. More concretely, if a frequency-dependent relapse rate γ2 A I_t / (1 + ω A) or γ2 A I_t/N (rather than γ2 A I_t) removes the backward bifurcation in the same parameter regime, the paper's central claim that peer influence per se sustains addiction below R0=1 would be falsified for those more realistic contact laws. Alternatively, an agent-based simulation with per-contact peer influence rather than density-dependent influence would show whether the bistability survives social heterogeneity.","tokens_in":13096,"feed_emoji":"📱","tokens_out":5351,"duration_ms":41080,"temperature":0.7,"pith_summary":"The paper builds an SAI_tI_p epidemic model of social media addiction that splits the population into susceptible, addicted, temporarily inert, and permanently inert individuals. Its central claim is that peer influence—written as a nonlinear relapse term proportional to the product of addicted and temporarily inert populations—turns the addiction-free state into a backward bifurcation and produces a bistable regime. In that regime the endemic addicted state coexists with the addiction-free state even when the basic reproduction number R0 is below 1, so a peer-pressure-driven relapse can keep addiction alive below the classic epidemic threshold. The paper also reports that in a reaction-diffusion version on a 2D lattice, the bistability combined with diffusion generates long-lived spatial clustering, which the authors interpret as social segmentation of addicted and inert populations.","feed_headline":"Peer influence keeps addiction alive below the epidemic threshold","feed_subtitle":"A relapse model shows addiction can persist even when the reproduction number is below one.","key_machinery":"The carrying object is the SAI_tI_p compartmental model (susceptible S, addicted A, temporary inert I_t, permanent inert I_p), specifically the nonlinear relapse term γ2 A I_t that makes the transition from temporary inert back to addicted proportional to the product of the addicted and inert subpopulations. This mass-action (bilinear) peer-influence term is what generates the backward (saddle-node) bifurcation and the bistable window R_c < R0 < 1, with thresholds γ2_th and R_c derived from the endemic equilibrium's quadratic pA²+qA+r=0. The analysis uses the next-generation matrix to define R0 and a reaction-diffusion extension (with diffusion only for A and I_t) to produce the reported spatiotemporal clustering.","core_discovery":"The paper's core discovery is that the peer-influenced relapse rate γ2 A I_t, which is absent from the reproduction number R0, controls the qualitative outcome of the model. When γ2 is small, the system shows a forward transcritical bifurcation at R0=1 and the addiction-free equilibrium is the only stable state below threshold. When γ2 is large enough (above a derived threshold γ2_th), a saddle-node bifurcation appears at R0=1, giving a backward bifurcation with a bistable interval R_c < R0 < 1. Consequently, even though R0 measures the average number of new addicts from one addict as below one, the endemic addiction state can persist, and the system exhibits hysteresis: whether the population is addicted or addiction-free depends on history and initial conditions. The authors further demonstrate that with diffusion, this bistability yields pattern formation and long transients of spatial clusters rather than fast convergence to a homogeneous steady state.","pith_inferences":["If peer influence is better represented by a saturating (frequency-dependent or Michaelis-Menten type) function instead of the bilinear mass-action term, the backward bifurcation and sub-threshold endemic state may disappear; testing this sensitivity would indicate how robust the central claim is to the form of peer pressure.","The reaction-diffusion model only lets A and I_t diffuse, so local population conservation fails (S and I_p remain fixed in space); permitting all compartments to diffuse could change the pattern formation and the interpretation of social segmentation.","The model's bistability suggests an intervention that reduces γ2 below γ2_th could eliminate the endemic state; this yields a testable prediction that anti-peer-pressure campaigns (e.g., reduced social media exposure) can act as a control parameter in the same way vaccination acts in epidemic models.","The same mechanism might apply to other self-reinforcing social behaviors with relapse, such as smoking, gambling, or misinformation sharing, where peer influence on relapse rather than initial exposure may be what sustains the behavior below the classic threshold."],"forward_implications":["Interventions that only aim to push the reproduction number below 1 will not eradicate addiction if peer-influenced relapse is strong, because the system can remain in the endemic state for R0 below 1.","Because the bistable region widens as γ2 increases, reducing peer pressure (lower γ2) shrinks the parameter window in which addiction can survive, and can restore the addiction-free state as the only outcome.","The model predicts hysteresis: after addiction has become endemic, temporarily reducing risk factors to bring R0 below 1 may not eliminate the addiction unless the system is pushed past the bistable fold point.","In a spatially structured society, diffusion of addictive behavior combined with bistability produces spatial segregation of addicted and inert clusters rather than a homogeneous mix, so local pockets of addiction can persist even as the overall population approaches an endemic steady state.","Network simulations with heterogeneous contacts show residual addictive populations even in parameter regions where the deterministic model predicts an addiction-free steady state, because some individuals never fully recover."],"supporting_citations":[{"why":"Provides the next-generation matrix technique used to compute the reproduction number R0 and to discuss sub-threshold endemic equilibria.","marker":"[11]"},{"why":"Supplies the standard reproduction-number formalism for compartmental models, which the paper extends with the temporary inert class.","marker":"[10]"},{"why":"Offers the bifurcation-analysis framework for epidemic models with generalized incidence, on which the backward-bifurcation study is based.","marker":"[1]"},{"why":"Demonstrates backward bifurcation in a smoking-cessation model with media campaigns, the precedent for bistability in addiction behavior.","marker":"[30]"},{"why":"Presents the authors' earlier quantitative model and network analysis for patterned proliferation of social media addiction, directly informing the spatiotemporal section.","marker":"[20]"},{"why":"Supplies the SIR epidemic backbone from which the SAI_tI_p compartment model is extended.","marker":"[17]"}],"fun_headline_variants":["Peer influence can sustain addiction below R0=1","Backward bifurcation: how peer relapse defeats the threshold","Spatial patterns in addiction from peer-driven relapse","Hysteresis in social contagion: peer effects rule below R0=1"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central result depends on the choice that peer influence drives relapse through the bilinear term γ2 A I_t — the product of the addicted and temporarily inert populations — and that only the addicted and temporarily inert populations diffuse in space; if peer influence saturates or all compartments diffuse, the sub-threshold endemic state and the spatial clustering may not occur.","fun_headline_variants_meta":{"raw":{"variants":["Peer influence can sustain addiction below R0=1","Backward bifurcation: how peer relapse defeats the threshold","Spatial patterns in addiction from peer-driven relapse","Hysteresis in social contagion: peer effects rule below R0=1"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000751,"raw_usage":{"total_tokens":3373,"prompt_tokens":1002,"completion_tokens":2371,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":618,"completion_tokens_details":{"reasoning_tokens":2301}},"tokens_in":618,"tokens_out":2371,"duration_ms":18429,"temperature":1.0,"reasoning_tokens":2301,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:58:29.615057+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test would be to fit the model to longitudinal social-media-use data with and without a saturating peer-relapse function and check whether the best-fitting interaction is linear in the product A I_t. More concretely, if a frequency-dependent relapse rate γ2 A I_t / (1 + ω A) or γ2 A I_t/N (rather than γ2 A I_t) removes the backward bifurcation in the same parameter regime, the paper's central claim that peer influence per se sustains addiction below R0=1 would be falsified for those more realistic contact laws. Alternatively, an agent-based simulation with per-contact peer influence rather than density-dependent influence would show whether the bistability survives social heterogeneity.","supporting_citations":[{"cited_title":"Reproductionnumbersandsub-thresholdendemicequilibriaforcompartmentalmodelsofdisease transmission","cited_arxiv_id":null,"evidence_quote":"Provides the next-generation matrix technique used to compute the reproduction number R0 and to discuss sub-threshold endemic equilibria."},{"cited_title":"Reproduction numbers of infectious disease models","cited_arxiv_id":null,"evidence_quote":"Supplies the standard reproduction-number formalism for compartmental models, which the paper extends with the temporary inert class."},{"cited_title":"Bifurcationanalysisofansirsepidemicmodelwithgeneralizedincidence","cited_arxiv_id":null,"evidence_quote":"Offers the bifurcation-analysis framework for epidemic models with generalized incidence, on which the backward-bifurcation study is based."},{"cited_title":"Backward bifurcation in a smoking cessation model with media campaigns","cited_arxiv_id":null,"evidence_quote":"Demonstrates backward bifurcation in a smoking-cessation model with media campaigns, the precedent for bistability in addiction behavior."},{"cited_title":"Visual representation for patterned proliferation of social media addiction: Quantitative model and network analysis","cited_arxiv_id":null,"evidence_quote":"Presents the authors' earlier quantitative model and network analysis for patterned proliferation of social media addiction, directly informing the spatiotemporal section."},{"cited_title":"Acontributiontothemathematicaltheoryofepidemics","cited_arxiv_id":null,"evidence_quote":"Supplies the SIR epidemic backbone from which the SAI_tI_p compartment model is extended."}],"review_version":1}