REVIEW 2 major objections 6 minor 1 cited by
Majority Dynamics on Assortative Sparse Stochastic Block Models
T0 review · 2 major / 6 minor · reviewed 2026-07-31 · grok-4.5
Pith's one-line read On sparse assortative block graphs, a weighted camp advantage—not raw majority—sets the time to opinion unanimity.
desk verdict Solid sparse-SBM majority-dynamics paper: weighted advantage drives sharp constant/subpoly/poly thresholds with matching I_0 bounds. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The weighted advantage Δ̃_t=b|B_t|−a|R_t| together with sharp one-vertex flip-probability estimates for sparse binomial differences. These estimates convert the local majority vote into Gaussian, moderate-deviation, or large-deviation asymptotics that feed one-step drift and concentration bounds, which in turn drive the constant-time, subpolynomial, and block-amplification arguments.
What would settle it
Simulate the process for large N with initial weighted advantage just below −N/√log N and check whether the median time to blue unanimity stays bounded by three steps or jumps into a visibly growing sub-polynomial regime; alternatively, fix a camp ratio bounded away below a/b and verify that the empirical exponent of the extinction time tracks I_0 from below.
Extended reading notes
Core claim
When majority dynamics is run on a resampled sparse assortative stochastic block model with a>b>1, the weighted advantage Δ̃_t=b|B_t|−a|R_t| (rather than the unweighted size difference alone) governs the high-probability time to blue unanimity. Three nested regimes appear: Δ̃_0≳−N/√ log N yields unanimity in three updates; a o(N) weighted deficit yields N^{o(1)} updates; a linear weighted deficit with Δ_0≫√(N/log N) yields N^{I_0+o(1)} updates, where I_0 is the large-deviation rate (ReLU(√(a|R_0|/N)−√(b|B_0|/N)))^2, and the same exponent is necessary when the initial camp ratio stays a fixed distance below a/b.
Load-bearing premise
The edge density between opposite camps is forced above the connectivity threshold for every possible opinion split that can appear, so the graph never becomes disconnected.
Editorial extensions
If this is right
- Near the weighted threshold a constant-size or slowly growing deficit is erased in three updates, giving an explicit sparse-block analogue of the “power of few.”
- The polynomial exponent I_0 coincides with the information-theoretic mismatch rate of community detection in the same SBM, linking consensus time to exact-recovery thresholds.
- Away from the weighted threshold the same exponent is both necessary and sufficient, so the time to unanimity is tightly characterized by a single large-deviation function of the initial sizes.
- The one-vertex flip-probability estimates for sparse binomial differences are stated in a form reusable for other sparse majority or bootstrap processes.
Reading between the lines
- Because the connectivity assumption is stronger than the partition-dependent threshold, temporary disconnection might allow minority pockets to persist; checking whether unanimity still occurs when b中1 would test the robustness of the weighted-advantage picture.
- The same weighted advantage should control the disassortative regime once the sign of the feedback is reversed, potentially producing oscillation rather than absorption.
- Spatial or degree-corrected versions of the model would replace the global weighted advantage by a local or degree-weighted analogue, offering a concrete route to more realistic opinion networks.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies synchronous majority dynamics on a binary stochastic block model that is resampled at each step from the current opinion partition, in the sparse assortative regime α=a log N/N, β=b log N/N with a>b>1. The main conceptual claim is that the weighted advantage Δ̃_t = b|B_t| − a|R_t|, not the raw majority Δ_t, governs convergence to unanimity. Three regimes are established: (i) constant-time (one, two, or three updates) blue unanimity above sharp weighted-advantage thresholds at the N/√log N scale (Theorem 1.6); (ii) subpolynomial time N^{o(1)} when the weighted disadvantage is o(N) (Theorem 1.7); (iii) polynomial time N^{I_0+ε} when the weighted disadvantage is linear but Δ_0≫√(N/log N) (Theorem 1.8), with a matching lower bound N^{I_0−ε} when the initial camp ratio stays below a/b−κ (Theorem 1.9). The exponent I_0 = (ReLU(√(a|R_0|/N) − √(b|B_0|/N)))² is derived from the large-deviation rate function of a sparse binomial difference. The proofs combine one-step flip-ratio comparisons, Gaussian/moderate/large-deviation tail estimates for binomial differences, read-2 concentration, and a dyadic block-amplification argument built on a stopped supermartingale.
Significance. If correct (and the proofs check out on my reading), this is a solid contribution to the "power of few" literature: the first sharp analysis of majority dynamics at the connectivity scale on opinion-correlated random graphs, with matching polynomial-time upper and lower bounds. Strengths that weigh in the assessment: the exponent I_0 is derived, not fitted — it comes from the large-deviation rate of a binomial difference (Lemma A.3 via Gärtner–Ellis); the upper bound (Theorem 1.8) and lower bound (Theorem 1.9) pin down the same exponent up to the stated slack; the moderate-deviation estimate with Mills-ratio prefactor (Lemma A.5) is of independent technical interest; the read-k/Finner MGF route to the block-amplification supermartingale is clean; and numerical experiments corroborate both the constant-time transitions and the predicted exponent. The echo of the exact-recovery rate (√a−√b)² in Remark 1.10 raises the paper's interest beyond dynamics specialists.
major comments (2)
- [§1, Assumption 1.3; §3, Lemma 3.1, Eq. (1.11)] Assumption 1.3 (b>1) is presented as a convenience guaranteeing uniform connectivity, and §7 acknowledges it is stronger than the state-dependent threshold of Lemma H.1. But the main text never says where it is actually used. Inspection suggests the genuine entry point is Lemma 3.1: r*>0 in (1.11) is equivalent to (a+b)(b-1)>0, and the extinction step needs I(B^B_t,A^R_t)>1, while the isolated-vertex obstruction (ties preserve opinion) is only motivational. Please state explicitly which lemmas fail when b<=1; this is load-bearing for the advertised 'critical sparse regime' framing, since b>1 is precisely the connectivity scale.
- [§5, Theorem 1.8; Figure 1] The polynomial-time upper bound requires Δ_0≫√(N/log N), and the failure probability exp(-cΔ_0²log N/N) is Θ(1) at the boundary scale, so the band ẼΔ_0=-Θ(N) with 0<Δ_0≲√(N/log N) is left 'unexplored' in Figure 1 without even a conjecture. Is the obstruction purely the degeneration of exp(-cΔ_sθ_s) in Lemmas 5.5-5.6, or do you expect a different exponent there? Relatedly, the abstract claims unanimity within N^{I_0+o(1)} but Theorem 1.8 delivers N^{I_0+ε} for every fixed ε; please align the two statements (Remark 5.8 already explains the slack).
minor comments (6)
- [§2, Lemma 2.4] u_t = -ẼΔ_t√(log N)/N is positive only when ẼΔ_t<0; the hypothesis ℓ≤u_t implicitly assumes this. State it explicitly.
- [§1.1, Remark 1.10] The coincidence I=½(√a-√b)² with the exact-recovery threshold [1,17] is striking. One sentence on whether this is structural (e.g., both governed by the same binomial-difference rate) or coincidental would be valuable.
- [§1.2, Figures 2-3] Captions should define H_exp = ẼΔ_0√(log N)/N inline, and note that N=10⁴ is modest given log N≈9; the visible finite-N deviations near transition points deserve a sentence.
- [§1, Introduction] A brief comparison to the sparse Erdős-Rényi results [20,16] would orient the reader: what do the present techniques give there, and why does the SBM require the full block-amplification machinery rather than the simpler nesting of §3?
- [Appendix H, Lemma H.1] Since Lemma H.1 is used only as preparation for a future extension, say so when Assumption 1.3 is introduced, not just in §7.
- [Abstract; headings] Typographical: 'the detailed estimates' in the abstract should be 'detailed estimates'; several headings show stray spaces ('F uture Directions', 'T echnical Lemmas', 'W eighted Self-Opinions') in the compiled PDF.
Circularity Check
No circularity: theorems are proved from sparse binomial LD/MGF estimates; I_0 is a derived rate, not a fitted or self-defined input.
full rationale
This is a self-contained probability paper. The weighted advantage Δ̃_t = b|B_t| − a|R_t| is introduced as a bookkeeping device that matches the within/across edge intensities α = a log N/N and β = b log N/N; one-step drifts and flip probabilities are then bounded by Berry–Esseen, Chernoff, and moderate-deviation analysis of Bin(s,β)−Bin(r,α) (Appendix A), and the polynomial exponent I_0 is exactly the large-deviation rate I(a|R_0|/N, b|B_0|/N) of the red-to-blue flip. Upper and lower bounds (Theorems 1.8–1.9) are proved by block amplification and flip counting under that rate, not by fitting parameters to data or by importing an author-unique theorem that forces the conclusion. Remark 1.10’s coincidence with the community-detection exact-recovery exponent is an a-posteriori observation, not an input. Citations (Abbe–Bandeira–Hall, Mossel–Tamuz, etc.) are standard external background. No prediction reduces to its inputs by construction.
Assumptions & free parameters
assumptions (5)
- domain assumption Sparse assortative SBM: α=a log N/N, eta=b log N/N with fixed a>b>0 (Assumption 1.2).
- domain assumption Connectivity: b>1 so every resampled graph is connected whp uniformly over partitions (Assumption 1.3).
- domain assumption Synchronous majority update with tie-breaking by retaining current opinion (rule (1.3)).
- standard math Berry–Esseen, Chernoff, Gärtner–Ellis left-tail, and read-k Chernoff bounds (Appendices A–B, Lemmas I.3–I.4, B.2).
- domain assumption Initial blue unweighted advantage Δ_0>0 (standing orientation; Remark 1.4).
invented entities (2)
-
Weighted advantage Δ̃_t = b|B_t| − a|R_t|
independent evidence
-
Rate function I(x,y)=(ReLU(√x−√y))^2 and I_t = I(a|R_t|/N, b|B_t|/N)
independent evidence
Cite this review
Pith. "Pith review of Majority Dynamics on Assortative Sparse Stochastic Block Models." pith.science (2026). https://pith.science/paper/2JGIWXPU
@misc{pith2026260724652,
author = {Pith},
title = {Pith review of: Majority Dynamics on Assortative Sparse Stochastic Block Models},
year = {2026},
howpublished = {\url{https://pith.science/paper/2JGIWXPU}},
note = {Machine review of arXiv:2607.24652}
}
abstract
Majority dynamics is a two-opinion process in which each vertex repeatedly updates to the majority opinion among its neighbors. We study this process on a resampled sparse binary stochastic block model in the assortative regime. At each time step, a graph is sampled from the current opinion partition: vertices with the same opinion are joined with probability $\alpha=a\log N/N$, while vertices with differing opinions are joined with probability $\beta=b\log N/N$, where $a>b>1$. Let $B_t$ and $R_t$ denote the blue and red camps at time $t$. We show that the weighted advantage $\widetilde{\Delta}_t =b|B_t|-a|R_t|$, rather than the unweighted advantage $\Delta_t=|B_t|-|R_t|$ alone, governs the pace to unanimity. Our results, which hold with high probability as \(N\to\infty\), identify three regimes for blue unanimity under the initial blue advantage, i.e., $\Delta_0>0$: constant time, subpolynomial time, and polynomial time. First, when $\widetilde{\Delta}_0 \gtrsim -N/\sqrt{\log N}$, blue unanimity occurs within three updates. Second, when $\widetilde{\Delta}_0 < 0$ and $|\widetilde{\Delta}_0| = o(N)$, blue unanimity occurs within $N^{o(1)}$ updates. Furthermore, when $\widetilde{\Delta}_0 < 0$, $|\widetilde{\Delta}_0| = O(N)$, and $\Delta_0\gg\sqrt{N/\log N}$, blue unanimity still occurs within $N^{I_0+o(1)}$ updates, where \[ I_0= \left(\mathbf{ReLU}\Big(\sqrt{a\frac{|R_0|}{N}}-\sqrt{b\frac{|B_0|}{N}}\Big)\right)^2, \] and $\mathbf{ReLU}(x)=\max\{x,0\}$. Conversely, away from the weighted threshold, when $|B_0|/|R_0|\le a/b-\kappa$ and $\Delta_0>0$, $N^{I_0 - o(1)}$ updates are necessary for blue unanimity. Our analysis relies on detailed estimates for one-vertex flip probabilities in sparse binomial differences, which could be of independent interest.
Figures
Forward citations
Cited by 1 Pith paper
-
Majority Dynamics on Resampled Sparse Erd\H{o}s--R\'enyi Graphs: Gaussian Winner Selection and Pace to Unanimity
For majority dynamics on resampled sparse Erdős-Rényi graphs, the first update performs a Gaussian coin flip that decides the winner, and unanimity follows within (1+o(1)) log N / log log N rounds.
Reference graph
Works this paper leans on
-
[1]
Exact recovery in the stochastic block model
Emmanuel Abbe, Afonso S Bandeira, and Georgina Hall. “Exact recovery in the stochastic block model”. In:IEEE Transactions on Information Theory62.1 (2016), pp. 471–487.doi: 10.1109/TIT.2015.2490670(cit. on p. 7)
arXiv 2016
-
[2]
Anℓ p theory of PCA and spectral clustering
Emmanuel Abbe, Jianqing Fan, and Kaizheng Wang. “Anℓ p theory of PCA and spectral clustering”. In:The Annals of Statistics50.4 (2022), pp. 2359–2385.doi:https://doi.org/ 10.1214/22-AOS2196(cit. on p. 54)
-
[3]
Entrywise eigenvector analysis of random matrices with low expected rank
Emmanuel Abbe, Jianqing Fan, Kaizheng Wang, and Yiqiao Zhong. “Entrywise eigenvector analysis of random matrices with low expected rank”. In:The Annals of Statistics48.3 (2020), pp. 1452–1474.doi:https://doi.org/10.1214/19-AOS1854(cit. on p. 7)
-
[4]
Conver- gence, unanimity and disagreement in majority dynamics on unimodular graphs and random graphs
Itai Benjamini, Siu-On Chan, Ryan O’Donnell, Omer Tamuz, and Li-Yang Tan. “Conver- gence, unanimity and disagreement in majority dynamics on unimodular graphs and random graphs”. In:Stochastic Processes and their Applications126.9 (2016), pp. 2719–2733 (cit. on p. 3). 54
2016
-
[5]
Central limit theorem for majority dynamics: Bribing three voters suffices
Ross Berkowitz and Pat Devlin. “Central limit theorem for majority dynamics: Bribing three voters suffices”. In:arXiv preprint arXiv:2010.08172(2020) (cit. on p. 3)
arXiv 2010
-
[6]
Majority dynam- ics on sparse random graphs
Debsoumya Chakraborti, Jeong Han Kim, Joonkyung Lee, and Tuan Tran. “Majority dynam- ics on sparse random graphs”. In:Random Structures & Algorithms63.1 (2023), pp. 171–191 (cit. on p. 3)
2023
-
[7]
Spectral analysis of random graphs with skewed degree distributions
A. Dasgupta, J.E. Hopcroft, and F. McSherry. “Spectral analysis of random graphs with skewed degree distributions”. In:45th Annual IEEE Symposium on Foundations of Computer Science. 2004, pp. 602–610.doi:10.1109/FOCS.2004.61(cit. on p. 22)
-
[8]
A moment inequality with an application to the central limit theorem
Carl-Gustav Esseen. “A moment inequality with an application to the central limit theorem”. In:Scandinavian Actuarial Journal1956.2 (1956), pp. 160–170.doi:10.1080/03461238. 1956.10414946.url:https://doi.org/10.1080/03461238.1956.10414946(cit. on p. 53)
arXiv 1956
Show all 26 references
-
[9]
A Generalization of H¨ older’s Inequality and Some Probability Inequalities
Helmut Finner. “A Generalization of H¨ older’s Inequality and Some Probability Inequalities”. In:The Annals of Probability20.4 (1992), pp. 1893–1901.doi:10.1214/aop/1176989534 (cit. on p. 54)
1992
-
[10]
Resolution of a conjecture on majority dynamics: Rapid stabilization in dense random graphs
Nikolaos Fountoulakis, Mihyun Kang, and Tam´ as Makai. “Resolution of a conjecture on majority dynamics: Rapid stabilization in dense random graphs”. In:Random Structures & Algorithms57.4 (2020), pp. 1134–1156 (cit. on p. 3)
2020
-
[11]
Community detection in degree-corrected block models
Chao Gao, Zongming Ma, Anderson Y. Zhang, and Harrison H. Zhou. “Community detection in degree-corrected block models”. In:The Annals of Statistics46.5 (2018), pp. 2153 –2185. doi:10.1214/17-AOS1615.url:https://doi.org/10.1214/17-AOS1615(cit. on p. 22)
2018 doi
-
[12]
Exact Recovery in the Geometric SBM
Julia Gaudio and Andrew Jin. “Exact Recovery in the Geometric SBM”. In: (2025). arXiv: 2512.22773 [math.PR].url:https://arxiv.org/abs/2512.22773(cit. on p. 22)
2025
-
[13]
Exact community recovery in the geometric sbm
Julia Gaudio, Xiaochun Niu, and Ermin Wei. “Exact community recovery in the geometric sbm”. In:Proceedings of the 2024 Annual ACM-SIAM Symposium on Discrete Algorithms (SODA). SIAM. 2024, pp. 2158–2184 (cit. on p. 22)
2024
-
[14]
A Tail Bound for Read-kFamilies of Functions
Dmitry Gavinsky, Shachar Lovett, Michael E. Saks, and Srikanth Srinivasan. “A Tail Bound for Read-kFamilies of Functions”. In:Random Structures & Algorithms47.1 (2015), pp. 99 –108.doi:10.1002/rsa.20532.url:https://doi.org/10.1002/rsa.20532(cit. on p. 33)
2015 doi
-
[15]
A new bound in Majority Dynamics on Random Graphs
Sean Jaffe. “A new bound in Majority Dynamics on Random Graphs”. In:arXiv preprint arXiv:2503.14401(2025) (cit. on p. 3)
2025 arXiv
-
[16]
A new density limit for unanimity in majority dynamics on random graphs
Jeong Han Kim and BaoLinh Tran. “A new density limit for unanimity in majority dynamics on random graphs”. In:arXiv preprint arXiv:2503.07447(2025) (cit. on p. 3)
2025 arXiv
-
[17]
Consistency thresholds for the planted bisec- tion model
Elchanan Mossel, Joe Neeman, and Allan Sly. “Consistency thresholds for the planted bisec- tion model”. In:Electronic Journal of Probability21.none (2016), pp. 1 –24.doi:10.1214/16- EJP4185.url:https://doi.org/10.1214/16-EJP4185(cit. on p. 7)
2016 doi
-
[18]
Opinion exchange dynamics
Elchanan Mossel and Omer Tamuz. “Opinion exchange dynamics”. In:Probability Surveys 14 (2017), pp. 155–204 (cit. on p. 3)
2017
-
[19]
Majority dynamics: The power of one
Ashwin Sah and Mehtaab Sawhney. “Majority dynamics: The power of one”. In:Israel Journal of Mathematics(2024), pp. 1–49 (cit. on p. 3)
2024
-
[20]
The “Power of Few
BaoLinh Tran and Van Vu. “The “Power of Few” Phenomenon: The Sparse Case”. In:Random Structures & Algorithms66.1 (2025), e21260 (cit. on p. 3). 55
2025
-
[21]
Reaching a Consensus on Random Networks: The Power of Few
Linh Tran and Van Vu. “Reaching a Consensus on Random Networks: The Power of Few”. In:Theory of Computing19.6 (2023), pp. 1–21.doi:10.4086/toc.2023.v019a006.url: https://theoryofcomputing.org/articles/v019a006(cit. on p. 3)
2023 doi
-
[22]
New estimates of the convergence rate in the Lyapunov theorem
Ilya Tyurin. “New estimates of the convergence rate in the Lyapunov theorem”. In:arXiv preprint arXiv:0912.0726(2009) (cit. on p. 53)
2009 arXiv
-
[23]
Cambridge Series in Statistical and Probabilistic Mathematics
Roman Vershynin.High-Dimensional Probability: An Introduction with Applications in Data Science. Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, 2018 (cit. on p. 53)
2018
-
[24]
Consensus on dynamic stochastic block models: Fast convergence and phase transitions
Haoyu Wang, Jiaheng Wei, and Zhenyuan Zhang. “Consensus on dynamic stochastic block models: Fast convergence and phase transitions”. In:arXiv preprint arXiv:2209.03999(2022) (cit. on p. 3)
2022 arXiv
-
[25]
Opinion forming in Erd˝ os–R´ enyi random graph and expanders
Ahad N Zehmakan. “Opinion forming in Erd˝ os–R´ enyi random graph and expanders”. In: Discrete Applied Mathematics277 (2020), pp. 280–290 (cit. on p. 3)
2020
-
[26]
Minimax rates of community detection in stochas- tic block models
Anderson Y. Zhang and Harrison H. Zhou. “Minimax rates of community detection in stochas- tic block models”. In:The Annals of Statistics44.5 (2016), pp. 2252 –2280.doi:10.1214/15- AOS1428.url:https://doi.org/10.1214/15-AOS1428(cit. on p. 7). 56
2016 doi
Reviewed July 31, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.