[Paper Review] Analysis of GWTC-3 with fully precessing numerical relativity surrogate models

Tousif Islam, A. Vajpeyi+7 more|arXiv (Cornell University)|Jul 5, 2023

Pulsars and Gravitational Waves Research4 citations

TL;DR

This study re-analyzes 47 binary black hole events from GWTC-3 using the fully precessing NRSur7dq4 numerical relativity surrogate waveform model and its associated remnant properties model, NRSur7dq4Remnant. The analysis reveals significant differences in black hole spin, mass, and extrinsic parameter estimates compared to LVK’s semi-analytical models, with NRSur7dq4 showing higher signal-to-noise ratios and Bayes factors for several events, particularly in precession-sensitive systems.

ABSTRACT

Posterior samples from the NRSurCat-1. Refer to the paper for more details. Website hosting plots for this work can be found here.

Motivation & Objective

• To re-analyze 47 binary black hole events from GWTC-3 using a fully precessing numerical relativity surrogate waveform model, NRSur7dq4, to assess its accuracy and performance compared to existing models.
• To evaluate the impact of including all ℓ ≤ 4 spin-weighted spherical harmonic modes and full precession effects on parameter estimation of black hole masses, spins, and kick vectors.
• To compare the signal-to-noise ratio (SNR) and Bayes factors obtained with NRSur7dq4 against those from the LVK’s semi-analytical models (IMRPhenomXPHM and SEOBNRv4PHM) to assess model preference.
• To investigate whether model extrapolation or limited duration of the surrogate affects parameter estimation, particularly in high-mass-ratio or high-spin regimes.
• To identify events where the NRSur7dq4 model yields significantly different or more constrained astrophysical inferences, especially regarding effective spin and kick velocity.

Proposed method

• Utilizes the NRSur7dq4 surrogate model, trained on numerical relativity simulations, to generate fully precessing waveforms including all ℓ ≤ 4 spin-weighted spherical harmonic modes.
• Employs the NRSur7dq4Remnant surrogate model to estimate remnant black hole properties such as mass, spin vector, and kick velocity directly from the binary parameters.
• Performs Bayesian parameter estimation on 47 BBH events from GWTC-3 that lie within the validated domain of NRSur7dq4 (mass ratio q ≥ 1/6, total mass > 60 M☉).
• Computes signal-to-noise ratio (SNR) and log Bayes factor for each event using both NRSur7dq4 and IMRPhenomXPHM models to compare model performance.
• Analyzes potential sources of discrepancy by quantifying model extrapolation (q < 1/4 or χ₁ > 0.8) and the starting frequency of higher modes (e.g., (3,3)) in the NRSur7dq4 model.
• Uses synthetic injection studies and statistical correlation analysis to assess whether missing low-frequency content in higher modes or transient noise affects model preference.

Experimental results

Research questions

• RQ1 How do parameter estimates (mass, spin, inclination, distance) from the fully precessing NRSur7dq4 model differ from those obtained using semi-analytical models like IMRPhenomXPHM for GWTC-3 BBH events?
• RQ2 To what extent does the NRSur7dq4 model outperform IMRPhenomXPHM in terms of signal-to-noise ratio and Bayes factor for individual events?
• RQ3 Are discrepancies in parameter estimation driven by model extrapolation beyond the training domain or by missing low-frequency content in higher-order modes?
• RQ4 Which events show the most significant differences in inferred astrophysical properties, such as effective spin χ_eff or kick velocity, when using NRSur7dq4 versus LVK’s standard models?
• RQ5 How do transient detector noise events affect the reliability of model comparisons and parameter estimation with high-accuracy NR surrogate models?

Key findings

• For more than 20% of the 47 BBH events, NRSur7dq4 yields noticeably different measurements of black hole masses, spins, and extrinsic parameters (inclination, distance) compared to the LVK’s IMRPhenomXPHM and SEOBNRv4PHM models.
• GW150914_095045 exhibits significant differences in spin precession and spin magnitude estimates due to the inclusion of full precession and higher-mode content in NRSur7dq4.
• GW191109_010717 constrains the effective spin χ_eff to be negative at a 99.3% credible level, indicating a strong preference for anti-aligned spins.
• Two events—GW191109_010717 and GW200129_065458—show well-constrained kick velocities, with NRSur7dq4 providing more accurate and consistent estimates than semi-analytical models.
• NRSur7dq4 recovers higher signal-to-noise ratios and/or Bayes factors than IMRPhenomXPHM for several events, indicating a mild preference for the NR surrogate model in model selection.
• No clear correlation is found between model extrapolation (q < 1/4 or χ₁ > 0.8) and model performance, suggesting that the observed differences are not primarily due to extrapolation errors, though synthetic injections are needed for confirmation.