Pith. sign in

REVIEW 5 major objections 4 minor 55 references

The paper claims that a one-parameter Gaussian graphon, fitted only to a real network's mean interaction strength, reproduces that network's asymptotic opinion distribution to Fourier distances of order 10^-3.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 08:55 UTC pith:66OMNT2K

load-bearing objection Useful model and stationary-state analysis, but the convergence theorem is unproven for the actual constrained dynamics and the main empirical claim is calibration, not prediction. the 5 major comments →

arxiv 2607.02821 v2 pith:66OMNT2K submitted 2026-07-02 physics.soc-ph math-phmath.APmath.MPmath.OCnlin.AO

From graphons to real-world networks: kinetic opinion dynamics under selective media influence

Bertram D\"uring , Martina Fraia , Alessandro Licciardi This is my paper
classification physics.soc-ph math-phmath.APmath.MPmath.OCnlin.AO MSC 35Q2035Q7035Q9191D30
keywords opinion dynamicsgraphonBoltzmann-type equationskinetic theorymedia influencemodel predictive controlreal-world social networkssphere of legitimate controversy
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper builds a kinetic model of opinion formation on networks in which agents exchange opinions through bounded-confidence compromise plus noise, and a media controller steers only those agents whose opinions lie in a 'sphere of legitimate controversy' (selective media influence). From the microscopic rules the authors derive Boltzmann-type and Fokker-Planck equations, show that the stationary states are piecewise Beta distributions whose parameters the media shifts inside the controversy region, and prove exponential convergence to equilibrium in a Fourier metric. Their central numerical claim is that a Gaussian graphon with a single fitted bandwidth (sigma_W* = 0.4445) reproduces the asymptotic opinion distributions produced by the model on two real single-issue Twitter networks up to Fourier distances of 4.00e-3 and 5.18e-3. If true, this implies that the fine topology of a social network barely matters for the macroscopic opinion outcome; only the network's mean interaction strength does, making the graphon a cheap surrogate. The authors themselves stress that the match holds for the opinion marginal, not for the microscopic structure.

Core claim

Central claim: for this opinion model the network structure matters only through its mean interaction intensity. In the Fokker-Planck limit the graphon enters the stationary state only through the weighted mass and mean; the media shifts the Beta parameters inside the controversy set, giving a piecewise Beta distribution continuous at its boundary. The theorem states exponential Fourier-metric convergence to the stationary state at a rate that is strictly negative regardless of the graphon. Numerically, one fitted Gaussian graphon reproduces the asymptotic opinion marginal of two real Twitter networks to Fourier distances 4.00e-3 and 5.18e-3.

What carries the argument

The graphon W(x,y) as the Boltzmann interaction kernel; in the Fokker-Planck limit only the weighted mass rho_W(x) and mean m_W(x) survive. The media control is the closed-form projection (Lemma 3.1) of the unconstrained linear update onto the controversy set S. Convergence is measured in the Fourier metric D_s, which controls the opinion law at every graph position.

Load-bearing premise

The exponential-convergence theorem assumes the media update can be treated as the unconstrained linear map w''=(lambda_tilde w + delta^2 w_d)/(lambda_tilde + delta^2); if that substitution fails, or if the stationary state does not in fact solve the noiseless Boltzmann equation as asserted, the exponential convergence conclusion does not follow.

What would settle it

Rewire the gun-control network edges uniformly at random while preserving each node's degree (configuration model); run the model on the rewired matrix and on the fitted Gaussian graphon. If the asymptotic opinion marginals differ by more than the reported 10^-3 Fourier distance, the claim that only the mean interaction strength matters would be falsified.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the surrogate claim is right, one can simulate opinion dynamics on real networks without storing the adjacency matrix: a one-parameter graphon uses O(N) storage and allows arbitrarily many particles for smoother densities.
  • The near-identical calibrated bandwidths for the two networks suggest that in this model the asymptotic opinion distribution is determined by a single scalar, the mean interaction strength, rather than by community structure or degree distribution.
  • The model implies that to achieve consensus at the media target, either the compromise interactions must be global or the media must reach the whole opinion interval; if both are localised, extreme agents are shielded and polarisation persists.
  • The convergence theorem guarantees that every solution on a graphon relaxes to the piecewise-Beta stationary state exponentially fast in the Fourier metric, uniformly in the graph position, provided the media update is replaced by its unconstrained linearisation.
  • The piecewise-Beta characterisation gives an explicit formula for the stationary opinion distribution under selective media influence, which could be used to estimate media strength or target opinion from observed opinion surveys.

Where Pith is reading between the lines

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

  • The claim that only the mean interaction intensity matters suggests a stronger, testable version: any graphon family (e.g., a stochastic block model) whose mean interaction strength matches rho_W should produce the same asymptotic marginal; comparing such surrogates would delineate exactly what structure, if any, is washed out.
  • Since the match is only for the opinion marginal, the surrogate should fail for position-resolved observables such as the correlation between opinion and degree; measuring how the Fourier distance degrades when the population is split by connectivity would give a quantitative boundary of the surrogate's validity.
  • A natural extension the authors do not pursue is to let the media act on two competing targets on overlapping controversy sets; the piecewise-Beta framework would then yield a stationary state with two shifted Beta branches, providing a ready-made model for polarisation by adversarial media.
  • Because the convergence proof uses the unconstrained linearised media update, a direct Monte Carlo comparison of the clamped and unclamped evolutions would show whether the exponential rate holds for the actual model or only for its proxy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

5 major / 4 minor

Summary. The paper proposes a kinetic model of opinion formation on graphon-modulated networks in which a media influence acts selectively on a subset S of opinions (Hallin's spheres) through a model predictive control solved in closed form. The authors derive a Boltzmann-type equation, analyse moment evolution, pass to a Fokker-Planck equation in a quasi-invariant limit, characterise its stationary states as piecewise Beta-type distributions, and claim exponential convergence to equilibrium in a Fourier-based metric. Numerical experiments are run on Gaussian graphons and on two real Twitter networks (gun control, Obamacare), with initial opinions taken from Barberá ideal points. A one-parameter Gaussian graphon is fitted to each real network and the asymptotic opinion distributions are compared via the same Fourier metric.

Significance. If the convergence theorem were correct, it would be a useful extension of Fourier-metric contraction arguments to graphon-mediated kinetic opinion dynamics with a non-trivial constrained control. The closed-form solution of the MPC problem in Lemma 3.1 and the piecewise Beta stationary characterization in Section 6 are interesting and potentially reusable. The numerical setup is thoughtful: using Barberá ideal points to initialise opinions and comparing real adjacency matrices with a calibrated Gaussian graphon is a sensible test bed. However, the advertised analytical result (exponential convergence to the Section 6 stationary state) is not established, and the central empirical claim is based on an in-sample fit. The modelling and numerical parts have merit, but the headline claims need substantial reworking.

major comments (5)
  1. [Section 7, Eqs. (40)–(41)] The Fourier estimate of the media operator is not an estimate of the model's media operator. Lemma 3.1 defines the constrained update w''_* = sgn(w''_free) Proj_{[r1,r2]}(|w''_free|), but Eqs. (40)–(41) use the unconstrained affine update w''=(λ̃ w+δ² w_d)/(λ̃+δ²), leading to the factorisation cQ1(f)(ξ)=cM(...)∠f_S(...). The projection is nonlinear and does not factorise in Fourier space; the factor (λ̃/(λ̃+δ²))^s in κ_W(t) is the contraction rate of the linear map only. Moreover, the loss term in Eq. (32) is −f̂, although the media acts only on S; the effective loss should be confined to the support of χ_S. Thus Theorem 7.2 is not proved for the constrained model defined in Section 3.
  2. [Section 7, Corollary 7.3 and Assumption 7.1(iii)] Corollary 7.3 asserts that f∞ solves the same equation on the same W because ∂_t f∞=0 and the local masses agree. But f∞ is the stationary state of the Fokker–Planck equation (15), which contains diffusion, while Assumption 7.1(iii) sets η=0, so the relevant Boltzmann equation is noiseless. The absolutely continuous piecewise Beta density (24) is not, in general, a stationary solution of the noiseless first-order Boltzmann equation, whose equilibria are typically supported on invariant atoms. This identification is asserted, not proved, and is implausible. Even if Theorem 7.2 were valid, it would only establish convergence to some common asymptotic state, not to the particular f∞ characterised in Section 6.
  3. [Section 4, Lemma 4.1, Eq. (12)] The mean-evolution calculation uses the unconstrained media update: w''−w = δ²/(λ̃+δ²)(w_d−w). For agents whose unconstrained image falls outside S, Lemma 3.1 clamps w''_* to the boundary r1 or r2, so this equality fails. Consequently the condition w_d = m_S/M_S is not sufficient for total mean conservation under the constrained control, and Assumption 7.1(ii), on which Theorem 7.2 relies, is not guaranteed. The lemma needs either a constraint-aware computation or an explicit assumption that the projection is inactive on the support of f.
  4. [Section 7, Eq. (37)] The estimate for the binary interaction term is also not justified as written. The first term on the right of Eq. (37) has coefficient |∠f_1((1−γ)ξ,x0,t)|, which is at most ρ(x0,t), not ρ_W(x0,t). The text's assertion that ρ(x,t)≤1 is false: f(w,x,t) is a density in (w,x), so the x-marginal ρ(x,t) can exceed 1. The later step replacing the coefficient by ρ_W(t)=inf_x ρ_W(x,t) in (44)–(45) is therefore unsupported. This affects the binary part of the convergence theorem independently of the media-projection issue.
  5. [Section 8.4] The central empirical claim is an in-sample calibration. The bandwidth σ_W is selected by minimising the Fourier distance d_s between the synthetic and real asymptotic distributions, and the minimised distances (4.00e−3 and 5.18e−3) are then reported as evidence of descriptive power. This is circular: the minimised distance is not an independent measure of fit. The 'common' σ*_W=0.4445 is the coincidence of two separately fitted minimisers on a logarithmic grid, and no validation is provided on held-out networks, other issues, or out-of-sample time windows. The conclusion that macroscopic dynamics depends on the interaction structure only through its mean intensity is stronger than this experiment supports.
minor comments (4)
  1. [Notation in Section 7] The parameter λ in Eqs. (34) and (40)–(41) appears to be the control penalty λ̃ of Lemma 3.1, but Section 5 uses λ for the noise-variance ratio σ²/γ. This notational collision is confusing and should be fixed throughout.
  2. [Typos] There are several typographical errors: 'model predicative control' in the Introduction, 'correspondance' in the author footnote, and 'Fullcon-dence' in the axis label of Figure 5.
  3. [Algorithm 1] Lines 9–13 apply the media update to w_free^i, the opinion after the binary interaction, rather than to the pre-interaction w_i. This sequential-update convention should be stated explicitly in the text, since the analytical equations in Sections 2–5 treat the two mechanisms as separate weak-form operators.
  4. [Section 8.3] The rescaling of Barberá ideal points to [−1,1] is described as linear, but the specific affine map or normalization is not given. This makes the real-data initial conditions less reproducible.

Circularity Check

1 steps flagged

Empirical graphon-reproduction claim is a fitted-parameter report: σ_W is selected by minimizing the very Fourier distance later quoted as evidence of descriptive power.

specific steps
  1. fitted input called prediction [Section 8.4 (Graphon surrogate), Figure 10 caption, and Conclusions]
    "To recover the real dynamics we sweep σ_W and select, for each dataset, the value σ⋆_W that minimises the Fourier distance d_s between the asymptotic empirical distribution of the synthetic graphon and that of the real network. ... terminal Fourier distance d_s = 4.00×10^−3 for gun control and d_s = 5.18×10^−3 for Obamacare, both at the recovered bandwidth σ⋆_W = 0.4445."

    The single graphon parameter σ_W is calibrated by minimizing the same Fourier distance d_s that is then reported as the measure of descriptive power ('recovered up to Fourier distances...'). The quoted d_s values are therefore training errors of a one-parameter fit on the very asymptotic distributions used to choose the bandwidth, not out-of-sample predictions. For any one-parameter family, the minimized distance is small by construction at its minimizer; the near-agreement of the two minimizers is a property of the fitting procedure, not independent confirmation that the graphon 'recovers' the real dynamics.

full rationale

The analytic parts of the paper are not circular: the Boltzmann, Fokker–Planck, and stationary-state derivations are self-contained given their stated assumptions, and the exponential convergence argument, while containing a serious correctness gap (Section 7 estimates the media operator using the unconstrained linear update w''=(λ̃w+δ²w_d)/(λ̃+δ²) from eqs. (40)–(41), rather than the constrained projection of Lemma 3.1, and Corollary 7.3 asserts without proof that the Fokker–Planck stationary state f∞ solves the noiseless Boltzmann equation (32)), is not a definitional reduction to its inputs. Those are proof/assumption failures, not instances of circularity under the criteria here. The paper's own limitation statement concedes the data comparison is restricted to the asymptotic state from a single temporal snapshot, which further weakens the empirical claim but does not itself constitute circularity. The load-bearing circularity is the central empirical claim: a one-parameter Gaussian graphon is 'recovered' by sweeping σ_W to minimize the Fourier distance to the real asymptotic distributions, and the minimized distances (4.00e-3, 5.18e-3) are then presented as evidence of descriptive power. This is a fitted input presented as a reproduction/prediction, warranting the score 6.

Axiom & Free-Parameter Ledger

3 free parameters · 6 axioms · 0 invented entities

The formal model analysis rests on standard quasi-invariant kinetic assumptions plus several choices that are either hand-set or asserted without proof; the empirical graphon-recovery claim rests on a fitted bandwidth. The list above separates model inputs, analytic assumptions, and the unverified substitutions in the convergence proof.

free parameters (3)
  • Gaussian graphon bandwidth σ_W = 0.4445 (grid search, both datasets)
    Fitted by minimising Fourier distance d_s between synthetic-graphon and real-network asymptotic distributions; the central descriptive-power claim is evaluated at this fitted value.
  • Diffusion exponent α = 2 in numerics; 1/2 in explicit steady-state formulas
    Chosen by hand; controls boundary degeneracy of noise. The explicit piecewise-Beta stationary state is derived only for α = 1/2, while all simulations use α = 2, so the analytic steady state is not directly compared with numerics.
  • Controversy set boundaries r1, r2 = r1 = 0.05, r2 = 0.6 in real-network runs; varied in synthetic scans
    Hand-chosen shape of the sphere of legitimate controversy; qualitative outcomes such as polarisation persistence versus consensus depend on it.
axioms (6)
  • domain assumption Graphon W is a symmetric Lebesgue-measurable kernel representing interaction probabilities; real networks are approximated by a binary adjacency matrix.
    Symmetry is used in Lemma 4.1 and the graphon is the interaction kernel throughout; real networks are directed, so the analytical results do not apply to them directly (stated in Section 8.3).
  • domain assumption Noise truncation in Appendix A.1 guarantees opinions stay in [-1,1].
    The boundedness of post-interaction opinions is enforced by restricting the noise support; without this, the Fokker–Planck boundary structure is not justified.
  • domain assumption Quasi-invariant scaling with σ²/γ fixed and formal Taylor expansion yields the Fokker–Planck equation.
    The derivation is formal, following [47,52]; no uniform error estimates are given.
  • ad hoc to paper In Section 7, Assumption 7.1(iii) sets η = 0, and the FP stationary state f∞ is assumed stationary for the resulting noiseless Boltzmann equation.
    This disconnects the convergence theorem from the stochastic model for which f∞ was derived; the stationarity of f∞ for the noiseless equation is asserted in Corollary 7.3 without proof.
  • ad hoc to paper The media control in the Fourier estimate of Q+1 is treated as the unconstrained linear update, with the projection in Lemma 3.1 dropped.
    The proof of Theorem 7.2 uses w'' = (λ̃w + δ²w_d)/(λ̃ + δ²), ignoring the constraint r1 ≤ |w''| ≤ r2 that defines the actual control.
  • domain assumption Barberá ideal points, linearly rescaled, measure issue-specific initial opinions.
    Used to initialise real-network opinions; the paper acknowledges this is a general ideological position rather than a stance on the specific issue.

pith-pipeline@v1.3.0-alltime-deepseek · 27036 in / 22506 out tokens · 225651 ms · 2026-08-02T08:55:22.325258+00:00 · methodology

0 comments
read the original abstract

We propose a kinetic model of opinion dynamics under selective media influence on both graphon-based and real-world networks. The media action, inspired by Hallin's theory of spheres, is incorporated through a model predictive control strategy designed to steer agents' opinions toward a desired target opinion. For the resulting Boltzmann-type description, we analyse the evolution of the moments and, in the quasi-invariant interaction limit, derive a Fokker--Planck-type equation together with a characterisation of its stationary states. We also prove exponential convergence to equilibrium in the Fourier metric. Numerical experiments are performed on networks generated by a Gaussian graphon and on real-world, single-issue Twitter networks, allowing us to investigate the role of control and interaction parameters, as well as the impact of the subset of agents subject to media influence. Using real-world social networks data to initialise opinions and infer the interaction structure, we then compare the dynamics obtained on the original networks with those produced by Gaussian graphons fitted to their adjacency matrices, thereby assessing the descriptive power of the graphon approach for real-world opinion dynamics.

Figures

Figures reproduced from arXiv: 2607.02821 by Alessandro Licciardi, Bertram D\"uring, Martina Fraia.

Figure 1
Figure 1. Figure 1: also isolates the joint role of the compromise function P and of the controversy set S. The three panels share all parameters and differ only in this pair. In panel (a), bounded-confidence compromise combined with a strictly localised S leaves the population polarised: agents whose opinions fall inside S are drawn toward the target wd, while the extreme agents near w = ±1 neither interact with the moderate… view at source ↗
Figure 1
Figure 1. Figure 1: Schematic view of Hallin’s spheres of consensus, legitimate controversy and deviance. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: reports the asymptotic distributions for three nested choices of S, from the full interval to a narrow band. Enlarging S increases the fraction of the population reached by the control and produces a stronger concentration of opinions around the target wd: the peak at wd is highest for S = [−1, 1] and progressively lower as S shrinks, while the mass retained away from the target correspondingly grows. For … view at source ↗
Figure 2
Figure 2. Figure 2: Opinion interval [−1, 1], divided into spheres of consensus, deviance and legitimate controversy. In particular, depending on the strength of the control of the media and the size and shape of S, this effect can counteract against the symmetry of classical compromise models, similar as seen in [30] where an external effect is introduced to break the consensus in the system. The control z(t) in (3) is chose… view at source ↗
Figure 3
Figure 3. Figure 3: Asymptotic opinion distributions comparing complete interactions ( [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figure 3
Figure 3. Figure 3: also isolates the joint role of the compromise function P and of the controversy set S. The three panels share all parameters and differ only in this pair. In panel (a), bounded-confidence compromise combined with a strictly localised S leaves the population polarised: agents whose opinions fall inside S are drawn toward the target wd, while the extreme agents near w = ±1 neither interact with the moderate… view at source ↗
Figure 4
Figure 4. Figure 4: Initial opinion distributions obtained from the normalised Barber´a ideal points for the [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗
Figure 4
Figure 4. Figure 4: Opinion distributions f(w, t) at t = 1000 for three nested choices of the set S = [−r2, −r1] ∪ [r1, r2], from the full interval S = [−1, 1] to S = [−0.8, −0.2] ∪ [0.2, 0.8] and S = [−0.5, −0.3] ∪ [0.3, 0.5], under the bounded-confidence opinion dynamics with confidence radius r = 0.2. Parameters: γ = 0.4, σ = 0.05, σW = 0.2, λ˜ = 0.05, wd = 0.4, δ = 0.5, α = 2. The peak at wd is highest for S = [−1, 1] and… view at source ↗
Figure 5
Figure 5. Figure 5: Trajectories of the agents’ opinions on the two real networks, (a) gun control and (b) [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
Figure 5
Figure 5. Figure 5: Opinion distributions f(w, t) at t = 5 and t = 1000 for media strengths δ ∈ {0.1, 0.2, 0.3} under complete interactions (P ≡ 1). Remaining parameters: γ = 0.4, σ = 0.05, σW = 0.2, λ˜ = 0.05, wd = 0.4, S = [−0.5, −0.3] ∪ [0.3, 0.5], α = 2. A stronger media yields a faster and sharper concentration of the whole population at wd. -1 -0.5 0 0.5 1 w 0 10 20 30 40 50 60 70 f(w;t) / = 0:3; r = 0:2 t = 5 t = 1000 … view at source ↗
Figure 6
Figure 6. Figure 6: Evolution of the opinion density f(w, t) on a log-spaced time axis, comparing the calibrated synthetic graphon (σ ⋆ W = 0.445) with the real network: gun control (top, a synthetic and b real) and Obamacare (bottom, c synthetic and d real), all initialised from the Barber´a scores under identical opinion-space and media parameters. The bright vertical line marks the concentration of the mediated agents at t… view at source ↗
Figure 6
Figure 6. Figure 6: Opinion distributions f(w, t) at t = 5 and t = 1000 for media strengths δ ∈ {0.3, 0.6, 1} under bounded-confidence interactions with confidence radius r = 0.2. Remaining parameters as in [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Asymptotic opinion distributions of the calibrated synthetic graphon (orange) overlaid [PITH_FULL_IMAGE:figures/full_fig_p026_7.png] view at source ↗
Figure 7
Figure 7. Figure 7: Initial opinion distributions obtained from the normalised Barber´a ideal points for the [PITH_FULL_IMAGE:figures/full_fig_p025_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Trajectories of the agents’ opinions on the two real networks, (a) gun control and (b) [PITH_FULL_IMAGE:figures/full_fig_p026_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Evolution of the opinion density f(w, t) on a log-spaced time axis, comparing the calibrated synthetic graphon (σ ⋆ W = 0.445) with the real network: gun control (top, a synthetic and b real) and Obamacare (bottom, c synthetic and d real), all initialised from the Barber´a scores under identical opinion-space and media parameters. The bright vertical line marks the concentration of the mediated agents at t… view at source ↗
Figure 10
Figure 10. Figure 10: Asymptotic opinion distributions of the calibrated synthetic graphon (orange) overlaid [PITH_FULL_IMAGE:figures/full_fig_p028_10.png] view at source ↗

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

55 extracted references · 1 linked inside Pith

  1. [1]

    Albi et al

    G. Albi et al. ‘A data-driven kinetic model for opinion dynamics with social network con- tacts’.Eur. J. Appl. Math.36.2 (2025), pp. 264–290

    2025

  2. [2]

    Albi et al

    G. Albi et al. ‘Boltzmann games in heterogeneous consensus dynamics’.J. Stat. Phys.175.1 (2019), pp. 97–125

    2019

  3. [3]

    Albi et al

    G. Albi et al. ‘Boltzmann-type control of opinion consensus through leaders’.Phil. Trans. R. Soc. A372.2028 (2014)

    2028

  4. [4]

    Albi et al

    G. Albi et al. ‘On the optimal control of opinion dynamics on evolving networks’.System Modeling and Optimization. Ed. by L. Bociu et al. Vol. 494. IFIP Advances in Information and Communication Technology. Springer, 2016, pp. 1–12

    2016

  5. [5]

    Albi et al

    G. Albi et al. ‘Opinion dynamics over complex networks: kinetic modelling and numerical methods’.Kinet. Relat. Mod.10.1 (2017), pp. 1–32

    2017

  6. [6]

    Aletti and G

    G. Aletti and G. Naldi. ‘Opinion dynamics on graphon: The piecewise constant case’.Applied Math. Letters133 (2022), p. 108227

    2022

  7. [7]

    Aylaj et al

    B. Aylaj et al. ‘A unified multiscale vision of behavioral crowds’.Math. Models Methods Appl. Sci.30.1 (2020), pp. 1–22

    2020

  8. [8]

    Barber´ a

    P. Barber´ a. ‘Birds of the Same Feather Tweet Together: Bayesian Ideal Point Estimation Using Twitter Data’.Political Analysis23.1 (2015), pp. 76–91

    2015

  9. [9]

    Bayraktar et al

    E. Bayraktar et al. ‘Graphon mean field systems’.Ann. Appl. Probab.33.5 (2023), pp. 3587– 3619

    2023

  10. [10]

    Bellomo et al

    N. Bellomo et al. ‘From a systems theory of sociology to modeling the onset and evolution of criminality’.Netw. Heterog. Media10.3 (2015), pp. 421–441

    2015

  11. [11]

    Bellomo et al

    N. Bellomo et al. ‘What is life? A perspective of the mathematical kinetic theory of active particles’.Math. Models Methods Appl. Sci.31.9 (2021), pp. 1821–1866

    2021

  12. [12]

    Bonnet et al

    B. Bonnet et al. ‘Consensus formation in first-order graphon models with time-varying to- pologies’.Math. Models Methods Appl. Sci.32.11 (2022), pp. 2121–2188

    2022

  13. [13]

    Borgs et al

    C. Borgs et al. ‘AnL p theory of sparse graph convergence I: Limits, sparse random graph models, and power law distributions’.Trans. Am. Math. Soc.372.5 (2019), pp. 3019–3062

    2019

  14. [14]

    Boscheri et al

    W. Boscheri et al. ‘Modeling and simulating the spatial spread of an epidemic through multiscale kinetic transport equations’.Math. Models Methods Appl. Sci.31.6 (2021), pp. 1059– 1097

    2021

  15. [15]

    Boudin and F

    L. Boudin and F. Salvarani. ‘Opinion dynamics: kinetic modelling with mass media, applic- ation to the Scottish independence referendum’.Physica A444 (2016), pp. 448–457

    2016

  16. [16]

    M. Burger. ‘Network structured kinetic models of social interactions’.Vietnam J. Math. (2021)

    2021

  17. [17]

    Burger et al

    M. Burger et al. ‘Sorting phenomena in a mathematical model for two mutually attract- ing/repelling species’.SIAM J. Math. Anal.50.3 (2018), pp. 3210–3250

    2018

  18. [18]

    Caron et al

    F. Caron et al. ‘On sparsity, power-law, and clustering properties of graphex processes’.Adv. Appl. Probab.55.4 (2023), pp. 1211–1253

    2023

  19. [19]

    J. A. Carrillo, G. Toscani et al. ‘Contractive probability metrics and asymptotic behavior of dissipative kinetic equations’.Riv. Mat. Univ. Parma6.7 (2007), pp. 75–198

    2007

  20. [20]

    Chazelle et al

    B. Chazelle et al. ‘Well-posedness of the limiting equation of a noisy consensus model in opinion dynamics’.J. Differ. Equations263.1 (2017), pp. 365–397

    2017

  21. [21]

    Coculescu et al

    D. Coculescu et al. ‘Opinion dynamics in communities with major influencers and implicit social influence via mean-field approximation’.Math. Financ. Econ.18.2 (2024), pp. 333– 377. 30

    2024

  22. [22]

    F. Coppini. ‘A Note on Fokker-Planck Equations and Graphons’.J. Stat. Phys.187.2 (2022), p. 15

    2022

  23. [23]

    Cristiani and A

    E. Cristiani and A. Tosin. ‘Reducing complexity of multiagent systems with symmetry break- ing: an application to opinion dynamics with polls’.Multiscale Model. Simul.16.1 (2018), pp. 528–549

    2018

  24. [24]

    Degond et al

    P. Degond et al. ‘Continuum dynamics of the intention field under weakly cohesive social interaction’.Math. Models Methods Appl. Sci.27.1 (2017), pp. 159–182

    2017

  25. [25]

    D¨ uring and M

    B. D¨ uring and M. Wolfram. ‘Opinion dynamics: inhomogeneous Boltzmann-type equations modelling opinion leadership and political segregation’.Proc. R. Soc. A.471 (2015), p. 20150345

    2015

  26. [26]

    D¨ uring and O

    B. D¨ uring and O. Wright. ‘On a kinetic opinion formation model for pre-election polling’. Phil. Trans. R. Soc. A.380.2224 (2022), p. 20210154

    2022

  27. [27]

    D¨ uring and O

    B. D¨ uring and O. Wright. ‘PDE-constrained optimal control of a leader–follower opinion formation model’.Boll. Unione Mat. Ital.(2025), pp. 1–25

    2025

  28. [28]

    D¨ uring and O

    B. D¨ uring and O. Wright. ‘Voter demographics and socio-economic factors in kinetic models for opinion formation’.EMS Ser. Congr. Rep.23 (2026). Ed. by G. Albi et al., pp. 277–297

    2026

  29. [29]

    D¨ uring et al

    B. D¨ uring et al. ‘Boltzmann and Fokker-Planck equations modelling opinion formation in the presence of strong leaders’.Proc. R. Soc. A.465 (2009), pp. 3687–3708

    2009

  30. [30]

    D¨ uring et al

    B. D¨ uring et al. ‘Breaking Consensus in Kinetic Opinion Formation Models on Graphons’. Kinet. Relat. Mod.15.5 (2022), pp. 815–843

    2022

  31. [31]

    Furioli et al

    G. Furioli et al. ‘Fokker-Planck equations in the modeling of socio-economic phenomena’. Math. Models Methods Appl. Sci.27.1 (2017), pp. 115–158

    2017

  32. [32]

    Gabetta et al

    G. Gabetta et al. ‘Metrics for probability distributions and the trend to equilibrium for solutions of the Boltzmann equation’.J. Stat. Phys.81.5 (1995), pp. 901–934

    1995

  33. [33]

    Garimella et al

    K. Garimella et al. ‘Political discourse on social media: Echo chambers, gatekeepers, and the price of bipartisanship’.Proceedings of the 2018 world wide web conference. 2018, pp. 913– 922

    2018

  34. [34]

    Garimella et al

    K. Garimella et al. ‘Quantifying Controversy on Social Media’.ACM Trans. Soc. Comput. 1.1 (2018). Dataset:https://rutgers.box.com/s/9lavww3g0uyajkcvxwu66sxtauncq589, pp. 1–27

    2018

  35. [35]

    Garnier et al

    J. Garnier et al. ‘Consensus convergence with stochastic effects’.Vietnam J. Math.45 (2017), pp. 51–75

    2017

  36. [36]

    D. J. Glasscock. ‘What is a graphon?’Notices Amer. Math. Soc.66.5 (2019), pp. 733–738

    2019

  37. [37]

    B. D. Goddard et al. ‘Noisy bounded confidence models for opinion dynamics: the effect of boundary conditions on phase transitions’.IMA J. Appl. Math.87.1 (2021), pp. 80–110

    2021

  38. [38]

    D. C. Hallin.The Uncensored War: The Media and Vietnam. Oxford University Press, 1986

    1986

  39. [39]

    Hegselmann and U

    R. Hegselmann and U. Krause. ‘Opinion Dynamics and Bounded Confidence Models, Ana- lysis, and Simulation’.J. Artif. Soc. Soc. Simul.5.3 (2002)

    2002

  40. [40]

    Helfmann et al

    L. Helfmann et al. ‘Modelling opinion dynamics under the impact of influencer and media strategies’.Scientific Reports13.1 (2023), p. 19375

    2023

  41. [41]

    Jabin and S

    P.-E. Jabin and S. Motsch. ‘Clustering and asymptotic behavior in opinion formation’.J. Differ. Equations257.11 (2014), pp. 4165–4187

    2014

  42. [42]

    Lov´ asz.Large networks and graph limits

    L. Lov´ asz.Large networks and graph limits. Vol. 60. American Mathematical Society, 2012

    2012

  43. [43]

    Loy et al

    N. Loy et al. ‘Opinion polarization in social networks’.Phil. Trans. R. Soc. A380.2224 (2022)

    2022

  44. [44]

    Motsch and E

    S. Motsch and E. Tadmor. ‘Heterophilious dynamics enhances consensus’.SIAM Rev.56.4 (2014), pp. 577–621. 31

    2014

  45. [45]

    Nugent et al

    A. Nugent et al. ‘Opinion dynamics with continuous age structure’.Eur. J. Appl. Math. (2025), pp. 1–36

    2025

  46. [46]

    Nurisso et al

    M. Nurisso et al. ‘Network-based kinetic models: Emergence of a statistical description of the graph topology’.Eur. J. Appl. Math.(2024), pp. 1–22

    2024

  47. [47]

    Pareschi and G

    L. Pareschi and G. Toscani.Interacting Multiagent Systems: Kinetic equations and Monte Carlo methods. Oxford University Press, 2013

    2013

  48. [48]

    Pareschi et al

    L. Pareschi et al. ‘Hydrodynamic models of preference formation in multi-agent societies’. J. Nonlinear Sci.29.6 (2019), pp. 2761–2796

    2019

  49. [49]

    Pedraza et al

    L. Pedraza et al. ‘Measure-valued opinion dynamics’.Math. Models Methods Appl. Sci.30.2 (2020), pp. 225–260

    2020

  50. [50]

    P´ erez-Llanos et al

    M. P´ erez-Llanos et al. ‘Opinion formation models with heterogeneous persuasion and zealotry’. SIAM J. Math. Anal.50.5 (2018), pp. 4812–4837

    2018

  51. [51]

    Taricco and A

    G. Taricco and A. Tosin. ‘Homogeneous Boltzmann-type equations on dense graphs’.arXiv 2606.17638 (2026)

    Pith/arXiv arXiv 2026

  52. [52]

    G. Toscani. ‘Kinetic models of opinion formation’.Comm. Math. Sci.4.3 (2006), pp. 481– 496

    2006

  53. [53]

    Toscani et al

    G. Toscani et al. ‘Opinion modeling on social media and marketing aspects’.Phys. Rev. E 98.2 (2018), p. 022315

    2018

  54. [54]

    Van Der Hofstad.Random Graphs and Complex Networks

    R. Van Der Hofstad.Random Graphs and Complex Networks. Vol. 43. Cambridge University Press, 2016

    2016

  55. [55]

    Wang et al

    C. Wang et al. ‘Noisy Hegselmann-Krause systems: Phase transition and the 2R-conjecture’. J. Stat. Phys.166.5 (2017), pp. 1209–1225. 32

    2017