arXiv subjects

Alejandro Torres-Forné

Publications and source records attributed to Alejandro Torres-Forné.

At least 37 records · Page 2Linked to original sources

arXiv2023

Comparison of neural network architectures for feature extraction from binary black hole merger waveforms

We evaluate several neural-network architectures, both convolutional and recurrent, for gravitational-wave time-series feature extraction by performing point parameter estimation on noisy waveforms from binary-black-hole mergers. We build datasets of 100,000 elements for each of four different waveform models (or approximants) in order to test how approximant choice affects feature extraction. Our choices include \texttt{SEOBNRv4P} and \texttt{IMRPhenomPv3}, which contain only the dominant quadrupole emission mode, alongside \texttt{IMRPhenomPv3HM} and \texttt{NRHybSur3dq8}, which also account for high-order modes. Each dataset element is injected into detector noise corresponding to the third observing run of the LIGO-Virgo-KAGRA (LVK) collaboration. We identify the Temporal Convolutional Network (TCN) architecture as the overall best performer in terms of training and validation losses and absence of overfitting to data. Comparison of results between datasets shows that the choice of waveform approximant for the creation of a dataset conditions the feature extraction ability of a trained network. Hence, care should be taken when building a dataset for the training of neural networks, as certain approximants may result in better network convergence of evaluation metrics. However, this performance does not necessarily translate to data which is more faithful to numerical relativity simulations. We also apply this network on actual signals from LVK runs, finding that its feature-extracting performance can be effective on real data.

gr-qc↗

arXiv2023

Gradient-Annihilated PINNs for Solving Riemann Problems: Application to Relativistic Hydrodynamics

We present a novel methodology based on Physics-Informed Neural Networks (PINNs) for solving systems of partial differential equations admitting discontinuous solutions. Our method, called Gradient-Annihilated PINNs (GA-PINNs), introduces a modified loss function that requires the model to partially ignore high-gradients in the physical variables, achieved by introducing a suitable weighting function. The method relies on a set of hyperparameters that control how gradients are treated in the physical loss and how the activation functions of the neural model are dynamically accounted for. The performance of our GA-PINN model is demonstrated by solving Riemann problems in special relativistic hydrodynamics, extending earlier studies with PINNs in the context of the classical Euler equations. The solutions obtained with our GA-PINN model correctly describe the propagation speeds of discontinuities and sharply capture the associated jumps. We use the relative $l^{2}$ error to compare our results with the exact solution of special relativistic Riemann problems, used as the reference ``ground truth'', and with the error obtained with a second-order, central, shock-capturing scheme. In all problems investigated, the accuracy reached by our GA-PINN model is comparable to that obtained with a shock-capturing scheme and significantly higher than that achieved by a baseline PINN algorithm. An additional benefit worth stressing is that our PINN-based approach sidesteps the costly recovery of the primitive variables from the state vector of conserved ones, a well-known drawback of grid-based solutions of the relativistic hydrodynamics equations. Due to its inherent generality and its ability to handle steep gradients, the GA-PINN method discussed could be a valuable tool to model relativistic flows in astrophysics and particle physics, characterized by the prevalence of discontinuous solutions.

physics.comp-ph↗

arXiv2023

Convolutional Neural Networks for the classification of glitches in gravitational-wave data streams

We investigate the use of Convolutional Neural Networks (including the modern ConvNeXt network family) to classify transient noise signals (i.e.~glitches) and gravitational waves in data from the Advanced LIGO detectors. First, we use models with a supervised learning approach, both trained from scratch using the Gravity Spy dataset and employing transfer learning by fine-tuning pre-trained models in this dataset. Second, we also explore a self-supervised approach, pre-training models with automatically generated pseudo-labels. Our findings are very close to existing results for the same dataset, reaching values for the F1 score of 97.18% (94.15%) for the best supervised (self-supervised) model. We further test the models using actual gravitational-wave signals from LIGO-Virgo's O3 run. Although trained using data from previous runs (O1 and O2), the models show good performance, in particular when using transfer learning. We find that transfer learning improves the scores without the need for any training on real signals apart from the less than 50 chirp examples from hardware injections present in the Gravity Spy dataset. This motivates the use of transfer learning not only for glitch classification but also for signal classification.

gr-qc↗

arXiv2023

Impact of the wave-like nature of Proca stars on their gravitational-wave emission

We present a systematic study of the dynamics and gravitational-wave emission of head-on collisions of spinning vector boson stars, known as Proca stars. To this aim we build a catalogue of about 800 numerical-relativity simulations of such systems. We find that the wave-like nature of bosonic stars has a large impact on the gravitational-wave emission. In particular, we show that the initial relative phase $Δε=ε_1-ε_2$ of the two complex fields forming the stars (or equivalently, the relative phase at merger) strongly impacts both the emitted gravitational-wave energy and the corresponding mode structure. This leads to a non-monotonic dependence of the emission on the frequency of the secondary star $ω_2$, for fixed frequency $ω_1$ of the primary. This phenomenology, which has not been found for the case of black-hole mergers, reflects the distinct ability of the Proca field to interact with itself in both constructive and destructive manners. We postulate this may serve as a smoking gun to shed light on the possible existence of these objects.

gr-qc↗

arXiv2022

Dictionary learning: a novel approach to detecting binary black holes in the presence of Galactic noise with LISA

The noise produced by the inspiral of millions of white dwarf binaries in the Milky Way may pose a threat to one of the main goals of the space-based LISA mission: the detection of massive black hole binary mergers. We present a novel study for reconstruction of merger waveforms in the presence of Galactic confusion noise using dictionary learning. We discuss the limitations of untangling signals from binaries with total mass from $10^2 M_{\odot}$ to $10^4 M_{\odot}$. Our method proves extremely successful for binaries with total mass greater than $\sim 3\times 10^3$ $ M_{\odot}$ up to redshift 3 in conservative scenarios, and up to redshift 7.5 in optimistic scenarios. In addition, consistently good waveform reconstruction of merger events is found if the signal-to-noise ratio is approximately 5 or greater.

gr-qc↗

arXiv2022

Implementation of the rROF denoising method in the cWB pipeline for gravitational-wave data analysis

The data collected by the current network of gravitational-wave detectors are largely dominated by instrumental noise. Total variation methods based on L1-norm minimization have recently been proposed as a powerful technique for noise removal in gravitational-wave data. In particular, the regularized Rudin-Osher-Fatemi (rROF) model has proven effective to denoise signals embedded in either simulated Gaussian noise or actual detector noise. Importing the rROF model to existing search pipelines seems therefore worth considering. In this paper, we discuss the implementation of two variants of the rROF algorithm as two separate plug-ins of the coherent Wave Burst (cWB) pipeline designed to conduct searches of unmodelled gravitational-wave burst sources. The first approach is based on a single-step rROF method and the second one employs an iterative rROF procedure. Both approaches are calibrated using actual gravitational-wave events from the first three observing runs of the LIGO-Virgo-KAGRA collaboration, namely GW1501914, GW151226, GW170817, and GW190521, encompassing different types of compact binary coalescences. Our analysis shows that the iterative version of the rROF denoising algorithm implemented in the cWB pipeline effectively eliminates noise while preserving the waveform signals intact. Therefore, the combined approach yields higher signal-to-noise values than those computed by the cWB pipeline without the rROF denoising step. The incorporation of the iterative rROF algorithm in the cWB pipeline might hence impact the detectability capabilities of the pipeline along with the inference of source properties.

astro-ph.IM↗

arXiv2021

Inference of proto-neutron star properties from gravitational-wave data in core-collapse supernovae

The eventual detection of gravitational waves from core-collapse supernovae (CCSN) will help improve our current understanding of the explosion mechanism of massive stars. The stochastic nature of the late post-bounce gravitational wave signal due to the non-linear dynamics of the matter involved and the large number of degrees of freedom of the phenomenon make the source parameter inference problem very challenging. In this paper we take a step towards that goal and present a parameter estimation approach which is based on the gravitational waves associated with oscillations of proto-neutron stars (PNS). Numerical simulations of CCSN have shown that buoyancy-driven g-modes are responsible for a significant fraction of the gravitational wave signal and their time-frequency evolution is linked to the physical properties of the compact remnant through universal relations, as demonstrated in [1]. We use a set of 1D CCSN simulations to build a model that relates the evolution of the PNS properties with the frequency of the dominant g-mode, which is extracted from the gravitational-wave data using a new algorithm we have developed for our study. The model is used to infer the time evolution of a combination of the mass and the radius of the PNS. The performance of the method is estimated employing simulations of 2D CCSN waveforms covering a progenitor mass range between 11 and 40 solar masses and different equations of state. Considering signals embedded in Gaussian gravitational wave detector noise, we show that it is possible to infer PNS properties for a galactic source using Advanced LIGO and Advanced Virgo data at design sensitivities. Third generation detectors such as Einstein Telescope and Cosmic Explorer will allow to test distances of ${\cal O}(100\, {\rm kpc})$.

gr-qc↗

arXiv2021

Universal relations for gravitational-wave asteroseismology of proto-neutron stars

State-of-the-art numerical simulations of core-collapse supernovae reveal that the main source of gravitational waves is the excitation of proto-neutron star modes during post-bounce evolution. In this work we derive universal relations that relate the frequencies of the most common oscillation modes observed, i.e. g-modes, p-modes and the f-mode, with fundamental properties of the system, such as the surface gravity of the proto-neutron star or the mean density in the region enclosed by the shock. These relations are independent of the equation of state, the neutrino treatment, and the progenitor mass and hence can be used to build methods to infer proto-neutron star properties from gravitational-wave observations alone. We outline how these measurements could be done and the constraints that could be placed on the proto-neutron star properties.

gr-qc↗

arXiv2021

Classification of the core-collapse supernova explosion mechanism with learned dictionaries

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.

gr-qc↗

arXiv2021

Confusing Head-On Collisions with Precessing Intermediate-Mass Binary Black Hole Mergers

We report a degeneracy between the gravitational-wave signals from quasi-circular precessing black-hole mergers and those from extremely eccentric mergers, namely head-on collisions. Performing model selection on numerically simulated signals of head-on collisions using models for quasi-circular binaries we find that, for signal-to-noise ratios of 15 and 25, typical of Advanced LIGO observations, head-on mergers with respective total masses of $M\in (125,300)M_\odot$ and $M\in (200,440)M_\odot$ would be identified as precessing quasi-circular intermediate-mass black hole binaries, located at a much larger distance. Ruling out the head-on scenario would require to perform model selection using currently nonexistent waveform models for head-on collisions, together with the application of astrophysically motivated priors on the (rare) occurrence of those events. We show that in situations where standard parameter inference of compact binaries may report component masses inside (outside) the pair-instability supernova gap, the true object may be a head-on merger with masses outside (inside) this gap. We briefly discuss the potential implications of these findings for the recent gravitational-wave detection GW190521, which we analyse in detail in [Phys. Rev. Lett. 126, 081101].

gr-qc↗

arXiv2021

GW190521 as a merger of Proca stars: a potential new vector boson of $8.7 \times 10^{-13}$ eV

Advanced LIGO-Virgo reported a short gravitational-wave signal (GW190521) interpreted as a quasi-circular merger of black holes, one populating the pair-instability supernova gap, forming a remnant black hole of $M_f\sim 142 M_\odot$ at a luminosity distance of $d_L \sim 5.3$ Gpc. With barely visible pre-merger emission, however, GW190521 merits further investigation of the pre-merger dynamics and even of the very nature of the colliding objects. We show that GW190521 is consistent with numerically simulated signals from head-on collisions of two (equal mass and spin) horizonless vector boson stars (aka Proca stars), forming a final black hole with $M_f = 231^{+13}_{-17}\,M_\odot$, located at a distance of $d_L = 571^{+348}_{-181}$ Mpc. The favoured mass for the ultra-light vector boson constituent of the Proca stars is $μ_{\rm V}= 8.72^{+0.73}_{-0.82}\times10^{-13}$ eV. This provides the first demonstration of close degeneracy between these two theoretical models, for a real gravitational-wave event. Confirmation of the Proca star interpretation, which we find statistically slightly preferred, would provide the first evidence for a long sought dark matter particle.

gr-qc↗

arXiv2020

Exploring gravitational-wave detection and parameter inference using Deep Learning methods

We explore machine learning methods to detect gravitational waves (GW) from binary black hole (BBH) mergers using deep learning (DL) algorithms. The DL networks are trained with gravitational waveforms obtained from BBH mergers with component masses randomly sampled in the range from 5 to 100 solar masses and luminosity distances from 100 Mpc to, at least, 2000 Mpc. The GW signal waveforms are injected in public data from the O2 run of the Advanced LIGO and Advanced Virgo detectors, in time windows that do not coincide with those of known detected signals. We demonstrate that DL algorithms, trained with GW signal waveforms at distances of 2000 Mpc, still show high accuracy when detecting closer signals, within the ranges considered in our analysis. Moreover, by combining the results of the three-detector network in a unique RGB image, the single detector performance is improved by as much as 70%. Furthermore, we train a regression network to perform parameter inference on BBH spectrogram data and apply this network to the events from the the GWTC-1 and GWTC-2 catalogs. Without significant optimization of our algorithms we obtain results that are mostly consistent with published results by the LIGO-Virgo Collaboration. In particular, our predictions for the chirp mass are compatible (up to 3$σ$) with the official values for 90% of events

gr-qc↗

arXiv2020

Application of dictionary learning to denoise LIGO's blip noise transients

Data streams of gravitational-wave detectors are polluted by transient noise features, or "glitches", of instrumental and environmental origin. In this work, we investigate the use of total-variation methods and learned dictionaries to mitigate the effect of those transients in the data. We focus on a specific type of transient, "blip" glitches, as this is the most common type of glitch present in the LIGO detectors and their waveforms are easy to identify. We randomly select 80 blip glitches scattered in the data from advanced LIGO's O1 run, as provided by the citizen-science project Gravity Spy. Our results show that dictionary-learning methods are a valid approach to model and subtract most of the glitch contribution in all cases analyzed, particularly at frequencies below $\sim 1$ kHz. The high-frequency component of the glitch is best removed when a combination of dictionaries with different atom length is employed. As a further example, we apply our approach to the glitch visible in the LIGO-Livingston data around the time of merger of binary neutron star signal GW170817, finding satisfactory results. This paper is the first step in our ongoing program to automatically classify and subtract all families of gravitational-wave glitches employing variational methods.

gr-qc↗

arXiv2018

Classification of gravitational-wave glitches via dictionary learning

We present a new method for the classification of transient noise signals (or glitches) in advanced gravitational-wave interferometers. The method uses learned dictionaries (a supervised machine learning algorithm) for signal denoising, and untrained dictionaries for the final sparse reconstruction and classification. We use a data set of 3000 simulated glitches of three different waveform morphologies, comprising 1000 glitches per morphology. These data are embedded in non-white Gaussian noise to simulate the background noise of advanced LIGO in its broadband configuration. Our classification method yields a 96% accuracy for a large range of initial parameters, showing that learned dictionaries are an interesting approach for glitch classification. This work constitutes a preliminary step before assessing the performance of dictionary-learning methods with actual detector glitches.

astro-ph.IM↗

arXiv2018

Towards asteroseismology of core-collapse supernovae with gravitational-wave observations - I. Cowling approximation

Gravitational waves from core-collapse supernovae are produced by the excitation of different oscillation modes in the proto-neutron star (PNS) and its surroundings, including the shock. In this work we study the relationship between the post-bounce oscillation spectrum of the PNS-shock system and the characteristic frequencies observed in gravitational-wave signals from core-collapse simulations. This is a fundamental first step in order to develop a procedure to infer astrophysical parameters of the PNS formed in core-collapse supernovae. Our method combines information from the oscillation spectrum of the PNS, obtained through linear-perturbation analysis in general relativity of a background physical system, with information from the gravitational-wave spectrum of the corresponding non-linear, core-collapse simulation. Using results from the simulation of the collapse of a 35 $M_{\odot}$ presupernova progenitor we show that both types of spectra are indeed related and we are able to identify the modes of oscillation of the PNS, namely g-modes, p-modes, hybrid modes, and standing-accretion-shock-instability (SASI) modes, obtaining a remarkably close correspondence with the time-frequency distribution of the gravitational-wave modes. The analysis presented in this paper provides a proof-of-concept that asteroseismology is indeed possible in the core-collapse scenario, and it may serve as a basis for future work on PNS parameter inference based on gravitational-wave observations.

astro-ph.SR↗

arXiv2018

Towards asteroseismology of core-collapse supernovae with gravitational-wave observations - II. Spacetime perturbations

Improvements in ground-based, advanced gravitational wave (GW) detectors may allow in the near future to observe the GW signal of a nearby core-collapse supernova. For the most common type of progenitors, likely with slowly rotating cores, the dominant GW emission mechanisms are the post-bounce oscillations of the proto-neutron star (PNS) before the explosion. We present a new procedure to compute the eigenmodes of the system formed by the PNS and the stalled accretion shock in general relativity including spacetime perturbations. The new method improves on previous results by accounting for perturbations of both the lapse function and the conformal factor. We apply our analysis to two numerical core-collapse simulations and show that our improved method is able to obtain eigenfrequencies that accurately match the features observed in the GW signal and to predict the qualitative behaviour of quasi-radial oscillations. Our analysis is possible thanks to a newly developed algorithm to classify the computed eigenmodes in different classes (f-, p-, and g-modes), improving previous results which suffered from misclassification issues. We find that most of the GW energy is stored in the lowest order eigenmodes, in particular in the $^2g_1$ mode and in the $^2f$ mode. Our results also suggest that a low-frequency component of the GW signal attributed in previous works to the characteristic frequency of the Standing Accretion Shock Instability should be identified as the fundamental quadrupolar f-mode. We also develop a formalism to estimate the contribution of quasi-radial ($l=0$) modes to the quadrupolar component of the GW emission in the case of a deformed background, with application to rapidly rotating cores. This work provides further support for asteroseismology of core-collapse supernovae and the inference of PNS properties based on GW observations.

astro-ph.HE↗

arXiv2018

Total-variation methods for gravitational-wave denoising: performance tests on Advanced LIGO data

We assess total-variation methods to denoise gravitational-wave signals in real noise conditions, by injecting numerical-relativity waveforms from core-collapse supernovae and binary black hole mergers in data from the first observing run of Advanced LIGO. This work is an extension of our previous investigation where only Gaussian noise was used. Since the quality of the results depends on the regularization parameter of the model, we perform an heuristic search for the value that produces the best results. We discuss various approaches for the selection of this parameter, either based on the optimal, mean, or multiple values, and compare the results of the denoising upon these choices. Moreover, we also present a machine-learning-informed approach to obtain the Lagrange multiplier of the method through an automatic search. Our results provide further evidence that total-variation methods can be useful in the field of Gravitational-Wave Astronomy as a tool to remove noise.

astro-ph.SR↗

arXiv2016

Denoising of gravitational wave signals via dictionary learning algorithms

Gravitational wave astronomy has become a reality after the historical detections accomplished during the first observing run of the two advanced LIGO detectors. In the following years, the number of detections is expected to increase significantly with the full commissioning of the advanced LIGO, advanced Virgo and KAGRA detectors. The development of sophisticated data analysis techniques to improve the opportunities of detection for low signal-to-noise-ratio events is hence a most crucial effort. We present in this paper one such technique, dictionary-learning algorithms, which have been extensively developed in the last few years and successfully applied mostly in the context of image processing. However, to the best of our knowledge, such algorithms have not yet been employed to denoise gravitational wave signals. By building dictionaries from numerical relativity templates of both, binary black holes mergers and bursts of rotational core collapse, we show how machine-learning algorithms based on dictionaries can be also successfully applied for gravitational wave denoising. We use a subset of signals from both catalogs, embedded in non-white Gaussian noise, to assess our techniques with a large sample of tests and to find the best model parameters. The application of our method to the actual signal GW150914 shows promising results. Dictionary-learning algorithms could be a complementary addition to the gravitational wave data analysis toolkit. They may be used to extract signals from noise and to infer physical parameters if the data are in good enough agreement with the morphology of the dictionary atoms.

astro-ph.IM↗