A practical Bayesian method for gravitational-wave ringdown analysis with multiple modes

Gravitational-wave (GW) ringdown signals from black holes (BHs) encode crucial information about the gravitational dynamics in the strong-field regime, which offers unique insights into BH properties. In the future, the improving sensitivity of GW detectors is to enable the extraction of multiple quasi-normal modes (QNMs) from ringdown signals. However, incorporating multiple modes drastically enlarges the parameter space, posing computational challenges to data analysis. Inspired by the $F$-statistic method in the continuous GW searches, we develope an algorithm, dubbed as FIREFLY, for accelerating the ringdown signal analysis. FIREFLY analytically marginalizes the amplitude and phase parameters of QNMs to reduce the computational cost and speed up the full-parameter inference from hours to minutes, while achieving consistent posterior and evidence. The acceleration becomes more significant when more QNMs are considered. Rigorously based on the principle of Bayesian inference and importance sampling, our method is statistically interpretable, flexible in prior choice, and compatible with various advanced sampling techniques, providing a new perspective for accelerating future GW data analysis.

💡 Research Summary

The paper addresses the computational bottleneck in Bayesian inference of gravitational‑wave (GW) ringdown signals when multiple quasi‑normal modes (QNMs) are included. Each QNM contributes four physical parameters (amplitude, phase, damping time, frequency), but in General Relativity the latter two are functions of the remnant black‑hole mass and spin. Consequently, adding an overtone increases the dimensionality by two, and a realistic analysis with several overtones quickly becomes high‑dimensional and computationally expensive.

Inspired by the F‑statistic used in continuous GW searches, the authors develop an algorithm called FIRESFLY (F‑statistic Inspired REsampling For AnalYzing GW ringdown signals). The key idea is that, under the assumption of stationary Gaussian noise, the likelihood is a Gaussian function of the linear parameters B (the complex amplitudes encoding amplitude and phase). By choosing a broad, factorized prior on B (essentially flat in amplitude and phase), the authors can analytically marginalize over all 2N amplitude/phase parameters, leaving a marginal likelihood that depends only on the intrinsic parameters θ (e.g., final mass M_f and spin χ_f).

FIRESFLY proceeds in two stages. In the auxiliary inference stage, the marginal likelihood L_m∅(θ) is sampled with any stochastic sampler (the authors use the nested‑sampling code dynesty) to obtain posterior samples of θ and the auxiliary evidence Z∅. For each sampled θ_i, a corresponding B_i is drawn from the conditional Gaussian N(ĤB(θ_i), M⁻¹(θ_i)). In the second stage, importance sampling is applied to transform the auxiliary posterior into the target posterior defined by the desired prior π(θ,B|⊗). The importance weights are factorized: first a weight for θ based on the ratio of marginal priors, then a weight for B based on the ratio of conditional priors, with a Gaussian proposal used for the latter. The final output consists of posterior samples and evidence under the target prior, identical to what would be obtained by a full‑parameter Bayesian run.

The method is tested on simulated ringdown data injected into ET‑D noise using the SXS:BBH:0305 numerical‑relativity waveform (a GW150914‑like binary). Three scenarios are considered: (i) a single fundamental mode (N=1), (ii) one overtone added (N=2), and (iii) two overtones (N=3). For each case the start time of the analysis is chosen to capture the relevant overtones. The target priors are uniform in M_f, χ_f, ϕ, and A (with a generous amplitude ceiling). The auxiliary prior on B is taken proportional to the amplitude, ensuring analytic marginalization.

Results show excellent agreement between FIRESFLY and a conventional full‑parameter sampling: posterior contours overlap at the 1‑σ to 3‑σ levels, P‑P plots are flat, and the Wasserstein distance between one‑dimensional marginals is <0.1 σ. The computational gain is dramatic. For the most demanding case (N=3), the full‑parameter run takes about 5 hours 15 minutes, whereas FIRESFLY completes in roughly 3 minutes (2 min 49 s for auxiliary sampling, 5.7 s for evidence integration, and 7.9 s for importance sampling), a speed‑up factor of ≈ 100. The acceleration scales with the number of modes because each added mode contributes two analytically marginalized parameters.

Beyond speed, FIRESFLY retains full statistical interpretability: the marginalization is exact, and the importance‑sampling step correctly accounts for any difference between auxiliary and target priors, allowing flexible prior choices unlike traditional F‑statistic methods that implicitly fix priors. Moreover, the algorithm is sampler‑agnostic; any advanced stochastic sampler (nested sampling, Hamiltonian Monte Carlo, etc.) can replace dynesty, potentially yielding further gains.

In conclusion, FIRESFLY provides a principled, highly efficient framework for multi‑mode ringdown analysis, making Bayesian inference feasible for the high‑signal‑to‑noise, multi‑overtone detections anticipated with next‑generation ground‑based detectors (Cosmic Explorer, Einstein Telescope) and space‑based missions (LISA, Taiji, TianQin). Future work may extend the method to handle non‑Gaussian noise, waveform systematics, and joint inspiral‑merger‑ringdown analyses, as well as integrate automated prior optimization or machine‑learning‑driven proposals. The approach promises to be a cornerstone for real‑time or near‑real‑time black‑hole spectroscopy in the era of precision gravitational‑wave astronomy.