{"id":"c7a2d7a6-8407-4d61-8239-572ba33f0439","arxiv_id":"2502.03456","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Row sums of an estimated transition matrix give local inverse dissipation; column maxima give an upper bound on the inverse Jacobian, tested on 1D and 2D chaotic maps.","lead":"By counting how often a chaotic trajectory jumps between boxes in state space, the paper estimates local contraction and stretching rates from the resulting transition matrix. A generalist might care because it offers a data-only route to quantities usually computed from known equations, potentially useful for forecasting and control.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The B_j derivation yields an upper bound, not the claimed lower bound, on |J|; in weakly unstable or contracting regions B_j saturates at 1, so the advertised lower-bound claim is not supported.","rationale":"The reader correctly flagged the coarse-graining assumption as the weakest point, and the paper's numerical limitation for B_j in weakly unstable regions is visible in its own figures. However, the more precise load-bearing defect is the inequality direction: Eq. (9) gives B_j ≤ 1/|J|, so 1/B_j is an upper bound on |J|, not the lower bound claimed in the abstract. This is not repaired by finer partitions; it is a structural saturation when |J|<1, and an unjustified replacement of Σ_k p^k_ij by max_i p^{k*}_{ij} when multiple preimages share a cell. The A_i observable, by contrast, is derived without this problematic step and is well supported by the numerics. The paper's central claim should therefore be revised to state that B_j provides an upper bound on |J| that becomes an accurate estimate only in sufficiently expanding regions. This does not change the overall conditional verdict — the method still has value for dissipation fields and for instability in expanding regions — but it requires a substantive correction to the abstract and Section 2, plus a quantitative characterization of when B_j≈1/|J|.","tokens_in":11298,"tokens_out":27884,"duration_ms":230125,"concrete_test":"Run the proposed estimator on the piecewise-linear contracting map f(x)=x/2 on [0,1] with a uniform N-cell partition: compute the empirical transition matrix, evaluate B_j=max_i P_ij, and compare 1/B_j with |J|. The prediction is that B_j=1 for every source cell, hence 1/B_j=1 while |J|=1/2, demonstrating that 1/B_j is an upper bound, not a lower bound. Equivalently, on the Ulam map, restrict to cells where |J|<1 (near x=0) and verify B_j≈1 while 1/|J|>1. This check isolates the saturation mechanism and settles the inequality direction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the transition matrix yields a lower bound on the local Jacobian is contradicted by the paper's own inequality. From Eq. (9), B_j = max_i P_ij ≤ min{|J(C_j)|^{-1}, 1}, so the data-derived quantity 1/B_j satisfies 1/B_j ≥ max{|J(C_j)|, 1} — an upper bound on |J|, not a lower bound. The abstract and Section 2 state the opposite. The intermediate step B_j ≈ max_i p^{k*}_{ij}/|J(C_j)| drops the sum over k and assumes the maximizing target cell receives the full image of the source cell. When |J(C_j)| < 1, the image of a source cell is smaller than a target cell, so the transition probability saturates at 1 and B_j ≈ 1, independent of 1/|J|. The paper's own numerics confirm that B_j underestimates 1/|J| in weakly unstable regions, yet the text does not correct the bound direction. This is a genuine flaw in the central claim, not merely a convergence issue.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a data-driven method to estimate local dissipation and instability fields of a discrete-time dynamical system from a finite transition matrix approximating the Perron-Frobenius operator. The central objects are A_i = sum_j P_ij, which under a fine-partition and smooth-Jacobian approximation equals sum_k 1/|J(f^{-1}_k(C_i))|, and B_j = max_i P_ij, which the authors claim bounds the local inverse Jacobian. Numerical tests on one-dimensional maps (Ulam, continued fraction, cusp, Chebyshev) and two-dimensional coupled Chebyshev maps, with and without additive noise, compare both quantities with analytical Jacobians. The A_i estimates match the analytical values well, while B_j tracks the inverse Jacobian only in strongly unstable regions and saturates or underestimates in weakly unstable and contracting regions.","tokens_in":11528,"tokens_out":10808,"duration_ms":93222,"significance":"If the central claim holds, the method is valuable because it recovers local Jacobian information from orbit data alone, without fitted parameters and without knowledge of the governing equations. The A_i part is a clean and well-supported contribution, with a transparent derivation and extensive numerical validation, including a noise study. However, the B_j claim as stated is flawed: the inequality derived in the paper has the opposite direction from the advertised 'lower bound on the local Jacobian'. This is a load-bearing issue that affects the abstract and the interpretation of the instability-field estimates. The numerical evidence shows that B_j still provides a qualitative indicator in strongly unstable regions, so the paper can likely be repaired by restating the bound direction and the saturation limitations, but the current text must be corrected before acceptance.","major_comments":[{"comment":"The direction of the claimed bound is reversed. Equation (9) states B_j = max_i P_ij <= min{|J(C_j)|^{-1}, 1}, so B_j is a lower bound on the inverse Jacobian 1/|J(C_j)|, not an upper bound, and the derived quantity 1/B_j satisfies 1/B_j >= max{|J(C_j)|, 1}, which is an upper bound on |J|. This contradicts the abstract's claim that 'a lower bound for the local Jacobian J(x) may also be estimated from the entries of the transfer matrix' and the sentence after Eq. (9) calling B_j 'an upper bound on the derivative of the forward map'. The underestimation of B_j in weakly unstable regions reported in Sections 3.1.5 and 3.2 is a manifestation of this inequality, not merely a convergence or noise artifact. The text should correct the bound direction or explicitly reframe B_j as a lower bound on 1/|J| that is informative only when |J| is sufficiently larger than one.","section":"Section 2, Eq. (9) and Abstract"},{"comment":"The intermediate step B_j = max_i sum_k 1/|J(C_j)| p^k_ij ≈ 1/|J(C_j)| max_i p^{k*}_{ij} <= 1/|J(C_j)| drops the sum over k and assumes that the maximizing target cell receives the full image of the source cell. This approximation is uncontrolled: for |J(C_j)| < 1, the forward image of a source cell is smaller than a target cell, so the transition probability saturates at 1 and B_j becomes independent of 1/|J|. Thus Eq. (9) is not an equality-based estimator but a coarse upper bound with a gap that can be large. The paper should state this limitation explicitly and, ideally, give a quantitative condition on the partition size and Jacobian magnitude under which the gap is small.","section":"Section 2, derivation of Eq. (9)"},{"comment":"The coefficients p^k_ij are not defined unambiguously. The equality sum_j p^k_ij q^n_j := integral over X_i of rho_n(f^{-1}_k(x)) dx defines p^k_ij relative to a particular density rho_n, which does not imply that p^k_ij is a density-independent stochastic matrix. The subsequent interpretation of p^k_ij as the probability that f^{-1}_k(C_i) lies in X_j is a different, map-only object. The derivation of Eqs. (8) and (9) requires the latter; please either define p^k_ij directly as a Ulam-type transition probability and state the approximation, or prove that the two definitions agree in the fine-partition and smooth-density limit.","section":"Section 2, Eq. (5)"},{"comment":"The statement that when the Jacobian vanishes, 'the integral ... <= 1 now provides an upper bound for max_i P_ij' is uninformative as written, because the probability bound P_ij <= 1 already holds by definition. The more relevant consequence is that the approximation in Eq. (22) fails, and no quantitative Jacobian estimate can be extracted from that cell. The non-vanishing-Jacobian assumption is load-bearing for both A_i and B_j and should be stated in the main text before Eq. (5), not only in the appendix.","section":"Appendix, after Eq. (22)"}],"minor_comments":[{"comment":"The word 'discretetization' near the end of Section 2 is a typo; it should read 'discretization'.","section":"Section 2"},{"comment":"The reference title contains 'Qantum'; it should be 'Quantum'.","section":"Reference [24]"},{"comment":"The word 'coeffficients' is a typo; it should read 'coefficients'.","section":"Section 3.2"},{"comment":"The symbol N is used both for the number of control volumes in Eq. (1) and for the degree of the Chebyshev polynomial in Eq. (16). This creates confusion in Section 3.1.4, where N=100 control volumes are used alongside Chebyshev orders N=2,3,4,5; please use distinct symbols or explicitly state the context.","section":"Sections 2 and 3.1.4"},{"comment":"The analytical B_j curves are capped at unity before comparison; the captions should state that the comparison is against min{1/|J|, 1}, so that the reader is aware that saturation is part of the displayed quantity.","section":"Figures 1 and 2 captions"}],"recommendation":"major_revision","confidential_remarks":"The paper has a solid core in the A_i estimator, which is well derived and convincingly validated. The B_j bound direction is wrong as stated but appears fixable by rewording and by making the saturation behavior explicit. I would encourage the authors to also tighten the definition of p^k_ij, since it underpins the theoretical derivation. If these points are addressed, the paper could be a useful contribution to Physica D."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know before you read it. The A_i estimator is legit and numerically convincing; the B_j claim is mislabeled. From the paper's own inequality (Eq. 9), B_j = max_i P_ij ≤ min{|J(C_j)|^{-1}, 1}. So 1/B_j ≥ max{|J(C_j)|, 1} — an upper bound, not the lower bound advertised in the abstract and Section 2. When |J| < 1, B_j saturates at 1 and carries no information about the Jacobian, which is exactly what their numerics show in weakly unstable regions. That is a genuine flaw in the central claim, not a convergence issue.\n\nWhat is actually new: the specific repackaging of the standard Perron-Frobenius relation into the row-sum observable A_i and the column-max observable B_j. The A_i derivation is clean under the stated fine-partition and smooth-Jacobian assumptions, and the numerical tests on the Ulam, continued fraction, cusp, Chebyshev, and coupled Chebyshev maps support it strongly. The paper reproduces the derivation rather than just citing ChaosBook, so the self-citation is not a concern here.\n\nSoft spots beyond the B_j direction error: no code or data shipped, no error bars or convergence studies, and no comparison against existing Jacobian-estimation baselines. The demonstrated scope is low-dimensional discrete maps, and the fine-partition assumption with nonvanishing Jacobians is explicitly stated but not stress-tested near critical points. The A_i estimator is nevertheless simple, parameter-free, and reproducible from the text alone.\n\nThe paper is worth a serious referee: the A_i result is solid and useful for transfer-operator practitioners, and the B_j issue is fixable by correcting the direction, noting the saturation behavior, and being explicit that B_j is useful only in sufficiently expanding regions. The right outcome is major revision, not rejection. I would bring it to a reading group focused on data-driven dynamical systems, and I would cite the A_i estimator in my own work on transfer-operator methods.","headline":"A_i is a clean, useful data-driven estimator of inverse dissipation, but the paper's claimed lower bound on the local Jacobian via B_j is actually an upper bound — the central advertised claim needs correcting.","tokens_in":12079,"tokens_out":1384,"would_cite":true,"duration_ms":13557,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37M25","37D45","37C30","37E05"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that local dissipation and instability fields of a chaotic map can be estimated directly from a transition matrix built from orbit data, without knowing the governing equations.","keywords":["Perron-Frobenius operator","transition matrix","Jacobian estimation","dissipation field","instability field","chaotic maps","data-driven dynamics","state-space partition"],"falsifier":"Estimate $A_i$ and $B_j$ for a one-dimensional map with a critical point (where $|J|$ vanishes) using a partition that places the critical point inside a single cell. The analytical inverse Jacobian diverges at that cell while every transition-matrix entry satisfies $P_{ij}\\le 1$, so $B_j$ cannot reproduce the analytical value; the size of that discrepancy, and whether $A_i$ also fails there, settles how much of the central claim survives.","tokens_in":11088,"feed_emoji":"🌀","tokens_out":11676,"duration_ms":99642,"temperature":0.7,"pith_summary":"This paper aims to establish that the local stability landscape of a chaotic map can be extracted from a purely statistical object: the transition matrix that records how often orbits move between cells of a partition. If correct, local dissipation and instability fields are available for systems whose equations are unknown, because the matrix is estimated directly from trajectory data. The paper derives two estimators from the Perron-Frobenius transport formula: the row sum $A_i$ approximates the inverse dissipation $\\sum_k |J(f_k^{-1}(C_i))|^{-1}$, and the column maximum $B_j$ approximates the inverse Jacobian $|J(C_j)|^{-1}$ from above. Tests on one-dimensional maps and on two-dimensional coupled maps show that $A_i$ agrees with theory robustly while $B_j$ is reliable mainly where the map is strongly expanding.","feed_headline":"Chaos fields read off orbit data, no equations needed","feed_subtitle":"A row-sum and a column-max of the transfer matrix recover local dissipation and instability on chaotic maps.","key_machinery":"The machinery is the finite transition matrix $P_{ij}$ that approximates the Perron-Frobenius operator on a uniform partition $\\{X_i\\}$ of state space, with $P_{ij}$ the probability that mass in cell $j$ lands in cell $i$ in one step. The load-bearing step is to replace the integrand in the Perron-Frobenius transport formula by its value at the cell centroid $C_i$, which turns each matrix entry into a combination of inverse Jacobians and preimage transition probabilities. From that the row sum $A_i$ and the column maximum $B_j$ are exact algebraic targets: $A_i$ aggregates the inverse Jacobians over all preimages of $C_i$, while $B_j$ isolates the single dominant preimage and yields the bound $B_j\\le |J(C_j)|^{-1}$.","core_discovery":"The central claim is that the local Jacobian of a chaotic map is encoded in the transition matrix. For a partition of state space into $N$ equal control volumes with centroids $C_i$, the empirical transition matrix $P$ obeys $A_i=\\sum_j P_{ij}\\approx \\sum_k |J(f_k^{-1}(C_i))|^{-1}$, so the row sums are the local inverse dissipation field, and $B_j=\\max_i P_{ij}\\lesssim |J(C_j)|^{-1}$, so the column maxima give an upper bound on the inverse Jacobian and therefore a lower bound on local stretching. Because $P$ can be counted directly from orbit data, both fields are available without knowing the map. Numerical tests show $A_i$ matches the analytical dissipation field accurately even with noise, while $B_j$ is accurate where the map is strongly expanding and degrades in weakly unstable regions.","pith_inferences":["A natural extension, not tested here, is to apply the estimators to the time-one map of a continuous flow or to a Poincaré section, which would bring ODE systems and experimental data into the same framework.","The robustness of $A_i$ relative to $B_j$ suggests a practical two-step strategy: use $B_j$ to identify unstable regions, then use $A_i$ to quantify the dissipation there; this is a workflow the authors gesture toward but do not work out.","Because entries of $P$ are bounded by 1, cells where $B_j$ saturates at its upper bound flag near-critical points with vanishing Jacobian; the failure mode could itself be used as a detector of critical loci.","A finite-sample error analysis comparing the convergence of $A_i$ and $B_j$ as the orbit length and partition resolution increase would sharpen the method's practical limits; the paper's numerics do not provide such scaling laws."],"forward_implications":["For any discrete-time system with enough orbit data, the transition matrix alone yields a map of where infinitesimal perturbations grow and where phase-space volume contracts, without any model equation.","The $A_i$ field is insensitive to moderate observational noise, so it can serve as a data-driven check of whether a proposed model respects local conservation or dissipation.","The $B_j$ field, though weaker, localizes the strongly unstable regions where chaos control or sensitivity enhancement would be most effective.","The same estimators apply to experimental time series, provided the state space is partitioned consistently, making Jacobian fields accessible when equations are unavailable.","Because $A_i$ aggregates all preimages, the method handles maps with multiple branches, where explicit inverse maps are unavailable or expensive to compute."],"supporting_citations":[{"why":"Supplies the continuous Perron-Frobenius transport formula with preimage summands that the whole derivation starts from.","marker":"[24]"},{"why":"States the density-evolution form of the Perron-Frobenius operator used to define the transition matrix.","marker":"[37]"},{"why":"Provides the analytical Perron-Frobenius expressions for one-dimensional maps used as benchmark targets.","marker":"[43]"},{"why":"Defines the two-dimensional coupled map testbed and its locally expanding, non-hyperbolic regime.","marker":"[44]"},{"why":"Supports the claim that the weakly coupled testbed map is ergodic and mixing.","marker":"[45]"},{"why":"Supports learning the finite transition matrix from observed time-series data.","marker":"[34]"}],"fun_headline_variants":["Transition matrix extracts local chaos fields","Chaos fields from transfer matrix, no equations needed","Orbit data reveals dissipation and instability","Perron-Frobenius operator unlocks chaos structure","Local Jacobian estimated from orbit counts"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The derivation assumes the partition is fine enough and the Jacobian smooth enough that the Jacobian is nearly constant on each cell, and that the Jacobian does not vanish; if either fails, the estimated fields are biased.","fun_headline_variants_meta":{"raw":{"variants":["Transition matrix extracts local chaos fields","Chaos fields from transfer matrix, no equations needed","Orbit data reveals dissipation and instability","Perron-Frobenius operator unlocks chaos structure","Local Jacobian estimated from orbit counts"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000189,"raw_usage":{"total_tokens":1266,"prompt_tokens":809,"completion_tokens":457,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":425,"completion_tokens_details":{"reasoning_tokens":405}},"tokens_in":425,"tokens_out":457,"duration_ms":4596,"temperature":1.0,"reasoning_tokens":405,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T04:40:44.305435+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Estimate $A_i$ and $B_j$ for a one-dimensional map with a critical point (where $|J|$ vanishes) using a partition that places the critical point inside a single cell. The analytical inverse Jacobian diverges at that cell while every transition-matrix entry satisfies $P_{ij}\\le 1$, so $B_j$ cannot reproduce the analytical value; the size of that discrepancy, and whether $A_i$ also fails there, settles how much of the central claim survives.","supporting_citations":[{"cited_title":"Cvitanovi ´c, R","cited_arxiv_id":null,"evidence_quote":"Supplies the continuous Perron-Frobenius transport formula with preimage summands that the whole derivation starts from."},{"cited_title":"Cvitanovi ´c, R","cited_arxiv_id":null,"evidence_quote":"States the density-evolution form of the Perron-Frobenius operator used to define the transition matrix."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the analytical Perron-Frobenius expressions for one-dimensional maps used as benchmark targets."},{"cited_title":"Dettmann, D","cited_arxiv_id":null,"evidence_quote":"Defines the two-dimensional coupled map testbed and its locally expanding, non-hyperbolic regime."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supports the claim that the weakly coupled testbed map is ergodic and mixing."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supports learning the finite transition matrix from observed time-series data."}],"review_version":1}