{"id":"6891da12-3f4d-4355-bbfa-3501ebc57b52","arxiv_id":"1908.04871","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A variable-iteration-number Discrete Lagrangian Descriptor (VIN-DLD) is proposed for unbounded maps and validated on the Hénon map.","lead":"This paper modifies the Discrete Lagrangian Descriptor (DLD) method so it can reveal phase space structures of maps whose orbits escape to infinity, stopping each orbit once it leaves a fixed region. It demonstrates the modified method on the Hénon map, recovering chaotic saddles, KAM tori, and strange attractors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central transfer of manifold-detection properties from fixed-iteration DLDs to VIN-DLDs rests on an unproven large-region approximation that is never quantified or convergence-checked.","rationale":"The reader's weakest_assumption identifies exactly the same issue: the transfer of FIN-DLD properties to VIN-DLD is asserted as an expectation, not proven, and the boundary-crossing tail is not quantified. My independent reading confirms that this is the load-bearing point. The paper's demonstrations are plausible and the method may well work, but the central theoretical justification is a single unproven sentence. The concrete test of varying r would directly probe whether the approximation actually converges and whether r=100 is safe. Since the reader already assigned CONDITIONAL with high confidence, my assessment does not change the verdict; it reinforces the condition. I do not see a stronger objection: the numerical examples are qualitatively consistent with known Hénon map behavior, and the absence of code or reproducibility is secondary. The Birkhoff/KAM extrapolation is also informal, but it is a secondary point compared to the truncation-approximation assumption, because even if the KAM claim is weaker, the manifold-detection claim already rests on the unproven transfer. The paper would be materially strengthened by replacing the 'expected result' with a theorem or a numerically verified convergence estimate in R.","tokens_in":8667,"tokens_out":1851,"duration_ms":21953,"concrete_test":"Repeat the VIN-DLD computation for the Hénon map with A=9.5, B=-1, p=0.05, N=10 on the same initial-condition grid, using interaction radii r=50, 100, 200, 400, and 800. Compute the relative L1 and sup-norm differences between Dp fields for consecutive radii, and also compare the extracted manifold sets (e.g., the level sets of ||grad Dp||) using a discrete Hausdorff distance. If the differences do not decrease monotonically with r, or if the manifold locations shift by more than one grid cell between r=200 and r=400, the large-region approximation underlying the paper's central claim is not supported in the tested regime.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 2, immediately after Eq. (8), asserts: 'if one takes a large interaction region R of the planar phase space, the expected result is that this new variable iteration number definition of DLD approaches the values of the original DLD in Eq. (3). Therefore, the same mathematical properties that were proved for the DLD in Lopesino et al. (2015b) also hold in this case.' This is the load-bearing step of the paper: every later claim about detecting stable/unstable manifolds, chaotic saddles, KAM tori, and strange attractors inherits its validity from this transfer. The step is not proven and, as written, is not even stated as a theorem. The difficulty is concrete: VIN-DLD truncates the sum at the first exit from R, dropping a tail that includes the boundary-crossing jump. For an orbit escaping through the boundary of a circle of radius r=100, that final jump can be comparable to or larger than the entire retained sum, especially for p=0.05 where the contribution is weighted nonlinearly. More importantly, the stopping rule makes the number of terms variable, so Dp(x0,N) is not merely a partial sum of Eq. (3) plus a small error; it is a sum over a different set of terms, and the singular features (unbounded gradients) that encode manifolds could be created, shifted, or destroyed by the truncation. The paper provides no bound of the form |Dp(x0,N) - M_Dp(x0,N)| < C(R,N) tending to zero, and no numerical convergence check in R. The choice r=100 is presented without justification ('without loss of generality'), but the approximation error depends on the geometry of escaping orbits and cannot be dismissed a priori. Because this unproven transfer is the bridge from the established FIN-DLD theory to all VIN-DLD results, it is the single most load-bearing concern.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a modification of Discrete Lagrangian Descriptors (DLDs), termed Variable Iteration Number DLDs (VIN-DLDs), designed for unbounded maps whose orbits escape to infinity. Instead of iterating every initial condition for a fixed number N of forward and backward steps, the sum in Eq. (8) is truncated when the orbit leaves a fixed planar interaction region R (chosen as a circle of radius r=100 in the examples). The central claim is that for large enough R, VIN-DLDs approximate the fixed-iteration DLDs of Eq. (3), so the rigorous manifold-detection properties proved for DLDs in Lopesino et al. (2015b) transfer to VIN-DLDs. The method is illustrated on the Hénon map for three parameter regimes: a chaotic saddle (A=9.5, B=-1), KAM tori and stable/unstable manifolds (A=0.298, B=1), and a strange attractor (A=1.4, B=0.3). The VIN-DLD outputs are compared visually with transit-time (exit-time) distributions. The paper concludes that VIN-DLDs overcome the NaN/overflow problems of fixed-iteration DLDs on unbounded maps and provide a useful tool for phase-space visualization.","tokens_in":9024,"tokens_out":2115,"duration_ms":22812,"significance":"If the transfer of the manifold-detection property were rigorously established, the proposed VIN-DLD would be a practical and useful extension of DLDs to open and unbounded dynamical systems, a class that includes many physically relevant maps. The paper is clearly written and the numerical examples indeed recover the expected phase-space structures of the Hénon map, which lends empirical support to the method. The comparison with exit-time distributions is a helpful sanity check. However, the central mathematical claim that VIN-DLDs inherit the singular-feature properties of fixed-iteration DLDs is only asserted, not proved, and no quantitative validation (convergence in R, error bounds, comparison against known manifold shapes) is provided. The paper also extends a continuous-time ergodic-partition argument to discrete maps without proof. These gaps are significant but addressable within the scope of the manuscript.","major_comments":[{"comment":"The statement that for a large interaction region R the VIN-DLD approaches the fixed-iteration DLD of Eq. (3), and therefore inherits the properties proved in Lopesino et al. (2015b), is the load-bearing step of the paper, yet it is only phrased as an expectation ('the expected result is that...') and is not supported by any bound or proof. For an orbit that exits R, Dp(x0,N) is not a partial sum of Eq. (3) with a small omitted tail: the variable stopping rule changes the set of terms included, and the final jump across the boundary can be comparable to the retained sum. The authors should either prove a quantitative estimate of the form |Dp(x0,N)-MDp(x0,N)| < C(R,N) for suitable orbits, or provide a rigorous statement of when singular features are preserved under this truncation. Failing that, the manifold-detection capability of VIN-DLD remains an empirical observation, not a consequence of the earlier DLD theory.","section":"Section 2, after Eq. (8)"},{"comment":"The numerical results are not validated quantitatively. The paper does not provide a convergence study with respect to the interaction-region radius r (the choice r=100 is presented without justification), does not compare the detected manifolds against analytically known expressions (e.g., the stable/unstable manifolds of the Hénon map's fixed points), and gives no error metric for the claimed agreement with transit-time distributions. Since the central methodological claim is about approximation in R, a simple log-log plot of a suitable norm of (Dp - MDp) versus r, or a comparison of detected manifold points as r increases, would substantially strengthen the argument. Without such a check, the reader cannot assess whether r=100 is in the asymptotic regime or whether the observed structures are robust to the truncation.","section":"Section 3, Figures 2-5"},{"comment":"The claim that the iteration-averaged VIN-DLD <Dp> in Eq. (12) detects KAM tori via the connection with Birkhoff's Ergodic Partition Theorem is not justified. The cited theorem and its Lagrangian-descriptor application in Lopesino et al. (2017) are formulated for continuous-time systems, and the paper argues only that Eq. (3) 'can be interpreted as a discretized version' of the continuous descriptor. This is not a proof that the ergodic-partition result carries over to discrete maps, especially under the variable truncation convention. The authors should either provide a proof for the discrete case or clearly mark this as a heuristic analogy. Since KAM tori recovery is one of the three main advertised applications, this gap is load-bearing for the corresponding claim.","section":"Section 3, KAM tori discussion and Eq. (12)"}],"minor_comments":[{"comment":"There are numerous typographical errors and misspellings, e.g., 'dynamical behavior' instead of 'dynamical behavior' (Section 1), 'exapmple' and 'increaseing' (Section 3), 'iteartions' and 'demosntrate' (Section 3), 'corrsponds' in figure captions. A careful proofreading pass is needed.","section":"Throughout"},{"comment":"The definition of N±x0 as a maximum over k is slightly ambiguous: it implicitly requires that the orbit stays inside R for all intermediate steps. This should be stated explicitly, since the formula alone does not encode the 'until it leaves R' stopping rule.","section":"Section 2, Eq. (7)"},{"comment":"The claim that the shape of R is not important 'without loss of generality' is plausible but not demonstrated; for non-circular R, the boundary-crossing jump and the stopping rule change. A brief remark that the expected approximation property would hold for any bounded R with the same asymptotic reasoning would be helpful.","section":"Section 2, paragraph on the shape of R"},{"comment":"The figures are presented without colorbars or axis labels in several panels, making it difficult to interpret the Dp value ranges. Adding colorbars and labeled axes would improve the reproducibility and readability of the results.","section":"Figure 2, 3, and 5"},{"comment":"The comparison between VIN-DLD and transit-time distributions is purely visual. While the agreement is encouraging, a quantitative similarity measure (e.g., correlation or mutual information) would make the validation more convincing.","section":"Section 3, transit time comparison"}],"recommendation":"major_revision","confidential_remarks":"The paper is methodologically promising but the central transfer-of-properties claim is asserted rather than proven, and the numerical validation lacks convergence checks. The required fixes are within scope: adding a proof or a rigorous numerical convergence study in R, and justifying the discrete ergodic-partition analogy. I do not see indications of citation or authorship issues. The paper fits the journal's scope, but should not be published in its present form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a useful practical note, not a theorem paper. The variable-iteration stopping rule is the discrete analogue of what has already been done for continuous-time LDs, and the author says so. The contribution is Eq. (8), the demonstration on three Hénon parameter regimes, and the comparison with exit-time plots. That is worth having on record for people who use Lagrangian descriptors on open maps.\n\nWhat is genuinely good: the method is simple and clearly specified; the figures recover the expected Hénon structures—chaotic saddle for A=9.5, KAM islands and manifolds for the area-preserving case, and the strange attractor for A=1.4; and the exit-time comparison is a sensible sanity check. The author is honest that the variable-iteration idea is inspired by escape-time plots, and the citations cover the continuous-time precedents. I do not see a citation problem.\n\nThe soft spot is exactly the one flagged in the stress-test note. Section 2 says that for a large interaction region R, VIN-DLD approaches FIN-DLD and therefore inherits the manifold-detection properties proved in Lopesino et al. That is the load-bearing step, and it is not proved or even stated with a quantified error. The truncation is not a harmless tail: the first step that exits R is omitted from the sum, and for a chaotic orbit that jump can be of the same order as the whole retained sum, especially with p=0.05. So it is not obvious that the singular features survive unchanged. No convergence study in R is given, and r=100 appears without justification. This is the main reason I would not call the paper's central claim established. The Birkhoff/ergodic-partition argument for KAM tori is also extrapolated from continuous time to maps without proof, though that is a smaller issue and likely true in the intended sense.\n\nThere is also no code or data, and no quantitative validation against analytically known manifold shapes. Agreement with exit-time plots is reassuring but those plots obey the same truncation, so it is partly a self-consistency check.\n\nOverall, the paper knows what it is and the prose is clear. The central gap is addressable rather than fatal. If the author added a convergence check in R and, better, an error estimate or a proof under reasonable hypotheses, it would become a solid methods note. As it stands I would send it to referees with that request rather than desk reject it. The audience is applied dynamical-systems people who want a working recipe, not readers looking for new theory. I would not cite it for the theorem transfer, but I would point to it as a clean statement of the variable-iteration recipe.","headline":"A useful, clearly written practical extension of DLDs to unbounded maps, but the central claim that a large interaction region transfers the manifold-detection theorems is asserted, not proved, and the numerics are illustrative rather than validating.","tokens_in":9588,"tokens_out":3217,"would_cite":false,"duration_ms":35988,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37M99","37D45","37C05","37E30"],"pacs":["05.45.-a"],"model":"deepseek-v4-flash","headline":"By truncating each orbit's descriptor sum at the moment it exits a fixed interaction region, this paper extends Discrete Lagrangian Descriptors to unbounded maps and shows the truncated field still carries stable and unstable manifolds…","keywords":["Discrete Lagrangian descriptors","variable iteration number","unbounded maps","Hénon map","stable and unstable manifolds","chaotic saddle","KAM tori","strange attractor"],"falsifier":"On the Hénon map with $A=9.5$, $B=-1$, recompute $D_p$ with the same grid and $N=10$, $p=0.05$, for interaction radii $r=50$, $r=100$, and $r=200$; if the singular-feature curves extracted from $\\|\\nabla D_p\\|$ drift or disappear as $r$ grows, the large-region approximation is not working. A more direct check is to compare $D_p(x_0,N)$ with the fixed-iteration sum for an orbit whose escape happens exactly at iteration $k$: if the boundary-crossing jump is comparable to the retained sum, the transfer of the theorem lacks uniformity.","tokens_in":8411,"feed_emoji":"🌀","tokens_out":10099,"duration_ms":93553,"temperature":0.7,"pith_summary":"This paper proposes a way to apply Discrete Lagrangian Descriptors to maps whose orbits escape to infinity: instead of iterating every initial condition a fixed number of steps, it stops each orbit as soon as it leaves a chosen interaction region in the plane. The variable-iteration sum is expected to approach the fixed-iteration descriptor when that region is large, so the known result that descriptor singularities trace stable and unstable manifolds should carry over. The paper demonstrates the device on the Hénon map, recovering a chaotic saddle, the KAM tori of an area-preserving case, and a strange attractor, and it shows the same calculation yields the classical escape-time distribution used for comparison. A sympathetic reader would care because the fixed-iteration descriptor fails on unbounded maps once escaping orbits produce huge or undefined values, while this truncation converts escape into a benign stopping signal.","feed_headline":"Escaping orbits no longer blind the Hénon map's phase-space structure","feed_subtitle":"Stopping orbits at a fixed region reveals chaotic saddles, KAM tori, and strange attractors.","key_machinery":"The load-bearing mechanism is the stopping-time rule $N^\\pm_{x_0}=\\max_{k=1,\\dots,N}\\{k\\mid f^{\\pm k}(x_0)\\in R\\}$, which truncates the $\\ell^p$-arclength sum (with $0<p\\le 1$) at the last iterate inside a fixed interaction region $R$, here a circle of radius $r=100$. The rule turns escape into a legitimate end-of-sum signal and therefore prevents the NaN and overflow contamination produced by fixed-iteration descriptors on unbounded maps; it also makes the transit-time count $T_{x_0}=N^+_{x_0}+N^-_{x_0}$ available at no extra cost. The formal justification is transfer from the original descriptor: for large $R$ the new sum is expected to approach Eq. (3), so the previously proved singular-feature theorem is invoked for the variable-iteration version.","core_discovery":"The paper's central claim is that the Variable Iteration Number Discrete Lagrangian Descriptor, $D_p(x_0,N)=\\sum_{i=-N^-_{x_0}}^{N^+_{x_0}-1}\\|x_{i+1}-x_i\\|^p$ with $N^\\pm_{x_0}$ the number of forward/backward iterates of $x_0$ that stay inside a fixed planar region $R$, replaces the fixed-iteration descriptor $\\mathcal{M}^D_p(x_0,N)$ for unbounded maps without sacrificing the manifold-detection property. The argument is that for $R$ large enough the truncated sums approximate the fixed-iteration sums, so the points where $D_p$ is non-differentiable still locate stable and unstable manifolds, exactly as proved for the original descriptor. On the Hénon map the method is shown to reveal the chaotic saddle for parameters $(A,B)=(9.5,-1)$, the KAM tori and surrounding stable/unstable manifolds for $(0.298,1)$, and the strange attractor for $(1.4,0.3)$, and the gradient of $D_p$ extracts the manifolds directly.","pith_inferences":["I would expect the main uncontrolled error to be the single large jump an orbit takes when it crosses the boundary of $R$: that jump is present in $D_p$ for escaped orbits but absent for orbits that linger inside, so two nearby initial conditions can differ by an amount comparable to the whole retained sum.","A natural strengthening, not attempted here, would be to prove a bound such as $|D_p-\\mathcal{M}^D_p|\\le C(r,N)$ under hyperbolicity or a uniform escape bound; with that, the heuristic 'large $R$' transfer becomes a theorem rather than an expectation.","Because the stopping rule is purely spatial, it is plausible that the construction extends to non-autonomous or higher-dimensional invertible maps and to maps whose escape sets are anisotropic; those regimes are untested in the paper."],"forward_implications":["The same orbit data used for $D_p$ also produce the transit-time distribution $T_{x_0}$, so escape statistics for the map come from the same computation.","The gradient magnitude $\\|\\nabla D_p\\|$ yields stable and unstable manifolds directly, without separately integrating manifolds.","Iteration-averaged values $\\langle D_p\\rangle=D_p/N$ draw the KAM tori of regularity islands as smooth contours, where classical exit-time plots are flat and uninformative.","Because escape no longer contaminates the scalar field, larger maximum iteration counts $N$ can expose finer layers of chaotic-saddle structure.","The construction is stated for general invertible maps, so the Hénon tests instantiate a general recipe rather than a one-off calculation."],"supporting_citations":[{"why":"Supplies the original fixed-iteration DLD definition and the singular-feature theorem for stable and unstable manifolds that the extension preserves.","marker":"Lopesino et al. (2015b)"},{"why":"Defines the Hénon map and its strange-attractor parameter values used as the test problem.","marker":"Hénon (1976)"},{"why":"Proves the hyperbolic chaotic-saddle regime for the Hénon parameters, fixing the target structure.","marker":"Devaney and Nitecki (1979)"},{"why":"Gives the classical average/exit-time distribution used as the comparison diagnostic.","marker":"Meiss (1997)"},{"why":"Establishes KAM-tori trapping in the area-preserving Hénon case, grounding the regular-island test.","marker":"de Oliveira et al. (2019)"},{"why":"Provides the theoretical connection between averaged Lagrangian descriptors and invariant sets via the ergodic partition.","marker":"Lopesino et al. (2017)"},{"why":"Supplies the ergodic-partition theorem invoked for reading KAM tori from averaged descriptor contours.","marker":"Mezic and Wiggins (1999)"}],"fun_headline_variants":["Stopping orbits at a boundary reveals Hénon's hidden structure","Escape-time trick unveils manifolds in unbounded maps","Variable iteration descriptor reveals Hénon chaos","New escape-aware descriptor maps unbounded dynamics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that choosing the interaction region $R$ large enough makes the truncated variable-iteration sums close enough to the fixed-iteration descriptor that the proved manifold-detection property transfers; the paper states this as an expected result rather than proving a quantitative bound in $R$ and $N$.","fun_headline_variants_meta":{"raw":{"variants":["Stopping orbits at a boundary reveals Hénon's hidden structure","Escape-time trick unveils manifolds in unbounded maps","Variable iteration descriptor reveals Hénon chaos","New escape-aware descriptor maps unbounded dynamics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000491,"raw_usage":{"total_tokens":2406,"prompt_tokens":928,"completion_tokens":1478,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":544,"completion_tokens_details":{"reasoning_tokens":1414}},"tokens_in":544,"tokens_out":1478,"duration_ms":11604,"temperature":1.0,"reasoning_tokens":1414,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:29:01.459606+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On the Hénon map with $A=9.5$, $B=-1$, recompute $D_p$ with the same grid and $N=10$, $p=0.05$, for interaction radii $r=50$, $r=100$, and $r=200$; if the singular-feature curves extracted from $\\|\\nabla D_p\\|$ drift or disappear as $r$ grows, the large-region approximation is not working. A more direct check is to compare $D_p(x_0,N)$ with the fixed-iteration sum for an orbit whose escape happens exactly at iteration $k$: if the boundary-crossing jump is comparable to the retained sum, the transfer of the theorem lacks uniformity.","supporting_citations":[{"cited_title":"and Nitecki, Z","cited_arxiv_id":null,"evidence_quote":"Proves the hyperbolic chaotic-saddle regime for the Hénon parameters, fixing the target structure."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the classical average/exit-time distribution used as the comparison diagnostic."},{"cited_title":"J., Wiggins, S., and Mancho, A","cited_arxiv_id":null,"evidence_quote":"Provides the theoretical connection between averaged Lagrangian descriptors and invariant sets via the ergodic partition."},{"cited_title":"and Wiggins, S","cited_arxiv_id":null,"evidence_quote":"Supplies the ergodic-partition theorem invoked for reading KAM tori from averaged descriptor contours."}],"review_version":1}