{"id":"181379bf-9a24-48cd-b120-b841b7d30c82","arxiv_id":"1908.05313","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A linear-response approximation yields autocorrelation and variance formulas from which Van der Pol parameters can be estimated from noisy time series.","lead":"This paper derives approximate formulas for a noisy Van der Pol oscillator and uses them to estimate the oscillator's parameters from recorded time series. The method works on simulated data and on recordings of hair-cell oscillations from a bullfrog's ear.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Period averaging in Eqs. (9)-(11) discards the limit cycle's zero Floquet mode, so Eq. (13) uses a kernel with artificial damping; the inferred 15% accuracy rests on an undemonstrated insensitivity of the short-time fit window to that discarded mode.","rationale":"The paper's central claim is Eq. (13), and every derived statistic and the inference recipe flow from the period-averaged kernel g_xi. The weakest point is not the harmonic approximation per se but the averaging of the variational equation: the exact linearization about a limit cycle has a zero Floquet exponent (the phase mode), while the averaged oscillator has purely damped modes. Thus g_xi is not a controlled approximation to g1 for long times; this is a structural, not merely quantitative, mismatch. The paper explicitly limits use to short times and to moderate noise, and its simulation table is consistent with the claim under those restrictions. My proposed check would turn the current plausible-but-unproved status into quantitative support or refutation. I therefore keep the reader's CONDITIONAL verdict rather than escalating to rejection: the concern is real and load-bearing, but the simulation evidence and the paper's self-imposed restrictions prevent it from being a proven fatal flaw. Fixing the sign typo in Eq. (11) and adding the Floquet comparison would materially strengthen the paper.","tokens_in":16013,"tokens_out":7558,"duration_ms":83971,"concrete_test":"For mu = 1 with a = -1, b = 1, c = 1, solve the periodic-coefficient variational equation (6) with f(t) = delta(t) and zero initial conditions, using xi_0(t) from Eq. (A3). Compute the exact Floquet exponents of Eq. (6) and the exact response y_ex(t); compare y_ex(t) with g_xi(t) from Eq. (12) in relative L2 norm over [0, tau] and [0, 2 tau], where tau = pi/(2 sqrt(b)). If the relative error over the fitting window exceeds the 15% accuracy claimed in Table I, or if the exact exponents include a near-zero mode, the period-averaging step is the cause and Eq. (13) needs a phase-mode term or a proven error bound.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decisive step is the replacement of the time-periodic variational problem (6) by the averaged constant-coefficient oscillator (11). The two systems are not close in any norm that preserves the stability structure: because Eq. (6) is the linearization about a stable limit cycle, its Floquet spectrum contains a zero exponent corresponding to phase shifts, whereas the averaged oscillator has two exponents with real part -<a_xi>/2 < 0 (taking <a_xi> = -a > 0). Hence the exact linear response g1(t) contains a non-decaying component proportional to the derivative of the limit cycle, while g_xi(t) from Eq. (12) decays exponentially. For stochastic forcing the discarded mode generates phase diffusion and a contribution to <x^2> that is absent from the stationary Gaussian term in Eq. (13). The paper acknowledges phase fluctuations in Sec. II A and the Conclusion, and restricts the autocorrelation fit to short times, so the central claim is not trivially false; however, all inference formulas inherit the averaged kernel, and no bound or numerical comparison with the Floquet response is provided. The 15% claim in Table I is therefore supported only if the short-time window 0 <= t <= tau is insensitive to the discarded zero mode; that condition is asserted, not demonstrated. There is also a sign typo in Eq. (11) as printed: with <a_xi> > 0, the equation should read +<a_xi> dot tilde-gamma. This typo does not change the structural objection, but it should be fixed in any revision.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an approximate analytic treatment of the Van der Pol oscillator driven by white noise. The authors expand the solution in a Volterra series around the limit-cycle solution and approximate the limit cycle by a single harmonic ξ(t) = α cos(√b t) (Eq. A3). Linearizing around ξ(t) gives a time-periodic variational equation (Eq. 6), whose coefficients are then replaced by their period averages (Eqs. 9–11), yielding a damped harmonic oscillator with response kernel g_ξ(t) (Eq. 12) and approximate solution x_ξ(t) = ξ(t) + ∫ g_ξ(t−s) f(s) ds (Eq. 13). From this approximation the paper derives the autocorrelation function (Eqs. B7–B8), the characteristic function (Eq. B4), and a two-step parameter-inference scheme: fit the autocorrelation to obtain a and b, then fit half-period trajectory pieces to obtain c and A (Eqs. 15–18). The method is tested on simulations (Table I) and applied to hair-bundle experimental data, where the simple Van der Pol model fails but a hidden Van der Pol system (Eqs. 20–21) and an effective linear model reproduce the data.","tokens_in":16330,"tokens_out":4581,"duration_ms":46335,"significance":"Should the approximation hold, the paper offers a simple and broadly useful route: a nonlinear stochastic oscillator is reduced to a driven damped harmonic oscillator, giving closed-form statistics and a practical inference protocol. The paper's strengths are explicit analytical formulas, validation against simulations for μ = 1 and moderate noise, and an experimental application with a clear negative result for the simple model. The method is not circular: the parameter estimates are calibrated against simulated data with known parameters, and the experimental application is a genuine out-of-sample use. However, the central reduction is heuristic; the main value would be greatly strengthened by a quantitative justification of the period-averaging step and a demonstration of the robustness of the fitting window.","major_comments":[{"comment":"The replacement of the time-periodic coefficients a_ξ(t) and b_ξ(t) by their period averages is the load-bearing step of the paper, but its accuracy is not established. The exact variational equation about a stable limit cycle has a zero Floquet exponent corresponding to phase shifts; the averaged oscillator (11) has two complex exponents with real part −⟨a_ξ⟩/2 < 0. Hence the exact linear response contains a non-decaying component proportional to the derivative of the limit cycle that g_ξ(t) in Eq. (12) cannot represent. For stochastic forcing this discarded mode is responsible for the phase diffusion acknowledged in Sec. II A and Appendix B. The statement that the approximation is valid on short times does not by itself justify the fitting interval 0 ≤ t ≤ τ, because the discarded mode contributes on all time scales. I ask the authors to provide a quantitative comparison with the exact Floquet response of Eq. (6) (for example, by numerically computing its fundamental matrix) and to state the resulting error bounds on the inference formulas.","section":"II A, Eqs. (6)–(13)"},{"comment":"The claim that the inferred parameter values are accurate within 15% is not supported by an error analysis that propagates the approximations through the fitting procedure. The autocorrelation fit (Eq. 15) depends on two fitted nuisance amplitudes λ1 and λ2, the chosen window τ, and the initial guesses described in Sec. II B, while the amplitude/variance step (Eq. 18) uses the sample variance ⟨x²⟩ whose relation to the theoretical variance is affected by phase diffusion. The paper tests one value μ = 1 and a specific set of noise amplitudes, but it does not quantify how the bias in Table I changes with the fitting window or with μ near the stated range of validity. Adding a sensitivity study of the window length, noise amplitude, and initial guesses, together with fitting residuals, is necessary to support the stated accuracy.","section":"Table I and Sec. II B"},{"comment":"The single-mode approximation ξ0(t) for the limit cycle itself is uncontrolled. The authors provide two levels ξ0 and ξ1, but Fig. 4 shows that ξ1 is visibly closer to the simulated limit cycle, while the paper nevertheless uses ξ0 for the variance and amplitude step (Eq. 18) because ξ1 overestimates ⟨x²⟩. This mixed use of the two approximations is not systematically justified. A consistent error estimate for the limit-cycle approximation, for example from the next Picard iterate or from the omitted harmonic, is needed before the inference formulas can be regarded as quantitative.","section":"Appendix A and Sec. II B"}],"minor_comments":[{"comment":"There is a sign typo in Eq. (11): since ⟨a_ξ⟩ ≈ −a > 0, the equation should read +⟨a_ξ⟩ γ̇, not −⟨a_ξ⟩ γ̇, for consistency with the decaying response in Eq. (12).","section":"Eq. (11)"},{"comment":"The caption refers to “χ1(t) [Eq. (1)]”; this should be Eq. (B8).","section":"Fig. 6 caption"},{"comment":"There are several typographical errors: “Comparision” in the Fig. 2 caption, “expeirments” in the Fig. 3 caption, “self-sustatined” in the Conclusion, and a stray “v,” before the coefficient definitions in Appendix D. These should be corrected.","section":"Throughout"},{"comment":"Equation (15) is extremely dense and difficult to parse; splitting it into a prefactor and a bracket, or introducing auxiliary variables for Ω and the µ-dependent damping, would substantially improve readability.","section":"Eq. (15)"},{"comment":"Reference [43] is incomplete: it lacks journal, volume, and page information and should be completed.","section":"Reference [43]"}],"recommendation":"major_revision","confidential_remarks":"The paper sits at the boundary of cond-mat.stat-mech and nonlinear dynamics; its main contribution is methodological. I would not reject it outright, because the approximations are explicit and numerically testable, but the uncontrolled period-averaging step is a genuine gap in the central derivation. The revision should add the Floquet comparison and the sensitivity analysis described in the major comments, and should temper the accuracy claims until those checks are provided."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: this is a useful methods paper that extends the Volterra linear-response approach from the Duffing oscillator to the Van der Pol oscillator, and the proposed inference recipe appears to work at the 10-15% level for moderate noise. The main weakness is that the period-averaging step that turns the linearized equation into a constant-coefficient damped oscillator is not controlled. The exact linearization about a limit cycle has a zero Floquet mode, so the averaged kernel is missing a non-decaying component. The paper knows this—it discusses phase diffusion and restricts the autocorrelation fit to short times—but it never shows how much error that introduces. The sign typo in Eq. (11) (should be +⟨aξ⟩ times the velocity) and the one in Eq. (18) (the square root needs −ab, not ab) are minor but should be fixed.\n\nWhat's genuinely new is the Van der Pol extension itself. Unlike Duffing, the limit cycle has no closed form, the stationary density is singular, and phase diffusion is a real effect. The single-mode approximation plus period averaging is a pragmatic choice, and Table I shows the inference method works for moderate noise. The hidden Van der Pol model for hair-bundle oscillations is a nice demonstration that the approach can adapt to real data. The paper is also honest about its limits: it explicitly says the method degrades for small and large noise, and that phase fluctuations are the major challenge.\n\nThe big soft spot is the period averaging. The stress-test note is correct: the linear response of the true system contains a term proportional to the derivative of the limit cycle, which does not decay, while the averaged kernel has artificial damping. The paper's defense is that only short-time behavior is used, but it doesn't quantify the effect on the fit. Table I is a simulation check, which is useful, but it doesn't isolate this particular error. A referee could reasonably ask for a numerical comparison with the full Floquet response in a few parameter regimes. Also, the fitting procedure relies on nuisance parameters λ1 and λ2 in Eq. (15) that are fitted rather than predicted—acceptable for a practical tool, but it reduces inferential power.\n\nThe experimental section is qualitative: the simple Van der Pol model clearly fails, the hidden Van der Pol matches, but there is no quantitative goodness-of-fit measure. That's fine for a demonstration, but not a strong validation.\n\nIf I were editing, I'd send this to peer review. The central approximation is reasonable, the method is practical, and the paper is honest about its shortcomings. A referee should ask for a discussion of the zero-Floquet-mode issue (and preferably a numerical check), plus the typo fixes. For a reader working on stochastic oscillators or hair-cell biophysics, this is worth reading and citing. It doesn't reorganize the field, but it gives a workable tool.","headline":"A practical Volterra-based parameter-inference method for noisy Van der Pol oscillators, credible for moderate noise but with an uncontrolled period-averaging step that deserves scrutiny.","tokens_in":16850,"tokens_out":3851,"would_cite":true,"duration_ms":35818,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["34C15","34F05","60H10"],"pacs":["05.40.-a","05.45.-a","02.50.Ey"],"model":"deepseek-v4-flash","headline":"A white-noise-driven Van der Pol oscillator can be treated, to first order in the noise, as a single-mode relaxation cycle plus a Gaussian linear-response term, yielding explicit statistics and a parameter-inference method.","keywords":["Van der Pol oscillator","white noise","Volterra series","linear response","relaxation oscillations","parameter inference","hair bundle","autocorrelation function"],"falsifier":"Simulate Eq. (1) with $\\mu=2$ and $A=0.6$, then compare the empirical autocorrelation function against the theoretical expression of Eq. (B8); if the discrepancy exceeds the claimed 15% parameter error, the single-mode period-averaging assumption fails. Alternatively, measure the empirical linear response to a weak probe force and check whether it matches the kernel $g_\\xi(t)$ of Eq. (12).","tokens_in":15801,"feed_emoji":"🔁","tokens_out":4728,"duration_ms":42559,"temperature":0.7,"pith_summary":"This paper claims that the Van der Pol oscillator driven by white noise can be understood, to first order in the noise, as a deterministic relaxation cycle plus an independent Gaussian linear-response term. The deterministic part is approximated by a single harmonic mode, and the noise response by a damped harmonic oscillator whose friction and stiffness are the period-averaged coefficients of the linearized equation. From that decomposition the authors derive closed-form expressions for the autocorrelation function, variance, and characteristic function, and turn them into a two-step parameter-estimation procedure. If the approximation holds, a researcher can extract the oscillator's friction, stiffness, and damping parameters directly from a measured noisy time series, and can build a linear surrogate model that mimics the underlying nonlinear oscillator.","feed_headline":"Noisy oscillations reduce to a clean cycle plus linear noise","feed_subtitle":"A period-averaged response kernel lets researchers recover Van der Pol parameters from time series to within 15 percent.","key_machinery":"The load-bearing object is the first-order Volterra kernel $g_\\xi(t)=\\Omega^{-1}\\exp(-\\langle a_\\xi\\rangle t/2)\\sin(\\Omega t)$, the linear response of the period-averaged variational equation. The variational equation around the limit cycle has time-periodic coefficients; replacing them by their period averages turns the response into that of a constant-coefficient harmonic oscillator, so the noise-driven part becomes Gaussian with a known autocovariance. The single-mode approximation $\\xi_0(t)=\\alpha\\cos(\\sqrt{b}\\,t)$ with $\\alpha=2\\sqrt{-a/c}$ supplies the deterministic part and makes the averaging and the statistics explicit, including the arcsine law for the autonomous term and a Bessel-function characteristic function for the sum.","core_discovery":"The central discovery is that the stochastic Van der Pol solution admits the approximate representation $x(t)\\approx \\xi(t)+\\int_0^t g_\\xi(t-s)f(s)\\,ds$, in which $\\xi(t)=\\alpha\\cos(\\sqrt{b}\\,t)$ is the single-mode limit cycle and $g_\\xi$ is the Green function of a damped harmonic oscillator with effective coefficients $\\langle a_\\xi\\rangle$ and $\\langle b_\\xi\\rangle$ obtained by averaging the time-dependent linearized coefficients over one period. This reduces a nonlinear relaxation oscillator to a linear system driven by the same noise, and yields explicit formulas for the autocorrelation function and variance. The paper also claims that fitting measured autocorrelations and oscillatory trends to these formulas recovers $a$, $b$, $c$, and the noise amplitude $A$ within about 15% for moderate noise ($0.2<A<1.2$ at $\\mu=1$), with $b$ the most biased parameter and $\\mu$ within 10%.","pith_inferences":["The same period-averaging trick should extend to other Lienard-class oscillators, but the single-mode approximation will break down as the limit cycle develops strong harmonics; the paper's own $\\mu>1$ caveat suggests a quantitative boundary worth testing.","The phase-diffusion effect the paper identifies, noise-induced shifts near the origin of phase space, implies that the autocorrelation formulas are trustworthy only on time scales short compared with phase decorrelation; a polar-coordinate treatment would likely replace the persistent cosine term by an exponentially decaying envelope.","One could turn the linear surrogate into a generative model for synthetic data, simulating the effective driven oscillator with fitted parameters and using it for detector design or hypothesis testing in sensory neuroscience.","A direct experimental test would be to apply a weak sinusoidal probe force and measure the empirical linear response function; if it matches $g_\\xi(t)$ within the stated noise range, the core approximation is confirmed directly rather than through autocorrelation fits."],"forward_implications":["Any measured time series whose autocorrelation matches the derived expression over its early decay can be mapped to Van der Pol parameters without solving the nonlinear equation numerically.","The decomposition provides a linear surrogate: a driven harmonic oscillator plus periodic forcing reproduces the statistics of the nonlinear relaxation oscillations, which is useful for engineering and biological modeling.","The inference method cannot resolve very small noise amplitudes because the deterministic trend dominates, and its precision deteriorates as the nonlinearity parameter $\\mu$ grows.","For hair-bundle oscillations, the simple Van der Pol equation is rejected by the data, while a hidden Van der Pol oscillator linearly coupled to the measured coordinate reproduces the observations, with an effective linear model nearly indistinguishable from it."],"supporting_citations":[{"why":"Supplies the variational Volterra-series technique and the parameter-inference-by-fitting strategy that this paper extends to the Van der Pol oscillator.","marker":"[23]"},{"why":"Provide the Volterra functional-expansion theory that underpins the linear-response truncation.","marker":"[24, 25]"},{"why":"Supplies the harmonic-balance and Lindstedt-Poincare methods used for the single-mode and first-order limit-cycle approximations, plus the time-averaging rationale.","marker":"[26]"},{"why":"Gives the linear response function and autocovariance of a damped harmonic oscillator driven by white noise, which become the effective response kernel.","marker":"[28]"},{"why":"Provides the Lagarkov-Sergeev criterion used to select the time interval over which the theoretical autocorrelation is expected to be accurate.","marker":"[30]"},{"why":"Supplies the parsimonious hair-bundle model that the paper reduces to a hidden Van der Pol oscillator coupled to the measured coordinate.","marker":"[37]"},{"why":"Gives the arcsine-law distribution used to describe the statistics of the single-mode autonomous term.","marker":"[45]"},{"why":"Provides the canonical Van der Pol form, the Lienard-system context, and the scaling used to define the reduced units and the parameter $\\mu$.","marker":"[3]"}],"fun_headline_variants":["Noisy Van der Pol cycles collapse to a linear kernel","Volterra kernel linearizes noisy Van der Pol","From noisy relaxation to clear parameters: a Volterra view","Van der Pol under white noise: a linear response trick","Linearizing white-noise Van der Pol with a Volterra kernel"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"That the Van der Pol limit cycle can be represented by the single harmonic $\\xi(t)=\\alpha\\cos(\\sqrt{b}\\,t)$ and that the time-dependent linearized coefficients can be replaced by their period averages, an approximation the paper verifies only by simulation for $\\mu=1$ and moderate noise, without a rigorous error bound.","fun_headline_variants_meta":{"raw":{"variants":["Noisy Van der Pol cycles collapse to a linear kernel","Volterra kernel linearizes noisy Van der Pol","From noisy relaxation to clear parameters: a Volterra view","Van der Pol under white noise: a linear response trick","Linearizing white-noise Van der Pol with a Volterra kernel"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000512,"raw_usage":{"total_tokens":2470,"prompt_tokens":905,"completion_tokens":1565,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":521,"completion_tokens_details":{"reasoning_tokens":1485}},"tokens_in":521,"tokens_out":1565,"duration_ms":12162,"temperature":1.0,"reasoning_tokens":1485,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:18:16.714838+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate Eq. (1) with $\\mu=2$ and $A=0.6$, then compare the empirical autocorrelation function against the theoretical expression of Eq. (B8); if the discrepancy exceeds the claimed 15% parameter error, the single-mode period-averaging assumption fails. Alternatively, measure the empirical linear response to a weak probe force and check whether it matches the kernel $g_\\xi(t)$ of Eq. (12).","supporting_citations":[{"cited_title":"This approach unfortunately fails in the case of the noisy Van der Pol oscillator: such an expansion may not exist near the two singularities at which p(ξ0) tends to infinity","cited_arxiv_id":null,"evidence_quote":"Supplies the variational Volterra-series technique and the parameter-inference-by-fitting strategy that this paper extends to the Van der Pol oscillator."},{"cited_title":"Schetzen, The Volterra and Wiener Theories of Non- linear Systems (Krieger Pub., 2006)","cited_arxiv_id":null,"evidence_quote":"Supplies the harmonic-balance and Lindstedt-Poincare methods used for the single-mode and first-order limit-cycle approximations, plus the time-averaging rationale."},{"cited_title":"Jordan and P","cited_arxiv_id":null,"evidence_quote":"Gives the linear response function and autocovariance of a damped harmonic oscillator driven by white noise, which become the effective response kernel."},{"cited_title":"Belousov and E","cited_arxiv_id":null,"evidence_quote":"Provides the Lagarkov-Sergeev criterion used to select the time interval over which the theoretical autocorrelation is expected to be accurate."},{"cited_title":"Tinevez, F","cited_arxiv_id":null,"evidence_quote":"Supplies the parsimonious hair-bundle model that the paper reduces to a hidden Van der Pol oscillator coupled to the measured coordinate."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the arcsine-law distribution used to describe the statistics of the single-mode autonomous term."},{"cited_title":"van der Pol, The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science 2, 978 (1926)","cited_arxiv_id":null,"evidence_quote":"Provides the canonical Van der Pol form, the Lienard-system context, and the scaling used to define the reduced units and the parameter $\\mu$."}],"review_version":1}