Session G14: Gravitational Waves: Data Analysis Techniques and Parameter Estimation - II

Classifier for gravitational-wave inspiral signals in nonideal single-detector data

Gravitational waves from merging neutron star (NS) black hole (BH) binaries are a high-profile target of LIGO-Virgo searches. However, the sensitivity of templated searches for such signals is limited by "glitches": non-Gaussian noise artefacts in real detector data. These searches typically employ a ranking statistic, such as re-weighted SNR, built from the matched-filter signal-to-noise ratio (SNR) and chi-squared test computed at the time of occurrence of the individual candidate events. We propose instead a Random-Forest multivariate classifier that exploits features from times surrounding each candidate event, as well as properties constructed from two independent searches of the same data: one involving a bank of inspiral templates, and the other involving sine-Gaussian templates. Additional information thus provided to the classifier should significantly enhance its ability to discriminate signals from glitches. Indeed, when evaluated on realistic Advanced LIGO data injected with synthetic NSBH signals, we find that the new classifier detects 1.5-2 times more signals at low false positive rates than re-weighted SNR, without the need to compute the chi-squared test.   

Optimizing SEOBNRv3 for LIGO Parameter Estimation

The Spinning Effective One Body--Numerical Relativity (SEOBNR) series of gravitational wave approximants are among the best available for Advanced LIGO data analysis. Unfortunately, SEOBNR codes as they currently exist within LALSuite are typically too slow to be directly used for standard Markov-Chain Monte Carlo-based parameter estimation (PE). Reduced-Order Models (ROMs) of SEOBNR have been developed for aligned-spin SEOBNR approximants, but there is no known way to make efficient ROMs of the full eight-dimensional parameter space, modeled by SEOBNR version 3 (SEOBNRv3). We therefore focus on direct optimization of the SEOBNRv3 code, building on our experience optimizing the aligned-spin SEOBNRv2 code. We report on a host of new optimizations to SEOBNRv3, including a form of “guided automatic differentiation” and optimized interpolation routines, which together speed up SEOBNRv3 by two orders of magnitude. This brings direct SEOBNRv3-based PE within reach for many important LIGO sources. We also report on future plans for optimization, which may improve SEOBNRv3 performance by at least another order of magnitude.   

Comparing parameter estimation results using model and numerical relativity templates of binary black hole systems

The use of both state-of-the-art models and numerical relativity in unison allows for the most confident and comprehensive parameter estimation results. It is important to be able to use and compare the best models and numerical relativity simulations in a quick and comprehensive way. In this talk, I present both using our highly-parallelized parameter estimation code that allows for timely and accurate results. This extends to models and simulations that have higher modes and precessing spins.   

Deep Learning for Real-time Gravitational Wave Detection and Parameter Estimation: Results with Advanced LIGO Data

The recent detections of gravitational waves from merging black holes and the subsequent detection of the collision of two neutron stars in coincidence with electromagnetic observations have inaugurated a new era of multimessenger astrophysics. To enhance the scope of this emergent science, we proposed the use of deep learning with convolutional neural networks for detecting and characterizing gravitational wave signals in real-time. Here we present the first application of deep learning using continuous real data streams from multiple LIGO detectors for both detection and parameter estimation of gravitational waves from binary black hole mergers. We show for the first time that machine learning can detect and estimate the true parameters of real GW events. Our comparisons show that deep learning is far more computationally efficient than matched-filtering, while retaining similar accuracy, allowing real-time processing of weak time-series signals in non-stationary non-Gaussian noise, with minimal resources, and also enables the detection of new classes of gravitational wave sources that may go unnoticed with existing algorithms. This approach is uniquely suited to enable coincident detection campaigns of gravitational waves and their multimessenger counterparts in real-time.   

GWSurrogate: An easy-to-use interface to gravitational wave surrogate models

Recently there has been significant interest in building data-driven gravitational wave models directly from numerically generated data. These surrogate (or reduced-order) models can faithfully reproduce a parameterized gravitational wave model specified through computationally expensive ordinary or partial differential equations with significant speedups. Surrogates can be used, for example, to accelerate the generation of effective one body or numerical relativity (NR) waveform models thereby reducing the overall runtime of a multi-query data analysis study. For surrogates to be useful, it is necessary that they be publicly available, easy-to-use, and decoupled from the building codes which produce them. In this talk, I will describe a light-weight open-source code, GWSurrogate, which aims to address this issue. I will also briefly summarize the most recently built NR surrogate waveform models including those which describe precession.   

Swarm Intelligence to detect Gravitational Waves.

We investigate the use particle swarm optimization (PSO) algorithm in the detection of gravitational waves from compact binary coalescences. The main advantage of PSO is that it can be fast without hampering the effectiveness of the search. This property can be used to reduce the computational burden and speed up the detection process. An early detection helps electromagnetic follow-up of gravitational wave triggers. A PSO based detection search is a continuous search in parameter space and thus overcomes the hurdles in generation of template banks. In future, this can help speed up parameter estimation as well. We present results for coincident and coherent searches for CBCs with PSO algorithm.   

Applying the TwoSpect X-Statistic to Searches for Continuous Gravitational Waves

Non-axisymmetric spinning neutron stars are expected to emit near-monochromatic gravitational waves, which will be frequency modulated if the star is part of a binary system. This modulation requires a large expansion of the search parameter space for templated search methods, and brings with it greatly increased computational cost. Here we discuss a technique, developed for use with a pre-existing templated search method called TwoSpect, which samples parameter space more sparsely than the original method in order to achieve an intermediate balance of sensitivity and computational cost. We also discuss prospects for, and progress toward, the use of this technique in directed searches for high-latitude unassociated Fermi-LAT sources.   

Efficient Estimation of Barycentered Relative Time Delays for Distant Gravitational Wave Sources

Accurate determination of gravitational wave source parameters relies on transforming between the source and detector frames. All-sky searches for continuous wave sources are computationally expensive, in part, because of barycentering transformation of time delays to a solar system frame. We investigate approximations for determining time delays of signals received by a gravitational wave detector with respect to the solar system barycenter. A highly non-linear conventional computation is transformed into one that has a pure linear sum in its innermost loop.   

Highly-Spinning Data for Puncture-Based Codes

We describe the challenges and successes associated with our recent extension of HiSpID data to generate initial data for highly spinning binaries with unequal masses and misaligned spins. In the standard puncture approach to initial data, the linear momentum and spins of the two black holes are specified precisely, while the masses are implicitly specified using bare mass parameters (which can differ significantly from the actual measured black hole masses). In our approach, the black hole masses match the mass parameters to a high degree of precision, as do the spins and spin parameters, however, the linear momentum of each hole is an implicit function of boost-like parameters. To get the desired orbital angular momentum, we use an iterative procedure to fix the ADM linear and angular momentum of the spacetime to the desired values. We find that our approach yields binaries with both high spins and low eccentricity.