Authors

Abstract

We present a deep-learning artificial intelligence model (AI) that is capable of learning and forecastingthe late-inspiral, merger and ringdown of numerical relativity waveforms that describe quasicircular,spinning, nonprecessing binary black hole mergers. We used theNRHybSur3dq8 surrogate model toproduce train, validation and test sets of l jmj 2 waveforms that cover the parameter spaceof binary black hole mergers with mass ratios q 8 and individual spins jszf1;2gj 0.8. These waveformscover the time range t 5000 M; 130 M, where t 0M marks the merger event, defined as themaximum value of the waveform amplitude. We harnessed the ThetaGPU supercomputer at theArgonne Leadership Computing Facility to train our AI model using a training set of 1.5 millionwaveforms. We used 16 NVIDIA DGX A100 nodes, each consisting of 8 NVIDIA A100 Tensor CoreGPUs and 2 AMD Rome CPUs, to fully train our model within 3.5 h. Our findings show that artificialintelligence can accurately forecast the dynamical evolution of numerical relativity waveforms in thetime range t 100 M; 130 M. Sampling a test set of 190,000 waveforms, we find that the averageoverlap between target and predicted waveforms is 99% over the entire parameter space underconsideration. We also combined scientific visualization and accelerated computing to identify whatcomponents of our model take in knowledge from the early and late-time waveform evolution toaccurately forecast the latter part of numerical relativity waveforms. This work aims to accelerate thecreation of scalable, computationally efficient and interpretable artificial intelligence models forgravitational wave astrophysics