Schedule for: 21w5066 - Detection and Analysis of Gravitational Waves in the era of Multi-Messenger Astronomy: From Mathematical Modelling to Machine Learning

The direct detection of gravitational waves opened an unprecedented channel to probe fundamental physics. Proposed extensions of our current theories predict a dispersion of the gravitational waves during their propagation, distorting the signals observed by ground-based interferometers compared to their predictions from general relativity. In this talk, I present several analysis probing different alternative theories of gravitation. Using the multimessenger events consisting of gravitational waves and their electromagnetic counterpart, extra dimensions and scalar-tensor theories are constrained from the comparison of the luminosity distance inferred independently from both signals. Relying only on gravitational wave signals, a large class of proposed theories (e.g. massive gravity) predict a frequency-dependent dispersion of the gravitational waves breaking local CPT and/or Lorentz symmetry. Constraints on the corresponding effective field theories coefficients are obtained from the analysis of 31 events from the second LIGO-Virgo detections catalog.

We present a novel paradigm for the deployment of computing infrastructure for the low-latency analyses of gravitational wave (GW) data. Here, we specifically discuss two deep-learning algorithms, 'DeepClean' and 'BBHNet', which are used for GW data denoising and compact binary source identification respectively. Using “replayed” streams of the GW data of LIGO Hanford and Livingston from their third observing run, we demonstrate the subtraction of stationary and non-stationary noise sources followed by identification of candidates for the astrophysical transient detections at low latency. Depending on the computing platform used, we show that it is possible to achieve these at latencies of ~ a few hundreds of milliseconds to a few seconds. Real-time delivery of gravitational-wave alerts is important for enabling rapid multi-messenger follow-up, especially to capture short-lived electromagnetic counterparts. Additionally, our implementation offers seamless incorporation of hardware accelerators and enables the use of as-a-service computing, which would be capable of meeting the future needs of gravitational-wave data analysis.

Gravitational waves have opened a new window on the Universe and paved the way to multimessenger astronomy. Advanced LIGO and Advanced Virgo interferometers are probing an increasingly larger volume of space, discovering more and more signals produced by the coalescence of compact binary systems. Characterizing these detectors and their noise is crucial to optimize the sensitivity. In particular, glitches are transient noise events impacting the data quality, and their detection and classification is very important to improve the performance of the interferometers. Deep learning techniques are a promising approach to recognize and classify glitches and to study noise in general. We will present how machine learning can help in investigating the time-frequency evolution of glitches and thus contribute to the low-latency characterization of gravitational wave detectors.

Gravitational wave measurements provide a new opportunity to determine whether dark matter is truly a cold, collisionless particle. Intermediate mass-ratio inspirals (IMRIs) embedded in dark matter halos are particularly promising targets. These IMRIs can compress their dark halos to extreme densities as they form, leading to distinctive gravitational wave signals. In this talk I will show that future interferometers could observe and characterize these "dark dress" systems over large swaths of their parameter space. After explaining their waveform modeling, I will map out which dark dresses could be detected, distinguished from astrophysical IMRIs and accurately measured.

Understanding the aftermath, or ringdown, of a binary black hole merger is crucial in understanding astrophysical black holes. As gravitational-wave detectors increase their sensitivity, they will be able to detect more ringdown frequencies that correspond to higher order multipole modes. However, most current models focus on fitting just for the dominant mode. In this talk, I will present recent work on multimode ringdown modeling and the importance of being in the correct frame of reference when performing ringdown fits. Also, I will briefly mention an example of how machine learning is applied to ringdown modeling. With these tools ready, we will be able to take full advantage of the gravitational-wave data when third generation detectors begin their observing runs.

Instrumental and environmental transient noise bursts in gravitational-wave detectors, or glitches, may impair astrophysical observations by adversely affecting the sky localization and the parameter estimation of gravitational-wave signals. Denoising of detector data is especially relevant during low-latency operations because electromagnetic follow-up of candidate detections requires accurate, rapid sky localization and inference of astrophysical sources. NNETFIX is a machine learning-based algorithm designed to remove glitches detected in coincidence with transient gravitational-wave signals. NNETFIX uses artificial neural networks to estimate the portion of the data lost due to the presence of the glitch, which allows the recalculation of the sky localization of the astrophysical signal. The sky localization of the denoised signal may be significantly more accurate than the sky localization obtained from the original data or by excising the portion of the data impacted by the glitch. We test NNETFIX in simulated scenarios of binary black hole coalescence signals and discuss the potential for its use in future low-latency LIGO-Virgo-KAGRA searches.

The interplay between dark matter and black holes remains largely unexplored. Dark matter can in principle be *made of* black holes, as long as these are primordial, i.e. they are formed in the very early universe. Dark matter can also accumulate *around* black holes, be them astrophysical or primordial, and modify the rich phenomenology exhibited by these objects. After a brief overview of the status of dark matter searches, I will discuss the prospects for detecting primordial black holes, or robustly ruling them out as dark matter candidates. I will then discuss the prospects for characterizing and identifying dark matter using gravitational waves, covering a wide range of dark matter candidate types and signals.

The use of machine learning techniques in the analysis of the data produced by gravitational wave detectors is a very active field of research, with many promising results. In this talk however, I will discuss how machine learning could also be applied to the instrument science side of gravitational wave detectors, to improve the sensitivity with noise subtraction or to improve the detector robustness with advanced control systems. This is a less mature field of research, and results are only recently starting to appear.

Core collapse supernovae is among the most exciting events that we expect to observe in the future by gravitational wave interferometers. They provide a unique multi messenger opportunity with the combined emission of gravitational waves, neutrinos and electromagnetic waves. In this talk I will focus in the current understanding of core-collapse GW signals and how they can be modelled in terms of normal oscillations modes of proto-neutron stars excited during the post-bounce phase before the onset of the SN explosion. The observation of such modes in the future by gravitational wave observatories (Virgo, LIGO) may allow to infer the properties of proto-neutron stars and constrain the the equation of state of hot nuclear matter.

Distinguishing black hole – neutron star binaries from binary black holes mergers for high mass ratios could be challenging because the neutron star coalesces with the black hole without experiencing significant disruption. To investigate the transition of the behavior of a mixed binary merger into one like a black hole binary, we present results from a series of merger simulations for different mass ratios. We show how the degree of disruption of the neutron star impacts the inspiral and merger dynamics, the properties of the final black hole, the accretion disk formed from the circularization of the tidal debris, the gravitational waves, and the strain spectrum and mismatches. The simulations use initial data constructed with a method that generalizes the Bowen-York initial data for black hole punctures to the case of neutron stars.

Convolutional Neural Networks (CNNs) have demonstrated potential for the real-time analysis of data from gravitational-wave detector networks for the specific case of signals from coalescing compact-object binaries such as black-hole binaries. Unfortunately, training these CNNs requires a precise model of the target signal; they are therefore not applicable to a wide class of potential gravitational-wave sources, such as core-collapse supernovae and long gamma-ray bursts, where unknown physics or computational limitations prevent the development of comprehensive signal models. We demonstrate for the first time a CNN with the ability to detect generic signals -- those without a precise model -- with sensitivity across a wide parameter space. Our CNN has a novel structure that uses not only the network strain data but also the Pearson cross-correlation between detectors to distinguish correlated gravitational-wave signals from uncorrelated noise transients. We demonstrate the efficacy of our CNN using data from the second LIGO-Virgo observing run. We show that it has sensitivity approaching that of the "gold-standard'' unmodeled transient searches currently used by LIGO-Virgo, at extremely low (order of 1 second) latency and using only a fraction of the computing power required by existing searches, allowing our models the possibility of true real-time detection of gravitational-wave transients associated with gamma-ray bursts, core-collapse supernovae, and other relativistic astrophysical phenomena.

The panchromatic observations of the electromagnetic (EM) counterpart of the binary neutron star (BNS) merger, GW170817, marked the dawn of multi-messenger EMGW astronomy. But it remains the only success story, even after the LIGO/Virgo third observing run, which reported 56 GW discoveries publicly. This shows that while we may have reached an era of routine GW astronomy, the same for EMGW astronomy still awaits us. The first step is to develop low-latency data-products that aid the follow-up of interesting GW candidates. The low-latency nature of the problem, taking into account physically motivated models plus un-modeled or poorly understood sources of search biases makes the use of data-driven approaches and machine learning particularly suited for this problem. I will talk about such efforts using machine learning approaches both from the GW and EM standpoint for near real-time inference of EMGW counterparts.

A major boost in the understanding of the universe was given by the revelation of the first coalescence event of two neutron stars (GW170817) and the observation of the same event across the entire electromagnetic spectrum. With third-generation gravitational wave detectors and the new astronomical facilities, we expect many multi-messenger events of the same type. We anticipate the need to analyse the data provided to us by such events not only to fulfill the requirements of real-time analysis, but also in order to decipher the event in its entirety through the information emitted in the different messengers using machine learning. What we propose is the application of a multimodal machine learning approach to characterize these events.

The global network of ground-based gravitational wave (GW) detectors is expected to expand and increase in sensitivity by the middle of this decade. Effective characterization and calibration of near-future detectors will allow us to explore new physics and astrophysics, including the origin of detected GW sources and probes of cosmology. I will give an overview of recent efforts at UBC to develop new approaches to characterize the performance of current and near-future detectors, including novel metrics for detector performance and characterization, new methods for distinguishing between true GW signals and detector noise, and improved techniques for GW signal reconstruction.

After decades of preparation, the era of gravitational wave astronomy has begun. The gravitational wave detectors, LIGO and Virgo, have published a catalog of 50 events of coalescing compact objects including black holes and neutron stars. I will present the role that numerical relativity played in the unveiling of the gravitational wave sky and anticipate how it might improve our understanding of gravity as the detectors improve..

Multi-messenger observations of compact binary coalescences will enrich our understanding of the astrophysics behind such sources, especially binary neutron star mergers. The joint analysis of gravitational wave data in conjunction with electromagnetic observations is a crucial ingredient for multi-messenger astrophysics. This talk will present an overview of multi-messenger astrophysics research at the University of Glasgow, including the analysis of kilonova light curves for joint observations and a hierarchical Bayesian analysis, with a machine learning augmented sampler (Nessai), for combining gravitational wave and gamma-ray observations to identify plausible models for gamma-ray burst jet structures.

Bayesian inference is the foundational framework used to reason about models and extract the source properties of binary coalescence signals from gravitational wave data. Over the last decade and more, stochastic sampling approaches have demonstrated their ability to perform unbiased and robust computation inference at scale. I will review the state of the art in this field. I'll provide insights into the shortcomings of current approaches and highlight where new ideas are required.

The search of gravitational waves (GW) from core-collapse supernovae (CCSNe) relies on detection algorithms such as coherent WaveBurst (cWB). False alarm rates and statistical significance might be affected by blip and glitch noises embedded in the strain data, which survive rejection tests. It is of interest then to detect and discard surviving noise events. This work presents the use of supervised machine learning (ML) methods, in specific, Linear Discriminant Analysis and Support Vector Machines, to recognize between noise and signal events using a set of reconstruction parameters from cWB. We tested this ML follow-up method using strain data from the O3a run of advanced LIGO, and CCSNe GW signals extracted from 3D simulations. The ML model is learned using a dataset of noise and signal events extracted from a given on-source window, and then it is used to identify and discard noise events in cWB analyses in different on-source windows. Noise and signal reduction levels were assessed in single detector networks (L1 and H1) and two detector networks (L1H1). The results showed an effective enhancement of the statistical significance of cWB-based searches of GWs from CCSNe.

Multi-dimensional simulations of core-collapse supernovae are essential for providing waveforms that can assist the detection and interpretation of gravitational waves from a prospective nearby explosion. An emerging theme in core-collapse supernova modelling is the wider importance of magnetic fields. While magnetorotational explosions have long been investigated as a scenario for rare hypernova explosions, recent simulations suggest that magnetic fields may play an important role in normal explosions as well. To better understand the role of magnetic fields both in normal and hyperenergetic explosions, it is also imperative to revisit the interplay of convection, magnetic fields and rotation during the pre-supernova evolution of massive stars. I will discuss progress on 3D magnetohydrodynamic simulations of supernovae and their progenitors, possible implications for hypernovae and magnetar formation, and point to open issues in the current models

Gravitational wave detection is now commonplace and as the sensitivity of the global network of GW detectors improves, we will observe on the order of 100s of transient GW events per year. The current methods used to estimate their source parameters employ optimally sensitive but computationally costly Bayesian inference approaches where typical analyses have taken between 6 hours and 6 days. For binary neutron star (BNS) and neutron star black hole (NSBH) systems prompt counterpart electromagnetic (EM) signatures are expected on timescales of 1 second – 1 minute and the current fastest method for alerting EM follow-up observers, can provide estimates in on the order of 1 minute, on a limited range of key source parameters. Here we show that a conditional variational autoencoder (CVAE) pre-trained on binary black hole (BBH) signals can return Bayesian posterior probability estimates. The training procedure need only be performed once for a given prior parameter space and the resulting trained machine can then generate samples describing the posterior distribution ∼ 6 orders of magnitude faster than existing techniques.

The search for the counterpart to the extraordinary GW event GW170817 was a global effort. With many optical facilities rushing to first identify the counterpart, the first bonafide optical identification came nearly 11 hours post-trigger. Since then, the search for counterparts has become fairly routine, searching for sources using variations of the greedy algorithm. However, in these routine searches, there is still an inherently large number of optical astrophysical transients that arrive simultaneously and that must be vetted in real-time. I will give an overview of the current strategies of transient classification employed by optical observatories, such as Bayesian Convolution Neural Networks, Photometric Time-Series classification, and improvements on building balanced training sets for large sky-surveys in preparation for EM follow-up during O4.