{"id":"49871ecb-1996-48f6-90b9-d01fdbbe7971","arxiv_id":"2411.14132","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Attractive diffusive coupling of excitable units induces multistability by repeatedly reinjecting trajectories into the excitable region, a mechanism called trapping.","lead":"This paper shows that diffusive coupling between excitable units that individually do not oscillate can create many coexisting oscillations: periodic, quasiperiodic, and chaotic. The mechanism is a reinjection of the units into an excitability region of their state space, trapping them there and generating multistability.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The trapping mechanism is demonstrated only for N=2 periodic and quasiperiodic attractors; its extension to the N=10 chaotic attractor is asserted, not shown.","rationale":"The reader's weakest_assumption already identifies exactly the load-bearing concern: direct geometric verification of the reinjection mechanism exists only for the N=2 periodic and quasiperiodic cases, yet the paper generalizes it to the N=10 chaotic attractor and states it as universal. My reading of the text confirms this: Sec. III D and Fig. 5 provide the geometric evidence, but Sec. III A's chaotic attractor is supported only by Lyapunov exponents and overlays of all units' projected trajectories, with no check of the coupling-vector orientation relative to W^s(x_s^unc). The paper is otherwise strong: the numerical methodology is transparent (Attractors.jl, 5000 initial conditions for N=10, long integrations, publicly available code), the bifurcation analysis for N=2 is careful, and the existence of multistable periodic, quasiperiodic, and chaotic attractors is convincingly demonstrated. Therefore I would not change the verdict: CONDITIONAL remains appropriate, because the existence claim is solid while the universal mechanism claim is not yet fully evidenced. My proposed test is designed to settle exactly this: it quantifies whether the chaotic attractor uses the same trapping mechanism by measuring reinjecting crossings of the uncoupled stable manifold, and it directly compares the chaotic case with the verified N=2 cases.","tokens_in":18954,"tokens_out":3090,"duration_ms":34530,"concrete_test":"For the N=10 chaotic attractor at epsilon=0.15 (Fig. 2I), compute, along a long trajectory, every time a unit's projected state (x_i,y_i) crosses the uncoupled stable manifold W^s(x_s^unc) of the saddle in the two-dimensional projection. For each crossing, evaluate the sign of (epsilon h_i) · n_i, where n_i is the normal to W^s at the crossing point, to determine whether the coupling vector points into the excitability region. Compare the fraction of reinjecting crossings (and the distribution of crossing locations near the saddle) with the same statistic for the N=2 LA-LA attractor at epsilon=0.065 and epsilon=0.15.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that a reinjection/trapping mechanism explains all newly created oscillatory attractors, including the chaotic one: the abstract says 'in all cases the reinjection mechanism is present,' and the Discussion repeats that the attractors 'emerge in a similar way.' The direct geometric evidence, however, is confined to the N=2 case in Sec. III D, Fig. 5: only the LA-LA periodic attractor, the LA-SA periodic attractor, and briefly the quasiperiodic torus are examined by overlaying the uncoupled stable manifold W^s(x_s^unc) and the coupling vectors. The N=10 chaotic attractor in Sec. III A, Fig. 2I is characterized only by Lyapunov exponents and a projection of the trajectory; no analogous check shows that each unit is repeatedly reinjected across W^s(x_s^unc) by the diffusive coupling. The load-bearing assumption is that the geometric mechanism identified for two units survives in the 20-dimensional chaotic attractor, where chaos could plausibly arise from desynchronization or from stretching/folding dynamics unrelated to the saddle's stable manifold. If that assumption fails, the existence result (multistability) stands, but the universal mechanism claim would have to be weakened to 'the mechanism is present in the N=2 cases and is consistent with, but not directly verified for, the larger-network and chaotic cases.' This is an internal-evidence concern, not a correctness objection: the numerics for existence appear reproducible and carefully documented, including XPPAUT continuations and Lyapunov-exponent verification.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript investigates how attractive diffusive coupling between excitable units, each possessing only a stable equilibrium when isolated, can create multistability of oscillations. Using a Hodgkin-Huxley-type model (Eqs. (1)-(7)) with parameters from Ref. 46, the authors report coexistence of periodic, quasiperiodic, and chaotic attractors in a random N=10 network (Fig. 2) and in the N=2 case (Fig. 3), even though individual units do not oscillate. They identify a 'trapping' mechanism: near the uncoupled saddle x_s^unc, where local dynamics is slow, the diffusive coupling can overcome the local flow and reinject each unit across the stable manifold W^s(x_s^unc) into the excitability region, thereby turning transient excitable excursions into sustained oscillations. Continuation analysis (Fig. 4 and Fig. 8) characterizes the bifurcations: the LA-LA attractor arises via a saddle-node of limit cycles and is destabilized via a Neimark-Sacker (torus) bifurcation, while the LA-SA attractor arises via a homoclinic bifurcation and disappears via a saddle-node of limit cycles. The paper concludes that this reinjection mechanism is present for all observed attractors, including the chaotic one.","tokens_in":19209,"tokens_out":9417,"duration_ms":85063,"significance":"The existence results are supported by extensive attractor searches, Lyapunov exponent calculations, continuation analysis, and publicly available code, and they make a valuable contribution by showing that a simple linear diffusive coupling can create a rich multistable set of oscillations from non-oscillating excitable units. The N=2 case provides a concrete and well-documented example of a mechanism that has been invoked more generally, and the parameter dependence is tested in a targeted way through x-only versus y-only coupling. The main significance is conceptual: the paper proposes trapping in transient regions as a route to coupling-induced multistability, complementing earlier work on Turing/Hopf oscillations and on trapping in chaotic saddles or canards. The strengths are reproducible numerics and careful bifurcation analysis; the main weakness is that the trapping mechanism is argued qualitatively from selected trajectories rather than quantified, and the claim that the mechanism applies to the chaotic N=10 attractor is not directly verified.","major_comments":[{"comment":"The central claim that the reinjection/trapping mechanism is present 'in all cases' (Abstract) and that all attractors 'emerge in a similar way' (Discussion) is not directly supported for the N=10 chaotic attractor. The geometric evidence in Sec. III D and Fig. 5 is restricted to N=2: the LA-LA periodic attractor at epsilon=0.065 and epsilon=0.15, the x-only-coupled LA-LA at larger epsilon_1, and the LA-SA attractor; the quasiperiodic torus is discussed only verbally. The N=10 chaotic attractor in Fig. 2I is characterized by Lyapunov exponents and a projected trajectory, but no check is shown that the coupling repeatedly reinjects each unit across W^s(x_s^unc) near the saddle. Since the chaotic attractor is desynchronized and could in principle originate from stretching/folding dynamics unrelated to the manifold geometry, the universal mechanism claim requires either a quantitative check for the chaotic/large-network case or a corresponding restriction of the claim in the abstract and discussion. This missing check is load-bearing for the paper's main explanatory claim.","section":"Abstract and Sec. IV; evidence in Sec. III A and Sec. III D"},{"comment":"The reinjection mechanism is asserted on the basis of visual inspection of selected trajectories: a trajectory 'crosses W^s(x_s^unc) in this projection' and is 'effectively reinjected' into the excitability region. As the authors correctly note, this crossing in projection is not a crossing of an invariant manifold of the coupled system. To make the claim testable and applicable uniformly to all attractors, including those in Fig. 2, the manuscript should define an operational diagnostic of reinjection—for example, the sign of the transverse component of the coupled vector field near x_s^unc, or the fraction of time each unit spends inside the excitability region compared with the uncoupled transient—and report this diagnostic for every quoted attractor. Without such a definition, 'the reinjection mechanism is present' remains an interpretation of the geometry rather than a demonstrated property.","section":"Sec. III D, especially Fig. 5"}],"minor_comments":[{"comment":"The relation between oscillation amplitude and degree is stated inconsistently: the text first says the amplitude is inversely proportional to the number of neighbors, but later, for the quasiperiodic two-unit case, it says the amplitude is proportional to the number of neighbors. Please reconcile the wording; the earlier statement and Fig. 2 suggest the second occurrence should read 'inversely proportional'.","section":"Sec. III A"},{"comment":"The text refers to 'Figs.3D1', 'Fig.3D2', and 'Fig.3D3' when discussing the LA-SA attractor; these should be 'Fig. 5D1', 'Fig. 5D2', and 'Fig. 5D3' to match the actual figure numbering.","section":"Sec. III D and Fig. 5"},{"comment":"The caption contains a typo: 'Neimarck-Sacker' should be 'Neimark-Sacker'.","section":"Fig. 4 caption"},{"comment":"The formula for m_infinity has an unmatched parenthesis: it should presumably read 1/(1+exp((m_h - x_i)/k_m)) rather than 1/(1+exp(m_h - x_i)/k_m).","section":"Eq. (6)"},{"comment":"The phrase 'A stronger coupling coupling term' contains a duplicated word and should be corrected.","section":"Sec. IV"},{"comment":"The statement that 'for a range roughly between epsilon=0.05 and epsilon=0.1, more than 50 attractors can be found' is not documented by any figure or table for the N=10 case; consider adding a count curve analogous to Fig. 7 or specifying the exact range and how the count was obtained.","section":"Sec. III A"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a solid numerical study with publicly available code, and the existence results for N=2 and N=10 are convincing. My main concern is that the title, abstract, and discussion promise a universal trapping mechanism, while the direct geometric verification is limited to N=2 non-chaotic attractors. I recommend asking the authors either to add a quantitative check for the N=10 chaotic attractor or to visibly soften the universal claim. This is a fixable gap rather than a fundamental error, so I do not see a need for rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core result here is real and worth knowing: attractive diffusive coupling of excitable units, each individually at a stable equilibrium, can create coexisting periodic, quasiperiodic, and chaotic oscillatory attractors. That contradicts the impression one might take from Yanagita et al. 2005, who saw multistability only for repulsive coupling. The paper documents this carefully. The attractor searches are extensive, the Lyapunov exponents and continuation analysis back the stated bifurcation scenarios, and the code is available. I especially like the targeted tests with x-only versus y-only coupling, which show the mechanism is not an artifact of symmetric coupling.\n\nThe geometric story for N=2 is convincing. The panels overlaying the uncoupled stable manifold and the coupling vectors show clearly how the trajectory gets reinjected near the saddle and trapped in the excitability region. That part is well supported and the explanation is intuitive without being sloppy.\n\nThe soft spot is the universal mechanism claim. The abstract says the reinjection mechanism is present in all cases, and the discussion says the chaotic attractor emerges in a similar way. But the direct geometric evidence is only shown for the N=2 periodic and quasiperiodic attractors. The N=10 chaotic attractor is characterized only by a projection and Lyapunov exponents. I do not think the mechanism is wrong, but it is an extrapolation. The chaotic attractor could in principle arise from a different route, for instance desynchronization or stretching and folding unrelated to the saddle's stable manifold. If that happened, the existence result would stand but the mechanism claim would need to be weakened. This is a fixable issue: either soften the \"in all cases\" language or add a direct reinjection check for the chaotic case.\n\nA minor point: the mechanism is presented qualitatively rather than proven. That is acceptable for a numerical study, though I would like the paper to be more explicit about what is demonstrated and what is inferred.\n\nOverall, this paper is a genuine contribution to the literature on coupling-induced oscillations and multistability. It deserves a serious referee. The referee should ask for the mechanism scope to be tightened, but the existence results and the N=2 analysis are reproducible and worth publishing.","headline":"Solid numerical demonstration that attractive coupling creates multistable oscillations in excitable units; the universal trapping-mechanism claim runs ahead of the direct evidence, but the existence result and the N=2 geometry work are worth a careful look.","tokens_in":19753,"tokens_out":1688,"would_cite":true,"duration_ms":17308,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["34C15","34C23","37C29","37G15","37N25"],"pacs":[],"model":"deepseek-v4-flash","headline":"Diffusive coupling alone can create coexisting oscillatory states in networks of excitable units that do not oscillate by themselves.","keywords":["multistability","excitable dynamics","diffusive coupling","trapping mechanism","homoclinic bifurcation","saddle-node bifurcation of limit cycles","quasiperiodic attractors","chaotic oscillations"],"falsifier":"Project a long trajectory on the chaotic attractor of the ten-unit network onto the $(x_i, y_i)$ plane of one unit and mark each pass near the uncoupled saddle point $x_s^{\\mathrm{unc}}$; if the coupling vector $\\varepsilon h_i$ does not point across the stable manifold $W^s(x_s^{\\mathrm{unc}})$ into the excitability region at those passes, then the trapping mechanism is not producing the chaos.","tokens_in":18734,"feed_emoji":"🧠","tokens_out":10868,"duration_ms":88118,"temperature":0.7,"pith_summary":"The paper asks what happens when excitable units that never oscillate on their own are coupled by an attractive diffusive interaction. It shows that the coupling can trap the units inside the excitability region of their individual state space, repeatedly reinjecting them there and turning their transient spiking into permanent oscillations. As a result, networks of two or ten units display multistability: a stable equilibrium coexists with periodic, quasiperiodic, and chaotic oscillatory attractors, all produced by the same trapping mechanism. This matters because attractive diffusive coupling is a common interaction in neuronal, ecological, and chemical networks, and the paper identifies a simple geometric mechanism by which it can generate a rich variety of coexisting behaviors.","feed_headline":"Diffusive coupling alone creates coexisting oscillations","feed_subtitle":"Only two diffusively coupled neurons, each quiet alone, produce oscillatory attractors; ten produce chaos too.","key_machinery":"The central mechanism is the trapping mechanism, in which the coupling vector $\\varepsilon h_i$ of a unit points across the stable manifold $W^s(x_s^{\\mathrm{unc}})$ of the saddle point $x_s^{\\mathrm{unc}}$ of the uncoupled local dynamics. Because the local vector field nearly vanishes near the saddle, the relative effect of the coupling is strongest there, so the trajectory crosses the stable manifold and re-enters the excitability region instead of converging to the stable node. This reinjection converts transient excitations into recurrent oscillations and is claimed to underlie all the coexisting attractors, including those born in saddle-node-of-limit-cycles and homoclinic bifurcations.","core_discovery":"The central discovery is that in a network of diffusively coupled excitable neurons, the diffusive coupling can overcome the local dynamics in the slow region near the saddle point of each unit, pushing the trajectory across the stable manifold of that saddle and back into the excitability region. Once there, the unit performs a large-amplitude excitation instead of converging to the stable equilibrium; the units then repeatedly reinject each other, so the transient spiking of the uncoupled system becomes a permanent oscillation. This trapping mechanism generates the attractors regardless of the bifurcation that creates them: periodic attractors appear through either a saddle-node bifurcation of limit cycles or a homoclinic bifurcation, and the same reinjection is seen for the quasiperiodic torus. In the ten-unit network the mechanism yields up to 84 coexisting attractors for intermediate coupling strengths, including solitary states, two-unit cluster states, a torus, and a chaotically spiking state.","pith_inferences":["If the trapping mechanism is generic, then any excitable system whose state space contains a saddle whose stable manifold separates a region of long transients should be able to host multistable oscillations under attractive coupling, not just the specific neuron model studied here.","The mechanism suggests a practical diagnostic: to decide whether experiments or simulations are seeing trapping-induced multistability, look for trajectories that spend long periods near the saddle and cross its stable manifold at each cycle.","The same reasoning applied to ecological metapopulations predicts that diffusive migration of prey between excitable predator-prey patches should generate coexisting oscillatory attractors of the type described here.","If the chaotic states indeed share the trapping mechanism, then coexistence of chaotic and regular attractors may be a robust property of excitable networks, which would affect predictions of noise-induced tipping between behavioral regimes."],"forward_implications":["A purely attractive diffusive coupling, with no repulsive or asymmetric terms, is sufficient to turn a quiescent excitable network into a multistable oscillator network.","For ten units, the coexisting attractors include periodic solitary states, two-unit cluster states, a quasiperiodic torus, and a chaotic state, with up to 84 attractors found at intermediate coupling strengths.","The same trapping mechanism is compatible with different bifurcation routes: the periodic attractors are born either in saddle-node bifurcations of limit cycles or in homoclinic bifurcations, and the torus appears via a Neimark-Sacker bifurcation.","The amplitude of an oscillating unit decreases as its number of neighbors grows, because a stronger coupling term pulls the trajectory away from the saddle more effectively.","Coupling restricted to the x-direction alone still produces the large-amplitude periodic attractor, while the asymmetric large-amplitude / small-amplitude attractors require the y-direction."],"supporting_citations":[{"why":"Supplies the excitable neuron model with three equilibria and defines the excitability region that the trapping mechanism relies on.","marker":"[46]"},{"why":"Classic result showing diffusion can create oscillations from non-oscillating units with a single stable equilibrium; the present setting adds a saddle and multi-equilibria.","marker":"[42]"},{"why":"Recent rigorous conditions for chaos from diffusive coupling; serves as the benchmark the paper distinguishes its trapping-based chaos from.","marker":"[45]"},{"why":"Numerical framework for global stability analysis used to find and classify the coexisting attractors reported here.","marker":"[55]"},{"why":"Monte Carlo basin bifurcation analysis method, one of the algorithms the paper uses to enumerate attractors.","marker":"[48]"},{"why":"Prior example of coupling-induced trapping in chaotic saddles; the paper extends this to multistability of periodic, quasiperiodic and chaotic oscillations.","marker":"[49]"}],"fun_headline_variants":["Trapping mechanism turns quiet neurons into oscillators","Diffusive coupling traps neurons into multistable oscillations","Quiet units, coupled, yield chaos and coexisting oscillations","Coupling alone produces multistable oscillations in quiet neurons"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reinjection across the stable manifold is verified geometrically for two coupled units in the periodic and quasiperiodic attractors, and the paper assumes it also generates the chaotic attractor and all attractors in the ten-unit network.","fun_headline_variants_meta":{"raw":{"variants":["Trapping mechanism turns quiet neurons into oscillators","Diffusive coupling traps neurons into multistable oscillations","Quiet units, coupled, yield chaos and coexisting oscillations","Coupling alone produces multistable oscillations in quiet neurons"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000504,"raw_usage":{"total_tokens":2442,"prompt_tokens":907,"completion_tokens":1535,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":523,"completion_tokens_details":{"reasoning_tokens":1470}},"tokens_in":523,"tokens_out":1535,"duration_ms":11310,"temperature":1.0,"reasoning_tokens":1470,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:30:16.127755+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Project a long trajectory on the chaotic attractor of the ten-unit network onto the $(x_i, y_i)$ plane of one unit and mark each pass near the uncoupled saddle point $x_s^{\\mathrm{unc}}$; if the coupling vector $\\varepsilon h_i$ does not point across the stable manifold $W^s(x_s^{\\mathrm{unc}})$ into the excitability region at those passes, then the trapping mechanism is not producing the chaos.","supporting_citations":[{"cited_title":"Dynamical systems in neuroscience","cited_arxiv_id":null,"evidence_quote":"Supplies the excitable neuron model with three equilibria and defines the excitability region that the trapping mechanism relies on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Classic result showing diffusion can create oscillations from non-oscillating units with a single stable equilibrium; the present setting adds a saddle and multi-equilibria."},{"cited_title":"Queiroz, and Dmitry Turaev","cited_arxiv_id":null,"evidence_quote":"Recent rigorous conditions for chaos from diffusive coupling; serves as the benchmark the paper distinguishes its trapping-based chaos from."},{"cited_title":"Framework for global stability analysis of dynamical systems","cited_arxiv_id":null,"evidence_quote":"Numerical framework for global stability analysis used to find and classify the coexisting attractors reported here."},{"cited_title":"Monte Carlo basin bifurcation analysis","cited_arxiv_id":null,"evidence_quote":"Monte Carlo basin bifurcation analysis method, one of the algorithms the paper uses to enumerate attractors."},{"cited_title":"Medeiros, Rene O","cited_arxiv_id":null,"evidence_quote":"Prior example of coupling-induced trapping in chaotic saddles; the paper extends this to multistability of periodic, quasiperiodic and chaotic oscillations."}],"review_version":1}