{"id":"b91e766e-a2f2-4471-aff6-64961c98a3bf","arxiv_id":"2507.08900","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"For noisy Hegselmann-Krause dynamics, bounded spaces guarantee finite-time quasi-synchronization in all dimensions, while unbounded spaces only do so in dimensions one and two, with non-integrable waiting times.","lead":"A mathematical study shows that noisy opinion dynamics in a bounded space always reach quasi-synchronization in finite time, for any dimension. In an unbounded space, that only happens almost surely in one and two dimensions, and fails for three or more.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (3.15) falsely identifies cube clipping with projection onto B(epsilon/2); Theorem 3.1 is therefore unproven.","rationale":"The Reader's verdict is REJECT with high correctness risk, and my independent reading confirms the most load-bearing flaw: equation (3.15). The bounded model's coordinate-wise clipping to the cube is not equivalent to Euclidean projection onto B(epsilon/2), so the contraction result Corollary 3.4 does not apply. The proof's one-line derivation of the contraction inequality from (2.2) is unjustified for the ball. This directly undermines Theorem 3.1, the central positive result and the foundation of the claimed boundedness/dimension dichotomy. I agree with the reader's weakest_assumption. The paper may be salvageable by a different argument using the cube projection and diameter, but as submitted the stated result is not established. Issues in the unbounded part (e.g., the deferred d=2 recurrence argument) are secondary and do not affect this conclusion.","tokens_in":12883,"tokens_out":14381,"duration_ms":167007,"concrete_test":"Analytical one-step test: fix n=2, d=1, epsilon=0.5, x(0)=(1,1), and set noise xi=(0,0). Compute the right-hand side of (2.2): P_C((1,1))=(1,1). Compute the right-hand side of (3.15): P_{B(0.25)}((1,1))=(0.177,0.177). Since the claimed rewriting (3.15) must hold for every admissible noise realization but gives different states, it is false. Repeating with any xi=(a,a), |a|<=0.25, yields the same separation; this settles that the theorem's proof does not apply to the stated bounded HK model.","verdict_should_be":"REJECT","load_bearing_attack":"Equation (3.15) rewrites the bounded update (2.2) as x(t+1)=P_{B(epsilon/2)}(F(x(t))+xi(t+1)). This is not correct. The update (2.2) clips each coordinate to [-1,1], so it is the Euclidean projection P_C onto the cube C=[-1,1]^{n x d}; it is not projection onto the origin-centered ball B(epsilon/2) in R^{n x d}. The two maps disagree already at one step. Take n=2, d=1, epsilon=0.5, x(0)=(1,1), and any admissible noise value xi=(a,a) with |a|<=0.25 (for instance a=0). Update (2.2) gives x(1)=(min(1,1+a), min(1,1+a)), while P_{B(0.25)}(1+a,1+a) ~ (0.177,0.177); these are never equal. Hence Corollary 3.4 is applied to a different stochastic process. The inequality ||F(x)-P_D(F(x))|| <= ||x-P_D(x)|| asserted from (2.2) is natural only for D=C, not for D=B(epsilon/2). Because Theorem 3.1 is the sole bounded-space result and the basis of the bounded/unbounded dichotomy, this is a load-bearing correctness gap, not a stylistic issue.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies the noisy Hegselmann-Krause model in R^d, in both bounded and unbounded state spaces. It defines a stopping time T at which the configuration diameter first falls below the confidence threshold ε and claims: (i) in bounded space, E T < ∞ for every dimension d; (ii) in unbounded space, P{T < ∞} = 1 for d = 1, 2 with E T = ∞ for some initial states; and (iii) for d ≥ 3 with symmetric noise, P{T < ∞} < 1 for some initial states. The proofs combine a general contraction lemma for projected stochastic systems with random-walk recurrence and transience arguments.","tokens_in":13260,"tokens_out":30532,"duration_ms":315185,"significance":"If the results were correct, the bounded/unbounded and low/high-dimensional dichotomy would be a valuable contribution to the theory of noisy opinion dynamics, and the distinction between finite almost-sure hitting and integrability is natural. The paper correctly builds on the authors' earlier one-dimensional results for the almost-sure claims, and the d ≥ 3 transience argument contains a substantive idea. However, the current manuscript does not establish the main claims: the bounded-space proof rests on a false identification of the update rule with a ball projection, the one-dimensional non-integrability proof misapplies a key lemma, and the two-dimensional almost-sure claim is deferred to a reference to a similar argument. The central results are therefore unproven in this version.","major_comments":[{"comment":"The equality x(t+1) = P_{B(ε/2)}(F(x(t)) + ξ(t+1)) is false. The bounded update (2.2) is the coordinate-wise projection onto the cube C = [-1,1]^{n×d}, not the Euclidean projection onto the ball B(ε/2) in R^{n×d}. For example, with n = 2, d = 1, ε = 0.5, x(0) = (1,1) and ξ(1) = 0, the left side is (1,1), while P_{B(0.25)}(1,1) = (0.25/√2, 0.25/√2). The two projections also differ for ε > 2, where B(ε/2) is not even contained in C. Consequently, Corollary 3.4, which applies to the ball-projection system (3.2), does not apply to the actual bounded HK model, and Theorem 3.1 is unproven.","section":"3.1, Eq. (3.15), proof of Theorem 3.1"},{"comment":"Even if (3.15) were replaced by the correct cube projection, the hypotheses of Corollary 3.4 are not met by the bounded HK model. The corollary assumes ∥ξ_i(t)∥ ≤ r0/2 a.s.; with D = B(ε/2) this means r0 = ε/2, hence δ ≤ ε/4, whereas Theorem 3.1 only assumes δ ≤ ε/2. In addition, Lemma 3.2's condition (3.4) requires the noise to charge every measurable subset of B(r), which cannot hold when the noise is a.s. bounded by δ < r. Corollary 3.4 asserts without proof that bounded nondegenerate noise suffices for (3.11)-(3.12), but the sets appearing there can lie outside the noise support for states far from D. The proof of Theorem 3.1 therefore needs a different argument even beyond the correction of (3.15).","section":"3.1, application of Corollary 3.4"},{"comment":"Lemma 3.8 is not applicable to the process Q_ij. The lemma requires S(t+1) = g(U(t)) + h(ξ(t+1)), where U(t) = ∑_{k=1}^t ξ(k) and h is a function of the same increment ξ(t+1) that drives U. In the application, U(t) = Z(t) = ∑_{k=1}^t y(k), while the additive term is ξ_i(t+1) - ξ_j(t+1). This term is not a function of y(t+1), nor is it the increment of Z. In particular, the proof of Lemma 3.8 uses the event {ξ(1)>a, ..., ξ(L)>a} to conclude h(ξ(j)) > 0 for j ≤ L; no such conclusion is valid when the conditioning is on y(j) > a and the noise entering the additive term is ξ_i(j) - ξ_j(j). Hence E T_Q = ∞ is not established, and the non-integrability statements for d = 1, as well as the d = 2 reduction, are unsupported.","section":"3.2, Lemma 3.8 and Eqs. (3.27)-(3.31)"},{"comment":"The proof of P{T < ∞} = 1 for d = 2 consists of the sentence 'using the recurrence of random walk in R^2 (Lemma 3.9) and the homogenous ε of HK model (2.3), then following a similar argument of Proposition 3.1 of [11], we can obtain the conclusion.' This is not a proof. The reduction to a two-dimensional random walk, the role of the confidence threshold ε, and the interaction structure between the two groups are not written down. Since the two-dimensional almost-sure claim is a central part of Theorem 3.5(a), it requires a complete argument.","section":"3.2, proof of Theorem 3.5 for d = 2"}],"minor_comments":[{"comment":"The phrase 'Morkov property' should be 'Markov property'; it appears at least twice in Section 3.2.","section":"Throughout"},{"comment":"The time indexing is inconsistent: d_ij(t) in (3.24) contains ξ_i(t) - ξ_j(t), while (3.27) defines Q_ij(t+1) = Z(t) + (ξ_i(t+1) - ξ_j(t+1)). Please align the indices so that (3.28) holds as written.","section":"3.2, Eqs. (3.24) and (3.27)"},{"comment":"The notation U_k^L and U^{j-L}_L is used without a clear definition of the shifted sums; please define these partial sums explicitly before (3.16).","section":"3.2, Lemma 3.8 proof"},{"comment":"Reference [30] contains a typo: 'mordern probability' should be 'modern probability'.","section":"References"},{"comment":"The sentence 'Give any O1 ∈ R^{n×d}' should read 'Given any ...'.","section":"3.2, d ≥ 3 paragraph"},{"comment":"The vector clipping notation (y_1,...,y_d)^T_{[-1,1]} is introduced with a missing bracket in the displayed definition; please make clear that each coordinate is clipped separately.","section":"2, Eq. (2.2)"}],"recommendation":"reject","confidential_remarks":"The paper addresses an interesting question, and the d ≥ 3 transience argument contains a plausible idea. However, the bounded-space theorem is based on an invalid identification of the model, the non-integrability proof for d = 1 misapplies Lemma 3.8, and the d = 2 proof is deferred. These are load-bearing gaps that cannot be repaired by local edits; a resubmission with a correct proof of Theorem 3.1 under the actual cube-clipping dynamics and a complete d = 2 argument would be needed before the paper could be reconsidered."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this paper gives a clean, new story about noise-induced quasi-synchronization in HK dynamics, but the proof of the main bounded-space theorem (Theorem 3.1) is built on an incorrect rewriting of the model, and the d=1 non-integrability proof misapplies its own Lemma 3.8. As it stands, the results are not established.\n\nWhat is genuinely new: the dimension-dependent dichotomy (bounded: integrable hitting time for all d; unbounded: almost sure but non-integrable for d=1,2, and failure for d≥3 under symmetric noise) is not in the prior literature. The random-walk arguments for the unbounded case are creative, and the paper is clearly written.\n\nThe main problem is equation (3.15), which rewrites the bounded update (2.2) as projection onto the ball B(ε/2). Update (2.2) clips each coordinate to [−1,1], i.e. projection onto the cube C in R^{nd}; this distance is not the projection onto the ball. A one-step counterexample shows the two processes are never equal. So Corollary 3.4 is applied to a different stochastic process, and Theorem 3.1 is unproven. That is load-bearing, since Theorem 3.1 is the only bounded-space result and the bounded/unbounded contrast rests on it.\n\nThe unbounded part has further soft spots. Lemma 3.8 requires deterministic g,h with h(x)x>0 for x≠0; the application to Q_ij(t) treats the difference ξ_i(t+1)−ξ_j(t+1) as if it satisfied that sign condition, which it does not. The d=2 almost-sure claim is deferred to a prior paper without a detailed argument.\n\nCredit where due: the questions are good, the paper honestly engages the literature, and there is no fitting or circularity. But the central proofs have load-bearing gaps. Someone working on stochastic multi-agent systems or opinion dynamics would want to know the claimed dichotomy, and the mistakes are instructive. This deserves a serious referee: the editor should send it out, but the authors will need to repair the proof of Theorem 3.1 and fix the Lemma 3.8 application. I would not cite it as a proven result until then.","headline":"Interesting and plausible high-dimensional HK dichotomy, but the main theorem's proof rests on a false cube-vs-ball projection equivalence and the d=1 proof misapplies Lemma 3.8.","tokens_in":13681,"tokens_out":7014,"would_cite":false,"duration_ms":68978,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["34D06","60G40","60G50","91D30"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that for high-dimensional noisy Hegselmann-Krause opinion dynamics, bounded state space guarantees quasi-synchronization in finite time with finite expected waiting time for every dimension, while in unbounded space the…","keywords":["opinion dynamics","Hegselmann-Krause model","quasi-synchronization","noise-induced synchronization","stopping time","random walk recurrence","high-dimensional consensus"],"falsifier":"For the bounded model (2.2), a Monte Carlo estimate of the expected quasi-synchronization time in dimension $d=3$ from adversarial initial states should show a finite mean; if the empirical mean grows without bound as the number of samples increases, the theorem is contradicted. Separately, for the unbounded model in dimension 3 with symmetric bounded noise, simulate many runs starting from two separated clusters; the paper predicts a strictly positive fraction of non-synchronizing trajectories, so observing every run synchronize would contradict Theorem 3.5(b).","tokens_in":12657,"feed_emoji":"🗣️","tokens_out":7501,"duration_ms":79159,"temperature":0.7,"pith_summary":"The paper asks when noisy multi-agent opinion dynamics reach quasi-synchronization, meaning all agents' opinions lie within the confidence radius, and how long the random waiting time is. It claims the answer is determined by two features of the state space: whether it is bounded and what dimension it has. In the bounded model with states confined to $[-1,1]^d$, it proves for all $d\\ge 1$ and every initial state that the stopping time $T$ has finite expectation. In the unbounded model, it proves a sharp split: for $d=1,2$ quasi-synchronization occurs almost surely from every initial state but with infinite expected time for some initial states, whereas for $d\\ge 3$ with symmetric noise there are initial states from which synchronization is not certain. The upshot, if correct, is that spatial boundedness is a genuine driver of noise-induced order and that the one-dimensional picture does not extend to higher-dimensional unbounded spaces.","feed_headline":"Dimension and boundedness decide when noisy opinions sync","feed_subtitle":"In unbounded space, only dimensions 1 and 2 synchronize almost surely; dimensions 3 and higher may never sync from some states.","key_machinery":"The load-bearing object is the stopping time $T$, the first time the diameter $d_V(t)$ falls to at most $\\epsilon$; by Proposition 2.2, under $\\delta \\le \\epsilon/2$ this coincides almost surely with the permanent quasi-synchronization time $T'$. For the bounded result, the key identity is the rewriting $x(t+1)=P_{B(\\epsilon/2)}(F(x(t))+\\xi(t+1))$, which turns coordinate-wise clipping into projection onto a ball, together with the contraction condition (3.3) with $\\alpha=1$, meaning the update shortens distance to the target ball. Lemma 3.2 and Corollary 3.4 convert a positive probability of making progress into a geometric tail bound and hence a finite expectation. For the unbounded result, the key mechanism is a random-walk decomposition of the gap between two clusters: the between-cluster distance evolves as a centered random walk plus bounded increments, and recurrence in $d=1,2$ versus transience in $d\\ge 3$ decides whether that gap must eventually hit the confidence threshold.","core_discovery":"The central discovery is a dimensional phase transition in the quasi-synchronization time $T=\\inf\\{t\\ge 0: \\max_{i,j}\\|x_i(t)-x_j(t)\\|≤ \\epsilon\\}$. Theorem 3.1 asserts that for the bounded noisy HK model (2.2), under zero-mean nondegenerate noise with $\\delta \\le \\epsilon/2$ and $\\epsilon \\in (0,2\\sqrt{d}]$, one has $E T<\\infty$ for all $d\\ge 1$ and all initial states. Theorem 3.5 asserts that for the unbounded model (2.3), for $d=1,2$, $P\\{T<\\infty\\}=1$ for all initial states while $E T=\\infty$ for some initial states; and, when the noise is symmetric, for $d\\ge 3$ there are initial states with $P\\{T<\\infty\\}<1$. The bounded-space proof rewrites the clipped average update as projection onto a Euclidean ball of radius $\\epsilon/2$ and shows the distance to that ball contracts, while the unbounded proof reduces the two-cluster case to the hitting behavior of a centered random walk, which is recurrent in dimensions 1 and 2 and transient in dimension 3 and higher.","pith_inferences":["A direct check of equation (3.15) is warranted: the model's coordinate-wise cube clipping is not obviously the same as projection onto the Euclidean ball of radius $\\epsilon/2$, and if the two projections differ, the contraction argument does not automatically cover the stated model even though the theorem might still be true. ","The proof establishes finiteness of $E T$ in the bounded case but leaves the rate implicit; a natural next step is to derive explicit bounds on $E T$ in terms of the number of agents $n$, dimension $d$, confidence radius $\\epsilon$, and noise bound $\\delta$. ","The recurrence-versus-transience mechanism suggests that other bounded-confidence opinion models, such as asynchronous or inertial variants, may exhibit the same dimensional split in unbounded space. ","The paper assumes $\\delta \\le \\epsilon/2$ throughout, which is what makes the absorption property in Proposition 2.2 work; an extension could explore what happens with stronger noise, where leaving permanent quasi-synchronization becomes possible. "],"forward_implications":["The bounded HK model reaches quasi-synchronization almost surely in finite time in every dimension, and the expected waiting time is finite, so bounded opinion ranges guarantee noise-induced order. ","In unbounded spaces of dimension 1 and 2, quasi-synchronization still occurs almost surely from any initial state, but the waiting time has infinite expectation for some initial configurations, so almost-sure eventual synchronization does not mean typically fast synchronization. ","In unbounded dimension 3 and higher with symmetric noise, the system can fail to synchronize with positive probability from separated initial clusters, so high-dimensional unbounded opinion spaces can prevent consensus. ","The unbounded behavior is governed by recurrence versus transience of a random walk, connecting the synchronization question to classical random-walk return probabilities. ","In the bounded case the dimension plays no role in the qualitative conclusion, since all dimensions behave the same way. "],"supporting_citations":[{"why":"Supplies the definition of quasi-synchronization and the one-dimensional bounded result that this paper extends.","marker":"[9]"},{"why":"Supplies the one-dimensional unbounded result that $P\\{T<\\infty\\}=1$, which the paper uses and generalizes to dimension 2.","marker":"[11]"},{"why":"Provides the random-walk hitting-time result used in Lemma 3.7 to prove non-integrability of the stopping time.","marker":"[29]"},{"why":"Provides the recurrence of random walks in dimensions 1 and 2 and transience in dimension 3 and higher, used in Lemma 3.9.","marker":"[30]"}],"fun_headline_variants":["Bounded space guarantees sync, unbounded only in low dims","High-dim noisy opinions sync only if bounded or d≤2","Unbounded space: only dimensions 1-2 sync almost surely","Bounded space syncs all dims; unbounded only d=1,2"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise of the bounded-space proof is that the model's coordinate-wise clipping to the cube is equivalent to projecting onto the Euclidean ball $B(\\epsilon/2)$ as written in equation (3.15); if these two projections differ, the contraction argument behind Theorem 3.1 does not apply to the stated model.","fun_headline_variants_meta":{"raw":{"variants":["Bounded space guarantees sync, unbounded only in low dims","High-dim noisy opinions sync only if bounded or d≤2","Unbounded space: only dimensions 1-2 sync almost surely","Bounded space syncs all dims; unbounded only d=1,2"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000728,"raw_usage":{"total_tokens":3271,"prompt_tokens":966,"completion_tokens":2305,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":582,"completion_tokens_details":{"reasoning_tokens":2226}},"tokens_in":582,"tokens_out":2305,"duration_ms":19906,"temperature":1.0,"reasoning_tokens":2226,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:22:38.091736+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the bounded model (2.2), a Monte Carlo estimate of the expected quasi-synchronization time in dimension $d=3$ from adversarial initial states should show a finite mean; if the empirical mean grows without bound as the number of samples increases, the theorem is contradicted. Separately, for the unbounded model in dimension 3 with symmetric bounded noise, simulate many runs starting from two separated clusters; the paper predicts a strictly positive fraction of non-synchronizing trajectories, so observing every run synchronize would contradict Theorem 3.5(b).","supporting_citations":[{"cited_title":"Noise leads to quasi-consensus of Hegselmann-Krause opinion dynamics","cited_arxiv_id":null,"evidence_quote":"Supplies the definition of quasi-synchronization and the one-dimensional bounded result that this paper extends."},{"cited_title":"Noise-induced synchronization of Hegselmann- Krause dynamics in full space","cited_arxiv_id":null,"evidence_quote":"Supplies the one-dimensional unbounded result that $P\\{T<\\infty\\}=1$, which the paper uses and generalizes to dimension 2."},{"cited_title":"Chow and H","cited_arxiv_id":null,"evidence_quote":"Provides the random-walk hitting-time result used in Lemma 3.7 to prove non-integrability of the stopping time."},{"cited_title":"Kallenberg, Foundations of mordern probability, Springer, 2002","cited_arxiv_id":null,"evidence_quote":"Provides the recurrence of random walks in dimensions 1 and 2 and transience in dimension 3 and higher, used in Lemma 3.9."}],"review_version":1}