Schedule for: 24w5177 - Detection and Analysis of Gravitational Waves in the era of Multi-Messenger Astronomy

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Meet and Greet at the BIRS Lounge (PDC, 2nd Floor)

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

A brief introduction to BIRS with important logistical information, technology instruction, and opportunity for participants to ask questions.

With the detection of compact binary coalescences and their electromagnetic counterparts by gravitational-wave detectors, a new era of multi-messenger astronomy has begun. In this talk, I will describe how the gravitational-wave community is using machine learning to increase the etection prospects for ffuture mergers. I will then discuss how current ground based optical surveys and dedicated follow-up systems are being used to identify more of these, and how we are developing models to test what we find. We will close with near-term prospects for the field.

The detection of the electromagnetic (EM) emission following the gravitational wave (GW) event GW170817 opened the era of multi-messenger astronomy with GWs and provided the first direct evidence that at least a fraction of binary neutron star (BNS) mergers are progenitors of short Gamma-Ray Bursts (GRBs). GRBs are also expected to emit very-high energy (VHE, > 100 GeV) photons, as proven by the recent MAGIC, H.E.S.S. And LHAASO observations and one of the challenges for future multi-messenger observations will be the detection of such VHE emission from GRBs in association with GWs. In this talk I will review the challenges and the status of the searches for VHE EM counterparts to GWs and discuss the prospects for future detections with next generation instruments such as the Cherenkov Telescope Array. The implications that future joint GW and VHE EM observations could have on the understanding of GRB physics will also be discussed.

Since before the first Gravitational Wave Transient Catalog (GWTC-1), the LIGO/Virgo/KAGRA collaboration has shared its probabilistic catalogs as collections of posterior samples for each confident observation of a compact binary merger.  While this has enabled a slew of impactful population studies, limitations of these samples and the importance integrals they are typically used for, are becoming increasingly relevant.  As the catalog continues to grow, so will the need for a more scalable solution.  I will show how a normalizing-flow-based catalog overcomes these limitations, while also providing a more usable catalog for the broader community.

The population-level properties of merging compact object binaries encode critical information on the poorly constrained astrophysics of their formation. Previous studies have attempted to measure the astrophysical parameters and initial conditions of known formation models by comparing the predictions of population synthesis simulations with an ensemble of gravitational wave observations. However, the vast collection of known formation channels each characterized by several unknown parameters makes existing methods challenging and costly to implement for realistic population synthesis simulations and growing gravitational wave catalogs. In this work, we rely on neural posterior estimation to construct an emulator that can efficiently train on the predictions of existing population synthesis simulations to learn the mapping between astrophysical input parameters and the resulting compact binary population. Using our emulator we convert data-driven reconstructions of the compact binary population from growing gravitational wave catalogs directly into measurements of astrophysical parameters that characterize known formation channels. By analyzing GWTC-3 data and a mock universe of simulated compact binaries, we demonstrate the accuracy of our method and discuss future applications for larger gravitational wave catalogs and more realistic population synthesis simulations.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

The joint analysis of gravitational wave data in conjunction with electromagnetic observations is a crucial ingredient for multi-messenger astrophysics. These joint analyses can computationally intensive, leading to latencies that make the analysis unfeasible for near-realtime follow-ups. This talk will present machine learning tools that have been deployed to facilite rapid computation for multi-messenger astronomy, including population analyses for GRB jet structures, rapid EOS inference and kilonova light curve predictions.

Matched-filtering is a long-standing technique for the optimal detection of known signals in stationary Gaussian noise. It enjoys deterministic behaviour and provides theoretical guarantees for that behaviour. However, it has known departures from optimality when operating on unknown signals in real noise and suffers from computational inefficiencies in its pursuit to near-optimality. A compelling alternative that has emerged in recent years to address this problem is deep learning. Although it has shown significant promise when applied to the search for gravitational-waves in detector noise, we demonstrate that the existence of a multitude of biases hinder generalisation and detection performance. Our work identifies the sources of a comprehensive set of biases present in the supervised learning of the gravitational-wave detection problem, and contributes mitigation tactics and training strategies to concurrently address them. We introduce, Sage, a machine-learning based binary black hole search pipeline. An injection study in O3a noise showed that Sage detects ~11% more signals than matched filtering at a false alarm rate of one per month. Moreover, we also show that it can detect ~48% more signals than the previous best performing machine- learning pipeline on the same dataset. We empirically prove that our pipeline has the capability to effectively handle out-of-distribution noise power spectral densities and reject non-Gaussian transient noise artefacts. The investigation ends with an ablation study of our pipeline, where we systematically remove important components to better understand their contribution.

Meet in foyer of TCPL to participate in the BIRS group photo. The photograph will be taken outdoors, so dress appropriately for the weather. Please don't be late, or you might not be in the official group photo!

A neutron star binary merges somewhere in the Universe approximately every 10 to 1000 seconds, creating violent explosions potentially observable in gravitational waves and across the electromagnetic spectrum. The transformative coincident gravitational wave and electromagnetic observations of the binary neutron star merger GW170817 gave invaluable insights into these cataclysmic collisions and fundamental astrophysics. However, despite our high expectations, we have failed to see any other event like it. In this talk, I will highlight what we can learn from other observations of mergers seen directly in gravitational waves or indirectly as a gamma-ray burst and/or kilonova. I will also discuss the diversity in electromagnetic and gravitational-wave emission we can expect for future mergers and showcase tools to help maximally extract physics from existing and future observations.

The detection of gravitational waves has revealed an invisible side of the Universe, allowing us to test our understanding of dynamical spacetime, study matter at extreme densities, and measure the expansion history of our Universe in new ways. Gravitational wave data also provides opportunities for unexpected discoveries. In this talk I will discuss one avenue for future discovery with gravitational waves: direct searches for exotic stars.

"A small fraction (~0.1% — 1%) of the gravitational-wave (GW) signals detectable by ground-based detectors will undergo strong gravitational lensing by intervening galaxies and clusters. In addition, compact objects with sizes comparable to the GW wavelength can produce micro-lensing effects that can distort the observed signal. I will talk about how the observation (as well as non-observation) of lensed GWs can probe different aspects of cosmology, including primordial black holes, dark matter and cosmic expansion rate.

The general theory of relativity (GR) has been a highly successful theory in explaining current astronomical observations and laboratory experiments. However, there is a general consensus that GR is at best incomplete, representing an approximation to a more complete theory that cures some or all of its problems. Due to GR’s enormous success, the physics community will not be ready to accept a violation unless the statistical confidence associated with the result is very high. No violation of GR has been reported so far but there have been gravitational wave events that hint towards a violation though further analyses are needed to understand them. This presentation will discuss various causes that could potentially lead to a false GR violation and what we can do to mitigate them. We will discuss recent results from the study on the effect of missing eccentricity and strong lensing on tests of GR and how they could lead to false GR violations. We argue that these causes should be thoroughly investigated, quantified, and ruled out before claiming a GR violation in gravitational wave observations.

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

TBA

Future observing runs with ground-based gravitational-wave interferometers promise to rapidly increase the detection rate of signals and the breadth of science we can probe with gravitational waves. However, these discoveries are often hampered by the realistic features of detector data that violate many of the core assumptions of our current data analysis methods. To address these problems, numerous techniques are employed to identify and mitigate problems in the data that may impact the detection or analysis of gravitational wave signals. As the event rate grows, these procedures will also need to evolve to quickly and generically address the wide variety of problems that may impact each new signal. In this talk, I will detail the current state-of-the-art methods to address common data issues that impact the analysis of transient gravitational-wave signals and explore how the field can adapt to the big data challenges on the horizon.

The background noise in gravitational wave interferometers is commonly assumed to be Gaussian and stationary during data analysis, though deviations from these assumptions may arise from gravitational wave signals or transient terrestrial noise artifacts known as glitches. Reconstructing these transient features, whether originating from signals or glitches, is essential for a range of scientific analyses. In this study, we introduce DeepExtractor, a deep learning method based on the U-Net architecture, designed to reconstruct power excesses above Gaussian noise, independent of their origin. Signal reconstructions can serve as precursors to parameter estimation pipelines, while glitch reconstructions can enable large-scale simulations or mock data challenges involving realistic noise artifacts. We validate DeepExtractor’s efficacy across three experiments: (1) reconstruction accuracy with simulated glitch injections into detector noise, (2) a direct comparison with the state-of-the-art BayesWave algorithm on a reduced test set, and (3) analysis of real data from the Gravity Spy dataset, demonstrating effective glitch subtraction from LIGO strain data.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

The LIGO-Virgo-KAGRA (LVK) collaboration has published about 90 gravitational wave observation since the first detection in 2015, majority of which have been from black hole binaries. With such a statistically significant number of events it is possible to infer astrophysical distribution of black hole masses and spins, provided the signal models used in data analysis are accurate enough. Phenomenological gravitational waveform models of binaries are routinely used in LVK data analysis. These models are modular by construction which not only allows incremental improvements in the different sectors to be easily ported in, but also the addition of new physics. I will discuss one such physical effect that was recently added to these models, notably the asymmetric emission of gravitational waves between the +m and -m multipoles of the harmonic decomposition of the signal, and other recent improvements to the state-of-the-art. I will also discuss the amenability of these models to consider scenarios other than black hole binaries.

The degree of alignment of binary black hole spins with its orbital angular momentum is an important discriminator of formation channels. In a binary with a significant misalignment, the gravitational-wave waveform can be impacted by precession. Unfortunately, the most promising events that show signs of precession thus far also have data quality issues that have been shown to impact measurement of in-plane spin. In the first part of the talk, I will demonstrate new parameter estimation methods that can more robustly account for the presence of these glitches and show results when applied to these events. In the second part of the talk, I will zoom out to consider evidence of spin misalignment at the population level. There has been considerable debate in the literature about whether the data supports the presence of a population with negative tilts or a population with an excess of non-spinning binaries. After a survey of these studies, I will argue that a suitably flexible model of the effective inspiral spin distribution can allow us to robustly limit both the anti-aligned population and a preferentially aligned population.

Spin tilts of binary black holes—the angles between each black hole's spin and the binary's orbital angular momentum—are crucial for distinguishing different astrophysical formation channels, such as isolated and dynamical formation. However, the tilts observed when the binary enters the frequency band of ground-based gravitational-wave detectors can differ significantly from the tilts at formation. Therefore, evolving the binary backward in time from its detection to its formation is essential for understanding its origins. There exist codes to perform this evolution for quasi-circular precessing binary black holes, but not yet any for eccentric precessing binaries, even though dynamical formation channels can result in non-negligible eccentricities at detection. To address this gap, we have developed a code to evolve eccentric and precessing binary black holes using orbit-averaged post-Newtonian equations at small orbital separations and precession-averaged equations at large orbital separations. Applying this code, we have studied how the parameters of eccentric binaries at some reference frequency are mapped to their spin tilts at formation. The inclusion of eccentricity in this evolution is even important for cases where the eccentricity at detection is well below our current ability to measure it.

Next-generation gravitational-wave observatories will detect thousands of compact binary coalescences daily, with some signals lasting from minutes to hours. Consequently, multiple signals will overlap in the time-frequency plane and form a persistent 'foreground noise', particularly affecting the low-frequency range where binary neutron star inspirals evolve gradually. This talk delves into how foreground noise impacts parameter estimation for short-duration binary black hole signals, especially those with high detector-frame masses and/or at large cosmological distances. Our study reveals that detection sensitivity can decrease by up to 25% when the noise power spectrum deviates by 50% from Gaussian due to foreground contamination. Despite this, even without subtracting overlapping signals, the influence on parameter estimation remains minimal, primarily impacting measurement precision. These findings suggest that robust recovery of system parameters is achievable without global-fit techniques or signal subtraction, even amid substantial foreground noise, with only a slight reduction in precision.

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

Convolutional neural networks (CNNs) and other deep learning architectures are often considered "black boxy"; it's difficult to tell why they are making decisions. "Explainability" seeks to crack open the black box and help humans understand and interpret machine learning algorithms. I'll introduce the algorithms GWSkyNet and GWSkyNet-Multi and report how the GWSkyNet team, in collaboration with computer scientists, neuroscientists, cosmologists, and data scientists (the ML-ESTEEM network), interrogated the good performance of these algorithms to develop a framework for probing explainability. I'll close with suggested best practices for developing and testing explainable deep learning algorithms that human researchers can more easily trust and interpret.

In my talk I will present how AI aids in alert monitoring, decision-making, image analysis, and parameter extraction of multi-messenger sources. Multi-messenger studies, using for example gravitational waves, require real-time observations that mobilize numerous ground-based telescopes at any given moment after alerts are produced. To meet these demands, the scientific community has long used an on-call/shift system that redirects observations toward follow-up of candidate sources, for example in optical. In collaborations like GRANDMA, researchers are trained to respond quickly, but this requires complex logistics that evolves as soon as monitoring tools evolves. In my talk, I will present the use of AI to support real-time decision-making in these observation campaigns, to assist “shifters” by guiding observation strategies, providing responsive documentation, and streamlining the alert system. Secondly, AI also aids in online source characterization to identify potential counterparts of multi-messenger sources, that can be critical with the upcoming Vera Rubin that will generate vast volumes of alerts. On telescope front, AI enhances image analysis by detecting new sources and adjusting calibrations between current and reference images and aids in measuring magnitudes accurately. Finally, AI further facilitates the extraction of physical parameters of the multi-messenger events using Bayesian approaches and model grids, selecting optimal fits regarding the data. At the end of my talk, I will also adress on a more personnal view, the limits of AI and risks that we can balance on these various topics.

A new method for searching for transient long gravitational wave signals in interferometric detector data will be presented. These signals are expected from rapidly rotating newborn magnetars, and they can last from minutes to hours and change rapidly in frequency (up to 1 Hz in 1 s). For this reason, the matched filter techniques - used for standard semi-periodic persistent signals - are either not usable or are computationally extremely demanding. We explored a different approach through machine learning paradigms, with the goal of achieving a fast and computationally sustainable solution. Results obtained with different network models will be presented, along with prospects for application on real LVK data.

Gravitational wave event validation currently relies on humans to cross-check Data Quality Report tasks, determining whether data quality issues are associated with an event and identifying the affected time-frequency regions. This process is time-consuming and results are often difficult to reproduce. We extend the glitch-gravitational wave classifier: Gravity Spy Convolutional Neural Network Decision Tree (GSpyNetTree), by developing GSpyNetTreeS, which incorporates YOLO-based machine learning segmentation. GSpyNetTreeS not only identifies and classifies multiple glitches and gravitational waves in detector data simultaneously, but also locates them in frequency and time within seconds. This is our first step toward establishing an automated machine learning framework for noise removal and subtraction.

The spins of merging black holes are key information to reveal their formation history. To accurately infer spin distribution from gravitational waves, we revisited the validity of approximations employed in the current analysis framework and investigated their impact on recovered spin distribution. Especially, we have found that approximating spin prior distribution with kernel density estimation introduces systematic errors in recovered spin distribution and it can be solved by analytically calculating spin prior distribution. We have also investigated the effects of spin prior distribution assumed in parameter estimation of each event. In this talk, I will share the results of those studies.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

Enjoy beautiful Banff on your own!

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

With the fourth observing run of the International Gravitational-Wave Network (O4) underway, the search for gravitational-wave counterparts, including gamma-ray bursts and kilonovae, continues. Observations of gravitational waves and their counterparts, like with GW170817 and AT 2017gfo, are crucial for our understanding of the neutron star equation of state and of r-process nucleosynthesis expected to take place in the ejecta of compact binary mergers with at least one neutron star. Here, we present a summary of open public alerts in the first half of O4, the current data products used to classify compact binary mergers, and those under current development. These include, amongst others, exciting proposed data products designed to maximize multi-messenger follow-up opportunities, such as estimates of the likelihood that a candidate will produce a kilonova, estimates of the mass ejecta and light curves produced by such a kilonova, as well as estimates for the binary viewing angle.

Because of the electromagnetic (EM) radiation produced during the merger, compact binary coalescences with neutron stars may result in multi-messenger observations. To follow up on the gravitational-wave (GW) signal with EM telescopes, it is critical to promptly identify the properties of these sources. This identification must rely on the properties of the progenitor source, such as the component masses and spins, as determined by low-latency detection pipelines in real-time. The output of these pipelines, however, might be biased, which could decrease the accuracy of parameter recovery. In this work, we revisit this problem and discuss two new implementations of supervised machine learning algorithms, K-nearest neighbors and random forest, which can predict the presence of a neutron star and post-merger matter remnant in low-latency searches. Additionally, we present a novel approach for calculating the Bayesian probabilities for these two metrics. Our scheme is designed to provide the astronomy community with well-defined probabilities. This would deliver a more direct and easily interpretable product to assist EM telescopes in deciding whether to follow up on GW events in real-time.

Low-latency pipelines analyzing gravitational waves from Compact Binary Coalescence (CBC) events rely on matched filtering techniques, where the best matching template that yields the highest signal-to-noise ratio (SNR) is identified as the gravitational wave (GW) candidate. However, limitations in the template bank, waveform modeling, and non-stationary detector noise often cause discrepancies between the template parameters and the true event parameters, particularly for events with higher chirp masses. This work aims to quantify the extent of these discrepancies across the parameter space using machine learning. We present a Quantile Regression Neural Network (QRNN) model that provides dynamic confidence bounds on key parameters such as chirp mass, mass ratio, and total mass, using the best matching template parameters as inputs. The model demonstrated over 95% accuracy on the testing set and performed similarly during the recent LVK Mock Data Challenge (MDC) when input parameters were within the training range. Additionally, incorporating these bounds as priors for online parameter estimation (PE) in 100 MDC events resulted in similar skymap statistics while reducing the number of likelihood iterations by over 10%, directly decreasing the overall time required for PE runs.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

I will highlight a sparse dictionary learning approach, as a novel tool for reconstruction of merger waveforms in the presence of Galactic confusion noise [1], rapid detection of GWs from BBHs [2], and reconstruction of longf-duration GWs from extreme mass ratio inspirals[3]. references: [1] Dictionary Learning: A Novel Approach to Detecting Binary Black Holes in the Presence of Galactic Noise with LISA, Phys.Rev.Lett. 130 (2023) 9, 091401, e-Print: 2210.06194 [gr-qc]; [2] Rapid detection of gravitational waves from binary black hole mergers using sparse dictionary learning, e-Print: 2405.17721 [gr-qc]; [3] High-speed reconstruction of long-duration gravitational waves from extreme-mass-ratio inspirals using sparse dictionary learning, Phys.Rev.D 110 (2024) 6, 064074.

Non-gaussian, transient bursts of noise in gravitational wave (GW) interferometers, also known as glitches, hinder the detection and parameter estimation of short- and long-lived GW signals in the main detector strain. Glitches, come in a wide range of frequency-amplitude-time morphologies and may be caused by environmen- tal or instrumental processes, so a key step towards their mitigation is to understand their population. Current approaches for their identification use supervised models to learn their morphology in the main strain with a fixed set of classes, but do not consider relevant information provided by auxiliary channels that monitor the state of the interferometers. In this work, we present an unsupervised algorithm to find anomalous glitches. Firstly, we encode a subset of auxiliary channels from LIGO Livingston in the fractal dimension, which mea- sures the complexity of the signal. For this aim, we speed up the fractal dimension calculation to near-real time. Secondly, we learn the underlying distribution of the data using an autoencoder with cyclic periodic convolutions. In this way, we learn the underlying distribution of glitches and we uncover unknown glitch morphologies, and overlaps in time between different glitches and misclassifications. This led to the discovery of 6.6% anomalies in the input data. The results of this investigation stress the learnable structure of auxiliary channels encoded in fractal dimension and provide a flexible framework for glitch discovery.

The success of gravitational wave astronomy hinges on precise data quality assessment and the meticulous validation of detected events. This presentation highlights the critical role of these processes within the ongoing O4 observational campaign by the LIGO, Virgo, and KAGRA collaborations. We begin by introducing detector data and the concept of data quality. Next, we examine how common data-quality issues impact the detection of astrophysical signals, affecting both their significance and the reliability of astrophysical parameter estimates. We then describe the statistical methods used to identify and mitigate these issues, followed by an overview of the event validation framework employed in O4 to confirm the astrophysical origins of candidate signals. Finally, we discuss how advances in signal processing and artificial intelligence are poised to enhance these procedures in future observational campaigns.

Machine-learning applications within the gravitational-wave community have exploded in recent years. Many of these works have tackled big problems in the field, ranging from detection, to glitch classification, to near-instantaneous parameter estimation. Here, we instead motivate the application of ML to small problems where effective modeling is necessary, but a detailed understanding is not. In these scenarios, machine learning is a powerful tool that can enhance our understanding of local measurements. In particular, we examine two distinct applications of machine learning to detection pipeline outputs. First, we show that simple neural networks can accurately interpolate across the gravitational-wave signal space used by search pipelines, facilitating local signal-to-noise ratio maximization. Second, we show that the search response encodes the nature of observed transients, and that convolutional neural networks can accurately classify signals and noise in this parameterization.

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

Various locations

5-day workshop participants are welcome to use BIRS facilities (TCPL ) until 3 pm on Friday, although participants are still required to checkout of the guest rooms by 11AM.