{"id":"5d7ea182-c403-4322-8cb9-dc68c1682d4e","arxiv_id":"2508.04354","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":2,"one_line_summary":"Partially symmetric synaptic connectivity in balanced spiking networks can produce slow, non-ergodic collective dynamics.","lead":"This paper uses computer simulations of spiking neuron networks to argue that a realistic feature of brain wiring, partially symmetric connections, can make network activity slow and long-lasting, even breaking the usual assumption of ergodicity. A reader might care because it offers a candidate explanation for why the brain shows slow fluctuations on timescales much longer than individual neurons.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The ergodicity-breaking claim rests on finite-time finite-size simulations; without finite-size/time scaling it could be metastable transient, not non-ergodicity.","rationale":"Only the abstract was available, so I cannot check equations or simulations. The strongest claim as stated is about a regime where initial-state memory persists 'for very long times and ergodicity is broken.' In any finite stochastic network, all memories are transient; the key scientific question is whether the relaxation time diverges with system size (or at least exceeds all observed windows in a way that is robust). The reader's weakest_assumption exactly captures this: finite-time/finite-size artifacts. I agree with that read. The concrete test would separate a genuine slow phase from a transient by looking at scaling of switching time with N. If the scaling is absent, the central claim should be weakened to 'slow transients in finite networks.' This does not change the reader's UNVERDICTED verdict, because the evidence needed is not in the abstract; if forced to choose a final verdict on the current information, no verdict can be reached.","tokens_in":613,"tokens_out":4728,"duration_ms":52296,"concrete_test":"Run the same balanced spiking network for N = 500, 1000, 2000, 4000, 8000, rescaling synaptic weights so the effective self-coupling remains constant (e.g., J ~ g/√N). For each N, initialize two replicas in distinct basins of the order parameter (e.g., population-rate or first-PC coordinate) and compute the mean first-passage time τ_switch(N) for one replica to enter the other basin, over durations at least 10^5 membrane time constants. If τ_switch fails to grow with N or the replica-overlap decays to zero at the largest N, the 'broken ergodicity' is a finite-size/transient effect. If τ_switch grows as N^α with α>0 and exceeds the simulation span at N=8000, the claim passes this one check.","verdict_should_be":"UNCHANGED","load_bearing_attack":"On the evidence provided (the abstract), the central inference is from long-lived fluctuations in finite spiking networks to 'ergodicity is broken.' Broken ergodicity is an asymptotic property: for finite systems, any initial-state memory is transient, so the observed slow dynamics must be shown to persist in the thermodynamic or infinite-time limit. The abstract does not report finite-size scaling, escape-time growth, or a control with symmetric/random connectivity. The proposed mechanism (partial symmetry creating an effective excitatory self-coupling) could produce multiple equilibrium states in a finite realization, but in the N→∞ limit these states can be washed out by fluctuations; hence the 'multiple equilibria' and 'memory of initial state' claims are exactly the ones needing a scaling test. Additionally, 'ergodicity is broken' requires an invariant distribution to be defined; if the network is not stationary, 'memory of the initial state' may just be slow non-stationarity. These are standard, falsifiable requirements, not an accusation. If the full text contains scaling analyses and stationarity checks, the concern is resolved; from the submitted abstract it is not.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript, as available to me, consists solely of the abstract of arXiv:2508.04354. It claims that slow dynamics in cortical activity can be explained by partial symmetry in the synaptic connectivity of spiking neural networks, which generates an effective excitatory self-coupling. This coupling is said to produce long-lived fluctuations, and, when strong enough, multiple equilibrium states, initial-state memory, and broken ergodicity. No equations, simulation parameters, controls (e.g., symmetric vs. random connectivity), or quantitative results are presented in the submitted text.","tokens_in":924,"tokens_out":2730,"duration_ms":32090,"significance":"If substantiated, the claim would provide a novel and potentially important mechanism for slow cortical timescales and would bridge experimental observations of partial connectivity symmetry with network dynamics. The abstract also articulates a falsifiable prediction: that broken ergodicity appears only for sufficiently strong effective self-coupling. However, the present submission contains only the abstract, so the technical content and evidence base are absent. The significance is therefore conditional on a full manuscript that is not currently before me.","major_comments":[{"comment":"The manuscript provides no methods, equations, or parameter definitions. The terms \"partial symmetry\", \"effective excitatory self-coupling\", and \"dynamically-balanced network\" are used without formal definitions. The central causal claim—that partial symmetry generates an effective self-coupling that leads to slow dynamics—is asserted but not demonstrated. Without the full text, the claim is unsupported.","section":"Abstract (entire)"},{"comment":"The claim \"ergodicity is broken\" is an asymptotic statement about the invariant measure. Finite-time simulations of finite networks cannot establish broken ergodicity unless supported by finite-size scaling (e.g., divergence of escape times or mixing times with system size) and explicit checks for stationarity. The abstract reports no such analysis, so the load-bearing inference from observed long-lived fluctuations to non-ergodicity is not justified on the evidence provided.","section":"Abstract, final sentence"},{"comment":"The statement that \"long-lived fluctuations\" persist \"for very long times\" is qualitative. No data, error bars, or comparison with a control (e.g., fully random connectivity) are shown. Without these, the slow dynamics could be a finite-size or finite-time transient. The distinction is central to the paper's thesis.","section":"Abstract, sentence 4"},{"comment":"The abstract does not mention any limitations or the need for future work. Given that the claims about ergodicity are inherently asymptotic, the absence of a scaling analysis or a statement about the thermodynamic limit is a notable omission. This missing support must be supplied before the central claims can be evaluated.","section":"Abstract (missing limitation statement)"}],"minor_comments":[{"comment":"The abstract is clearly written and the narrative is easy to follow. However, it would benefit from a citation for the experimental evidence of partial symmetry in local cortical networks, and from a one-sentence definition of 'effective self-coupling' to orient the reader.","section":"Abstract (style)"}],"recommendation":"uncertain","confidential_remarks":"The submission as provided to me contains only the abstract; the full text is empty. This may be a technical error in the review package. If the full text is available, I will be happy to review it. Based solely on the abstract, I cannot assess the validity of the claims. The authors should be asked to provide the complete manuscript, including simulation details and scaling analyses, before a soundness judgment can be made."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Interesting mechanism, but the core claim is under-supported by the abstract. The idea—partial symmetry in balanced spiking networks creates an effective excitatory self-coupling that slows the dynamics, and with enough strength produces multiple equilibria and broken ergodicity—is genuinely original as far as I know. It's the kind of clean, testable mechanism that could be cited for years if the simulations back it up.\n\nWhat the paper does well: the abstract is honest. It says 'numerical study,' it states the connectivity feature is observed experimentally, and it doesn't oversell a direct match to data. The mechanism is concrete enough to simulate and the predicted regime is distinct from the usual balanced-network picture. That's worth referee time.\n\nThe soft spot is exactly what the stress-test note says: ergodicity breaking is an asymptotic property. Long-lived single-realization fluctuations can just be a long transient on a finite network. To make the claim stick, the paper needs finite-size scaling showing that the escape time grows with system size (or doesn't, in a way that still leaves multiple attractors), a control with symmetric or random connectivity, and a stationarity check. The abstract shows none of that. 'Multiple equilibrium states' also needs the thermodynamic limit to make sense; a finite network can have many stable states that are washed out as N grows. I'd also want a precise definition of ergodicity breaking for this stochastic spiking network—non-uniqueness of invariant measure, or just slow mixing? The abstract doesn't say.\n\nThis isn't an accusation that the full paper is missing these controls. It may well have them. But on the evidence here, I can't verify the central claim, and the reader's low-soundness score is fair. I'd still send it to review, because the question is important and the proposed mechanism is specific enough to be tested by a careful referee.\n\nBottom line: worth a serious look, but require scaling analysis and a clear definition of ergodicity before accepting. I wouldn't cite it until I see the full-text methods.","headline":"A promising mechanism for slow cortical dynamics, but the ergodicity-breaking claim needs finite-size/time scaling that the abstract doesn't show; worth reviewing.","tokens_in":1313,"tokens_out":2403,"would_cite":false,"duration_ms":24395,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that slow dynamics and broken ergodicity emerge in balanced spiking networks from partial symmetry in synaptic connectivity.","keywords":["spiking neurons","balanced networks","partial symmetry","slow dynamics","ergodicity breaking","long-lived fluctuations","multiple equilibria","cortical dynamics"],"falsifier":"Run the same network architecture at increasing system sizes and simulation durations. If the overlap of the activity with the initial state decays to zero for sufficiently large systems or long times, so that the apparent multiple equilibria are merely slow transients, then the claimed ergodicity breaking is not a true thermodynamic property.","tokens_in":564,"feed_emoji":"🧠","tokens_out":2892,"duration_ms":32847,"temperature":0.7,"pith_summary":"This paper tries to establish that slow, long-lived fluctuations in neural activity need no special slow membrane or synaptic mechanisms: they can arise purely from partial symmetry in how neurons are wired. In simulated spiking networks that maintain a balance of excitation and inhibition, a partially symmetric connectivity acts as an effective excitatory self-coupling of the population. This coupling generates long-lived collective fluctuations, and when strong enough, produces multiple equilibrium states and a memory of the initial condition that persists for very long times. Because partial symmetry is a known feature of local cortical circuits, the result offers a plausible, minimal explanation for the surprisingly slow dynamics observed in cortex.","feed_headline":"Partial synaptic symmetry explains slow cortical dynamics","feed_subtitle":"A new mechanism for the brain's slow activity emerges from wiring symmetry alone.","key_machinery":"The central object is partially symmetric synaptic connectivity, meaning the weight matrix can be decomposed into a symmetric and an antisymmetric part. The symmetric part acts as a positive feedback or effective excitatory self-coupling of the population activity, and this is the mechanism that carries the argument. Tuning the strength of this self-coupling controls the timescale of fluctuations and, at sufficiently high values, induces multi-stability and ergodicity breaking.","core_discovery":"The central claim is that in a dynamically balanced network of spiking neurons, partial symmetry of the synaptic matrix creates an effective excitatory self-coupling at the population level. This self-coupling is sufficient to produce slow fluctuations whose relaxation times vastly exceed the single-neuron time constants. As the effective self-coupling grows, the network undergoes a change from a single fluctuating equilibrium to multiple coexisting equilibria; the system then retains memory of its initial state and does not sample all configurations, so ergodicity is broken. The paper demonstrates this through numerical simulation of spiking neuron networks.","pith_inferences":["A direct experimental extension would be to engineer in vitro spiking networks with controlled partial symmetry and test whether their activity correlation times grow with the symmetric component of the connectivity.","The effect may be a general property of dynamical systems rather than a spiking-specific one; if so, analogous slow dynamics and ergodicity breaking should appear in rate-based networks with partially symmetric interactions.","The numerical claim of broken ergodicity could be sharpened by a finite-size scaling analysis: if the apparent memory of the initial state disappears as system size grows or simulation time is extended, the phenomenon would be a long transient rather than true ergodicity breaking."],"forward_implications":["Slow cortical dynamics could be explained by network structure alone, without requiring slow intrinsic or synaptic time constants.","Partial symmetry, already observed in local cortical circuits, becomes a candidate mechanism for long-lasting memory of initial activity patterns.","The broken-ergodicity regime offers a dynamical basis for persistent and multi-stable activity states, as seen in working memory.","The findings give a statistical-physics perspective on neural variability: long timescales may signal a multi-stable or near-critical collective regime.","The mechanism suggests that the degree of synaptic symmetry is a controllable parameter that can tune a network between fast fluctuation and long-memory regimes."],"supporting_citations":[],"fun_headline_variants":["Partial wiring symmetry drives slow cortical dynamics","Symmetry-induced slow dynamics in spiking networks","Broken ergodicity from partial synaptic symmetry","How wiring symmetry gives rise to slow brain activity"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The central claim rests on the assumption that the long-lived fluctuations and memory of the initial state seen in finite simulations are genuine properties of the infinite network, not artifacts of finite network size or finite simulation time.","fun_headline_variants_meta":{"raw":{"variants":["Partial wiring symmetry drives slow cortical dynamics","Symmetry-induced slow dynamics in spiking networks","Broken ergodicity from partial synaptic symmetry","How wiring symmetry gives rise to slow brain activity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000236,"raw_usage":{"total_tokens":1289,"prompt_tokens":642,"completion_tokens":647,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":386,"completion_tokens_details":{"reasoning_tokens":590}},"tokens_in":386,"tokens_out":647,"duration_ms":7329,"temperature":1.0,"reasoning_tokens":590,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T00:40:57.254924+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same network architecture at increasing system sizes and simulation durations. If the overlap of the activity with the initial state decays to zero for sufficiently large systems or long times, so that the apparent multiple equilibria are merely slow transients, then the claimed ergodicity breaking is not a true thermodynamic property.","supporting_citations":[],"review_version":1}