{"id":"ce5a7d6b-bfaf-41a4-9028-5db162a1e72e","arxiv_id":"2412.12134","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Electrically coupling nonchaotic Rulkov neurons in a ring produces chaotic spiking, synchronized bursting, and hyperchaos, with Lyapunov dimensions reaching 45 of 60 state-space dimensions.","lead":"Coupling simple nonchaotic neuron models into a ring makes the whole network chaotic, with computed attractor dimensions reaching 45 of the system's 60 state-space dimensions. The study maps how chaos and attractor complexity change with electrical coupling strength in a tractable model system.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The d_L up to 45/60 claim rests on Lyapunov spectra computed from single 1000-step orbits with no burn-in; for µ=0.001 this is likely insufficient, and the Table 1 box-counting validation is under-specified and cannot be assessed.","rationale":"The reader's weakest assumption identifies exactly the load-bearing point: the assertion that 1000 iterations are sufficient for Lyapunov exponent convergence. My stress-test agrees and sharpens it: because d_L depends on the entire Lyapunov spectrum, not just λ1, the slow-variable timescale argument is even more consequential than it is for the maximal exponent alone. The Table 1 box-counting validation is a related second pillar; it is under-specified and, for dimensions near 43, computationally implausible as described, so it does not currently rescue the claim. I do not see an internal inconsistency in the Jacobian derivation or in the qualitative regime descriptions; the model and methods are mostly standard, and the GitHub availability of data is a positive. The appropriate disposition is therefore the same conditional one the reader reached: accept the qualitative finding as plausible, but require convergence checks, error bars, and documented box-counting before the quantitative 45/60-dimensionality claim can be taken as established.","tokens_in":21740,"tokens_out":6155,"duration_ms":66607,"concrete_test":"Recompute the full Lyapunov spectrum for the homogeneous case at g=0.1, 0.3, 0.6, 0.9: discard the first 10^5 iterations, then compute d_L from Eq. (13) using orbits of length T=10^4, 10^5, and 10^6 with the same QR scheme (Appendix B), and also repeat over at least 10 random initial conditions to obtain mean and spread. If d_L at any of these g changes by more than 1 relative to the published 1000-step value, the convergence assumption fails and Figures 3 and 8 plus the abstract's dimensionality claim need revision. If d_L is stable, re-run the box-counting on the same longer trajectories with all ε values and N(ε) data documented, and check whether N(ε) follows a power law over at least one decade; otherwise Table 1 cannot validate the Kaplan–Yorke conjecture for this system.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline quantitative claim—attractors occupying as many as 45 of 60 dimensions—is a Lyapunov dimension d_L from Eq. (13), computed from the full 60-exponent Lyapunov spectrum. Every exponent, and therefore every point in Figure 8, comes from one orbit O(X0) of length 1000, with no transient discarded and no averaging over initial conditions (Section 2 and Appendix B). Because the slow variable evolves with µ=0.001, the slow timescale is ~1000 iterations; a single 1000-step orbit covers barely one such period, so convergence of finite-time Lyapunov exponents, especially for the near-zero and negative middle spectrum that determines κ in Eq. (12), is not credible. A small error in these exponents can change κ and shift d_L by several units. The paper's own validation of the Kaplan–Yorke approximation, Table 1, is also not reproducible as stated: it reports box-counting estimates in a 60-dimensional space without giving the box side lengths ε, the number of sampled points, or the number of independent orbits used, and for d≈43 a naive box count would require ~ε^(-43) points. While the qualitative emergence of chaos from coupled nonchaotic neurons is plausible and visually supported, the central fractal-geometry claim is carried by these two numerical pillars, and neither is presently secured.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript studies a ring of 30 electrically coupled nonchaotic Rulkov neurons (60-dimensional state space) in three parameter regimes: homogeneous, partially heterogeneous in sigma_i, and fully heterogeneous in sigma_i and alpha_i. The author computes maximal Lyapunov exponents as a function of coupling strength g, identifies regimes of chaotic spiking, synchronized chaotic bursting, and synchronized hyperchaos, and estimates attractor dimensions via the Kaplan-Yorke formula applied to the full 60-exponent Lyapunov spectrum. The headline quantitative claim is that strange attractors can occupy up to about 45 of the 60 dimensions, supported by a comparison of Lyapunov dimension to estimated box-counting dimension for four values of g.","tokens_in":1410,"tokens_out":2017,"duration_ms":107184,"significance":"The qualitative phenomenon---that coupling individually regular-spiking Rulkov neurons can produce chaotic and hyperchaotic collective dynamics---is plausible, visually documented, and of interest for map-based neuronal network studies. The paper contributes an explicit Jacobian for the ring system (Appendix A), pseudocode, and a public GitHub repository, and it systematically scans the coupling strength, going beyond the earlier qualitative discussion in Ref. [40]. If the Lyapunov spectra and box-counting estimates were properly converged and fully documented, the dimension trends (including the non-monotonic relation between lambda_1 and d_L) would be a substantial contribution. At present, however, the quantitative pillars supporting the 45/60 claim and the Kaplan-Yorke validation are not numerically secured.","major_comments":[{"comment":"The claim that an orbit of length 1000 is sufficiently long for Lyapunov exponent convergence is load-bearing and unsupported. With mu=0.001, the slow variable evolves on a timescale of order 1/mu = 1000 iterations, so the first 1000 iterates from the initial condition (A12) include the transient toward the attractor. No burn-in, no averaging over initial conditions, and no error bars are reported. Since Eq. (A11) estimates each lambda_i as (1/t) sum ln|r_ii^(k)| with t=1000, finite-time fluctuations in the near-zero and negative exponents that determine kappa in Eq. (12) can be comparable to the exponent magnitudes; an error of only a few percent in the 43rd-45th exponents shifts d_L by several units. All points in Figures 3, 6, and 8 and all rows of Table 1 inherit this uncertainty. Please add convergence tests (lambda_1 and d_L versus t for t=10^3, 10^4, 10^5) and ensemble averages over initial conditions, with transients discarded.","section":"Section 2 / Appendix B"},{"comment":"The box-counting validation of the Kaplan-Yorke approximation is not reproducible as written. The text states that points are sampled by generating many orbits of length 10^7 and that close values of epsilon are chosen, but it does not give the epsilon values, the total number of sampled points, the number of orbits, the box-counting algorithm in 60 dimensions, or uncertainties on d_L and d. For a claimed dimension near 43, a literal box count at resolution epsilon would require N(epsilon) ~ epsilon^(-43) sampled points, which the stated sampling cannot provide unless the attractor is much lower-dimensional at the scales used. Without this information the 5% agreement in Table 1 cannot be assessed, and the conclusion that the Kaplan-Yorke conjecture does hold for this system is not supported. Please report the full estimation protocol and uncertainties, or use a dimension estimator that is feasible and fully specified in high dimensions.","section":"Section 3, Table 1"},{"comment":"The statement that By using numerical simulations to systematically vary zeta (see Appendix C), it can be found that for zeta>=4, varying zeta has no effect on the qualitative behavior is not supported by the cited appendix, which contains only pseudocode. If the zeta-independence claim is used to generalize from zeta=30 to other ring sizes, please provide the numerical evidence (e.g., lambda_1 or d_L versus zeta); otherwise restrict the conclusions to the zeta=30 system.","section":"Section 2, choice of zeta=30"}],"minor_comments":[{"comment":"The last row of the displayed iteration function is missing the factor g/2 in the coupling term of the y-update for neuron zeta-1; compare with Eq. (8) and Eq. (A2), where the factor is present. Please correct the displayed function.","section":"Equation (10)"},{"comment":"The stopping condition if S<=0 is not the exact implementation of Eq. (12), which defines kappa as the largest index with cumulative sum >=0; the two agree generically but differ when a partial sum is exactly zero. Please align the boundary case.","section":"Algorithm A5"},{"comment":"The terminology nonchaotic Rulkov neurons is used for all three regimes, but in the partially and fully heterogeneous cases some individual neurons are chaotic even at g=0 (lambda_1 approx 0.0644 and 0.0469 in Figures 4a and 5a). Please qualify the terminology.","section":"Title and Abstract"},{"comment":"The phrase close values of epsilon are chosen, where the sampled points scale according to their attractor is circular as written; please state explicitly how the scaling region was selected and how many points were used at each epsilon.","section":"Section 3, box-counting description"}],"recommendation":"major_revision","confidential_remarks":"The manuscript has already been published in Fractal Fract, but as a referee assessment of the submitted version I focus on the numerical evidence. The main risk is that the 45/60-dimensional claim is driven by unconverged finite-time Lyapunov exponents or by an unverifiable box-counting comparison. The author's GitHub repository would allow a quick check of the convergence behavior. I would not reject on the basis of the qualitative claims, but the quantitative claims need the requested convergence tests and estimation details before they can be considered established."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the qualitative story here is plausible and worth having on record — electrically coupled nonchaotic Rulkov neurons on a ring produce chaotic spiking, chaotic bursting, and hyperchaos — and the full Lyapunov spectrum analysis is new for this specific system. But the headline claim that attractors occupy up to 45 of 60 dimensions rests on numerical pillars that don't yet hold: Lyapunov spectra from single 1000-step orbits without burn-in, and a box-counting validation that is under-specified and likely infeasible at those dimensions.\n\nWhat's genuinely new: the Jacobian for the ring is derived cleanly and the QR-based spectrum calculation is implemented for the full 60-dimensional system. The systematic g-scans across homogeneous and two heterogeneous regimes extend the qualitative observations of Ref. [40] into quantitative form. The comparison between the maximal Lyapunov exponent and Lyapunov dimension peaks — left peak higher for d_L, right peak higher for λ1, explained by 18 vs 9 positive exponents — is a nice observation that gives the paper a real point beyond \"chaos emerges.\"\n\nSoft spots, in order of severity. First, the 1000-step orbit length. With µ=0.001, the slow variable evolves over a timescale of ~1000 iterations; a single orbit of that length covers barely one slow period. Finite-time Lyapunov exponents for near-zero and negative exponents cannot be trusted, and those are exactly the ones that determine κ and hence d_L in Eq. (13). The paper states \"sufficiently long\" without any convergence check, averaging, or error bars. This undermines every quantitative claim in Figures 3, 6, and 8, not just the dimensions. Second, Table 1's box-counting validation is not reproducible: no ε values, no point counts, no orbits used. And for d≈43 in a 60-dimensional space, the required point count is astronomically large; the described \"many orbits of length 10^7\" likely cannot resolve that. So the Kaplan–Yorke check doesn't currently validate anything.\n\nThe qualitative emergence of chaos, however, is visually supported and likely robust. The paper is honest about what was computed, and the code and data are on GitHub. But the central fractal-geometry claim needs stronger numerical evidence. This deserves peer review — a serious referee could demand the needed convergence studies — but I wouldn't accept the dimension claims as-is.","headline":"Qualitatively plausible and a genuine extension of Ref. [40], but the headline '45 of 60 dimensions' claim rests on 1000-step Lyapunov spectra and an under-specified box-counting check.","tokens_in":22574,"tokens_out":2397,"would_cite":false,"duration_ms":22998,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37D45","37C45","37M25","37N25"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that electrically coupling 30 individually nonchaotic Rulkov neurons in a ring produces chaotic spiking, synchronized chaotic bursting, and synchronized hyperchaos, with strange attractors that can occupy up to 45 of the…","keywords":["neuronal dynamics","nonchaotic Rulkov model","high-dimensional systems","chaotic dynamics","Lyapunov exponents","strange attractors","fractal dimension","Kaplan-Yorke conjecture"],"falsifier":"Recompute the full Lyapunov spectrum for the homogeneous ring at $g = 0.1$, $0.25$, $0.6$, and $0.9$ using orbits of length 100,000 with the first 10,000 steps discarded as transient, then compare the Kaplan–Yorke dimensions with those in Table 1; a shift beyond a few percent would show the convergence assumption fails.","tokens_in":1894,"feed_emoji":"🧠","tokens_out":1892,"duration_ms":83586,"temperature":0.7,"pith_summary":"Individually nonchaotic Rulkov neurons, which spike regularly when uncoupled, become chaotic when wired into a ring with electrical coupling. The paper shows that varying the coupling strength drives the ring through distinct regimes—unsynchronized chaotic spiking, synchronized chaotic bursting, and synchronized hyperchaos—even though every isolated neuron behaves regularly. Using the full Lyapunov spectrum and the Kaplan–Yorke conjecture, the author estimates the fractal dimension of the resulting attractors and finds that they occupy as many as 45 of the 60 dimensions of state space, with Lyapunov-dimension estimates matching box-counting estimates to within a few percent. The significance is that high-dimensional chaos and fractal geometry emerge from a lattice of simple, individually regular units, making the system a tractable model for collective behavior in neuronal networks.","feed_headline":"Ring of 30 simple neurons turns chaotic and fills 45 of 60 dimensions","feed_subtitle":"Electrical coupling makes regular-spiking neurons chaotic, with attractors spanning most of 60 dimensions","key_machinery":"The load-bearing object is the piecewise nonchaotic Rulkov map, where the fast variable evolves by $f(x,y;\\alpha)$, and the electrical ring coupling enters as $C_i = \\frac{g}{2}(x_{i-1}+x_{i+1}-2x_i)$ added to both the fast and slow update of each neuron. From this, the paper constructs the $60\\times60$ piecewise Jacobian of the ring and feeds it into the QR-factorization algorithm for the Lyapunov spectrum. The Kaplan–Yorke dimension $d_L = \\kappa + \\frac{1}{|\\lambda_{\\kappa+1}|}\\sum_{i=1}^\\kappa \\lambda_i$, where $\\kappa$ is the largest index with cumulative Lyapunov sum nonnegative, then converts the spectrum into an estimate of the attractor's fractal dimension, bypassing box-counting, which would require on the order of $10^{36}$ points in 60-dimensional space.","core_discovery":"The central discovery is that electrical coupling alone is enough to make a ring of nonchaotic Rulkov neurons chaotic, and that the strange attractors produced are genuinely high-dimensional. For a homogeneous ring with $\\sigma=-0.5$ and $\\alpha=4.5$, coupling $g=0.05$ already yields $\\lambda_1 \\approx 0.0491$; $g=0.25$ yields synchronized chaotic bursting; and $g=1$ yields synchronized hyperchaos with 11 positive Lyapunov exponents out of 60. The author computes all 60 Lyapunov exponents via QR factorization of the piecewise ring Jacobian and uses the Kaplan–Yorke formula to obtain Lyapunov dimensions $d_L$. In the homogeneous regime, $d_L$ reaches values close to 45 at moderate coupling and is validated against direct box-counting estimates at four coupling strengths with errors between 0.04% and 2.65%. A striking result is that the maximal Lyapunov exponent and the attractor dimension do not track each other: synchronized hyperchaos has the largest $\\lambda_1$ but a smaller dimension than the weaker, unsynchronized chaotic spiking regime, because synchronization concentrates expansion in few directions.","pith_inferences":["Because the slow variable evolves with $\\mu = 0.001$, the reported orbits of length 1000 cover roughly one slow oscillation; recomputing spectra with $10^4$–$10^5$ iterations and a discarded transient would directly test whether the Lyapunov exponents and dimension curves are numerically converged.","The same Lyapunov-dimension workflow could be applied to other lattice topologies the author mentions—mesh, torus, all-to-all coupling—to see whether the dimension-versus-coupling pattern (left peak higher than right in homogeneous rings) is generic.","If the synchronization-lowers-dimension effect is robust, it suggests a practical design lever for neuromorphic or reservoir-computing hardware: tuning coupling or heterogeneity can expand or contract the effective dimensionality of a fixed-size network without altering the number of neurons.","The numerical confirmation of Kaplan–Yorke for this piecewise-smooth system hints that the conjecture may extend to a broader class of hybrid or hysteretic maps, where piecewise Jacobians make analytic dimension estimates difficult."],"forward_implications":["Coupling strength $g$ acts as a single control parameter that moves a homogeneous ring through unsynchronized chaotic spiking, synchronized chaotic bursting, and synchronized hyperchaos.","The Kaplan–Yorke dimension gives a practical proxy for the true fractal dimension of high-dimensional neuron-lattice attractors, where box-counting would need roughly $10^{36}$ points.","Synchronization lowers an attractor's fractal dimension even while raising the maximal Lyapunov exponent, because it reduces the number of positive Lyapunov exponents.","Partially and fully heterogeneous rings are chaotic even at zero coupling, yet their Lyapunov-dimension curves become similar to the homogeneous case as coupling increases.","A 60-dimensional ring of simple maps supports attractors spanning about three quarters of the state space, showing that high-dimensional collective chaos needs no chaotic individual units."],"supporting_citations":[{"why":"Supplies the two-dimensional Rulkov map and the parameter roles ($\\alpha$, $\\sigma$, $\\mu$) that define the nonchaotic spiking regime used throughout.","marker":"[13]"},{"why":"Supplies the modified map form and the electrical coupling parameterization $C_{i,x}$ and $C_{i,y}$ used in the ring model.","marker":"[15]"},{"why":"Qualitatively describes complex dynamics in a similar Rulkov ring; this paper's quantitative Lyapunov analysis extends it.","marker":"[40]"},{"why":"Provides the full derivation of the piecewise ring Jacobian that the Lyapunov spectrum calculation relies on.","marker":"[41]"},{"why":"Supplies the QR-factorization method used to compute all 60 Lyapunov exponents.","marker":"[43]"},{"why":"States the Kaplan–Yorke conjecture and the Lyapunov-dimension formula used to approximate attractor dimensions.","marker":"[47]"},{"why":"Supports the claim that the Kaplan–Yorke conjecture holds in almost all cases, justifying its use here.","marker":"[49]"}],"fun_headline_variants":["Coupling nonchaotic neurons yields 60D chaos with fractal attractors","Simple neuron ring produces hyperchaos spanning 45 of 60 dimensions","How regular-spiking neurons become chaotic: a 60D ring study","Nonchaotic Rulkov ring turns chaotic, attractors fill most dimensions"],"cache_read_input_tokens":24576,"weakest_assumption_plain":"The reported chaos depends on assuming that a single 1000-step run, with no initial transient discarded and no averaging over starting states, already captures the long-term behavior at every coupling strength.","fun_headline_variants_meta":{"raw":{"variants":["Coupling nonchaotic neurons yields 60D chaos with fractal attractors","Simple neuron ring produces hyperchaos spanning 45 of 60 dimensions","How regular-spiking neurons become chaotic: a 60D ring study","Nonchaotic Rulkov ring turns chaotic, attractors fill most dimensions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000203,"raw_usage":{"total_tokens":1424,"prompt_tokens":1025,"completion_tokens":399,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":641,"completion_tokens_details":{"reasoning_tokens":318}},"tokens_in":641,"tokens_out":399,"duration_ms":4592,"temperature":1.0,"reasoning_tokens":318,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T20:50:13.039220+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the full Lyapunov spectrum for the homogeneous ring at $g = 0.1$, $0.25$, $0.6$, and $0.9$ using orbits of length 100,000 with the first 10,000 steps discarded as transient, then compare the Kaplan–Yorke dimensions with those in Table 1; a shift beyond a few percent would show the convergence assumption fails.","supporting_citations":[{"cited_title":"Modeling of spiking-bursting neural behavior using two-dimensional map.Phys","cited_arxiv_id":null,"evidence_quote":"Supplies the two-dimensional Rulkov map and the parameter roles ($\\alpha$, $\\sigma$, $\\mu$) that define the nonchaotic spiking regime used throughout."},{"cited_title":"Map-based models in neuronal dynamics.Phys","cited_arxiv_id":null,"evidence_quote":"Supplies the modified map form and the electrical coupling parameterization $C_{i,x}$ and $C_{i,y}$ used in the ring model."},{"cited_title":"Synchronized chaotic intermittent and spiking behavior in coupled map chains","cited_arxiv_id":null,"evidence_quote":"Qualitatively describes complex dynamics in a similar Rulkov ring; this paper's quantitative Lyapunov analysis extends it."},{"cited_title":"Ergodic theory of chaos and strange attractors.Rev","cited_arxiv_id":null,"evidence_quote":"Supplies the QR-factorization method used to compute all 60 Lyapunov exponents."},{"cited_title":"Chaotic behavior of multidimensional difference equations.Funct","cited_arxiv_id":null,"evidence_quote":"States the Kaplan–Yorke conjecture and the Lyapunov-dimension formula used to approximate attractor dimensions."},{"cited_title":"The dimension of chaotic attractors.Phys","cited_arxiv_id":null,"evidence_quote":"Supports the claim that the Kaplan–Yorke conjecture holds in almost all cases, justifying its use here."}],"review_version":1}