Pith. sign in

REVIEW 4 major objections 6 minor 77 references

A machine-learning surrogate can predict chaotic three-body scattering outcomes with 88% accuracy, and its remaining errors trace to chaos itself, not model failure.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 20:00 UTC pith:SDMGLQRO

load-bearing objection A genuinely new ML dataset and classifier for binary–single scattering outcomes, worth refereeing, but the 88% accuracy is on finite-time labels and the chaos-vs-model-inadequacy conclusion is not yet established. the 4 major comments →

arxiv 2607.16763 v1 pith:SDMGLQRO submitted 2026-07-18 astro-ph.GA astro-ph.SR

Predicting the unpredictable: binary--single scattering with machine learning

Ahmad Farhani Asl (1) , David Fonseca Mota (2) , Dennis Fremstad (2) , Fatemeh Rahimi (1) ((1) Department of Physics , Institute for Advanced Studies in Basic Sciences (IASBS) , Zanjan , Iran , (2) Institute of Theoretical Astrophysics
show 3 more authors
University of Oslo Oslo Norway)
This is my paper
classification astro-ph.GA astro-ph.SR
keywords binary–single scatteringthree-body problemchaotic scatteringmachine learningXGBoostbasin entropystellar dynamicshierarchical triples
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to show that the final state of a binary–single stellar encounter — ionization, flyby, exchange, or hierarchical triple — can be predicted accurately and quickly from the initial conditions, despite the three-body problem being chaotic. The authors train a gradient-boosted tree classifier on about a million numerically integrated scattering experiments and report 88.32% test accuracy and 98.77% top-2 accuracy, with well-calibrated probability estimates. They argue that the residual misclassifications are not a failure of the model but the imprint of intrinsically ambiguous regions of parameter space near the chaotic boundary between exchange and hierarchical outcomes, as quantified by basin entropy. If the claim holds, it means a five-orders-of-magnitude speedup over direct integration is available for predicting encounter outcomes in dense stellar systems, filling a gap left by analytic cross-section approximations.

Core claim

The central claim is that a supervised machine-learning model can learn the mapping from initial encounter parameters to final-state class in binary–single scattering with high accuracy, and that the errors that remain are due to the intrinsic chaos of the three-body problem, not to model inadequacy. Concretely, an XGBoost classifier — a gradient-boosted ensemble of decision trees — trained on a class-balanced dataset of about a million numerical simulations achieves 88.32% accuracy and 98.77% top-2 accuracy on a held-out test set, with per-class F1 scores between 0.85 and 0.93 and top-label expected calibration error near 0.0018. Feature importance shows that binary hardness H is the domina

What carries the argument

The central machinery is a gradient-boosted decision-tree ensemble (XGBoost) trained on a set of physically motivated dimensionless features of the encounter: binary hardness H, intruder mass fraction q3, focusing parameter R, Safronov number, energy penetration, angular momentum ratio, and others. The load-bearing identity is the empirical dominance of hardness H, which carries roughly 30% of the model's decision power and aligns with the Heggie–Hut hard/soft binary dichotomy. The complementary mechanism is basin entropy, computed in the H–q3 plane, which quantifies the fractal ambiguity of the exchange–hierarchical boundary and provides a dynamical explanation for where and why the classif

Load-bearing premise

The load-bearing premise is that the final states assigned at the fixed stopping time of three encounter time units are the true permanent outcomes; the paper notes that about 15% of simulations — the hierarchical configurations — are still evolving when labeling occurs, so if those labels are time-dependent, the reported accuracy and the attribution of errors to chaos only describe the classification at that finite horizon.

What would settle it

Take a random sample of simulations classified as hierarchical (and a control sample from the other three classes), continue the integrations for ten to a hundred encounter times, and record how many change final-state class. If a substantial fraction of hierarchical labels later ionize, exchange, or fly by, then the reported 88% accuracy and the 'errors come from chaos' conclusion characterize a finite-time labeling convention rather than the asymptotic scattering problem.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Star cluster Monte Carlo simulations that currently use analytic cross-section formulas could replace per-encounter outcome probabilities with this surrogate's calibrated predictions at negligible computational cost.
  • The roughly 10^5 speedup per encounter makes on-the-fly generation of encounter outcome tables across continuous parameter grids practical, rather than precomputing millions of integrations.
  • A hybrid scheme becomes attractive: route high-confidence predictions through the classifier and reserve direct N-body integration for low-confidence, ambiguous encounters near the chaotic boundary.
  • The calibrated probabilities enable the surrogate to serve as an uncertainty-aware predictor, not just a labeler, within the sampled parameter space.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The paper's fixed stopping time of three encounter times leaves the 15% of cases labeled 'hierarchical' still evolving; an immediate empirical test is to integrate those cases far longer and measure how many later change class — if many do, the reported accuracy and chaos-attribution conclusion describe a finite-time classification, not asymptotic final states.
  • The same feature set and training pipeline could be extended to predict continuous post-encounter quantities — final semi-major axis, eccentricity, recoil velocity — which are more directly useful to population-synthesis models than discrete class labels.
  • Because the training set is class-balanced, the model's calibrated probabilities reflect the branching ratios under the paper's specific parameter sampling; applying the surrogate to a different encounter distribution (e.g., different mass function or velocity dispersion) would require re-validation or re-calibration, as the paper itself notes.
  • The elevated basin entropy at the exchange–hierarchical boundary suggests a fundamental performance ceiling for any classifier in those regions; confidence-based selective prediction is therefore a more natural deployment than expecting the model to be certain everywhere.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper trains an XGBoost multiclass classifier on approximately one million numerically integrated binary–single scattering experiments to predict one of four outcome classes (ionization, flyby, exchange, hierarchical). The model achieves 88.32% test accuracy and 98.77% top-2 accuracy on a class-balanced holdout set. The authors report well-calibrated probabilities, identify binary hardness as the dominant feature, and interpret remaining misclassifications as originating from intrinsically chaotic regions near the exchange–hierarchical boundary. They propose the surrogate as a fast replacement for direct N-body integrations in cluster simulations. The paper also includes a validation of Heggie's law accounting for gravitational focusing and a basin-entropy analysis of the chaotic boundary.

Significance. If the central claim holds—that a fast ML surrogate can predict binary–single final states with high accuracy and that residual errors are dominated by intrinsic chaos—this would be a practically valuable tool for Monte Carlo cluster simulations, where millions of encounters must be evaluated quickly. The paper's strengths include a genuinely held-out test set, a large simulation dataset, explicit class-balanced training, and careful calibration diagnostics. The reported 10^5 speedup and the inclusion of resonant outcomes that analytic cross-section formulae miss are notable. However, the significance is conditional on the ground-truth labels representing true asymptotic final states; the current stopping criterion leaves 15% of simulations still evolving, which undermines the interpretation of both accuracy and the chaos-attribution claim.

major comments (4)
  1. [Section 2.2] The stopping-time convention is the paper's central methodological choice. The text states: 'We adopt a fixed integration time of tmax = 3 tenc ... At this time, approximately 85% of simulations have already reached a permanent final state ... the remaining 15% (hierarchical configurations) are still evolving but are classified accordingly.' The labels are therefore finite-time classifications, not asymptotic final states. The reported test accuracy (Section 4.1), confusion matrix (Figure 3), and calibration metrics (Section 4.2) all characterize prediction of these finite-time labels. The conclusion that residual errors arise from chaotic regions rather than model inadequacy (Section 5, bullet 4) is only meaningful if the labels at tmax are stable; otherwise a substantial fraction of 'hierarchical' labels are transient and may later ionize or exchange. The authors acknowledge ambiguity
  2. [Section 4.4] The 'chaotic boundary' is operationally defined as 'misclassified samples with confidence < 0.7' (Table 5). This is a model-based definition: it selects samples where the classifier is both wrong and uncertain. Low confidence can arise from label noise—especially from the non-asymptotic hierarchical labels of Section 2.2—from overlapping feature distributions, or from model miscalibration. Using this definition to conclude that misclassifications are 'due to intrinsically ambiguous regions of parameter space' is therefore circular. The basin entropy (0.32±0.14 bits) computed in the (H, q3) plane measures label sensitivity to initial conditions only if the labels themselves are converged; with finite-time labels it can be inflated by transient hierarchical states. A concrete non-circular test: retrain the same model on labels obtained with a stricter convergence criterion (or longer integ
  3. [Section 4.1 / Section 2.3] The performance numbers are computed on a class-balanced test set (each class ~49,500 samples), obtained by downsampling to the minority class (ionization). The natural encounter distribution in the dataset is highly imbalanced (flyby 46.4%, exchange 30.8%, hierarchical 16.1%, ionization 5.1% as stated in Section 2.3). The reported 88.32% accuracy and 0.288 log loss are therefore balanced-test metrics, not expected performance under the physical encounter distribution that the surrogate would face in cluster simulations (Section 6). The paper should report either accuracy, log loss, and calibration under the original class distribution, or a clearly labeled set of natural-distribution metrics. This is load-bearing for the claimed practical utility.
  4. [Section 2.3] The class counts and percentages are internally inconsistent. The text says '4.8×10^6' simulations are generated, 'less than 2% (86,772)' are excluded, and then lists: flyby 2,274,194 (46.4%), exchange 1,506,638 (30.8%), hierarchical 781,969 (16.1%), ionization 247,684 (5.1%). The sum of these four counts is 4,810,485, which exceeds 4.8×10^6 and also exceeds the number of retained simulations (4.8×10^6 − 86,772 = 4,713,228). The percentages also do not match the counts (e.g., flyby is 48.3% of the retained total, not 46.4%). Please correct the total number and class counts, or explain the discrepancy, because the training-set size and class-balance procedure depend on these numbers.
minor comments (6)
  1. [References] Fregeau et al. (2004) appears twice with different page numbers (ApJ 600, 101 and MNRAS 352, 1) but the same author list and year; Pinheiro et al. (2025) is also duplicated (A&A 693, A246 appears twice). Please check and consolidate.
  2. [Section 6 vs Section 4.2] Section 6, bullet 2 refers to 'After post-hoc calibration', but Section 4.2 does not describe a post-hoc calibration step; it reports the raw XGBoost outputs as already well-calibrated. Either describe the calibration procedure in Section 3.2 or remove 'post-hoc'.
  3. [Table 2] The 'Binary phase Φ' is defined as 'Binary true anomaly', but the initial condition places the binary at pericenter, so Φ=0 initially. The text later mentions that Φ and the separation ζ are 'at encounter'. Please clarify whether these are computed at the moment of closest approach of the intruder, and how.
  4. [Appendix B / Table D.1] Some feature definitions are abbreviated (A, S, etc.) and would benefit from explicit signposting to Table 2. Also, the caption says 'brought in' instead of 'given in' or 'defined in'.
  5. [Appendix C] The mean Δa/a values in Table C.2 are strongly affected by outliers (e.g., median is ~0.0004 while mean is 24.98 in the first bin). Consider reporting robust statistics to support the claimed transition at H≈2.25.
  6. [Section 5] The statement 'Zero-probability predictions correspond to phase space regions dynamically forbidden for that final-state class' is too strong; the paper itself notes that zero probability is 'not proof of physical impossibility'. Rephrase to avoid implying dynamical exclusion from a purely empirical classifier.

Circularity Check

1 steps flagged

No significant circularity: holdout evaluation and independent basin-entropy support the central claim; one minor self-definitional phrasing around 'chaotic boundary'.

specific steps
  1. self definitional [Section 4.4 (Table 5 and basin-entropy paragraph)]
    "Defining the chaotic boundary as misclassified samples with confidence < 0.7, we identify 12,153 such cases (6.1% of the test set). ... The fractal nature of the chaotic boundary is quantified by basin entropy (Daza et al. 2016). In the two-dimensional phase space of hardness H and intruder mass ratio q3, the basin entropy is 0.32±0.14 bits."

    The explanatory category 'chaotic boundary' is first operationalized as the model's own misclassified low-confidence predictions, so the later statement that misclassifications are caused by the chaotic boundary is partly true by definition. The independent basin-entropy calculation for the exchange–hierarchical pair provides external support and prevents the central claim from collapsing entirely, but the wording shifts 'chaotic boundary' from a model-error set to an intrinsic data property, creating a mild self-referential conflation.

full rationale

The main result — 88.32% test accuracy and 98.77% top-2 accuracy — is evaluated on an unseen holdout set explicitly withheld from training and hyperparameter tuning, so the prediction is not a refit of training data. The target labels come from REBOUND/IAS15 integrations with a fixed classification procedure, not from the ML model, so there is no fitted-input-called-prediction structure. The basin-entropy analysis (Daza et al. 2016) is an external metric computed on the outcome labels in the H–q3 plane, independent of the XGBoost model's confidence outputs; this gives genuine, non-circular support to the claim that residual errors concentrate in intrinsically ambiguous regions. The single mild issue is the phrase 'chaotic boundary' being defined in Section 4.4 as misclassified samples with confidence < 0.7 and then used rhetorically as if it were an independent physical property; however, the basin-entropy numbers are computed separately, so the central argument does not reduce to the definition. Self-citations (e.g., Rostami Shirazi et al. 2024) appear only in introductory context and are not load-bearing for the ML derivation or the chaos interpretation. The finite-time classification of hierarchical systems (tmax = 3 tenc, ~15% still evolving) is a real validity limitation for the claim about true asymptotic final states, but it concerns target-label semantics, not circularity between inputs and predictions.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central result rests on simulation labels and modeling choices (integration horizon, balanced sampling, feature set) rather than new physical postulates. No new particles, forces, or conserved quantities are introduced.

free parameters (4)
  • Integration horizon factor tmax/tenc = 3 = 3
    Chosen by hand; defines when outcomes are labeled. 15% of runs still evolving are labeled hierarchical, so the horizon directly shapes ground-truth labels and all accuracy numbers.
  • Chaotic-boundary confidence threshold = 0.7
    Defines “chaotic boundary” samples in Table 5 (N = 12,153); not derived. Changing the threshold changes the boundary population and the chaos interpretation.
  • Class-balance downsampling size = 247,684 per class
    All classes are downsampled to the ionization class size; evaluation on a balanced test set is not weighted by physical encounter-rate priors.
  • XGBoost hyperparameters = max_depth=18, lr=0.02, n_estimators=2000, subsample=0.9, colsample_bytree=0.8, min_child_weight=8, gamma=0.4, alpha=8e-4
    Tuned by TPE on validation; standard ML fitting choices, not physical constants, but the central performance claim depends on them.
axioms (5)
  • domain assumption Newtonian point-mass gravity is sufficient for the sampled parameter range
    Used throughout; relativistic effects are claimed negligible for pericenters >10 Rs (§2.1).
  • domain assumption IAS15 integration with ε = 10^-9 yields correct scattering outcomes
    Assumes integrator truncation errors do not change final-state labels; no convergence cross-check is reported (§2.2).
  • ad hoc to paper Instantaneous pairwise energy/escape criteria at tmax identify asymptotic final states
    Load-bearing for ground truth; contradicted in part by the paper's own statement that 15% of hierarchical systems are still evolving (§2.2, §2.3).
  • domain assumption Sampled parameter distributions represent the encounter distribution of interest
    Log-uniform ranges over masses, separations, velocities, and isotropic angles; model validity outside this distribution is untested, as the authors note (§2.1, §6).
  • ad hoc to paper Low-confidence misclassification implies intrinsic dynamical ambiguity rather than model deficiency
    No baseline rerun or model-capacity experiment is provided; basin entropy is suggestive but does not separate label noise from model error (§5).

pith-pipeline@v1.3.0-alltime-deepseek · 14904 in / 14872 out tokens · 130550 ms · 2026-08-01T20:00:54.754983+00:00 · methodology

0 comments
read the original abstract

Binary--single encounters occur frequently in stellar systems, constitute a non-integrable chaotic three-body problem, and are computationally expensive when treated with N-body codes. We aim to construct an accurate and physically interpretable surrogate model for binary--single scattering final states (ionization, flyby, exchange, or hierarchical), evaluate the reliability of its probabilistic predictions, and trace its predictive failures to the underlying chaotic nature of the scattering problem. We trained an XGBoost multiclass classifier on a class-balanced dataset of numerically integrated binary--single scattering experiments, using physically motivated features. The model achieves a test accuracy of 88.32 percent and a top-2 accuracy of 98.77 percent. Feature importance indicates binary hardness as the dominant predictor. Misclassifications concentrate near the chaotic boundary separating exchange and hierarchical outcomes. Basin-entropy analysis reveals that such misclassifications are due to intrinsically ambiguous regions of parameter space. Therefore, a fast machine-learning surrogate can predict binary--single scattering final states with high accuracy, and residual errors arise from chaotic regions rather than model inadequacy.

Figures

Figures reproduced from arXiv: 2607.16763 by (2) Institute of Theoretical Astrophysics, Ahmad Farhani Asl (1), David Fonseca Mota (2), Dennis Fremstad (2), Fatemeh Rahimi (1) ((1) Department of Physics, Institute for Advanced Studies in Basic Sciences (IASBS), Iran, Norway), Oslo, University of Oslo, Zanjan.

Figure 1
Figure 1. Figure 1: Initial configuration of the binary–single scattering ex [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Top 10 feature importances from the XGBoost model. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Confusion matrix for the four-class classification on the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

77 extracted references · 1 canonical work pages

  1. [1]

    and Henry, Todd J

    Raghavan, Deepak and McAlister, Harold A. and Henry, Todd J. and Latham, David W. and Marcy, Geoffrey W. and Mason, Brian D. and Gies, Douglas R. and White, Russel J. and ten Brummelaar, Theo A. , title =. The Astrophysical Journal Supplement Series , volume =. 2010 , doi =

    2010

  2. [2]

    , keywords =

    Mind Your Ps and Qs: The Interrelation between Period (P) and Mass-ratio (Q) Distributions of Binary Stars. , keywords =. doi:10.3847/1538-4365/aa6fb6 , archivePrefix =. 1606.05347 , primaryClass =

    Pith/arXiv arXiv

  3. [3]

    Milone, A. P. and Piotto, G. and Bedin, L. R. and others , title =. A&A , volume =. 2012 , doi =

    2012

  4. [4]

    Milone, A. P. and Piotto, G. and Renzini, A. and Marino, A. F. and Bedin, L. R. and Anderson, J. and others , title =. MNRAS , volume =. 2016 , doi =

    2016

  5. [5]

    and Gillessen, S

    Pfuhl, O. and Gillessen, S. and Eisenhauer, F. and Genzel, R. and others , title =. ApJ , volume =. 2015 , doi =

    2015

  6. [6]

    and Sollima, A

    Lucatello, S. and Sollima, A. and Gratton, R. and Vesperini, E. and D'Orazi, V. and Carretta, E. and Bragaglia, A. , title =. A&A , volume =. 2015 , doi =

    2015

  7. [7]

    , keywords =

    The fraction of binary systems in the core of five Galactic open clusters. , keywords =. doi:10.1111/j.1365-2966.2009.15676.x , archivePrefix =. 0909.1277 , primaryClass =

    arXiv 2009

  8. [8]

    and Milone, A

    Simioni, M. and Milone, A. P. and Bedin, L. R. and Aparicio, A. and Piotto, G. and Vesperini, E. and Hong, J. , title =. ApJ , volume =. 2016 , doi =

    2016

  9. [9]

    I - Numerical experiments for equal masses

    Binary-single star scattering. I - Numerical experiments for equal masses. , keywords =. doi:10.1086/160956 , adsurl =

    doi:10.1086/160956

  10. [10]

    , title =

    Hut, P. , title =. ApJ , volume =

  11. [11]

    , keywords =

    Binary--Single Star Interactions in Globular Clusters. , keywords =. doi:10.1086/173190 , adsurl =

    doi:10.1086/173190

  12. [12]

    , keywords =

    Binary evolution in stellar dynamics. , keywords =. doi:10.1093/mnras/173.3.729 , adsurl =

    doi:10.1093/mnras/173.3.729

  13. [13]

    , keywords =

    The binary zoo: the calculation of production rates of binaries through 2+1 encounters in globular clusters. , keywords =. doi:10.1093/mnras/276.3.887 , archivePrefix =. astro-ph/9507026 , primaryClass =

    Pith/arXiv arXiv

  14. [14]

    , keywords =

    Formation of massive black holes through runaway collisions in dense young star clusters. , keywords =. doi:10.1038/nature02448 , archivePrefix =. astro-ph/0402622 , primaryClass =

    Pith/arXiv arXiv

  15. [15]

    Monte Carlo Simulations of Globular Cluster Evolution. III. Primordial Binary Interactions. , keywords =. doi:10.1086/376593 , archivePrefix =. astro-ph/0301521 , primaryClass =

    Pith/arXiv arXiv

  16. [16]

    Fregeau, J. M. and Cheung, P. and Portegies Zwart, S. F. and Rasio, F. A. , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2004 , doi =

    2004

  17. [17]

    Binary-single interactions with different mass ratios: Implications for gravitational waves from globular clusters , journal =

    Rando Forastier, Bruno and Mar. Binary-single interactions with different mass ratios: Implications for gravitational waves from globular clusters , journal =. 2024 , doi =

    2024

  18. [18]

    Monthly Notices of the Royal Astronomical Society , year =

    Codazzo, Elena and Di Giovanni, Matteo and Harms, Jan , title =. Monthly Notices of the Royal Astronomical Society , year =

  19. [19]

    Astronomy & Astrophysics , year =

    Rostami Shirazi, Ali and Haghi, Hosein and Hasani Zonoozi, Akram and Farhani Asl, Ahmad and Kroupa, Pavel , title =. Astronomy & Astrophysics , year =

  20. [20]

    and Hut, Piet , title =

    Heggie, Douglas C. and Hut, Piet , title =. The Astrophysical Journal Supplement Series , volume =

  21. [21]

    and Hut, Piet , title =

    Heggie, Douglas C. and Hut, Piet , title =. 2003 , address =

    2003

  22. [22]

    Valtonen, M. J. and Myll. Dynamics of rotating triple systems: statistical escape theory versus numerical simulations , journal =. 2005 , volume =

    2005

  23. [23]

    and Hut, Piet and McMillan, Stephen L

    Heggie, Douglas C. and Hut, Piet and McMillan, Stephen L. W. , title =. The Astrophysical Journal , volume =. 1996 , doi =

    1996

  24. [24]

    and Boekholt, Tjarda C

    Portegies Zwart, Simon F. and Boekholt, Tjarda C. N. and Heggie, Douglas C. , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2023 , month =

    2023

  25. [25]

    , title =

    Ginat, Yonadav Barry and Perets, Hagai B. , title =. Physical Review Letters , volume =. 2021 , doi =

    2021

  26. [26]

    , title =

    Ginat, Yonadav Barry and Perets, Hagai B. , title =. Monthly Notices of the Royal Astronomical Society: Letters , volume =. 2022 , doi =

    2022

  27. [27]

    Acta Mathematica , year =

    Poincaré, Henri , title =. Acta Mathematica , year =

  28. [28]

    , title =

    Zotos, Euaggelos E. , title =. Celestial Mechanics and Dynamical Astronomy , volume =. 2015 , doi =

    2015

  29. [29]

    Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages =

    Optuna: A Next-generation Hyperparameter Optimization Framework , author =. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages =. 2019 , doi =

    2019

  30. [30]

    NASA TT F-677 , year =

    Evolution of the Protoplanetary Cloud and Formation of the Earth and Planets , author =. NASA TT F-677 , year =

  31. [31]

    Binary-single-star scattering. IV. Analytic approximations and fitting formulae for cross sections and reaction rates , author =. The Astrophysical Journal Supplement Series , volume =. 1993 , doi =

    1993

  32. [32]

    Basin entropy: a new tool to characterize uncertainty in dynamical systems , author =. Phys. Rev. Lett. , volume =

  33. [33]

    A&A , volume =

    Spatial mixing of stellar populations in globular clusters via binary-single star scattering , author =. A&A , volume =. 2025 , note =

    2025

  34. [34]

    Joint constraints on the field-cluster mixing fraction, common envelope efficiency, and globular cluster radii from a population of binary hole mergers via deep learning , author =. Phys. Rev. D , volume =

  35. [35]

    arXiv:1909.06373 , year =

    Machine-learning interpolation of population-synthesis simulations to interpret gravitational-wave observations , author =. arXiv:1909.06373 , year =

    Pith/arXiv arXiv 1909

  36. [36]

    , year = 2024, volume = 527, pages =

    Hierarchical triple formation in binary--single scattering. , year = 2024, volume = 527, pages =

    2024

  37. [37]

    , year = 2016, volume = 456, pages =

    On the formation of hierarchical triples from binary--single interactions. , year = 2016, volume = 456, pages =

    2016

  38. [38]

    , year = 2014, volume = 438, pages =

    Dynamical formation of hierarchical triples and its implications for stellar mergers. , year = 2014, volume = 438, pages =

    2014

  39. [39]

    Monthly Notices of the Royal Astronomical Society , volume =

    Kroupa, Pavel , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2001 , month = apr, doi =

    2001

  40. [40]

    Science , volume =

    Kroupa, Pavel , title =. Science , volume =. 2002 , month = jan, doi =

    2002

  41. [41]

    Publications of the Astronomical Society of the Pacific , volume =

    Chabrier, Gilles , title =. Publications of the Astronomical Society of the Pacific , volume =. 2003 , month = jul, doi =

    2003

  42. [42]

    and Umbreit, Stefan , title =

    Morscher, Meagan and Pattabiraman, Bharath and Rodriguez, Carl and Rasio, Frederic A. and Umbreit, Stefan , title =. The Astrophysical Journal , volume =. 2015 , doi =

    2015

  43. [43]

    The Astrophysical Journal , volume =

    Hut, Piet , title =. The Astrophysical Journal , volume =. 1983 , month = may, doi =

    1983

  44. [44]

    and Morscher, Meagan and Pattabiraman, Bharath and Chatterjee, Sourav and Haster, Carl-Johan and Rasio, Frederic A

    Rodriguez, Carl L. and Morscher, Meagan and Pattabiraman, Bharath and Chatterjee, Sourav and Haster, Carl-Johan and Rasio, Frederic A. , title =. Physical Review Letters , volume =. 2015 , doi =

    2015

  45. [45]

    The Astrophysical Journal , volume =

    Antonini, Fabio and Barausse, Enrico and Silk, Joseph , title =. The Astrophysical Journal , volume =. 2015 , doi =

    2015

  46. [46]

    , title =

    Samsing, Johan and D'Orazio, Daniel J. , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2018 , doi =

    2018

  47. [47]

    and Amaro-Seoane, Pau and Chatterjee, Sourav and Rasio, Frederic A

    Rodriguez, Carl L. and Amaro-Seoane, Pau and Chatterjee, Sourav and Rasio, Frederic A. , title =. Physical Review D , volume =. 2018 , doi =

    2018

  48. [48]

    and Amaro-Seoane, Pau and Chatterjee, Sourav and Kremer, Kyle and Rasio, Frederic A

    Rodriguez, Carl L. and Amaro-Seoane, Pau and Chatterjee, Sourav and Kremer, Kyle and Rasio, Frederic A. and Samsing, Johan and Ye, Claire S. and Zevin, Michael , title =. Physical Review D , volume =. 2018 , doi =

    2018

  49. [49]

    and Hut, Piet and McMillan, Steven L

    Heggie, Douglas C. and Hut, Piet and McMillan, Steven L. W. , title =. Monthly Notices of the Royal Astronomical Society , volume =. 1996 , doi =

    1996

  50. [50]

    Structure and Dynamics of Globular Clusters , editor =

    Pryor, Carlton and Meylan, Georges , title =. Structure and Dynamics of Globular Clusters , editor =. 1993 , publisher =

    1993

  51. [51]

    , title =

    Meylan, Georges and Heggie, Douglas C. , title =. The Stellar Dynamical Evolution of Globular Clusters , editor =

  52. [52]

    , title =

    Harris, William E. , title =. The Astronomical Journal , volume =. 1996 , doi =

    1996

  53. [53]

    The Astrophysical Journal , volume =

    Noyola, Eva and Gebhardt, Karl and Bergmann, Marcel , title =. The Astrophysical Journal , volume =. 2008 , doi =

    2008

  54. [54]

    and Lanzoni, Barbara and Vesperini, Enrico and Cadelano, Mario and Origlia, Livia and Fanelli, Cristiano and Dalessandro, Emanuele and Valenti, Elena , title =

    Leanza, Silvia and Pallanca, Cristina and Ferraro, Francesco R. and Lanzoni, Barbara and Vesperini, Enrico and Cadelano, Mario and Origlia, Livia and Fanelli, Cristiano and Dalessandro, Emanuele and Valenti, Elena , title =. Astronomy & Astrophysics , volume =. 2024 , doi =

    2024

  55. [55]

    and Cadelano, M

    Dalessandro, E. and Cadelano, M. and Della Croce, A. and Aros, F. I. and White, E. B. and Vesperini, E. and Fanelli, C. and Ferraro, F. R. and Lanzoni, B. and Leanza, S. and Origlia, L. , title =. Astronomy & Astrophysics , year =

  56. [56]

    and Lanzoni, Barbara and Dalessandro, Emanuele and Cadelano, Mario and Vesperini, Enrico and Origlia, Livia and Mucciarelli, Alessio and Valenti, Elena , title =

    Leanza, Silvia and Pallanca, Cristina and Ferraro, Francesco R. and Lanzoni, Barbara and Dalessandro, Emanuele and Cadelano, Mario and Vesperini, Enrico and Origlia, Livia and Mucciarelli, Alessio and Valenti, Elena , title =. Astronomy & Astrophysics , year =

  57. [57]

    and Lanzoni, Barbara and Dalessandro, Emanuele and Cadelano, Mario and Vesperini, Enrico and Origlia, Livia and Mucciarelli, Alessio and Valenti, Elena and Miola, Andrea , title =

    Pallanca, Cristina and Leanza, Silvia and Ferraro, Francesco R. and Lanzoni, Barbara and Dalessandro, Emanuele and Cadelano, Mario and Vesperini, Enrico and Origlia, Livia and Mucciarelli, Alessio and Valenti, Elena and Miola, Andrea , title =. Astronomy & Astrophysics , year =

  58. [58]

    and Genzel, R

    Krabbe, A. and Genzel, R. and Eckart, A. and Najarro, F. and Lutz, D. and Cameron, M. and Kroker, H. and Tacconi-Garman, L. E. and Thatte, N. and Weitzel, L. and Drapatz, S. and Geballe, T. and Sternberg, A. and Kudritzki, R. , title =. The Astrophysical Journal , volume =. 1995 , doi =

    1995

  59. [59]

    The nuclear star cluster of the Milky Way: proper motions and mass , journal =

    Sch. The nuclear star cluster of the Milky Way: proper motions and mass , journal =. 2009 , doi =

    2009

  60. [60]

    and Quirk, Amanda C

    Gilbert, Karoline M. and Quirk, Amanda C. N. and Guhathakurta, Puragra and Tollerud, Erik and Wojno, Jennifer and Dalcanton, Julianne J. and Durbin, Meredith J. and Seth, Anil and Williams, Benjamin F. and Fung, Justin T. and Tangirala, Pujita and Yusufali, Ibrahim , title =. The Astrophysical Journal , year =

  61. [61]

    Antognini, Joseph M. O. and Thompson, Todd A. , title =. The Astrophysical Journal , volume =. 2015 , doi =

    2015

  62. [62]

    Astronomy & Astrophysics , volume =

    Rein, Hanno and Liu, Shang-Fei , title =. Astronomy & Astrophysics , volume =. 2012 , doi =

    2012

  63. [63]

    , title =

    Rein, Hanno and Spiegel, David S. , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2015 , doi =

    2015

  64. [64]

    and Tamayo, Daniel and Brown, Garett and Eckels, Emily and Holmes, Emma and Lau, Michelle and Leblanc, R

    Rein, Hanno and Hernandez, David M. and Tamayo, Daniel and Brown, Garett and Eckels, Emily and Holmes, Emma and Lau, Michelle and Leblanc, R. Hybrid Symplectic Integrators for Planetary Dynamics , journal =. 2019 , doi =

    2019

  65. [65]

    Pinheiro, Tiago F. L. L. and Sfair, Rafael and Ramon, Giovana , title =. Astronomy & Astrophysics , year =

  66. [66]

    Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , year =

    Chen, Tianqi and Guestrin, Carlos , title =. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , year =

  67. [67]

    , keywords =

    The nuclear star cluster of the Milky Way: proper motions and mass. , keywords =. doi:10.1051/0004-6361/200810922 , archivePrefix =. 0902.3892 , primaryClass =

    Pith/arXiv arXiv

  68. [68]

    , keywords =

    Dynamical Simulation of the Interstellar Comet 3I/ATLAS. , keywords =. doi:10.3847/2041-8213/ae21cf , archivePrefix =. 2511.16247 , primaryClass =

    arXiv 2041

  69. [69]

    , keywords =

    Machine-learning approach for mapping stable orbits around planets. , keywords =. doi:10.1051/0004-6361/202451831 , archivePrefix =. 2412.04568 , primaryClass =

    Pith/arXiv arXiv

  70. [70]

    Aplicação de machine learning para mapeamento de regiões de estabilidade: Análise da performance do modelo através de ajuste de hiperparâmetros e threshold

  71. [71]

    , keywords =

    Globular cluster evolution with finite-size stars - Cross sections and reaction rates. , keywords =. doi:10.1086/163636 , adsurl =

    work page doi:10.1086/163636

  72. [72]

    Monte Carlo Simulations of Globular Cluster Evolution. III. Primordial Binary Interactions. , year = 2004, volume = 600, pages =

    2004

  73. [73]

    Comparison with N-body Simulations

    MOCCA Code for Star Cluster Simulations - I. Comparison with N-body Simulations. , year = 2013, volume = 429, pages =

    2013

  74. [74]

    Comparison with N-body Simulations

    MOCCA Code for Star Cluster Simulations - II. Comparison with N-body Simulations. , year = 2013, volume = 431, pages =

    2013

  75. [75]

    , year = 1996, volume = 282, pages =

    The effect of encounters on the eccentricity of binaries in clusters. , year = 1996, volume = 282, pages =

    1996

  76. [76]

    Monthly Notices of the Royal Astronomical Society , year = 2003, volume = 344, pages =

    A note on the eccentricity of binaries in star clusters. Monthly Notices of the Royal Astronomical Society , year = 2003, volume = 344, pages =

    2003

  77. [77]

    The Three-Body Problem