🔭 astrophysics

Bayesian parameter estimation for the Core-bounce phase of Rapidly Rotating Core-Collapse Supernovae in real interferometric data

This paper presents a novel Bayesian methodology using an improved phenomenological model to accurately estimate the rotational parameter and equation of state of rapidly rotating core-collapse supernovae from real gravitational-wave interferometric data, demonstrating significant improvements in fitting factors and parameter recovery accuracy over previous approaches.

Original authors: Emmanuel Avila, Michele Zanolin, Javier M. Antelis, Claudia Moreno

Published 2026-05-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Emmanuel Avila, Michele Zanolin, Javier M. Antelis, Claudia Moreno

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a star at the end of its life, a massive giant spinning so fast it's like a figure skater pulling in their arms. When this star runs out of fuel, its core collapses inward, hits a "wall" of resistance, and bounces back with a tremendous explosion. This moment, called the core bounce, is like a cosmic drumbeat. It sends out ripples in space and time called gravitational waves.

The paper you shared is about learning how to "listen" to these ripples to understand what the star was made of and how fast it was spinning before it exploded. Here is a breakdown of their work using simple analogies:

1. The Problem: A Noisy Room

Imagine trying to hear a specific drumbeat in a stadium full of cheering fans and construction noise. That's what scientists face when looking for these supernova signals. The detectors (like LIGO) are incredibly sensitive, but the "noise" of the universe and the machine itself can drown out the signal.

Previous methods were like trying to guess the drumbeat by just listening to the loudest part of the noise. The authors wanted a better way to tune in.

2. The Solution: A Better "Ear" (The Model)

The researchers created a new mathematical "template" or a soundtrack that predicts exactly what that core-bounce drumbeat should look like.

The Old Template: It was like a song with three distinct notes (peaks in the wave). It worked okay, but it was a bit rigid.
The New Template: They added a "knob" to the model. Imagine the old song had a fixed speed. The new model allows the duration of the notes to stretch or shrink slightly. This extra flexibility (a new parameter called 's') let the model fit the actual data much better, like a tailor adjusting a suit to fit a person perfectly rather than using a "one-size-fits-all" approach.

3. The Experiment: Testing in the Real World

To see if their new template worked, they didn't just use perfect, quiet computer simulations. They took their new model and tried to find signals inside real data from the LIGO detector (specifically from a run called "O3a").

The Test: They took thousands of simulated star explosions (from a database called "Abylkairov") and hid them inside the real, messy noise of the LIGO detector.
The Goal: Could their new method find the hidden signal and tell them how fast the star was spinning?

4. The Results: Finding the Spin

They found that their new method was much better at estimating the rotational speed (called $\beta$ ) of the star.

The Analogy: Imagine trying to guess how fast a spinning top is moving. The old method might guess "somewhere between 10 and 20 spins per second." The new method narrowed it down to "between 10.1 and 10.5."
The Numbers: In the real noisy data, their new method reduced the error significantly compared to older techniques. They could estimate the spin with a high degree of confidence, even when the signal was buried deep in the noise.

5. The "Recipe" for the Star (Equation of State)

The paper also looked at the Equation of State (EOS). In simple terms, this is the "recipe" for the star's core material. It tells us how stiff or squishy the matter is under extreme pressure.

The Discovery: They found a relationship between the star's spin speed and the shape of the third "note" in the gravitational wave.
The Metaphor: Think of it like a chef tasting a soup. If the soup is very salty (high spin), the flavor of the herbs (the EOS) changes in a predictable way. By measuring the spin and the "flavor" of the wave, they could group the different "recipes" (EOS models) into categories.
The Catch: They found that while they could distinguish between broad groups of recipes, it was hard to tell two very similar recipes apart, especially if the star was very far away (like trying to taste a soup from across the room).

6. The "Bias" Warning (Priors)

The authors also tested how their assumptions affected the results. In science, you have to start with a guess (a "prior").

The Analogy: If you guess a coin is fair before flipping it, you might interpret a few heads as just "luck." If you guess the coin is weighted, you might interpret the same heads as "proof."
The Finding: They showed that the choice of this starting guess can slightly skew the final answer. However, they found that using a specific type of guess (a "triangular" prior) gave the most accurate results with the least amount of bias.

Summary

In short, this paper is about tuning a radio to hear a faint cosmic drumbeat.

They built a better radio tuner (the new model with the extra parameter) that fits the signal better than before.
They proved it works in real, noisy conditions, not just in a quiet lab.
They showed that by listening to this beat, we can figure out how fast the star was spinning and get a clue about what the star was made of (its internal "recipe").

This work provides a more reliable tool for future astronomers to decode the secrets of exploding stars when the next one happens.

Technical Summary: Bayesian Parameter Estimation for the Core-Bounce Phase of Rapidly Rotating Core-Collapse Supernovae in Real Interferometric Data

Problem Statement
Core-Collapse Supernovae (CCSNe) are primary targets for Gravitational Wave (GW) astronomy. While the GW signal from these events is largely stochastic, the core bounce phase in rapidly rotating progenitors exhibits deterministic features, specifically a series of peaks in the time domain. Previous parameter estimation (PE) efforts have utilized numerical simulations to understand these signals, often relying on matched filtering or Maximum Likelihood Estimation (MLE) in Gaussian noise. However, there is a need to refine phenomenological analytical models to better fit numerical waveform databases and to assess the robustness of Bayesian PE methods when applied to real, non-Gaussian, and non-stationary interferometric noise (specifically LIGO O3aL1 data). Furthermore, the ability to constrain the nuclear Equation of State (EOS) using these signals remains a critical open question.

Methodology
The authors propose a novel methodology centered on an extended phenomenological analytical model for the core bounce phase.

Model Extension: Building on a previous model by Villegas et al. (2024), the authors introduce an additional free parameter, $s$ , representing the timescale (width) of the Gaussian peaks in the waveform. The original model fixed this value; the new model allows it to vary. The model describes the strain $h_+(t)$ as a sum of three Gaussian bells, where amplitudes depend on the rotational parameter $\beta$ (ratio of kinetic to gravitational potential energy) and a third peak amplitude parameter $\alpha$ (linked to EOS), while peak positions depend on the bounce time $\tau$ .
Data Sources: The study utilizes three numerical waveform catalogs:
- A subset of 126 waveforms from the Richers et al. (2017) catalog.
- 452 General Relativity (GR) waveforms from the Abylkairov et al. (2025) catalog (excluding GREP approximations).
- The Mitra et al. (2023) catalog for consistency checks.
  The authors verify consistency between catalogs using Fitting Factors (FF) and Bayesian model comparison.
Validation: The extended model is validated against the numerical databases using:
- Fitting Factor Analysis: Comparing the analytical model to numerical waveforms with $s$ fixed versus $s$ free.
- Bayesian Model Comparison: Using Nested Sampling to calculate the Bayes Factor between models with varying numbers of parameters (1 to 4 parameters) to determine the optimal model complexity.
Parameter Estimation (PE): PE is performed using a Markov Chain Monte Carlo (MCMC) implementation (Bilby sampler) in two noise environments:
- Gaussian colored noise based on LIGO Power Spectral Density (PSD).
- Real interferometric noise from the LIGO O3aL1 observing run.
  The study investigates the sensitivity of posterior distributions to different prior choices (Uniform, Log-Uniform, Triangular, and Uniform in $\beta^2$ ).
EOS Estimation: A power-law relationship between the estimated parameters $\alpha$ and $\beta$ is proposed to discriminate between different EOS models (SFHo, HSDD2, GShenFSU2.1).

Key Contributions and Results

Model Improvement: Extending the analytical model to include the free timescale parameter $s$ improved the median fitting factor with the Richers catalog waveforms from 88.88% to 90.83%. Bayesian model comparison (via Bayes Factors) and Fitting Factor analysis both support the preference for the four-parameter model ( $\alpha, \beta, \tau, s$ ) over simpler models.
Parameter Estimation Accuracy:
- Using real O3aL1 noise, the rotational parameter $\beta$ was estimated for 452 signals from the Abylkairov catalog. The median absolute relative error was 11.93% (95th percentile: 38.41%) at a distance of 10 kpc.
- The overall uncertainty was $\sigma_\beta = 1.083 \times 10^{-3}$ at 10 kpc. This represents an improvement over previous results using matched filtering and MLE in real noise, which reported $\sigma_\beta = 1.46 \times 10^{-2}$ .
Prior Sensitivity: The study quantified the bias introduced by different priors. In real O3 noise at 10 kpc, the bias in $\beta$ estimation ranged from a minimum of 0.6% (Triangular prior) to a maximum of 11.9% (Uniform in $\beta^2$ prior). The parameter $s$ was found to be the most sensitive to prior selection.
EOS Classification: A classifier based on the residue between estimated $\alpha$ and the fitted power-law curve for a given $\beta$ was tested. In the noiseless case, the classification accuracy was 76.2%. This degraded with distance and noise (44.0% at 10 kpc). However, when grouping similar EOSs (SFHo and GShenFSU2.1 as Group A; LS220 and HSDD2 as Group B), the accuracy improved to 66.3% at 10 kpc. The authors note that LS220 was excluded from the final analysis due to constraints on symmetry energy parameters.

Significance and Claims
The paper claims that Bayesian inference is an adequate tool for parameter estimation of CCSNe using physically motivated waveform models and real interferometric noise. The primary significance lies in demonstrating that:

Extending phenomenological models with physically motivated parameters (like the bounce timescale) improves the fit to numerical simulations.
Bayesian PE in real noise yields better constraints on the rotational parameter $\beta$ than previous MLE/matched filtering approaches in similar noise conditions.
The choice of Bayesian priors significantly impacts the bias of the estimated parameters, necessitating robustness studies when prior detections are unavailable.
While EOS discrimination is possible, the performance is currently limited by the similarity between certain EOS models and the degradation of signal-to-noise ratio with distance. The authors suggest that a network of interferometers could improve estimation results by a factor of $\sqrt{N}$ .

The work establishes a foundation for future GW searches and extends the framework to potential future detector networks (e.g., Cosmic Explorer, Einstein Telescope), noting that results at 1 kpc in current detectors may be representative of 10 kpc in next-generation instruments.