{"id":"41cd0c10-c8da-4f66-a28e-2874a61f906e","arxiv_id":"2507.09413","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A hybrid invariant-manifold and fluctuation-dissipation reduction is shown to produce a one-dimensional additive-noise surrogate that preserves the localization dynamics of a two-state quantum geometric Brownian motion.","lead":"This paper gives a step-by-step way to shrink a multi-variable random \"growth\" process into a single simple equation, then tests it on a noisy two-state quantum model. The simplified model keeps the slow localization behavior of the original while being much easier to analyze.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 2 estimates Wasserstein error using the IM closure moment p̄ as the covariance of z_red, but the actual reduced SDE (61) has covariance p̃(t)=1/3+(p̃0−1/3)e^{2Ā_z t}; the stated bounds do not apply to the true reduced process.","rationale":"The reader correctly identified Proposition 2 as imprecise, but its weakest assumption was the branch-selection rule for the affine closure (46)/(48). My reading confirms that the branch concern is real but repairable: the invariance equations (47) are exactly the eigenvector equations for Mε, and the condition lim_{ε→0} εa3(ε)=0 selects the eigenvalue branch emerging from the unperturbed zero eigenvalue, which is unique by analytic perturbation theory and matches the adiabatic rate −6α²ε/β² at leading order. The more load-bearing problem is in the error-estimation section: the actual reduced SDE (61) has a different second-moment law than the IM reduction (49), and Proposition 2 erroneously substitutes the IM moment for the covariance of z_red. This is not a fatal flaw in the construction—the reduced SDE still matches the deterministic slow drift and the stationary second moment exactly, and the minimization (60) is a legitimate heuristic—but it means the paper has not actually proved its claimed Wasserstein closeness between z(t) and z_red(t). The independent derivations of the invariance equations, Lemmas 1–2, and the reduced SDE itself appear consistent, so the correct remedy is a revision of the error estimate rather than rejection. Since the reader's verdict was already CONDITIONAL and my concern sharpens the same section without changing the overall assessment, I recommend UNCHANGED.","tokens_in":26844,"tokens_out":23054,"duration_ms":279321,"concrete_test":"Recompute Proposition 2 with the true covariance of z_red from Eq. (61). For the large-noise parameters α=0.5, β=1, initialize z(0)=z_red(0)=1 and p̃(0)=pzz(0)=1, solve p̃(t)=1/3+(2/3)e^{2Ā_z t} with Ā_z=−(β²−√(β⁴−4α²)), and compare the exact pzz(t) from Eq. (34)/(41) with p̃(t) and p̄(t). Then evaluate Lemma 4's upper and lower W2 bounds using Σ'=p̃−z̄² rather than p̄−z̄². If the corrected bounds are much wider than the printed ones, or if p̃ deviates substantially from pzz while p̄ does not, the claim that the reduced SDE preserves localization is quantitatively unsupported as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the reduced SDE (61) preserves the localization properties of z(t) is supported in Section 5.3 by Proposition 2, which purports to bound W2 between μ(t)=Law(z(t)) and μ'(t)=Law(z_red(t)). The proof sets Σ'=E(z_red²)−(E z_red)² = p̄zz−z̄², where p̄zz is the invariant-manifold/adiabatic second moment from Eq. (49)/(37). But this is not the second moment of the SDE (61). In Section 5.2 the minimizer of (60) is at b=b0=2Ā_z, hence B_z=0, and the reduced SDE is dz_red=Ā_z z_red dt+D_z du. Its second moment solves dp̃/dt=2Ā_z(p̃−1/3), so p̃(t)=1/3+(p̃0−1/3)e^{2Ā_z t}. This differs from p̄ generally because 2Ā_z ≠ 4αa3*: e.g. in the large-β limit 2Ā_z≈−4α²/β² while 4αa3*≈−6α²/β². Thus Proposition 2, as written, bounds W2 between Law(z) and an auxiliary law with covariance p̄, not Law(z_red). The paper's quantitative evidence that the actual reduced SDE captures localization is therefore missing, and the main error estimate needs to be recomputed with the correct covariance Σ'=p̃−z̄².","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a model-reduction framework for multivariate geometric Brownian motions with multiplicative noise, combining invariant-manifold (IM) closures and adiabatic elimination for the deterministic drift with a fluctuation-dissipation condition for the noise. The framework is applied to a two-state quantum model whose localization variable z satisfies the three-dimensional GBM system (12). The authors derive reduced ODEs for the mean and second moment of z (Eqs. (31c), (37), (45), (49)), assemble a one-dimensional surrogate SDE dz_red = A_bar_z z_red dt + D_bar_z du (Eq. (61)), and claim that this reduced process preserves the localization properties of the original dynamics. Section 5.3 then states Wasserstein error bounds between the laws of z(t) and z_red(t).","tokens_in":27130,"tokens_out":6849,"duration_ms":85243,"significance":"If the error estimates were correct, the paper would provide a useful, non-fitted reduction scheme for multiplicative-noise GBMs: the reduced drift and diffusion coefficients are derived from invariance and Lyapunov-type conditions rather than calibrated to the target solution, and the Chapman-Enskog expansion (51) correctly reproduces the adiabatic limit. The quantum localization application is also physically meaningful. The deterministic part of the reduction is internally consistent: the invariance equations (47) follow from the closure ansatz, and the large-beta asymptotics of the exact spectrum match the adiabatic chain. However, as detailed below, the main quantitative error statement (Proposition 2) is tied to the wrong covariance for the actual reduced process, so the current manuscript overstates the rigor of its central claim.","major_comments":[{"comment":"The proof of Proposition 2 identifies the covariance of z_red(t) with p_bar_zz(t)-z_bar(t)^2, where p_bar_zz is the IM/adiabatic second moment. But the reduced SDE actually derived in Section 5.2 is (61), with B_z=0, and its second moment solves d p_tilde/dt = 2 A_bar_z (p_tilde - 1/3), so p_tilde(t)=1/3+(p_tilde_0-1/3)e^{2 A_bar_z t}. Since 2 A_bar_z is generally different from 4 alpha a_3^*(epsilon) (for instance, in the large-beta limit 2A_bar_z ≈ -4 alpha^2/beta^2 while 4 alpha a_3^* ≈ -6 alpha^2/beta^2), p_bar_zz is not the second moment of z_red. Therefore the stated Wasserstein bounds apply to an auxiliary law with covariance p_bar_zz, not to Law(z_red(t)). The bounds need to be recomputed with Sigma' = p_tilde - z_bar^2; the same Lemma 4 argument would yield corrected bounds, but Proposition 2 as written is incorrect.","section":"Section 4.5, Eqs. (46)-(49)"},{"comment":"The IM reduction of the variance rests on the affine closure ansatz (46) and on selecting the branch a^* by continuity and the condition (48). The paper only states that 'numerically, we identify two such solutions' and gives no existence or uniqueness argument for the physically relevant branch over (0, epsilon''_c). Since Eq. (49) and the reduced second moment used in Section 5.2 feed on this branch, a proof or a precise continuation argument for a^* is needed to justify the IM part of the reduction, not merely the numerical evidence in Fig. 2.","section":"Section 4.5, Eqs. (46)-(49)"}],"minor_comments":[{"comment":"The IM drift is written as -(beta^2 - sqrt(beta^4 - 4 alpha^2)), whereas Eq. (45) gives 2 epsilon alpha a_+(epsilon) = -(beta^2 - sqrt(beta^4 - 4 epsilon^2 alpha^2)). The epsilon dependence (or the precise rescaling convention) should be restated here to make the two equations match.","section":"Section 5.2, Eq. (57)"},{"comment":"The phrase 'The second-moment equation for \\bar z' should read 'for \\tilde p_{zz}' or 'for the second moment', since the quantity being evolved is not the mean \\bar z.","section":"Section 5.2, before Eq. (59)"},{"comment":"Lemma 4 is quoted from [23], which concerns Gaussian measures, while Proposition 2 applies it to arbitrary probability measures. The two-sided estimate (64) should either be proved for general laws or attributed to a source that states it in that generality.","section":"Section 5.3, Lemma 4"},{"comment":"The text sometimes says 'multivariate Brownian motions' where 'geometric Brownian motions' is meant; please align the terminology throughout, including the abstract and the opening of the Conclusion.","section":"Abstract and Conclusion"},{"comment":"There is a typo, 'straightofrwardly' for 'straightforwardly', which should be corrected.","section":"Appendix E"}],"recommendation":"major_revision","confidential_remarks":"The main technical issue is localized: Proposition 2 uses the wrong covariance for the reduced process. A corrected version using p_tilde in place of p_bar for the covariance of z_red should be feasible within the manuscript's scope. The branch-selection gap in Section 4.5 also needs attention before the IM reduction can be considered fully rigorous. I recommend major revision rather than rejection because the deterministic reduction and the construction of the reduced SDE are otherwise coherent."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The honest take: this paper does what it claims. It extends the authors' hybrid invariant-manifold/adiabatic/fluctuation-dissipation reduction program from additive-noise linear SDEs to multiplicative-noise GBMs, and the extension is real. The derived reduced SDE (61) for the localization variable z is obtained from equations, not fitted: the drift comes from the invariance or adiabatic calculation, the noise from the Lyapunov condition plus an L-∞ minimization that lands on B_bar = 0. I checked several load-bearing pieces — the invariance equations (47), the adiabatic hierarchy (34)-(37), the characteristic polynomials, and Lemmas 1 and 2 — and they are internally consistent. The quantum GBM application is a good test case because the exact moment dynamics are small enough to verify the closure against numerics. Credit where due: the higher-order ODE for p_zz and the IM closure equations appear to be genuinely new, and the paper is honest about what is inherited from the prior program.\n\nThe soft spot is real and it is in Section 5.3. Proposition 2 claims to bound W_2 between Law(z(t)) and Law(z_red(t)), but the covariance it plugs in for the reduced process is p_bar_zz, the invariant-manifold closure moment. The actual reduced SDE (61) has B_z = 0, so its second moment solves dp_tilde/dt = 2 A_bar_z (p_tilde - 1/3), giving p_tilde(t) = 1/3 + (p_tilde(0) - 1/3) e^{2 A_bar_z t}. That is not the same as p_bar_zz: in the large-beta limit, 2 A_bar_z ≈ -4 alpha^2/beta^2 while the IM closure rate is 4 alpha a_3* ≈ -6 alpha^2/beta^2. So the bounds in Proposition 2 are for an auxiliary process, not for the reduced SDE the paper says it is analyzing. This does not sink the reduction scheme — the qualitative localization claim, driven by the mean decay and the stationary variance 1/3, still holds — but the quantitative error estimates need to be redone with Sigma' = p_tilde - z_bar^2. The fix is straightforward but the current text is incorrect as written.\n\nTwo smaller issues. First, the physically relevant IM branch is selected numerically (continuity plus the epsilon a_3 -> 0 condition) without a proof of existence or uniqueness over the whole admissible interval, and no code is shipped; that should be addressed in revision. Second, the paper cites balanced truncation work on GBMs but gives no comparison; a small benchmark table would make the contribution clearer. Neither is a refutation.\n\nWho should read it: people working on model reduction of stochastic systems, especially multiplicative-noise dynamics, and the noise-induced localization community. It deserves a serious referee — the core method is sound and the error-estimation flaw is repairable. I would send it to peer review with a request for a revised Proposition 2, a statement on the branch selection, and ideally the missing numerical artifacts.","headline":"The reduction scheme holds up and the derivations are consistent, but the Wasserstein error bounds in Proposition 2 are stated for the wrong covariance and need to be recomputed for the reduced SDE (61) the paper actually defines.","tokens_in":906,"tokens_out":3135,"would_cite":true,"duration_ms":75080,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["34C45","60H10","60J60"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper shows that a multivariate geometric Brownian motion with multiplicative noise can be reduced to a one-dimensional mixed Ornstein–Uhlenbeck and geometric process that preserves the localization properties of the original…","keywords":["model reduction","geometric Brownian motion","invariant manifolds","adiabatic elimination","fluctuation-dissipation relation","two-state quantum system","localization","Wasserstein distance"],"falsifier":"Solve the exact four-variable second-moment system (41) for parameter values such as $\\alpha=0.5$, $\\beta=1$, and $\\epsilon=0.5$, and compare the resulting $p_{xx}(t)$, $p_{yy}(t)$, $p_{yz}(t)$ with the affine closure (46) evaluated at the computed $a^*_i(\\epsilon)$; if the mismatch exceeds the error allowed by the Wasserstein bounds of Proposition 2, the reduction is not sustaining the localization claim at that noise strength.","tokens_in":26532,"feed_emoji":"⚛️","tokens_out":11248,"duration_ms":116987,"temperature":0.7,"pith_summary":"This paper develops a systematic way to shrink a multivariate geometric Brownian motion (a stochastic process whose noise multiplies the state) to a lower-dimensional surrogate without giving up the dynamics that matter. The scheme first reduces the deterministic drift using invariant-manifold or adiabatic-elimination closures, then fixes the reduced noise through a generalized fluctuation-dissipation relation combined with a closest-second-moment optimization. Applied to the localization variable $z$ of a two-state quantum system driven by white noise, the scheme yields a single driven SDE, $dz_{\\mathrm{red}} = \\bar A_z z_{\\mathrm{red}}\\, dt + \\bar D_z\\, du$ with $\\bar D_z = \\sqrt{-(2/3)\\bar A_z}$, and the paper argues that this one-variable process preserves the localization behavior of the full three-variable model. This matters because multiplicative-noise models appear throughout finance, population biology, and statistical physics, and the reduction gives an explicit, closed-form route from a coupled system to a solvable scalar process.","feed_headline":"One variable replaces a 3D noisy quantum model","feed_subtitle":"Reducing multiplicative-noise Brownian motions to a single driven variable still captures localization in the strong-noise regime.","key_machinery":"The load-bearing identity is the generalized fluctuation-dissipation, or Lyapunov, relation $(2\\bar A + \\bar B^2)\\bar p^{\\infty} + \\bar D^2 = 0$ for a one-dimensional mixed Ornstein–Uhlenbeck and geometric process, which determines the reduced noise once the reduced drift and the exact stationary second moment are known. Around it sit the invariance equations (43) and (47), which fix the reduced drift through the closure $m_y = a m_z$ and the affine second-moment closure $p_{xx} = a_1(p_{zz}-1/3)+1/3$, $p_{yy} = a_2(p_{zz}-1/3)+1/3$, $p_{yz} = a_3(p_{zz}-1/3)$; and Lemmas 1–2, which show that the closest-exponential minimization in (60) is achieved at the boundary $b = b_0$, forcing $\\bar B_z = 0$. Together these convert a coupled three-variable SDE into a single autonomous SDE while matching the stationary second moment and the slow drift, with the error controlled by the covariance estimates of Proposition 2.","core_discovery":"The central claim is that for the two-state quantum GBM (12), the resolved localization variable $z$ can be represented by the one-dimensional mixed Ornstein–Uhlenbeck and geometric process $dz_{\\mathrm{red}} = \\bar A_z z_{\\mathrm{red}}\\, dt + \\bar D_z\\, du$, where $\\bar A_z$ is either $-2\\alpha^2/\\beta^2$ (adiabatic elimination) or $-\\bigl(\\beta^2 - \\sqrt{\\beta^4 - 4\\alpha^2}\\bigr)$ (invariant manifold reduction), and $\\bar D_z = \\sqrt{-(2/3)\\bar A_z}$. The noise coefficient is fixed so that the stationary second moment of the reduced process equals the exact value $p^{\\infty}_{zz}=1/3$, and the minimization of the $L^\\infty$ (or $L^2$) distance between the reduced and the already-reduced second-moment dynamics selects $\\bar B_z=0$, so the reduced process is driven only by additive noise. The paper states that this reduced dynamics preserves the localization property of the original system characterized by Eq. (19), and supplies explicit Wasserstein-distance bounds (Proposition 2) between the full distribution of $z$ and the distribution of $z_{\\mathrm{red}}$.","pith_inferences":["An extension the paper leaves implicit is that the same recipe applies to any scalar observable of a GBM whose slow drift and stationary second moment can be computed: minimal additive driving is the universal way to match a given equilibrium variance and a prescribed relaxation rate, without carrying over multiplicative noise.","The appearance of the critical values $\\epsilon'_c$ and $\\epsilon''_c$ suggests a testable prediction the paper does not pursue: near those noise strengths the reduced model should lose quantitative accuracy precisely because the invariant-manifold branch ceases to be real, and measured localization relaxation should show a qualitative change there.","In the editor's reading, the affine second-moment closure is the part most likely to limit the scheme in practice, and the paper's own Fig. 1 shows that the analogous adiabatic closure fails for moderate noise, so a finite-$\\epsilon$ check of the ansatz against the exact moment equations would be the natural stress test before porting the method to higher-dimensional observables."],"forward_implications":["The one-dimensional process (61) reproduces the localization relaxation rate of the full model in the large-noise regime, and in the invariant-manifold version it remains valid across the whole interval $0<\\epsilon\\le \\epsilon''_c$, beyond what adiabatic elimination offers.","Because the reduced process is scalar and driven by additive noise, closed-form expressions for the mean, second moment, and localization probability follow directly, avoiding simulation of the full three-variable system.","Both the $L^\\infty$ and $L^2$ refinements select $\\bar B_z=0$, so the optimal scalar surrogate carries no multiplicative noise of its own; the multiplicative structure of the original GBM is fully absorbed by the drift and the additive noise scale.","Proposition 2 gives an a priori, computable two-sided bound on the Wasserstein distance between the original and reduced distributions in terms of their means and variances, so the reduction comes with a quantitative error certificate.","For $k$-dimensional reduced descriptions the same construction requires solving a matrix Lyapunov equation and the corresponding invariance equations, which is feasible numerically but no longer closed-form."],"supporting_citations":[{"why":"Supplies the two-state quantum model whose localization variable $z$ is the paper's target observable.","marker":"[7]"},{"why":"Provides the rescaling and invariant-manifold reduction scheme that the paper extends from additive to multiplicative noise.","marker":"[17]"},{"why":"Introduces the hybrid two-stage reduction, deterministic invariant manifold plus fluctuation-dissipation noise fixing, that Section 5 generalizes to GBMs.","marker":"[20]"},{"why":"Gives the Lyapunov stationarity condition used to fix the reduced noise amplitude via $2\\bar A_z p^{\\infty}_{zz}+\\bar D_z^2=0$.","marker":"[66]"},{"why":"Supplies the standard mean and second-moment dynamics for geometric Brownian motions on which the reduction is built.","marker":"[60]"},{"why":"Yields the covariance-based upper and lower bounds used in Proposition 2 to estimate Wasserstein distance.","marker":"[23]"},{"why":"Provides the mean-variance decomposition lemma that allows Proposition 2 to separate drift and covariance contributions to the Wasserstein error.","marker":"[38]"}],"fun_headline_variants":["Single-variable reduction preserves quantum localization","Reducing multivariate GBM to one variable still shows localization","One-driver model captures two-state quantum localization","Invariant manifold reduction keeps quantum localization","Quantum GBM shrinks to 1D while localization holds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reduction assumes that the three unresolved second moments $p_{xx}$, $p_{yy}$, and $p_{yz}$ stay locked to the resolved second moment $p_{zz}$ through a fixed affine relation, and that the numerically selected branch of the invariance equations is the one the true dynamics follows; if this manifold fails to attract the exact moment flow at finite noise strength, the reduced process and its error bounds inherit the wrong covariance.","fun_headline_variants_meta":{"raw":{"variants":["Single-variable reduction preserves quantum localization","Reducing multivariate GBM to one variable still shows localization","One-driver model captures two-state quantum localization","Invariant manifold reduction keeps quantum localization","Quantum GBM shrinks to 1D while localization holds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0002,"raw_usage":{"total_tokens":1361,"prompt_tokens":919,"completion_tokens":442,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":535,"completion_tokens_details":{"reasoning_tokens":370}},"tokens_in":535,"tokens_out":442,"duration_ms":5827,"temperature":1.0,"reasoning_tokens":370,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:00:02.854256+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve the exact four-variable second-moment system (41) for parameter values such as $\\alpha=0.5$, $\\beta=1$, and $\\epsilon=0.5$, and compare the resulting $p_{xx}(t)$, $p_{yy}(t)$, $p_{yz}(t)$ with the affine closure (46) evaluated at the computed $a^*_i(\\epsilon)$; if the mismatch exceeds the error allowed by the Wasserstein bounds of Proposition 2, the reduction is not sustaining the localization claim at that noise strength.","supporting_citations":[{"cited_title":"Blanchard, G","cited_arxiv_id":null,"evidence_quote":"Supplies the two-state quantum model whose localization variable $z$ is the paper's target observable."},{"cited_title":"Colangeli, M","cited_arxiv_id":null,"evidence_quote":"Provides the rescaling and invariant-manifold reduction scheme that the paper extends from additive to multiplicative noise."},{"cited_title":"Colangeli and A","cited_arxiv_id":null,"evidence_quote":"Introduces the hybrid two-stage reduction, deterministic invariant manifold plus fluctuation-dissipation noise fixing, that Section 5 generalizes to GBMs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the Lyapunov stationarity condition used to fix the reduced noise amplitude via $2\\bar A_z p^{\\infty}_{zz}+\\bar D_z^2=0$."},{"cited_title":"Dowson and B","cited_arxiv_id":null,"evidence_quote":"Yields the covariance-based upper and lower bounds used in Proposition 2 to estimate Wasserstein distance."},{"cited_title":"On Wasserstein distances for affine transformations of random vectors","cited_arxiv_id":"2310.03945","evidence_quote":"Provides the mean-variance decomposition lemma that allows Proposition 2 to separate drift and covariance contributions to the Wasserstein error."}],"review_version":1}