{"id":"e4506e30-c30f-4a83-8eda-af3d53656068","arxiv_id":"2607.06428","paper_version":1,"verdict":"CONDITIONAL","confidence":"UNKNOWN","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":5,"one_line_summary":"Scaling laws for early-warning signals near bifurcations are derived for non-Markovian noise, showing Hopf rotation suppresses memory dependence while spectral density bypasses masking.","lead":"This paper derives how memory in stochastic noise (fractional Brownian motion) changes early-warning signals for tipping points, and shows rotational dynamics near Hopf bifurcations suppress memory effects. It matters for predicting abrupt transitions in climate, ecology, and other complex systems with correlated noise.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"The H-independence claim for Hopf scaling relies on B(λ*)≠0, but the transition regime where rotation dominates is not characterized, leaving the practical applicability window unclear.","rationale":"The reader correctly identifies the linearization approximation as a key limitation, and this is a real constraint acknowledged in (A2). However, this is a standard caveat in the EWS literature—the scaling laws are derived for the linearized system, and their practical utility depends on detecting the increase before nonlinear effects dominate. The authors are transparent about this. My concern is more specific: the H-independent scaling for Hopf bifurcations (the paper's headline contribution) holds only in an intermediate regime whose boundaries are not analytically characterized. The asymptotic result at λ=λ* is correct, but the finite-distance behavior shows H-dependent scaling persists until |A(λ)| ~ |B(λ)|, and the admissible τ values for the autocovariance scaling depend on eigenvector structure through condition (4.3). This does not invalidate the theorems—the proofs are detailed and the key identity (B.2) is sound—but it does mean the practical guidelines would benefit from explicit characterization of the crossover regime. The numerical simulations (Figs. 10-12) are informative and show the predicted scaling in the intermediate regime, providing reasonable empirical support. The proofs extend existing methods (Lemma 3.1 via Mandelbrot-van-Ness representation, Theorem 4.1 via spectral decomposition) in a nontrivial way to the complex-eigenvalue setting. The spectral density results (Theorem 4.5, Lemma 4.4) are cleaner because they avoid the 'almost every τ' qualifier. Overall, the theoretical contributions are sound, the concern about the applicability window is real but does not rise to the level of changing the verdict from CONDITIONAL. The paper delivers what it promises: asymptotic scaling laws with clear derivation. The gap between asymptotic theory and finite-time practice is inherent to the EWS enterprise and is honestly discussed in Section 6.1.","tokens_in":53088,"tokens_out":1072,"duration_ms":958445,"concrete_test":"For the Hopf system (6.5) with ω₀=1, compute the autocovariance V∞(τ)[v1,v2] for a grid of (λ, τ) values with H∈{0.25, 0.5, 0.75}, specifically tracking the crossover point λ*(H) where the local log-log slope transitions from -2H to -1. If this crossover point depends significantly on H, the practical window for H-independent EWS detection is H-dependent, weakening the guideline that memory effects are universally suppressed by rotation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that rotational dynamics in Hopf bifurcations produce H-independent scaling laws (|A(λ)|^{-1} for autocovariance in Thm 4.1, |A(λ)|^{-2} for spectral density at ±B(λ) in Thm 4.5). The mathematical argument hinges on the memory term P(ζ_{j1}, ζ_{j2}, H, τ) remaining bounded as λ→λ* because B(λ*)≠0 (eq B.2). This is correct at the limit. However, the numerical simulations (Fig. 10) reveal three regimes: (1) far from criticality, slopes follow the 1D scaling |A|^{-2H}; (2) an intermediate regime with slope -1; (3) nonlinear damping. The H-independent scaling only appears in regime (2), which exists when |A(λ)| becomes comparable to |B(λ)|. The paper does not rigorously characterize when regime (1) transitions to regime (2). For the autocovariance, the cross-modal terms V∞(τ)[e*_1, e*_2] are O(1) (Table 6, column (1,2)), while the diagonal terms diverge as |A|^{-1}. The claim |V∞(τ)[v1,v2]| ≍ |A(λ)|^{-1} (eq 4.2) requires condition (4.3) to hold, which involves a specific phase relationship e^{±iB(λ*)τ}. The 'almost every τ' qualifier means there exist τ values where the leading terms cancel, collapsing the scaling. This is not a measure-zero artifact: at τ=0, the mixed-mode autocovariance in Fig. 10(c) shows no -1 regime at all. The practical utility of these EWSs depends on knowing which τ values are admissible, but the set of excluded τ values is only implicitly characterized through condition (4.3), which depends on the eigenvector structure of M(λ*) and the noise covariance Q. The reader's concern about the linearization boundary is valid but is acknowledged by the authors (A2) and is standard in the EWS literature. The more specific concern is the gap between the asymptotic claim (valid at λ=λ*) and the finite-distance behavior where the H-dependent regime persists.","agreement_with_reader":"partial"},"referee_report":{"model":"glm-5.2","summary":"This paper studies early-warning signals (EWSs) for critical transitions in fast-slow stochastic systems driven by non-Markovian noise (fractional Brownian motion, fractional Ornstein-Uhlenbeck processes, and red noise). The main contribution is a comparison of scaling laws for autocovariance, autocorrelation, and spectral density near codimension-1 bifurcations (fold, transcritical, pitchfork) versus Hopf bifurcations. For codimension-1 bifurcations, the authors extend existing results to the full range H∈(0,1) and show that scaling exponents depend explicitly on the Hurst index H. For Hopf bifurcations, they prove that rotational dynamics induce H-independent scaling laws (|A(λ)|^{-1} for autocovariance, |A(λ)|^{-2} for spectral density at ±B(λ)). The analysis covers fBm, red noise, and fOU forcings, and includes numerical validation on a Stommel-Cessi AMOC model and a Hopf normal form system. The proofs in Appendices A and B use the Mandelbrot-van Ness representation, Itô isometry, and Taylor expansion techniques.","tokens_in":54001,"tokens_out":1723,"duration_ms":239762,"significance":"The paper makes a genuine contribution by systematically comparing how bifurcation type and noise memory interact to shape EWS scaling laws. The H-independence result for Hopf bifurcations (Theorems 4.1 and 4.5) is a non-trivial finding: the rotational dynamics suppress the memory signature of the driving fBm, which has practical implications for EWS interpretation. The extension to H∈(0,1/2) (anti-persistent noise) for codimension-1 bifurcations fills a gap in the literature. The decomposition of the autocovariance into masking, memory, and mitigation terms (Lemma 3.1, equation 3.3) is a useful conceptual framework. The spectral density results for fOU noise (showing that S(δ) remains effective even when variance-based EWSs fail due to color blindness) are practically relevant. The numerical simulations on the AMOC model and Hopf system provide reasonable cross-validation of the analytic predictions, including the three-regime behavior (1D scaling, H-independent scaling, nonlinear damping) visible in Figure 10.","major_comments":[{"comment":"§4.1, Theorem 4.1, Eqs. (4.2)–(4.3): The H-independent scaling |V_∞(τ)[v1,v2]| ≍ |A(λ)|^{-1} is stated to hold for 'almost every τ≥0' and for v1,v2 satisfying condition (4.3). The numerical simulations in Figure 10(c) show that at τ=0, the mixed-mode autocovariance exhibits no intermediate regime with slope -1 at all, which is consistent with (4.3) failing at that specific τ. However, the set of excluded τ values is characterized only implicitly through (4.3), which depends on the eigenvector structure of M(λ*) and the noise covariance Q. The paper would benefit from a more explicit characterization of when (4.3) fails, or at minimum a sharper statement of the measure of the excluded set. As it stands, a practitioner cannot determine a priori which lag times are admissible for a given system. This is load-bearing for the paper's central claim because the practical utility of the autocovE","section":null},{"comment":"§6.1, discussion following (A2): The linearization approximation (2.10) is acknowledged to be valid only in a neighborhood of the stable branch, and the authors note that nonlinear effects eventually halt the divergence of EWSs near the critical threshold. The numerical results (Figure 10) reveal three regimes: (1) 1D scaling |A|^{-2H} when |A(λ)| >> |B(λ)|, (2) H-independent scaling |A|^{-1} when |A(λ)| ~ |B(λ)|, and (3) nonlinear damping. The transition from regime (1) to regime (2) is not rigorously characterized. Since the H-independent scaling is the paper's central novel claim, the absence of any analytic bound on when regime (2) begins (in terms of |A(λ)|/|B(λ)|) weakens the practical applicability of the result. The authors should at least provide a heuristic criterion or an explicit estimate for the crossover.","section":null}],"minor_comments":[{"comment":"Table 1: The entry for fOU variance reads 'Conv(2-2H)'. For H∈(0,1/2), the exponent 2-2H is in (1,2), so the variance converges to 0 as λ→λ*. This is consistent with the color blindness discussed in Section 5, but a brief footnote in the table caption clarifying that 'Conv' here means 'converges to zero' (as opposed to converges to a nonzero constant) would help the reader.","section":null},{"comment":"§3.1, Eq. (3.6): The phrase 'almost every τ≥0' is used without specifying the measure. It should be clarified that this refers to Lebesgue-a.e. τ.","section":null},{"comment":"§2.4: The notation ζ_1(λ) = A(λ) + iB(λ) = ζ_2(λ) appears to contain a typo; presumably ζ_2(λ) = A(λ) - iB(λ), consistent with the complex conjugate eigenvalue pair.","section":null},{"comment":"§5.2, Table 5 caption: The caption states the table applies to 'equivalently for (5.7)', but the text preceding the table discusses only (5.6). A cross-reference clarifying that the same exponents hold for the fOU system (5.7) in the (S2) setting would improve readability.","section":null},{"comment":"§6.2, Fig. 4(c): The text states that for H>1/2, the maximum of the SD 'is not attained' analytically but is computed numerically, yielding a slope of -2. This should be stated more carefully: the numerical maximum is finite due to discretization, but the true SD has a singularity at ω=0 for H>1/2.","section":null},{"comment":"Several typos: 'reults' (abstract), 'Ornestein-Uhlenbeck' (Section 5.1), 'explicitate' (§6.1, A2), 'ulterior complexity' (§4.1). Also, 'Gausssian' in the reference to Rosenblatt processes on p.3.","section":null},{"comment":"References [5] and [40] appear to be by one of the authors (Bernuzzi) and are cited as PhD thesis and recent preprint respectively. Ensuring these are accessible (e.g., via arXiv or repository) would be helpful.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The stress-test concern about the 'almost every τ' qualifier and the implicit characterization of excluded lag times is valid and is addressed in my first major comment. However, I assess this as a minor revision issue rather than a major one: the mathematical result is correct at the limit, the excluded set is genuinely measure-zero for generic Q and eigenvector configurations, and the numerical simulations transparently show the practical implications (including the τ=0 cancellation in Fig. 10(c)). The authors are aware of the limitation and discuss it in §6.1. The linearization caveat (second major comment) is also acknowledged by the authors in (A2) and is standard in the EWS literature. The paper's core mathematical contributions are sound; the issues are about sharper characterization of applicability windows, which can be addressed in revision."},"author_rebuttal":{"model":"glm-5.2","summary":"We thank the referee for a careful and constructive report. The referee correctly identifies the paper's main contributions: the H-independence result for Hopf bifurcations, the extension to anti-persistent noise, and the decomposition into masking/memory/mitigation terms. Both major comments concern the practical applicability of the autocovariance EWS in the Hopf setting. We address each below and indicate the revisions we will make.","responses":[{"response":"The referee is correct that condition (4.3) characterizes the admissible lag times only implicitly, and that this limits practical applicability. We will revise the manuscript to address this in two ways. First, we will add an explicit sufficient condition: condition (4.3) is guaranteed to hold whenever v1 and v2 both have nonzero projection onto at least one of the eigenvectors e_1(λ*) or e_2(λ*) and the corresponding diagonal masking terms ⟨e*_j(λ*), Q e*_j(λ*)⟩ are nonzero (which holds generically when Q is positive definite). This covers the practically relevant case of observing along coordinate directions when the noise is non-degenerate. Second, we will state more sharply that the excluded set of τ values is discrete (it consists of isolated points where a specific oscillatory combination vanishes), so its Lebesgue measure is zero and any randomly chosen τ is admissible with probability one. We will also note that the numerical observation in Figure 10(c) at τ=0 for the mixed mode is precisely one such isolated exclusion, and that τ=0.5 recovers the H-independent regime, consistent with the theory. We agree that the original presentation was insufficiently explicit for practitioners and will revise accordingly.","revision_made":"partial","referee_comment":"§4.1, Theorem 4.1, Eqs. (4.2)–(4.3): The H-independent scaling holds for 'almost every τ≥0' and v1,v2 satisfying (4.3), but the excluded set is characterized only implicitly. A practitioner cannot determine a priori which lag times are admissible. Request for a more explicit characterization or a sharper statement of the measure of the excluded set."},{"response":"The referee identifies a genuine gap between the analytic results and their practical applicability. We will add a heuristic crossover criterion. The key observation is that the memory term P(ζ_1(λ), ζ_2(λ), H, τ) in (4.1) converges to a finite, nonzero limit as λ→λ* precisely because B(λ*)≠0, whereas in the one-dimensional case the analogous term diverges. The crossover from regime (1) to regime (2) occurs when the imaginary part B(λ) of the eigenvalue becomes comparable in magnitude to the real part |A(λ)|, because at that point the oscillatory phase e^{iB(λ)τ} in the memory term begins to produce cancellations that suppress the H-dependent growth. Concretely, the heuristic criterion is |A(λ)| ≲ |B(λ)|, or equivalently |A(λ)|/|B(λ)| ≲ 1. We will verify that this is consistent with the numerical results in Figure 10, where ω_0 = B(λ*) = 1 and the crossover occurs near |A(λ)| ≈ 1. We will also note that the width of the intermediate regime depends on the noise intensity σ, as visible in the comparison between panels (a) and (b) of Figure 10. We agree that a fully rigorous bound on the crossover would require controlling the remainder terms in the Taylor expansion underlying Lemma 4.2 uniformly in the ratio |A(λ)|/|B(λ)|, which is beyond the scope of the current analysis. We will state this limitation explicitly.","revision_made":"partial","referee_comment":"§6.1, discussion following (A2): The transition from regime (1) (1D scaling |A|^{-2H} when |A(λ)| >> |B(λ)|) to regime (2) (H-independent scaling |A|^{-1} when |A(λ)| ~ |B(λ)|) is not rigorously characterized. No analytic bound on when regime (2) begins. Request for a heuristic criterion or explicit estimate for the crossover."}],"tokens_in":53138,"tokens_out":916,"duration_ms":166869,"standing_objections":[]},"desk_editor":{"model":"glm-5.2","letter":"The main thing to know: this paper extends early-warning signal scaling laws to the full Hurst range H∈(0,1) and, more interestingly, shows that rotational dynamics near a Hopf bifurcation wash out the Hurst dependence entirely. The Hopf result is the real contribution here — the memory term P(α,β,H,τ) stays bounded as λ→λ* because B(λ*)≠0, so the autocovariance diverges as |A(λ)|^{-1} and the spectral density at ±B(λ) as |A(λ)|^{-2}, both independent of H. This is a clean, non-obvious result and the proofs in Appendices A and B are detailed and careful, using the Mandelbrot-van Ness representation and Itô isometry in a standard but correct way. The extension to H∈(0,1/2) for the 1D case is also new, filling a gap left by [31]. The systematic comparison across fBm, red noise, and fOU forcing, with the spectral density workaround for color blindness, is useful for practitioners. The numerical validation on the Stommel-Cessi AMOC model and a Hopf system is informative and consistent with the theory. Credit where earned: the decomposition into masking/memory/mitigation terms is a nice organizing principle, and the spectral density at fixed frequency δ as an EWS that bypasses masking is a practical insight. Tables 1 and 2 are a good summary. Now the soft spots. The stress-test concern about the 'almost every τ' qualifier is real but not as sharp as it sounds. The condition (4.3) is a phase condition on e^{±iB(λ*)τ}, and the authors are upfront that at τ=0 the mixed-mode autocovariance can cancel — Figure 10(c) confirms this. The excluded τ values are not a measure-zero artifact in the trivial sense; they depend on the eigenvector structure and noise covariance Q. But the authors state the condition explicitly and the numerics bear it out. The more substantive gap is that the paper does not characterize the transition between the 1D scaling regime (slope -2H, when |A|≫|B|) and the H-independent regime (slope -1, when |A|~|B|). Figure 10 shows three regimes clearly, but there's no rigorous bound on when the crossover happens. This matters for practical use but is a reasonable omission given the asymptotic nature of the results. The linearization limitation is standard in the EWS literature and acknowledged honestly in (A2). No code or data shipped — minor for a theory paper, but the numerical figures would benefit from reproducibility. Overall: the theoretical core is sound, the Hopf H-independence result is genuinely new, and the paper is written for people working on critical transitions in non-Markovian settings. It deserves a serious referee who can check the gamma function manipulations in Appendix B and the eigenvector decomposition in the extended systems of Section 5.","headline":"Solid contribution: new scaling laws for non-Markovian EWSs across full Hurst range, with the Hopf H-independence result being the standout. Deserves a serious referee.","tokens_in":54002,"tokens_out":1270,"would_cite":true,"duration_ms":160279,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G22","60H10","37H20"],"pacs":[],"model":"glm-5.2","headline":"Spinning erases memory from early-warning signals","keywords":["early-warning signals","Hopf bifurcation","fractional Brownian motion","non-Markovian noise","autocovariance","spectral density","fast-slow systems","critical transitions"],"falsifier":"If the memory term P failed to converge at the Hopf threshold for some H in (0,1), the H-independence of the scaling laws would break down.","tokens_in":53138,"feed_emoji":"🌀","tokens_out":1067,"duration_ms":99236,"temperature":0.7,"pith_summary":"This paper proves that when a system approaches a Hopf bifurcation, the rotational motion inherent to that bifurcation acts as a mixing mechanism that erases the memory of the driving noise. For one-dimensional bifurcations (fold, transcritical, pitchfork), the scaling laws of early-warning observables like autocovariance and spectral density depend explicitly on the Hurst index H, which parametrizes how much memory the noise carries. The authors show that near a Hopf bifurcation this dependence vanishes: the autocovariance diverges as |A(λ)|^{-1} and the spectral density at the rotational frequencies diverges as |A(λ)|^{-2}, both independent of H. The paper establishes these results for fractional Brownian motion across the full range H in (0,1), extending prior work that was restricted to H > 1/2, and further extends the analysis to red noise and fractional Ornstein-Uhlenbeck forcings. The central mechanism is that the imaginary part B(λ) of the eigenvalue, which governs rotation, remains nonzero at the bifurcation threshold, causing the memory term P in the autocovariance decomposition to converge rather than diverge, leaving only the mitigation term (loss of stability) to control the scaling.","feed_headline":"Spinning erases memory from early-warning signals","feed_subtitle":"Near a Hopf bifurcation, rotational dynamics strip early-warning scaling laws of their dependence on noise memory, making signals easier to","key_machinery":"The autocovariance decomposition into three factors, the memory function P, and the spectral density formula","core_discovery":"The decomposition of the time-asymptotic autocovariance into three factors (masking, memory, mitigation) reveals that rotational dynamics near a Hopf bifurcation force the memory factor to remain bounded, so the scaling exponent of early-warning signals becomes independent of the Hurst index. This is in sharp contrast to one-dimensional codimension-1 bifurcations where the memory factor diverges and the exponent is governed by H. The spectral density develops peaks at the rotational frequencies ±B(λ) that diverge as |A(λ)|^{-2} regardless of noise coloring, providing an observable that is immune to the masking (color blindness) that can defeat variance-based indicators when the noise is a fO","pith_inferences":["The memory-erasure mechanism should generalize to any bifurcation where eigenvalues cross the imaginary axis away from the origin, including certain codimension-2 bifurcations with rotational components.","The three-stage regime observed in simulations (H-dependent scaling, then H-independent scaling, then nonlinear damping) suggests that the practical detectability window for Hopf early-warning signals is widest when the noise intensity is small enough that the linearized regime persists close to the bifurcation threshold.","If the rotational frequency B(λ*) is very small, the memory-erasure effect weakens and the system may behave more like the one-dimensional case, creating a continuous interpolation between the two scaling regimes."],"forward_implications":["For systems approaching a Hopf bifurcation, the Hurst index of the driving noise need not be estimated to interpret early-warning signals, simplifying practical deployment.","The spectral density evaluated at the rotational frequency ±B(λ) provides a noise-robust early-warning signal that works even for fractional Ornstein-Uhlenbeck noise where variance-based indicators fail due to masking.","The transition between the H-dependent scaling regime (far from bifurcation) and the H-independent regime (closer to bifurcation) occurs when the real part of the eigenvalue becomes comparable in magnitude to the imaginary part, giving a concrete criterion for when rotation dominates.","The unmasking effect of rotation suggests that Hopf bifurcations in systems with colored noise are intrinsically more detectable than fold-type bifurcations, which has implications for monitoring tipping elements in climate and ecological systems."],"fun_headline_variants":["Rotation near Hopf bifurcation overrides noise memory in early-warning signals","Memory-independent early-warning scaling near Hopf bifurcations","Spectral peaks at rotational frequencies bypass noise-color masking","Hurst index loses grip on early-warning exponents under rotational dynamics","Three-factor decomposition isolates rotational mitigation of noise memory"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The entire analysis rests on the linearization of the fast subsystem along the attracting critical manifold, which is valid only in a neighborhood of the stable branch. Since early-warning signals are most needed near the bifurcation threshold where nonlinear effects become important, the derived scaling laws hold only in a regime bounded away from the critical point, and the transition between the linearized and nonlinear regimes is not rigorously characterized.","fun_headline_variants_meta":{"raw":{"variants":["Rotation near Hopf bifurcation overrides noise memory in early-warning signals","Memory-independent early-warning scaling near Hopf bifurcations","Spectral peaks at rotational frequencies bypass noise-color masking","Hurst index loses grip on early-warning exponents under rotational dynamics","Three-factor decomposition isolates rotational mitigation of noise memory"]},"model":"glm-5.2","effort":"high","cost_usd":0.0,"raw_usage":{"total_tokens":568,"prompt_tokens":484,"completion_tokens":84,"prompt_tokens_details":null},"tokens_in":484,"tokens_out":84,"duration_ms":32103,"temperature":1.0,"reasoning_tokens":null,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-08T05:57:41.390255+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"If the memory term P failed to converge at the Hopf threshold for some H in (0,1), the H-independence of the scaling laws would break down.","supporting_citations":[],"review_version":1}