{"id":"0caf1d35-b3b4-4df0-9ab7-831f6ea802ca","arxiv_id":"2504.12741","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"FLIWB and REM chaos indicators applied to realistic HL-LHC lattices can highlight chaotic regions and give a power-law link between Lyapunov time and stability time, supporting cheaper dynamic aperture estimates.","lead":"This paper tests two fast chaos-detection tools, FLIWB and REM, on realistic HL-LHC accelerator lattices and finds they can flag chaotic particle orbits much faster than traditional long tracking. If the tools hold up, accelerator designers could screen magnet error configurations cheaply before doing the expensive long-term tracking used to estimate dynamic aperture.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Power-law TS–TL fit is performed only on orbits escaping within 10^7 turns, so the claimed extrapolation to higher TS (Section 5) is not supported by the fitted population.","rationale":"The paper's strongest claim is that chaos indicators, specifically the TS–TL power law, can substitute for expensive long-term tracking. This claim breaks if the power law cannot be extrapolated beyond the 10^7-turn horizon used for fitting. The most load-bearing weakness is not merely statistical transferability to new seeds or optics, but a selection effect internal to the fitted dataset: censoring removes exactly the stable orbits one wants to predict, so the fitted relation describes only orbits that escape within the simulation window. The reader identified the restriction to TS < 10^7 as part of the transferability assumption, but did not emphasize that this is a censoring bias that invalidates extrapolation to the uncensored population; hence partial agreement. The paper's own caveat in Section 4.2 supports keeping a CONDITIONAL verdict rather than ACCEPT. Since the demonstrated ability of FLIWB and REM to resolve chaotic structures at moderate turn counts stands independently, REJECT is not warranted. The appropriate verdict remains CONDITIONAL, which is what the reader already assigned, so no change is needed.","tokens_in":28026,"tokens_out":5619,"duration_ms":57849,"concrete_test":"Fit the power law using only data with TS < 10^6 turns, then use the fitted law to predict the mean TS in the band 10^6 < TS < 10^7 for the same seeds and quantify the prediction error. To test the actual >10^7 extrapolation, track a few hundred boundary orbits to 5×10^7 turns (or use the Hénon model of [35] where TS can be computed to 10^8 or beyond) and compare measured TS with predictions from TL evaluated at 10^6 turns. If the power law systematically overpredicts or underpredicts survival, the extrapolation claim in Section 5 fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.2 fits TS = α TL^β after deliberately excluding all initial conditions with TS = 10^7 turns (the 'white region' in Fig. 8), i.e., all orbits that remain stable for the full simulation. The fit therefore characterizes only the sub-population that escapes within the tracking horizon. Section 5 then claims the power law 'may offer a reliable extrapolation to higher TS values', which is the basis for using Lyapunov time as a cheap proxy for long-term dynamic aperture. For that extrapolation to hold, the conditional scaling of escaped orbits must coincide with the survival-time scaling of exactly the censored, longer-lived orbits that are the target of the prediction; no evidence or argument is given for this. Moreover, the weakly chaotic case ζ0 = 0, closest to standard DA computations, yields fit exponents β = 2.0 ± 0.8 (best seed) and β = 2 ± 2 (worst seed), essentially unconstrained, and the TL-based DA fits are poor (χ2_ν ≈ 5). Thus the extrapolation claim is both internally unvalidated and least reliable in the most relevant regime. The paper itself concedes in Section 4.2 that extrapolation abilities need further examination, which tempers the conclusion but does not resolve the selection-bias issue.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper applies the Birkhoff-weighted Fast Lyapunov Indicator (FLIWB) and the Reverse Error Method (REM) to realistic HL-LHC lattice models that include synchrotron motion, with the goal of assessing whether these indicators can serve as cheap proxies for long-term dynamic aperture (DA). The authors implement the indicators in the Xsuite GPU tracking code, study the influence of the shadow-particle parameters (initial displacement epsilon_0 and renormalization interval tau), and produce phase-space maps of stability time, FLIWB, REM, and FMA for two representative magnetic error seeds and three longitudinal amplitudes zeta_0. They then compare DA curves derived from stability time TS and Lyapunov time TL, fit a Nekhoroshev-type scaling law to both, and fit a power-law relation TS = alpha TL^beta to radially averaged data in an attempt to connect short-time chaos indicators to long-term stability. The central conclusion is that the power law may offer reliable extrapolation to high TS values and that TL-based stability regions could substitute for DA in lattice optimization.","tokens_in":28323,"tokens_out":10226,"duration_ms":93510,"significance":"If the extrapolation claim were valid, this would be a practically valuable way to reduce the cost of DA optimization in hadron colliders. The paper is solid in its descriptive content: it reports a careful implementation of FLIWB and REM within a modern GPU tracking framework, includes a parameter-sensitivity study, and is transparent about limitations (e.g., the absence of ground truth for chaotic regions acknowledged in Appendix B.3 and the explicit caveat in Section 4.2 on extrapolation). The numerical evidence, however, does not yet establish the central predictive claim. The power-law fit is conditional on escape within the tracking horizon, the parameters are largely unconstrained for zeta_0 = 0, and the Nekhoroshev-like scaling for TL-based DA fails for zeta_0 = 0. Thus the paper's main value lies in its demonstration of feasibility and in the phase-space maps, rather than in a validated extrapolation tool.","major_comments":[{"comment":"The power-law fit TS = alpha TL^beta is restricted to initial conditions with TS < 10^7 turns, i.e., the 'white region' in Fig. 8, deliberately excluding all orbits that survive the full 10^7-turn tracking horizon. The extrapolation to higher TS values claimed in Section 5 concerns precisely the censored population (TS >= 10^7), yet the fit provides no information about the scaling of that population; the conditional distribution of TS given escape is not the same as the unconditional survival distribution. The text in Section 4.2 correctly identifies the bias and excludes the censored data 'to cope with this problem', but this procedure converts a censored-data problem into a selection-biased regression and does not resolve the issue. To support the extrapolation statement, the analysis would need to include the censored observations (e.g., via survival-analysis methods) or the conclusion should be explicitly limited to the range over which the fit is valid.","section":"4.2 and Fig. 8"},{"comment":"For the configuration closest to standard DA computations, zeta_0 = 0, the power-law fit parameters are effectively unconstrained: the worst seed yields log10(alpha) = -3 +/- 4 and beta = 2 +/- 2, and the best seed yields beta = 2.0 +/- 0.8. With a 100% relative uncertainty on beta, the data at zeta_0 = 0 cannot discriminate between the claimed beta in [2,4] and other scalings, so the statement that the action dynamics 'could have features compatible with diffusion regimes' is not supported by the reported fit in this case. Since zeta_0 = 0 corresponds to the case most relevant to conventional betatron-only dynamic aperture studies, the universality of the relation across all zeta_0 is not established.","section":"4.2 and Figs. 9-10"},{"comment":"The conclusion in Section 5 that 'both measures follow a similar scaling law, effectively described by a Nekhoroshev-like scaling law' is not consistent with Table 2 for the TL-based DA in the zeta_0 = 0 rows, where the reduced chi-squared is chi2_nu = 4.7 (best seed) and 5.01 (worst seed), compared to chi2_nu = 0.03 for the TS-based fits. The model does not describe the TL-based DA data in the standard case without longitudinal coupling, and the text itself acknowledges that 'the quality of the fit is much worse for zeta_0 = 0'. The similar-scaling conclusion should either be restricted to the zeta_0 != 0 cases or the model's failure in the most relevant regime should be addressed before claiming substitutability of TL-based SR for DA.","section":"Table 2 and Section 5"}],"minor_comments":[{"comment":"The phrase 'for the luminosity the beam lifetime optimisation' should read 'for the luminosity and the beam lifetime optimisation'.","section":"Abstract"},{"comment":"The sentence 'hence, for the reminder of this study' contains a typo; 'reminder' should be 'remainder'. The same typo appears later in the same section.","section":"Section 2.4"},{"comment":"The phrase 'This arbitrarily threshold strategy' should be 'This arbitrary threshold strategy'.","section":"Appendix B.1"},{"comment":"Equation (1) appears incorrectly typeset in the preprint (e.g., the '1h' following the multiplication sign and the broken exponent in the Lambert W expression), obscuring the definition of the DA model. The equation should be verified in the final version.","section":"Equation (1)"},{"comment":"The choice of Delta r = 0.01 sigma is described as the 'best compromise' between statistical fluctuations and information loss, but no quantitative criterion is stated; specifying the metric used would improve reproducibility.","section":"Section 4.2 and Fig. 7"},{"comment":"Appendix B.3 correctly notes that no ground truth for chaotic regions is available for the realistic HL-LHC lattice; the wording in Section 3.1 that the indicators 'demonstrate the efficiency' in identifying chaotic dynamics should be tempered to reflect this acknowledged limitation.","section":"Appendix B.3"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take: this is a competent, honest application of two established chaos indicators (FLIWB and REM) to a realistic HL-LHC lattice with synchrotron dynamics, and that is genuinely new relative to the toy-map work in [35]. The paper does several things well: it shows the indicators resolve chaotic layers and regular islands at moderate turn counts, it includes a careful convergence study for the shadow-particle parameters epsilon0 and tau, and it documents the expected agreement between REM and FLIWB while also showing that FMA produces artefacts when the tune is modulated. The DA scaling-law fits are also a plus: the TS-based DA follows the Nekhoroshev-like model well, and the TL-based DA behaves comparably when zeta0 is nonzero. The authors are transparent about their limitations, which I appreciate.\n\nThe soft spots are real but not fatal. The central TS–TL power law is fitted only on orbits that escape within 10^7 turns, after explicitly excluding the censored stable orbits. That is exactly the population the extrapolation is supposed to predict, and no argument is given that the conditional scaling of escapers carries over to the longer-lived orbits. The stress-test concern about selection bias holds up. The fit uncertainties in the weakly chaotic zeta0 = 0 case are enormous (log10(alpha) = -3 +/- 4, beta = 2 +/- 2 for the worst seed), and the TL-based DA fit is poor there (chi2_nu ~ 5). The paper does concede in Section 4.2 that the extrapolation abilities need further examination, which tempers the claim, but the conclusion in Section 5 still leans on that extrapolation as a motivation. The computational-cost advantage of indicators over long tracking is asserted repeatedly but never quantified, and the whole study rests on two seeds and three zeta0 values, which is a narrow base for generalization. None of these are circular or dishonest; the TS and TL measurements are independent and the fits are free-parameter fits.\n\nWho is this for? Accelerator physicists working on DA evaluation and lattice optimisation, especially for HL-LHC and FCC. It is not a breakthrough, but it is a useful step toward cheap chaotic-region proxies. I would send it to a serious referee, with the expectation of major revision: reframe the extrapolation as a hypothesis, validate it on held-out seeds or with longer tracking on a subset, quantify the speedup, and soften Section 5 accordingly.\n\nI would bring it to a reading group focused on accelerator nonlinear dynamics, and I would cite it if I were working on chaos indicators for storage rings.","headline":"A solid exploratory application of FLIWB and REM to a realistic HL-LHC lattice, with a genuinely new TS–TL power-law observation whose extrapolation claim outruns the fit; worth refereeing, but the authors should be pushed to validate the extrapolation on censored and held-out data.","tokens_in":28912,"tokens_out":1263,"would_cite":true,"duration_ms":16938,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper establishes a power-law relation between stability time and Lyapunov time in a realistic LHC lattice, and argues that cheap chaos-indicator scans can replace long-term tracking for dynamic-aperture assessment.","keywords":["chaos indicators","Lyapunov time","stability time","dynamic aperture","FLIWB","reverse error method","HL-LHC","non-linear beam dynamics"],"falsifier":"Take a magnetic-error seed not used in the fit, compute $T_L$ from FLIWB or REM at $10^6$ turns and $T_S$ from tracking to $10^7$ turns, bin by radial amplitude, and check whether $T_S = \\alpha T_L^{\\beta}$ with $\\beta$ in [2,4] reproduces the data throughout the chaotic range; a clear deviation, or a threshold below which the log-log relation bends, would remove the extrapolation claim.","tokens_in":1648,"feed_emoji":"🌀","tokens_out":3017,"duration_ms":88670,"temperature":0.7,"pith_summary":"The paper aims to show that two chaos indicators, FLIWB and REM, computed over about a million turns, can substitute for much longer tracking campaigns when assessing the dynamic aperture of a realistic hadron collider lattice. It argues that chaotic regions identified by these indicators are the sites where slow diffusion eventually drives particles out of the stable region, so the Lyapunov time measured there carries information about the stability time. The central quantitative finding is a power-law relation between stability time and Lyapunov time, $T_S = \\alpha T_L^{\\beta}$ with $\\beta$ in [2,4], which could in principle extrapolate stability estimates toward the multi-hour time scales relevant for operation. If this holds, lattice optimisation could be done with cheap chaos-indicator scans rather than CPU-intensive million-turn tracking.","feed_headline":"One power law links chaos time to beam stability time in the LHC","feed_subtitle":"Fast chaos indicators map the LHC's phase space and could cut million-turn tracking campaigns.","key_machinery":"The load-bearing object is the empirical power law $T_S = \\alpha T_L^{\\beta}$ relating the first-passage stability time to the Lyapunov time, fitted to amplitude-binned averages in the weakly chaotic region. The Lyapunov time itself comes from FLIWB or REM: FLIWB uses a shadow particle whose displacement is renormalised every $\\tau$ turns and weighted with Birkhoff averaging, while REM measures the distance after forward then backward tracking with round-off noise acting as the perturbation. A secondary mechanism is a Nekhoroshev-like formula $D(n)$ for the dynamic aperture as a function of the number of turns, which is fitted to both the $T_S$-based and $T_L$-based stability regions.","core_discovery":"This paper demonstrates that two fast chaos indicators, the Birkhoff-weighted Fast Lyapunov Indicator and the Reverse Error Method, can be applied to a realistic model of the high-luminosity LHC lattice, including synchrotron motion, and reproduce the phase-space structure that would otherwise require tracking up to $10^7$ turns. It establishes that the stability time $T_S$ and the Lyapunov time $T_L$ are related by a power law $T_S = \\alpha T_L^{\\beta}$ with $\\beta$ between 2 and 4 across two magnetic-error seeds and three longitudinal amplitudes, and that a dynamic aperture defined through $T_L$ follows the same Nekhoroshev-like scaling law as the standard $T_S$-based aperture when longitudinal dynamics is present. The authors conclude that Lyapunov-based stability regions can stand in for conventional long-term dynamic-aperture estimates at a fraction of the computational cost, while cautioning that extrapolation of the power law requires further tests.","pith_inferences":["If the power-law extrapolation survives additional magnetic-error seeds, statistical screening of the many error realisations used in collider design becomes feasible: one-million-turn indicator maps could rank order stability across thousands of lattice variants.","The diffusion reading of $\\beta \\in [2,4]$ suggests a testable link to beam-halo loss measurements: loss rates from collimator scans should match the action-diffusion coefficient inferred from Lyapunov-time maps.","The same methodology should port to other storage rings where long tracking is prohibitive, but the paper's own caveat that no ground truth exists for realistic lattices means each new lattice needs a subsample cross-check against long tracking.","A natural extension is to exploit REM's sharp bimodal distribution to set an automatic, iteration-independent chaos threshold and feed the resulting binary maps directly into an optimisation objective for tuning non-linear magnet families."],"forward_implications":["A dynamic aperture evaluated through Lyapunov time follows the same Nekhoroshev-like scaling as the standard stability-time aperture when longitudinal dynamics is present, so Lyapunov-based stability regions can substitute for conventional dynamic-aperture estimates.","REM yields a sharper bimodal distribution of indicator values with a threshold that is nearly independent of the number of turns, enabling fast binary classification of regular and chaotic orbits in realistic lattices.","FLIWB converges more rapidly than plain FLI for regular orbits, improving the separation between regular and chaotic initial conditions at fixed computational cost.","The power law $T_S = \\alpha T_L^{\\beta}$ with $\\beta \\in [2,4]$ may provide extrapolation to higher stability times critical for lattice optimisation, subject to further validation across settings.","Frequency Map Analysis is shown to be unreliable for locating chaotic regions when tune modulation from longitudinal dynamics is present, whereas FLIWB and REM remain informative."],"supporting_citations":[{"why":"Supplies the theoretical regimes (power law and exponential) connecting Lyapunov time to macroscopic instability time that the fitted $T_S$-$T_L$ relation follows.","marker":"[40]"},{"why":"Prior performance study that selected FLIWB and REM as the best indicators for chaotic-orbit classification with few iterations.","marker":"[35]"},{"why":"Provides Birkhoff weighted averaging, the mechanism that gives FLIWB its faster convergence for regular orbits.","marker":"[15, 16]"},{"why":"Defines the Reverse Error Method used for the sharp bimodal regular-chaos classification.","marker":"[2, 17]"},{"why":"Describes the shadow-particle method that estimates Lyapunov indicators when no analytic tangent map is available.","marker":"[16, 56]"},{"why":"Gives the Nekhoroshev-like dynamic-aperture scaling law fitted to both $T_S$- and $T_L$-based stability regions.","marker":"[10, 11, 71]"},{"why":"Underpins the Nekhoroshev theorem behind the long-term stability estimates and dynamic-aperture extrapolation model.","marker":"[67-70]"},{"why":"Supplies the GPU parallel tracking framework that makes the large grids of initial conditions computationally feasible.","marker":"[45, 46]"}],"fun_headline_variants":["Power law links chaos time to LHC beam stability","Lyapunov time predicts LHC beam stability","Fast chaos indicators cut LHC tracking cost","Chaos metrics reveal power law for LHC stability","Birkhoff FLI and Reverse Error speed up LHC analysis"],"cache_read_input_tokens":30976,"weakest_assumption_plain":"The power-law relation is fitted on just two magnetic-error seeds and three longitudinal starting amplitudes, and it is assumed to keep holding for other error realisations, optics, energies, and configurations with beam-beam effects.","fun_headline_variants_meta":{"raw":{"variants":["Power law links chaos time to LHC beam stability","Lyapunov time predicts LHC beam stability","Fast chaos indicators cut LHC tracking cost","Chaos metrics reveal power law for LHC stability","Birkhoff FLI and Reverse Error speed up LHC analysis"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0011,"raw_usage":{"total_tokens":4585,"prompt_tokens":940,"completion_tokens":3645,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":556,"completion_tokens_details":{"reasoning_tokens":3569}},"tokens_in":556,"tokens_out":3645,"duration_ms":30150,"temperature":1.0,"reasoning_tokens":3569,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:23:42.693739+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a magnetic-error seed not used in the fit, compute $T_L$ from FLIWB or REM at $10^6$ turns and $T_S$ from tracking to $10^7$ turns, bin by radial amplitude, and check whether $T_S = \\alpha T_L^{\\beta}$ with $\\beta$ in [2,4] reproduces the data throughout the chaotic range; a clear deviation, or a threshold below which the log-log relation bends, would remove the extrapolation claim.","supporting_citations":[{"cited_title":"Morbidelli and C","cited_arxiv_id":null,"evidence_quote":"Supplies the theoretical regimes (power law and exponential) connecting Lyapunov time to macroscopic instability time that the fitted $T_S$-$T_L$ relation follows."},{"cited_title":"Bazzani, M","cited_arxiv_id":null,"evidence_quote":"Prior performance study that selected FLIWB and REM as the best indicators for chaotic-orbit classification with few iterations."}],"review_version":1}