Basilic: An end-to-end pipeline for Bayesian burst inference and model classification in gravitational-wave data

Iuliu Cuceu, Marie Anne Bizouard · 2026

Read Basilic: An end-to-end pipeline for Bayesian burst inference and model classification in gravitational-wave data by Iuliu Cuceu, Marie Anne Bizouard (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-13839.txt.

Introduction

Cuceu and Bizouard offer Basilic, a pipeline for Bayesian analysis of short gravitational-wave bursts. Built on bilby, it handles model selection and parameter estimation with modular burst models and HTCondor support, so analyses run fast. A case study pits a black hole merger against a cosmic string signal in injection campaigns. Besides the known high-mass shape degeneracy, high anti-aligned spins at moderate mass also mimic strings. The authors propose a data-driven degeneracy study for inconclusive Bayes factors, the likely regime for future low-SNR detections.

Catalog record

CREATORIuliu Cuceu, Marie Anne Bizouard
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56674
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.13839

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Basilic: An end-to-end pipeline for Bayesian burst inference and model classification in gravitational-wave data Iuliu Cuceu, Marie Anne Bizouard Abstract We present Basilic , a dedicated pipeline for Bayesian model selection and parameter estimation of short-duration gravitational-wave burst signals observable with ground-based detectors. Built on top of the bilby framework, Basilic combines modularity, pre-implemented burst models, and HTCondor integration to enable rapid, user-friendly analyses with minimal technical overhead. This work outlines the design philosophy, operational flow, and a set of example use cases demonstrating its scientific potential. As a case study, we also undertake an in-depth exploration of the comparison between a binary black hole merger and a cosmic string signal, through a parameter space exploration injection campaign. In addition to the well-known high-mass binary black-hole signal morphology degeneracy with cosmic string-like signals, we find that high anti-aligned component spins, even at moderate mass, can result in a similar degeneracy. Motivated by the likely low-SNR expected regime of possible future detections, we propose a data-driven study of model degeneracy, to be employed in the event of an inconclusive Bayes factor. I Introduction The LIGO-Virgo-KAGRA (LVK) 1 ; 9 ; 10 collaboration has propelled the field of gravitational-wave (GW) astronomy for over a decade, with detections of compact-binary coalescences (CBCs). At the same time, a wide and astrophysically well-motivated class of short-duration transient (burst) signals remains unexplored observationally. Dedicated unmodeled searches, designed to be robust against waveform uncertainty, have been carried out across observing runs, and no confident detection of a non-CBC burst source has yet been reported in the all-sky short-duration and long-duration burst analyses of recent LVK runs 6 ; 3 ; 2 ; 1 . This reflects the inherently challenging nature of detecting and interpreting short-duration GW transients in nonstationary detector noise due to the difficulty of distinguishing between instrumental artifacts and potential astrophysical signals. From an inference perspective, burst signals occupy a regime in which detection and classification decouple more sharply than for CBCs. Unmodeled (largely frequentist) burst pipelines, such as coherent WaveBurst (cWB) 19 ; 24 ; 23 , X-Pipeline 36 and PySTAMPAS 30 , are optimized to identify excess coherent power with minimal assumptions, reconstruct time-frequency structure, and deliver rapid detection statistics and sky localization. These tools are indispensable for discovery, but by construction they do not provide a complete Bayesian treatment of physical hypotheses. For short signals, this distinction matters: with few observed cycles, qualitatively different physical models can yield similar detector-frame morphologies once projected through the network response and filtered by the detector noise curve. Consequently, the relevant question for a candidate or sub-threshold event is often not only whether the data contain some coherent signal, but which of several plausible models is supported, how strongly, and with what parameter constraints. In the burst context, Bayesian techniques 13 are already widely used for waveform-agnostic reconstruction and glitch mitigation (e.g. BayesWave) 14 ; 27 , and past work has illustrated that physically motivated burst templates can be meaningfully classified against one another in realistic noise 18 . For CBCs, end-to-end inference is routinely performed with standardized, production-tested pipelines built atop community libraries. For bursts, it is similarly impractical to perform multi-hypothesis Bayesian inference via bespoke scripts and ad hoc orchestration. This slows iteration, complicates reproducibility, and raises the practical barrier to systematic studies (e.g. injection campaigns), precisely in the regime where large ensembles of runs are needed to separate noise-driven ambiguity from genuine model overlap. To address this gap, we present Basilic ( BA yesian tran SI ent mode L Ing and C lassification), an end-to-end pipeline for Bayesian inference and model classification of short-duration GW transients, designed to complement the LVK burst-detection suite. Basilic is built on top of bilby 12 and inherits its likelihood infrastructure and samplers, while adding the missing pipeline layer for bursts: automated run setup from a single configuration file, standardized data/PSD handling (including integration with GWpy 29 ), scalable orchestration for single-event analyses and injection campaigns, and a coherent post-processing stage that produces model-comparison summaries and diagnostic products. A central design goal is to make cross-hypothesis checks routine: a user specifies what to analyze and which models to compare, and Basilic handles the reproducible mechanics of running, aggregating, and reporting results. In parallel, Basilic ships with a growing registry of parametric burst models drawn from the literature, including core-collapse supernova phenomenological templates 31 ; 39 , cosmic-string burst models and memory-inspired waveforms 18 , and standard CBC approximants through LALSuite when burst-CBC comparisons are scientifically relevant 26 ; 40 . Beyond presenting the software, this paper uses Basilic to produce a methodological outcome that is directly relevant for burst inference in current detectors. We perform a controlled injection campaign focused on a concrete ambiguity that has already motivated targeted analyses of heavy, short CBC-like events: the potential confusion between high-mass binary black hole (BBH) mergers and cosmic-string cusp waveforms when only a handful of cycles are in band. Such comparisons have been discussed in the context of specific events (e.g. GW190521 5 and GW231123 4 ; 16 ), and here we expand the study from single-event follow-up to a broader mapping over BBH parameter combinations at fixed detectability. In light of the results presented and burst signals in general, we also discuss explicitly some quantitative measures that can be adopted when Bayesian model selection alone is inconclusive, namely through posterior predictive checks 20 and waveform match. The remainder of this paper is structured as follows. In Sec. II we summarize the Bayesian burst model selection problem and introduce the Basilic pipeline, its workflow, and the burst model catalog currently implemented. In Sec. III we define the BBH–cosmic-string confusion study and its injection-campaign design. Sec. IV presents the main results of the campaign, including regimes of enhanced ambiguity and their parameter dependence. Sec. V discusses implications for short-transient interpretation and outlines how predictive-degeneracy and posterior-driven morphology studies can guide follow-up strategy. We conclude in Sec. VI. II Burst Bayesian Inference with Basilic Short duration GW signals are well detected by search pipelines based on coherent analysis of a multi-detector streams of strain timeseries 23 ; 36 ; 14 . From the perspective of source identification, however, a Bayesian analysis performed on interesting candidates provided by such detection pipelines remains of paramount importance. We outline below Bayesian model selection and checking with Basilic and showcase its features that aim to bring burst model selection to an ease of use comparable to CBCs. II.1 Burst model selection and checking Unlike longer-duration signals, which can accumulate signal-to-noise ratio (SNR) through the many observed in-band cycles and thus allow for more robust model discrimination, burst source candidates are expected to have low SNR and are likely confused with noise fluctuations. Considering then the intrinsically challenging nature of a burst signal exhibiting similar signal morphology and low SNR, a complete Bayesian study is often needed. Model selection proceeds first through Bayes’ theorem, by updating the belief in parameters \boldsymbol{\theta} of a model M given observed data strain d^{\ast} 13 \mathcal{P}(\boldsymbol{\theta}\mid d^{\ast},M)=\frac{\mathcal{L}(d^{\ast}\mid\boldsymbol{\theta},M)\pi(\boldsymbol{\theta}\mid M)}{\mathcal{Z}(d^{\ast}\mid M)}, (1) with the a priori knowledge of said model’s parameters encoded through the prior \pi(\boldsymbol{\theta}\mid M) and the Gaussian-noise GW strain likelihood \mathcal{L}(d^{\ast}\mid\boldsymbol{\theta},M) . To enable selection between considered models M_{1}

Read the complete text