{"id":"14ce4e3f-e61f-429e-bf65-3cf42bcc8cc9","arxiv_id":"2507.17443","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"In a leaky integrate-and-fire network, slow Hebbian-Oja adaptation of couplings sweeps the system through chimera and bump states with changing domain multiplicity, while fast adaptation suppresses these transients.","lead":"This paper simulates a ring of spiking neurons whose connection strengths change over time using a Hebbian rule and finds that slow changes let the network pass through chimera and bump states of changing size and number. Fast changes skip these intermediate states and go straight to the final synchronized or desynchronized regime.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claimed adiabatic passage through frozen-coupling chimera/bump states is unverified: no comparison with constant-coupling attractors at matched σeff, and the nonuniform, history-dependent weights make the mean coupling an insufficient state descriptor.","rationale":"I read the paper as a numerical study of a concrete model; the slow versus fast timescale contrast is interesting, and the reported spatiotemporal patterns are visually plausible. The load-bearing condition for the headline claim is the adiabatic or continuation assumption: the slowly drifting, nonuniform couplings must track the attractors of the corresponding constant-coupling system. This is exactly the reader's weakest_assumption, and I find it genuine. Section VI explicitly compares the co-evolution to continuation, but no timescale-separation estimate, no convergence statistic, and no frozen-coupling benchmark is supplied. The nonuniformity of the weights (Sec. V, Fig. 9) means the mean coupling is not a sufficient parameter: two systems with the same mean σeff can have different dynamics, so the statement that domain size and multiplicity follow the average coupling is not supported by the mean alone. The proposed frozen-coupling control at matched mean and matched snapshot would settle whether the adaptive states coincide with genuine constant-coupling chimera and bump states. I also note the abstract's timescale wording is reversed relative to the simulations (large τσ corresponds to slow coupling), an internal inconsistency that should be corrected regardless. Since the paper's data do not rule out the alternative interpretation, the reader's CONDITIONAL verdict is appropriate; my read does not change it.","tokens_in":16401,"tokens_out":4138,"duration_ms":47511,"concrete_test":"At t=3000 and t=4200 of the slow run in Fig. 1 (reported two-headed and single-headed chimeras), read off the instantaneous network mean σeff(t). Re-initialize the same LIF ring (N=1024, R=350, same parameters) with uniform constant couplings set to those means, starting from (i) the same potential snapshot and (ii) fresh random potentials, and integrate the frozen system (Eqs. (3a)-(3b) with constant σ) for several thousand TUs. Count the number of incoherent domains and compare r(t). If the frozen uniform system at the matched mean does not reproduce the same multiplicity and domain structure, the slow-adaptive trajectory is not an adiabatic continuation through the constant-coupling phase diagram, and the claim should be weakened to 'transients resembling chimeras appear during adaptation.'","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that during slow Oja adaptation the LIF network 'transits through chimera state regimes with different multiplicity' whose size and multiplicity follow the average coupling strength—requires that at each instant the network is near the attractor of the frozen-coupling system with the same mean coupling. This is the continuation assumption the authors invoke in Sec. VI, but it is never tested. Three gaps make it load-bearing. (1) Eq. (3c) with τσ=1000 produces a slowly drifting but nonuniform coupling matrix; Sec. V shows the weights are not identical, and in the chimera-to-bump direction their distribution broadens considerably (Fig. 9). Hence the scalar mean σeff(t) does not uniquely specify the dynamical system, and 'following the average coupling' is only a correlation, not a demonstrated functional relation. (2) The observed patterns are single realizations from random initial potentials; no ensemble, no convergence check, and no comparison with the constant-coupling phase diagram at matched mean coupling. (3) The abstract states the timescale condition backwards ('time scales governing the link dynamics are relatively small' versus the large-τσ slow case actually used), which further obscures which regime the claim applies to. If the frozen system at σeff≈-0.3 does not itself exhibit a two-headed chimera, then the adaptive state is an adaptive transient, not a transition through the constant-coupling chimera regime.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies a ring of N=1024 leaky integrate-and-fire neurons with nonlocal coupling (Eq. 3a) whose synaptic strengths evolve according to the Hebbian/Oja rule (Eq. 3c). It reports that when the weight dynamics are slow (large τσ, here 1000), the mean effective coupling σ_eff drifts monotonically from positive to negative (or vice versa), and the network is claimed to pass through chimera states with different numbers of incoherent domains and through bump states with different numbers of active and subthreshold domains; the size and multiplicity of these domains are said to follow the evolution of the average coupling strength. When the weight dynamics are fast (τσ = 2), these intermediate hybrid states are suppressed and the system quickly reaches its asymptotic state. The paper interprets the slow case as a kind of continuation process along the drifting coupling parameter.","tokens_in":16734,"tokens_out":4754,"duration_ms":53565,"significance":"If the central claim were established quantitatively, the paper would offer a useful demonstration that chimera and bump states can arise naturally during slow synaptic adaptation in a spiking network, rather than only under carefully prepared constant-coupling conditions. The qualitative observation of state changes in a single trajectory is suggestive, and the comparison between slow and fast adaptation time scales is conceptually interesting. However, the current evidence is based on single realizations and visual classification, the key adiabatic assumption is not tested, and the scalar mean coupling is not shown to be a sufficient descriptor of the nonuniform coupling matrix; the significance of the claimed transitions therefore remains unproven.","major_comments":[{"comment":"The central interpretation that the adaptive network 'transits through' chimera regimes with different multiplicity relies on the unstated assumption that, during slow adaptation, the network remains close to the attractor of the constant-coupling system with the same instantaneous mean coupling. This continuation assumption is explicitly invoked in Sec. VI but never tested. I ask the authors to compare the adaptive trajectories against simulations of the frozen-coupling system with uniform σ_eff equal to the instantaneous mean from Eq. (5b) at matching times. Without such a comparison, the observed spacetime patterns could be adaptive transients unique to the time-varying coupled system, and the phrase 'transits through' is not supported.","section":"Sec. VI (and Figs. 1-2)"},{"comment":"Every scenario description is based on a single realization starting from one draw of random initial potentials; no ensemble statistics, standard deviations, or error bars are provided. The claims that the number and size of coherent/incoherent domains 'follow' the average coupling strength (Abstract, Sec. III) require quantitative definitions of domain multiplicity and domain boundaries, such as a local Kuramoto order parameter with a threshold or firing-rate-based criteria, applied over many initial conditions. As presented, the trajectory-level narrative is anecdotal and cannot support the general quantitative claim.","section":"Sec. III, Figs. 1-2"},{"comment":"The time-scale condition is stated backwards in the abstract: the paper's slow-adaptation regime corresponds to large τσ = 1000 (Sec. IV), where the link dynamics are slower than the potential dynamics, yet the abstract says transitions occur 'provided that the time scales governing the link dynamics are relatively small compared to the time scales of the potential evolution.' This contradicts the model setup and should be corrected to refer to slow (large-τσ) link evolution; the claim as written would falsely predict transitions only in the fast-adaptation case.","section":"Abstract (and Sec. IV)"},{"comment":"Figures 8-9 show that the coupling matrix is strongly nonuniform and history-dependent, with broad distributions of σ_eff values during the chimera-to-bump transition. This undermines the use of the scalar mean σ_eff(t) (Eq. 5b) as the state descriptor that controls the observed chimera multiplicity. The paper should either demonstrate that the dynamics depend on the mean alone (e.g., by comparing against a system with uniform but time-varying coupling at the same mean), or replace the scalar 'following' statement with a more precise description that accounts for the distribution width and spatial structure of the couplings.","section":"Sec. V B, Figs. 8-9"},{"comment":"The manuscript does not report the numerical integrator, time step, event-detection method for the reset condition (Eq. 3b), or the scheme used to integrate the stiff coupling equation (Eq. 3c). Because the model is event-driven, the discretization and the handling of simultaneous or near-simultaneous threshold crossings can affect whether chimera and bump transients appear. The authors should describe the algorithm in sufficient detail and, ideally, provide code or data to support reproducibility; the current 'data available upon request' statement is not enough for a computational study.","section":"Sec. II (numerical methods)"}],"minor_comments":[{"comment":"The phase definition θ_j = 2π u_j / u_th is introduced without explanation; please clarify that it is a linear rescaling of the potential used for the Kuramoto order parameter and is not the natural phase of the LIF oscillator, and note how this choice affects r values near reset.","section":"Sec. II C, Eq. (7)"},{"comment":"The linear fit τ_steady = A τσ + B with A = 6.4, B = 5.3 is presented as empirical, and the claim that B should vanish is argued from τσ = 0. Since the data have a quoted error of ±10, this consistency check is fine, but it should be phrased as a compatibility statement rather than a derivation.","section":"Sec. IV, Fig. 5"},{"comment":"The first sentence 'Adaptive link sizes is a major breakthrough step in evolving networks' is awkwardly phrased and contains a grammatical error; please revise for clarity.","section":"Abstract"},{"comment":"In the description of Fig. 6, the sentence 'the uncoupled connections are all colored brown-red' refers to zero-weight entries; please clarify that the color scale includes σ_eff = 0 and that this component of the matrix is time-invariant.","section":"Sec. V A"},{"comment":"References [17] and [19] both list the article number 033146 (Entropy 35 and Chaos 35, respectively); please verify that these are correct and not a typographical duplication.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses an interesting question and the qualitative observations are plausible, but the technical gap is the missing verification of the continuation assumption and the lack of quantitative, ensemble-based evidence. These issues are fixable within the scope of a numerical study: adding a frozen-coupling comparison, ensemble statistics, and quantitative chimera/bump measures would convert the paper from an anecdotal report into a defensible claim. I do not see grounds for rejection, because the observation of intermediate states under slow adaptation is likely reproducible even if the interpretation needs refinement. I would also encourage the editor to request that the authors make their integration scheme explicit, given the event-driven nature of the model."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague—\n\nThis paper does something genuinely new: it puts Oja-rule adaptive coupling on a nonlocally coupled LIF ring and shows that when the link dynamics are slow, the network sweeps through chimera and bump states of different multiplicity on the way to its asymptotic state, while fast adaptation skips those transients. That contrast is the real contribution, and the coupling-matrix distributions in Sec. V give it more texture than the usual order-parameter plots. The work is clearly presented and the model is well-defined.\n\nThe soft spots are real but fixable. The main interpretive claim—that the system 'transits through chimera state regimes' and that domain size/multiplicity 'follow the average coupling strength'—is not actually demonstrated. Every scenario is a single realization, there are no ensemble statistics, and there is no comparison with the frozen-coupling phase diagram at matched mean coupling. That last gap is load-bearing: the weights are nonuniform and history-dependent, so the scalar mean coupling does not uniquely specify the system. The stress-test challenge is on point. If the constant-coupling system at σeff ≈ −0.3 does not itself show a two-headed chimera, then the adaptive pattern is an adaptive transient, not a passage through the frozen-coupling regime. The paper's Sec. VI analogy to continuation is suggestive but not a substitute for checking.\n\nTwo smaller things. The abstract states the timescale condition backwards: it says the link dynamics should be 'relatively small' compared to the potential dynamics, but the slow-adaptation case that produces the transitions uses τσ = 1000, much larger than the potential timescale. That needs fixing. And the numerical details are thin—no integrator, no tolerance, no code or data beyond 'available upon request.' For a claim built on visual classification of chimera multiplicity, that is too little.\n\nI don't think any of this sinks the paper. The observation is plausible and interesting, and the slow/fast contrast is likely robust. But the central claim as worded needs the frozen-coupling comparison and some ensemble statistics before I'd treat it as established. It deserves a serious referee; I'd send it to review with a request for major revision.","headline":"Novel adaptive-LIF result with a real gap between observation and interpretation; deserves review but needs the frozen-coupling comparison and ensemble statistics.","tokens_in":17249,"tokens_out":2375,"would_cite":true,"duration_ms":25774,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"In a ring of leaky integrate-and-fire neurons with Hebb-Oja adaptive couplings, slow weight evolution makes the network pass through chimera states of changing multiplicity and through bump states with different numbers of active domains…","keywords":["leaky integrate-and-fire neurons","chimera states","bump states","adaptive coupling","Hebbian learning","Oja learning rule","Kuramoto order parameter","synchronization transitions"],"falsifier":"Run the same LIF ring with the couplings frozen at the instantaneous mean effective value $\\sigma_{\\rm eff}(t)$ at each time and compare the spacetime plots and Kuramoto order parameter; if the adaptive trajectory's intermediate chimera multiplicities and bump configurations are not reproduced by the frozen runs with the same mean coupling, the claimed transitions between constant-coupling regimes are not established.","tokens_in":16198,"feed_emoji":"🧠","tokens_out":7108,"duration_ms":70175,"temperature":0.7,"pith_summary":"The paper asks what happens to hybrid synchronization states—chimera states with coexisting coherent and incoherent domains, and bump states with active and subthreshold regions—when the coupling strengths between neurons are not fixed but evolve by a Hebbian learning rule with Oja normalization. It studies a one-dimensional ring of leaky integrate-and-fire neurons with nonlocal diffusive coupling and argues numerically that when the couplings evolve slowly relative to the potentials, the mean effective coupling drifts across zero and the network visits chimera states of different multiplicity and bump states with different numbers of active domains along the way. When the couplings evolve on the same timescale as the potentials, the network jumps directly to the final asymptotic state and these intermediate hybrid states are not observed. A sympathetic reader would care because it suggests that partial synchronization states are not artifacts of hand-tuned constant coupling, but states a plastic network passes through as its synapses slowly strengthen or weaken.","feed_headline":"Slow synapses steer spiking networks through chimera states","feed_subtitle":"With slow Hebb-Oja plasticity, a 1D spiking ring visits chimera and bump states transiently; fast plasticity skips them.","key_machinery":"The load-bearing object is the time-dependent coupling matrix $\\sigma_{jk}(t)$ with the Oja-normalized Hebbian rule $d\\sigma_{jk}/dt = (1/\\tau_\\sigma)(u_j u_k - \\alpha \\, u_j u_j \\, \\sigma_{jk})$, which drives the mean effective coupling $\\sigma_{\\rm eff}(t)$ monotonically toward $c_u/\\alpha$ while keeping the weights finite. The key control is $\\tau_\\sigma$, the ratio of the coupling-evolution timescale to the potential-evolution timescale. For $\\tau_\\sigma=1000$ the mean coupling sweeps slowly enough for the system to relax into each intermediate state; for $\\tau_\\sigma=2$ the sweep is too fast and the network lands directly in the asymptotic state. The Kuramoto order parameter $r(t)$ is the diagnostic that marks the transitions, especially the abrupt drop as $\\sigma_{\\rm eff}$ crosses zero.","core_discovery":"The central claim is that an adaptive LIF network governed by the coupled potential and coupling equations crosses genuine synchronization regimes as the average effective coupling $\\sigma_{\\rm eff}(t)$ changes: starting from $\\sigma_{\\rm eff}=+2.1$ (bump regime) and ending at $\\sigma_{\\rm eff}=-0.7$ (chimera regime), the spacetime evolution shows a traveling bump state, then a drop in the Kuramoto order parameter near $\\sigma_{\\rm eff}=0$, then a two-headed chimera, a single-headed chimera, and finally a disorganized fluctuating phase; the reverse route from $-2.1$ to $+0.7$ passes from a two-headed chimera through incoherence to a one-headed chimera and then to two-bump and multi-bump states. The size and multiplicity of the coherent and incoherent, or active and subthreshold, domains change following $\\sigma_{\\rm eff}(t)$. The authors propose that slow adaptation acts like a continuation method, dragging the system through the states of the frozen-coupling system, whereas fast adaptation with $\\tau_\\sigma=2$ suppresses all intermediate states.","pith_inferences":["A direct test the authors did not report: freeze the couplings at the current mean effective value at each instant and compare the resulting spacetime pattern with the adaptive run; agreement would confirm the continuation picture, while disagreement would show that the nonuniform, history-dependent weights themselves create the observed multiplicities.","If the continuation picture is right, the frozen-coupling phase diagram of the LIF ring could be used predictively: the sequence of chimeras and bumps should be readable from the trajectory of $\\sigma_{\\rm eff}(t)$ alone.","The asymmetry between the two drift directions implies that an adaptive network's synchronization history is path-dependent; one could test this by reversing the sign of $c_u$ midway through a slow run and checking whether the state returns along a different sequence.","The same slow-adaptation mechanism could be probed in other oscillator families, such as FitzHugh-Nagumo or Kuramoto networks, to see whether the traversal of hybrid states is generic or specific to integrate-and-fire dynamics."],"forward_implications":["If the claim holds, chimera and bump states are not confined to networks with carefully tuned constant coupling; they appear as transient stages of a plastically adapting spiking network.","Slow synaptic adaptation expands the repertoire of synchronization states the network can visit, while fast adaptation collapses that repertoire to the final asymptotic state.","The route from positive to negative effective coupling is not the time-reverse of the route from negative to positive: the two directions pass through different sequences of chimera multiplicities and bump configurations.","The Kuramoto order parameter provides a sharp, easily measurable marker of the crossing of zero effective coupling, which could be used to detect these transitions in simulation or experiment.","The time needed to reach the asymptotic coupling state scales linearly with $\\tau_\\sigma$, so the duration of each intermediate synchronization regime can be controlled by adjusting the plasticity timescale."],"supporting_citations":[{"why":"Supplies the baseline result that a constant-coupling LIF ring supports bump states and chimera states, the regimes the adaptive network is claimed to traverse.","marker":"[15]"},{"why":"Reports chimera states in the LIF model with constant inhibitory coupling, anchoring the one- and two-headed chimeras seen during adaptation.","marker":"[14]"},{"why":"Introduces the Oja normalization term that keeps Hebbian coupling growth bounded and drives couplings to the finite steady state used here.","marker":"[30]"},{"why":"Provides the further formulation of the Oja rule that justifies the specific form of the learning equation.","marker":"[31]"},{"why":"Establishes the Kuramoto order parameter as the synchrony measure used to detect the transitions.","marker":"[47]"}],"fun_headline_variants":["Slow Hebbian adaptation drags neurons through chimera states","Adaptive synapses unveil chimera transitions in spiking rings","Time-scale separation unlocks chimera multiplicity in LIF networks","Slow plasticity reorders synchronization regimes in adaptive rings","Adaptive coupling steers spiking networks across chimera and bump states"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that during slow coupling drift the network stays near the state that the constant-coupling system would have at the current mean effective coupling, even though the individual couplings are nonuniform and remember their history.","fun_headline_variants_meta":{"raw":{"variants":["Slow Hebbian adaptation drags neurons through chimera states","Adaptive synapses unveil chimera transitions in spiking rings","Time-scale separation unlocks chimera multiplicity in LIF networks","Slow plasticity reorders synchronization regimes in adaptive rings","Adaptive coupling steers spiking networks across chimera and bump states"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000444,"raw_usage":{"total_tokens":2327,"prompt_tokens":1104,"completion_tokens":1223,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":720,"completion_tokens_details":{"reasoning_tokens":1149}},"tokens_in":720,"tokens_out":1223,"duration_ms":10863,"temperature":1.0,"reasoning_tokens":1149,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T14:47:32.298718+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same LIF ring with the couplings frozen at the instantaneous mean effective value $\\sigma_{\\rm eff}(t)$ at each time and compare the spacetime plots and Kuramoto order parameter; if the adaptive trajectory's intermediate chimera multiplicities and bump configurations are not reproduced by the frozen runs with the same mean coupling, the claimed transitions between constant-coupling regimes are not established.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the baseline result that a constant-coupling LIF ring supports bump states and chimera states, the regimes the adaptive network is claimed to traverse."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Reports chimera states in the LIF model with constant inhibitory coupling, anchoring the one- and two-headed chimeras seen during adaptation."},{"cited_title":"Oja, Journal of Mathematical Biology 15, 267 (1982)","cited_arxiv_id":null,"evidence_quote":"Introduces the Oja normalization term that keeps Hebbian coupling growth bounded and drives couplings to the finite steady state used here."},{"cited_title":"Oja, International Journal of Neural Systems 1, 61 (1989)","cited_arxiv_id":null,"evidence_quote":"Provides the further formulation of the Oja rule that justifies the specific form of the learning equation."}],"review_version":1}