[Paper Review] Classification of the core-collapse supernova explosion mechanism with learned dictionaries

Ainara Saiz-Pérez, A. Torres-Forné+1 more|arXiv (Cornell University)|Oct 25, 2021

Gamma-ray bursts and supernovae4 citations

TL;DR

This paper proposes a supervised dictionary-learning approach using LASSO regression to classify core-collapse supernova (CCSN) gravitational wave signals by morphology. Trained on neutrino-driven (Mur), magneto-rotational (Dim), and sine-Gaussian (SG) waveforms injected into Advanced LIGO-like noise, the method achieves ~85% correct classification for Mur signals and near-perfect classification for Dim and SG signals at SNR=15–20, demonstrating strong performance comparable to the SMEE algorithm.

ABSTRACT

Core-collapse supernovae (CCSN) are a prime source of gravitational waves. Estimations of their typical frequencies make them perfect targets for the current network of advanced, ground-based detectors. A successful detection could potentially reveal the underlying explosion mechanism through the analysis of the waveform. This has been illustrated using the SupernovaModel Evidence Extractor (SMEE; Logue et al. (2012)), an algorithm based on principal-component analysis and Bayesian model selection. Here, we present a complementary approach to SMEE based on (supervised) dictionary-learning and show that it is able to reconstruct and classify CCSN signals according to their morphology. Our waveform signals are obtained from (a) two publicly available catalogs built from numerical simulations of neutrino-driven (Mur) and magneto-rotational (Dim) CCSN explosions and (b) from a third 'mock' catalog of simulated sine-Gaussian (SG) waveforms. Those signals are injected into coloured Gaussian noise to simulate the background noise of Advanced LIGO in its broadband configuration and scaled to a freely-specifiable signal-to-noise ratio (SNR). We show that our approach correctly classifies signals from all three dictionaries. In particular, for SNR=15-20, we obtain perfect matches for both Dim and SG signals and about 85% true classifications for Mur signals. These results are comparable to those reported by SMEE for the same CCSN signals when those are injected in only one LIGO detector. We discuss the main limitations of our approach as well as possible improvements.

Motivation & Objective

• To develop a supervised dictionary-learning method as a complementary approach to SMEE for classifying CCSN gravitational wave signals by morphology.
• To evaluate the performance of dictionary learning in reconstructing and classifying CCSN waveforms under realistic Advanced LIGO noise conditions.
• To compare the method’s classification accuracy with SMEE using the same waveforms and signal-to-noise ratios (SNR).
• To assess the impact of signal complexity, catalog size, and noise transients on classification reliability.
• To identify limitations such as spurious signal generation in pure noise and the need for larger, more accurate waveform catalogs.

Proposed method

• The method employs supervised dictionary learning to train on three waveform types: neutrino-driven (Mur), magneto-rotational (Dim), and sine-Gaussian (SG) waveforms.
• Waveforms are injected into colored Gaussian noise mimicking Advanced LIGO’s broadband configuration and scaled to user-defined SNR levels.
• A LASSO regression-based reconstruction framework is used to extract signals from noise using learned dictionaries, with classification based on the best-matching dictionary.
• The Wasserstein distance is used as a quantitative metric to evaluate reconstruction fidelity and construct confusion matrices across SNR levels.
• The algorithm classifies signals by selecting the dictionary that produces the lowest reconstruction error, enabling morphology-based identification.
• The method includes a regularization parameter (λ^denoise) to control noise suppression, though it affects false positive and false negative rates.

Experimental results

Research questions

• RQ1 Can a supervised dictionary-learning approach accurately reconstruct and classify CCSN gravitational wave signals based on their morphological features in noisy environments?
• RQ2 How does the classification performance of the proposed method compare to SMEE across different signal types (Mur, Dim, SG) at varying SNR levels?
• RQ3 What factors contribute to misclassification, particularly for the Mur signal class, and how can they be mitigated?
• RQ4 To what extent does the algorithm generate spurious signals in pure-noise conditions, and how does this affect detection reliability?
• RQ5 Can the method be improved by incorporating a dedicated 'noise' class to reduce false positives and improve robustness?

Key findings

• At SNR=15–20, the method achieves perfect classification for all Dim and SG signals, indicating high reliability for these morphological classes.
• For Mur signals, the method achieves approximately 85% true classification rate, with around 15% misclassified, reflecting the class’s inherent complexity and low catalog size.
• The algorithm shows a strong tendency to misclassify signals as Mur at low SNR (e.g., ~70% of signals classified as Mur at lowest SNR), suggesting sensitivity to noise and signal structure.
• Spurious signal generation occurs even in pure-noise conditions, particularly when the regularization parameter λ^denoise is suboptimal, leading to false positives.
• The method’s performance is comparable to SMEE for the same waveforms and SNR levels, validating its potential as a complementary classification tool.
• The scarcity of training data—especially only 11 Mur waveforms—limits classification accuracy, highlighting the need for larger, more diverse waveform catalogs.