{"id":"b337473e-6a23-483b-b6c7-db4be852281b","arxiv_id":"2601.21023","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"An anticonformity opinion model yields, in the infinite-population limit, stationary opinion distributions that are generalized Bernoulli-convolution fractals when the pull toward extremes is strong.","lead":"This paper analyzes a new model of opinion formation in which people move away from the opinions they hear, rather than toward them. In the large-population limit, the model's long-run opinion distribution can become a fractal, a self-similar Cantor-like set, under strong polarization parameters.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Asymmetric finite-agent equilibria are not rigorously linked to the fractal ρ∞: only finite-time propagation of chaos is proved, so the opinion-fragmentation claim may describe the PDE, not the N-agent system.","rationale":"The reader's weakest assumption coincides with this concern, and I agree with CONDITIONAL. The paper's quantitative PDE results (Theorems 10, 11, 13, 14) are plausible and appear internally consistent; Theorem 15 is a short and standard consequence of the self-similar support equation (4.3) once disjointness holds, so I do not see a fatal flaw in the fractal-geometry argument itself. The real soft spot is the modeling claim: the agent-based system is the paper's motivation, but the asymmetric regime lacks uniform-in-time propagation of chaos and finite-N stationary results. The finite-time PoC bound grows exponentially, so it cannot justify taking t→∞ before N→∞. This is not a mathematical inconsistency in the PDE theorems, but it is load-bearing for the paper's advertised phenomenon. The proposed analytical check—uniform second-moment estimate for the asymmetric case—would settle whether this is a technical omission or a genuine obstacle. I also note a minor boundary issue: Theorem 15 states D∈(0,1) but for μ+=1 the equation gives D=0; this is easily fixed by assuming μ±∈(0,1) and does not affect the main regime. Overall, CONDITIONAL is the right verdict.","tokens_in":20875,"tokens_out":16929,"duration_ms":175384,"concrete_test":"Attempt to prove the asymmetric analogue of Corollary 4: for μ+≠μ−, show E[(A_i^{N,t} - m_t)^2] ≤ C/N uniformly in t, using the explicit exponential convergence of m_t to m∞ in Eq. (2.3). If this estimate holds, the coupling argument of Theorem 5 extends verbatim to μ+≠μ−, yielding uniform-in-time propagation of chaos and closing the gap. If the estimate fails (or requires an extra t-dependent factor), then the finite-N system is not uniformly close to the mean-field equilibrium at large times, and the fragmentation claim must be weakened to a PDE-level statement.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim (Theorem 15) is a statement about the mean-field PDE, not about the finite-agent Markov process that motivates the paper. The bridge from finite N to the PDE is incomplete exactly in the asymmetric regime μ+≠μ− that constitutes the new contribution. Theorem 1 gives only finite-time propagation of chaos, W1(L(X^N_t), ρ_t^{⊗N}) ≤ C/√N e^{(μ++μ−)t}, which is useless as t→∞; uniform-in-time propagation of chaos is proved only for μ+=μ− (Theorem 5), requiring the special second-moment estimate of Corollary 4. No such estimate is proved for μ+≠μ−, and Section 2.3 explicitly states that finite-N stationarity was not proved ('we were not quite able to prove it') even in the symmetric case. Consequently the limits t→∞ and N→∞ are not rigorously interchangeable in the asymmetric case. The paper's own numerical evidence (Figures 2 and 3) suggests the finite-N system approaches the PDE equilibrium, but this is not a proof. If the central claim is read strictly as 'the mean-field PDE has a fractal stationary measure,' the concern does not bite; but the abstract and conclusion present it as an explanation of opinion fragmentation in multi-agent systems, where this gap is load-bearing.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a multi-agent opinion dynamics model on [-1,1] with an 'anticonformity' update rule: in each pairwise encounter the listener moves away from the speaker's stated extreme by proportions μ+ or μ-. The main results are: (i) a finite-time propagation of chaos bound for the N-agent system (Theorem 1); (ii) a uniform-in-time propagation of chaos and long-time estimates in the symmetric case μ+ = μ- (Theorem 5, Proposition 6); (iii) quantitative convergence of the mean-field PDE to a unique stationary law in Wasserstein and Fourier metrics, in both symmetric and asymmetric regimes (Theorems 10, 11, 13, 14); and (iv) the headline result (Theorem 15) that for μ+ + μ- > 1 the stationary law is supported on a Cantor-type fractal set of Hausdorff dimension D solving (1-μ+)^D + (1-μ-)^D = 1, recovering and generalizing the Bernoulli convolution. Numerical experiments illustrate the fractal CDFs in the supercritical case and the apparently absolutely continuous behavior in the subcritical case.","tokens_in":21222,"tokens_out":6630,"duration_ms":70303,"significance":"If correct, the paper makes a surprising and attractive connection between opinion dynamics and Bernoulli convolutions, and it provides a new family of self-similar equilibrium measures in the asymmetric case. The explicit first-moment ODE, the second-moment estimate in Corollary 4, and the coupling arguments in Theorems 5 and 11 are clear and useful. The numerical simulations are extensive and appear to confirm the predicted fractal structure. At the same time, the manuscript currently falls short of fully supporting the agent-level interpretation of the fractal equilibrium: the only bridge from finite N to the stationary law in the asymmetric case is a finite-time propagation-of-chaos bound that diverges exponentially in t, and Section 2.3 explicitly states that finite-N stationarity was not proved even in the symmetric case. The PDE-level results are plausible, but several load-bearing proofs are omitted or delegated to the authors' prior work, so the present form is not yet self-contained.","major_comments":[{"comment":"Theorem 1 is the only quantitative mean-field limit in the general asymmetric regime, but its proof is omitted with the sentence 'follows along the same lines as the one given in [8]'. The update rule here (1.1) differs from that in [8], and m_t is time-dependent when μ+ ≠ μ-, so a direct reduction is not automatic. Since this theorem is the rigorous basis for the mean-field PDE (2.6), the full proof, or at least a precise statement of the analogous theorem with all constants checked, is required.","section":"Section 2.1, Theorem 1"},{"comment":"Existence and uniqueness of the stationary law ρ∞ is asserted via a 'simple contraction argument (which we omit)'. This object is the target of all the convergence theorems (Theorems 10, 11, 13, 14) and the input to Theorem 15. Without a proof of existence and uniqueness, the convergence results are statements about an object that is not rigorously constructed. Provide the metric and contractive estimate, including the asymmetric case μ+ ≠ μ-.","section":"Section 4, Eq. (4.1)"},{"comment":"The proofs of Theorems 13 and 14 are omitted. Theorem 14 is the only convergence-to-equilibrium result for the asymmetric case, and Theorem 13 gives the Fourier-metric analogue in the symmetric case. These are central to the paper's claim of convergence to the fractal equilibrium. Please include the full arguments, or at least a complete proof sketch with all constants and the treatment of the asymmetric mean dynamics, rather than referring to 'the same lines'.","section":"Section 3, Theorems 13 and 14"},{"comment":"The abstract, introduction, and conclusion present ρ∞ as the long-run opinion profile of the multi-agent system. For μ+ ≠ μ- this link is not established: Theorem 1 is finite-time with bound C/√N e^{(μ+ + μ-)t}, and Theorem 5 is restricted to μ+ = μ-. Section 2.3 explicitly says that finite-N stationarity was not proved ('we were not quite able to prove it') even in the symmetric case. Thus the interchange of t→∞ and N→∞ is missing exactly in the regime that contains the new fractal family. Either prove a uniform-in-time asymmetric propagation of chaos estimate, or restrict the fragmentation claims to the mean-field PDE and clearly label the agent-level interpretation as a numerically supported conjecture.","section":"Sections 2.2-2.3 and Conclusion"}],"minor_comments":[{"comment":"'collusion gain operator' should be 'collision gain operator'.","section":"Section 3, after Theorem 10"},{"comment":"The inline labels for μ in the figure captions appear corrupted (e.g. '7! = 0:6'); please fix the typography so that parameter values are legible.","section":"Figures 2 and 3"},{"comment":"The expression for C has a removable singularity when the initial mean m0 equals m∞; please state the limiting formula or treat that case separately.","section":"Equation (2.3)"},{"comment":"The statement 'by the obvious symmetry, we can assume μ- ≤ μ+' deserves a brief explanation: swapping +1 and -1 maps (μ-, μ+) to (μ+, μ-).","section":"Introduction"},{"comment":"The arXiv abstract cites [9] and [24,51], while the body and reference list use [8] and [23,44]; make the numbering consistent.","section":"General references"}],"recommendation":"major_revision","confidential_remarks":"The main risk is the asymmetric finite-N gap. If the authors can either close the uniform propagation-of-chaos gap or explicitly downgrade the agent-level interpretation to a conjecture, and if they supply the omitted proofs for the contraction argument and Theorems 1, 13, and 14, the paper would be much stronger. I would ask the editor to insist that the current heavy reliance on the authors' own prior work be reduced, since several of these omitted arguments are load-bearing for the paper's central claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know about this paper because it connects a contrarian opinion dynamics model to Bernoulli convolutions, which is new and, as far as the PDE is concerned, correct. The authors take an agent-based model where a listener moves away from the stated opinion, derive the mean-field PDE, and show that when μ+ + μ- > 1 the stationary law is supported on a self-similar set with Hausdorff dimension D solving (1-μ+)^D + (1-μ-)^D = 1. Equal parameters reduce it to the Bernoulli convolution from their earlier SIADS paper; unequal parameters give a new family of self-similar measures. That is a real contribution.\n\nThe model is natural and clearly distinguished from the conformist dynamics of [8]. The first-moment ODE is explicit, and the convergence results for the PDE (Theorems 10, 11, 14) are quantitative. Theorem 15 is terse but sound: the set equation plus disjointness gives the dimension. The numerics support the fractal/smooth dichotomy and look honest.\n\nThe soft spots are real but not disqualifying. Several proofs are omitted: Theorem 1's propagation of chaos, the contraction argument for existence/uniqueness of the stationary law, and Theorems 13 and 14. Referencing [8] for Theorem 1 is acceptable for a companion paper, but a referee should ask for at least a sketch. The bigger issue is the finite-N to PDE link when μ+ ≠ μ-. Theorem 1 gives a bound C/√N e^{(μ++μ-)t}, which is useless as t→∞; uniform-in-time propagation of chaos is only proved for μ+ = μ- (Theorem 5). So the fractal ρ∞ is rigorously the PDE equilibrium but not rigorously shown to describe long-run behavior of the N-agent system in the asymmetric case, which is the new regime. The authors are candid: Section 2.3 says they were 'not quite able to prove' finite-N stationarity even symmetric. The numerics suggest the link, but it is open. If the claim is read strictly as 'the PDE has a fractal stationary measure,' the paper stands. But the abstract sells it as explaining opinion fragmentation in multi-agent systems, where the gap is load-bearing.\n\nI would send this to referees with a request to close or prominently state the gap. It deserves serious referee attention. For a reading group it would generate good discussion. I would cite it if I worked on opinion dynamics or Bernoulli convolutions.","headline":"New family of fractal stationary laws for a contrarian opinion model, but the finite-N to PDE link is only rigorous when the two update rates are equal.","tokens_in":21645,"tokens_out":3129,"would_cite":true,"duration_ms":33160,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","28A80","35Q84"],"pacs":[],"model":"deepseek-v4-flash","headline":"Contrarian opinion dynamics can make the group's equilibrium distribution fractal, and the fractal dimension has a closed-form equation.","keywords":["opinion dynamics","contrarian behavior","mean-field limit","opinion fragmentation","Bernoulli convolution","fractal support","Hausdorff dimension","propagation of chaos"],"falsifier":"Fix μ_- = 0.4 and μ_+ = 0.8, run the N-agent process to very large times for large N, and check whether the empirical distribution concentrates on the self-similar set predicted by the dimension equation (1-μ_-)^D + (1-μ_+)^D = 1. If the finite-N stationary laws converge to a different limit, or if the empirical support's dimension does not match D, the central claim that the PDE fractal describes the long-run opinion distribution is wrong.","tokens_in":20783,"feed_emoji":"🕸️","tokens_out":5979,"duration_ms":66374,"temperature":0.7,"pith_summary":"This paper studies a crowd of agents whose opinions live in [-1,1] and who, on meeting, move away from the opinion they hear. The authors prove that in the infinite-population limit this contrarian interaction is described by a simple kinetic PDE, and that its solution converges to a unique equilibrium. The central result is that when the two update strengths μ_- and μ_+ sum to more than 1, the equilibrium opinion distribution is concentrated on a Cantor-like fractal set rather than spread over the interval. Its Hausdorff dimension is the unique solution of (1-μ_-)^D + (1-μ_+)^D = 1; when the two strengths are equal, the equilibrium is exactly the classical Bernoulli convolution. A sympathetic reader should care because the model gives a minimal mathematical mechanism by which perpetual disagreement and opinion fragmentation can be an equilibrium outcome, not a transient.","feed_headline":"Contrarian bias turns public opinion into a fractal","feed_subtitle":"A mean-field model shows fragmented opinion equilibria are self-similar, with a formula for their dimension.","key_machinery":"The stationary law is characterized by a distributional fixed point Z∞ = B((1-μ_+)Z∞ + μ_+) + (1-B)((1-μ_-)Z∞ - μ_-), where B is Bernoulli with parameter (1-m∞)/2 and m∞ = (√μ_+ - √μ_-)/(√μ_+ + √μ_-). Iterating this identity shows the support S is the self-similar set S = ((1-μ_+)S + μ_+) ∪ ((1-μ_-)S - μ_-); when μ_- + μ_+ > 1 the two branches are disjoint, so standard self-similar set theory gives the Hausdorff dimension. The rest of the argument is the mean-field passage: propagation-of-chaos bounds connect the finite system to the PDE, and coupling estimates give convergence to equilibrium.","core_discovery":"The paper claims that anticonformist, or contrarian, interactions are enough to produce opinion fragmentation as the stable long-run state of a large population. Concretely, it establishes that the mean-field limit of the N-agent process is a nonlinear jump process whose law solves a Boltzmann-type PDE; for any parameters μ_- and μ_+ in (0,1], the PDE has a unique stationary distribution ρ∞, and solutions converge to it with quantitative rates. When μ_- + μ_+ > 1, the support of ρ∞ is the attractor of two contracting maps and hence a fractal of Hausdorff dimension D solving the dimension equation; in the symmetric case ρ∞ is the Bernoulli convolution. The paper thus extends a known link betw","pith_inferences":["If the mean-field equilibrium transfers to finite systems — which the paper does not prove in the asymmetric case — then opinion gaps are a genuine infinite-population phenomenon, and large finite populations should show the fractal support only approximately, with finite-N fluctuations filling the gaps.","The asymmetric family (ρ∞) may be worth studying as a generalized Bernoulli convolution in its own right: the boundary case μ_- + μ_+ = 1, where the paper's numerics show increasingly irregular histograms, is a natural target for dimension and absolute-continuity analysis.","A direct testable extension is to run agent simulations with μ_- ≠ μ_+ and estimate the dimension of the empirical support; matching D from the dimension equation would confirm that the PDE fractal is visible in finite populations, while a mismatch would pinpoint the missing uniform-in-time propagation of chaos."],"forward_implications":["If μ_- + μ_+ > 1, the equilibrium opinion distribution has zero Lebesgue measure: a positive fraction of opinion space is never occupied, so in infinite populations some opinions are unobtainable.","The stronger the contrarian pull (larger μ's), the smaller the Hausdorff dimension, so D provides a quantitative index of fragmentation.","In the symmetric case μ_- = μ_+, the equilibrium is the classical Bernoulli convolution, so known results about Bernoulli convolutions transfer to this opinion model.","The convergence estimates imply that even in the asymmetric case the PDE solution approaches the fractal distribution at a quantified rate, justifying numerical observation of the fractal at large times.","Because the contrarian model admits a unique nontrivial equilibrium for every μ_- and μ_+, contrarianism sustains persistent opinion diversity more robustly than conformism."],"fun_headline_variants":["Contrarian bias makes public opinion fractal","Anticonformists drive opinion into fractal shapes","Fractal geometry emerges in opinion dynamics","Model shows contrarian turns opinion into fractal","Fragmented opinions take fractal form"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The finite-agent system converges to the mean-field PDE uniformly in time; this is proved only for μ_- = μ_+, while for μ_- ≠ μ_+ the propagation-of-chaos bound grows like e^{(μ_-+μ_+)t}, and Section 2.3 states that even finite-N stationarity was not proved.","fun_headline_variants_meta":{"raw":{"variants":["Contrarian bias makes public opinion fractal","Anticonformists drive opinion into fractal shapes","Fractal geometry emerges in opinion dynamics","Model shows contrarian turns opinion into fractal","Fragmented opinions take fractal form"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00017,"raw_usage":{"total_tokens":1110,"prompt_tokens":758,"completion_tokens":352,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":502,"completion_tokens_details":{"reasoning_tokens":286}},"tokens_in":502,"tokens_out":352,"duration_ms":4377,"temperature":1.0,"reasoning_tokens":286,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T07:08:06.945437+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix μ_- = 0.4 and μ_+ = 0.8, run the N-agent process to very large times for large N, and check whether the empirical distribution concentrates on the self-similar set predicted by the dimension equation (1-μ_-)^D + (1-μ_+)^D = 1. If the finite-N stationary laws converge to a different limit, or if the empirical support's dimension does not match D, the central claim that the PDE fractal describes the long-run opinion distribution is wrong.","supporting_citations":[],"review_version":1}