Efficient Eccentric Effective-One-Body Dynamics via Near-Identity Averaging Transformations

Source: arXiv:2606.30594 · Published 2026-06-29 · By Philip Lynch, Alessandra Buonanno, Aldo Gamboa, Maarten van de Meent

TL;DR

This paper addresses the computational challenge of accurately modeling long-lived gravitational wave inspirals from eccentric black hole binaries using the effective-one-body (EOB) framework. The EOB equations of motion require numerically resolving rapidly oscillating orbital dynamics on the orbital timescale, which is computationally expensive especially for low-mass or large mass-ratio binaries with many cycles in band. The authors present a method to recast nonspinning eccentric EOB dynamics into osculating orbital elements and apply near-identity averaging transformations (NITs) to remove the fast orbital timescale oscillations during inspiral. This yields averaged equations of motion that evolve on the long radiation-reaction timescale rather than the orbital timescale. Solving the averaged dynamics and then mapping back to full EOB dynamics for the plunge phase reduces inspiral cost by up to two orders of magnitude. They validate accuracy by comparing waveforms from averaged and full EOB dynamics over a broad parameter space, showing that including next-to-next-to-leading order (2PA) terms yields mismatches below 8.05×10^−5 for comparable-mass binaries. Overall waveform generation speedups of 1.5–8× are achieved with essentially no loss in accuracy. This method provides a practical and efficient approach for generating accurate eccentric waveforms for future gravitational wave detectors requiring long, precise inspiral signals.

Key findings

• Near-identity averaging transformations suppress orbital-timescale oscillations by factors of the ratio of orbital to radiation-reaction timescales at each order in the post-adiabatic expansion.
• The 2PA (next-to-next-to-leading order) averaged equations are needed to model comparable-mass binaries accurately, achieving waveform mismatches ≤ 8.05×10^−5 against full EOB evolutions.
• Inspiral dynamics computational cost is reduced by up to two orders of magnitude using averaged equations.
• Overall waveform generation speedup spans 1.5 to 8× depending on binary parameters and eccentricity.
• Fourier decomposition up to κ_max=30 modes is sufficient for accurately reconstructing forcing functions and performing regularization near radial turning points.
• Switching from averaged to full EOB dynamics is performed conservatively at ˜p=8.5, ˜e=0.035, or when the adiabaticity parameter Q < 100 to maintain accuracy near plunge.
• The method achieves good accuracy across a grid of 488,250 points in mass ratio, semilatus rectum, and eccentricity parameter space (ν ∈ [0.01, 0.25], e ∈ [0.01, 0.9], p ∈ [~6+2e_max, 289]).
• Neglecting the back-transformation from NIT to osculating orbital elements for waveform generation introduces negligible phase errors.

Threat model

N/A; the paper focuses on computational modeling and numerical efficiency for gravitational waveform generation without addressing security or adversarial threat scenarios.

Methodology — deep read

1. Threat Model & Assumptions: The physical system modeled is a nonspinning black hole binary, approximated by an effective-one-body framework with symmetric mass ratio ν = µ/M. The model assumes knowledge of the system's parameters (masses, eccentricity, orbital elements) and aims to efficiently generate inspiral waveforms for data analysis, assuming gravitational wave radiation reaction is the only dissipative force. No explicit adversarial threat model applies here, as the paper focuses on computational efficiency and accuracy.

2. Data: No empirical dataset is used; instead, the paper constructs a large numerical interpolation grid over the physical parameter space: mass ratio ν in [0.01,0.25] in steps of 0.01, eccentricity e in [0.01, 0.9] in steps of 0.01, and a transformed semilatus rectum coordinate u = log(p^{-3}) covering weak and strong-field regimes with 217 points spaced by 0.05. This results in 488,250 grid points. At each grid point, the authors compute Fourier coefficients of forcing functions via numerical evaluation of EOB dynamics and fast Fourier transform (FFT), building smooth interpolated maps for efficient evaluation later.

3. Architecture / Algorithm: The core innovation is recasting the nonspinning eccentric EOB equations of motion into osculating orbital elements (p, e, w) and radial phase ζ (χ - w). The equations have forcing functions periodic on orbital phase but with singularities at turning points. The authors regularize these via Fourier decomposition, reconstructing forcing terms by excluding singular points. They then perform near-identity averaging transformations (NITs) up to second post-adiabatic (2PA) order, expanding in the small parameter ϵ representing the ratio of orbital to radiation reaction timescales. NIT functions Y, X, Z relate original variables to averaged variables, removing orbital phase dependence from equations of motion. The resulting averaged equations evolve on the slow radiation-reaction timescale only. Finally, transformations map averaged variables back to full EOB variables for the plunge phase.

4. Training Regime: Not applicable as the work is a numerical modeling and mathematical derivation study. Numerical solving of ODEs for EOB and averaged equations uses Mathematica NDSolve with adaptive explicit Runge-Kutta (RK4 for EOB, RK8 for averaged equations), high accuracy (AccuracyGoal=9) and cubic spline interpolation of Fourier coefficients over the dense parameter grid.

5. Evaluation Protocol: Accuracy is measured by computing waveform mismatches between waveforms generated from full EOB dynamics and averaged dynamics over a broad range of ν, p, e values. Waveforms focus on the dominant (2,2) mode computed up to 3PN order. Mismatches ≤ 8.05×10^−5 for 2PA averaged model demonstrate excellent agreement. Runtime speedups from 1.5× to 8× are reported depending on parameters. Conservative handover conditions for switching to full EOB dynamics are used to avoid breakdown of averaged equations close to plunge. Ablations between 0PA, 1PA, and 2PA approximations show necessity of 2PA terms for comparable mass accuracy.

6. Reproducibility: The authors indicate ancillary Mathematica files for deriving transformations and provide complete descriptions of grids, numerical methods, and transformations. However, no public code repository or frozen models are explicitly referenced. The interpolation grid and numeric methods appear fully described to enable reproduction in principle, though the large parameter space and Mathematica dependence pose practical difficulties.

Example: For a given binary with ν=0.1, initial eccentricity e=0.4 and initial orbit frequency f0, the method finds corresponding initial osculating elements p, e, ζ, then applies NIT transformations to obtain averaged variables ˜p, ˜e, ˜ζ. The averaged equations of motion are integrated over the radiation-reaction timescale with RK8, yielding faster solutions for (˜p(χ), ˜e(χ), ˜ζ(χ)). These are then sampled finely in χ to reconstruct the waveform h22 mode, compared against the full EOB integration, obtaining mismatches <10^−4 with an 3–8× speedup.

Technical innovations

• Reformulating nonspinning eccentric EOB inspiral dynamics in terms of osculating orbital elements with quasi-Keplerian parametrization to separate fast orbital phase and slow orbital element evolution.
• Applying near-identity averaging transformations (NITs) up to second post-adiabatic order to remove orbital-timescale oscillations and evolve dynamics on radiation-reaction timescale without mass ratio expansion.
• Developing Fourier decomposition techniques with shifted sampling to regularize singular forcing terms at radial turning points for numerical stability and efficient interpolation over dense parameter grids.
• Mapping averaged inspiral solutions back to full EOB dynamics near plunge transition to retain accurate waveform modeling through merger.

Datasets

• Numerical interpolation grid — 488,250 points — constructed over (ν, p, e) parameter space from numerical EOB evaluations

Baselines vs proposed

• Full EOB dynamics: mismatch baseline with averaged 0PA model = >10^{−3}, improved to ≤8.05×10^{−5} with 2PA model
• Waveform generation runtime: Full EOB baseline = 1×; averaged NIT models speedup range from 1.5× (near plunge) up to 8× (inspiral phase)
• Accuracy: 1PA model no better than 0PA, 2PA model needed for mismatch ≤8.05×10^{−5} over comparable-mass binaries

Limitations

• Method only developed and validated for nonspinning binaries; extension to spin-aligned and precessing cases remains future work.
• Breakdown of averaged equations as eccentricity approaches zero due to singular behavior in fw forcing terms; awaiting regularized variable transformations in follow-up.
• Handover from averaged to full EOB dynamics is heuristic and conservative; possibility exists to optimize timing to further improve speed without accuracy loss.
• Waveforms only generated for dominant (2,2) spherical harmonic mode and up to 3PN in amplitude; higher modes and complete PN order treatment could affect accuracy.
• No explicit assessment of robustness under distribution shift (e.g., extreme mass ratios), or inclusion of noise, parameter uncertainties, or real detector effects.
• Numerical implementation depends on Mathematica and large interpolation grids, complicating practical deployment in realtime or online parameter estimation.

Open questions / follow-ons

• How to extend near-identity averaging transformations to include aligned and precessing spin effects in eccentric EOB models?
• Can the averaging method be modified to remain well-behaved and accurate at low eccentricities, including smooth e → 0 limits?
• What are optimal criteria and algorithms for transitioning between averaged and full EOB dynamics to balance speed and accuracy?
• How does this approach generalize to include higher waveform multipoles and full 3PN or beyond PN amplitude corrections?

Why it matters for bot defense

Although this work is centered on gravitational waveform modeling rather than bot defense directly, its techniques for averaging rapidly oscillatory dynamics over multiple timescales to achieve large efficiency gains are methodologically relevant. Bot defense and CAPTCHA systems often face challenges from high-frequency noise or oscillatory behavior in input patterns, where similar averaging or multi-timescale analysis might improve response speed or system stability. Furthermore, the approach of transforming variables to remove fast fluctuations while preserving slow dynamics could inspire analogous methods in detecting or filtering bot-like behavior profiles that oscillate rapidly on short timescales but evolve slowly overall. However, practical application requires adapting the physics-based transformations and analytical expansions here to digital or behavioral signal domains relevant to CAPTCHA or bot defense.

Cite

@article{arxiv2606_30594,
  title={ Efficient Eccentric Effective-One-Body Dynamics via Near-Identity Averaging Transformations },
  author={ Philip Lynch and Alessandra Buonanno and Aldo Gamboa and Maarten van de Meent },
  journal={arXiv preprint arXiv:2606.30594},
  year={ 2026 },
  url={https://arxiv.org/abs/2606.30594}
}

Read the full paper

• arXiv:2606.30594 (abstract)
• arXiv:2606.30594 (PDF)