I will discuss the use of automatic differentiation through the programming framework JAX for accelerating a variety of analysis tasks throughout gravitational wave (GW) science. Firstly, I will demonstrate that complete waveforms which cover the inspiral, merger, and ringdown of binary black holes (i.e. IMRPhenomD) can be written in JAX and demonstrate that the serial evaluation speed of the waveform (and its derivative) is similar to the lalsuite implementation in C. I will then focus on three applications where efficient and differentiable waveforms are essential. Firstly, I will demonstrate how gradient descent can be used to optimize the ∼ 200 coefficients that are used to calibrate the waveform model. Secondly, I will show that Fisher forecasting calculations can be sped up by ∼ 100× (on a CPU) with no loss in accuracy. This increased speed makes population forecasting substantially simpler. Finally, I will show that by combining likelihood heterodyning, automatically-differentiable and accelerator-compatible waveforms, and gradient-based Markov chain Monte Carlo (MCMC) sampling enhanced by normalizing flows, we achieve full Bayesian parameter estimation for real events like GW150914 and GW170817 within a minute of sampling time.
Name of software library or topic
ripple, jim, and flowMC
URL to software repository
https://github.com/tedwards2412/ripple https://github.com/kazewong/jim https://github.com/kazewong/flowMC
Name of person who will give the talk
Thomas Edwards
Brief abstract
I will discuss the use of automatic differentiation through the programming framework JAX for accelerating a variety of analysis tasks throughout gravitational wave (GW) science. Firstly, I will demonstrate that complete waveforms which cover the inspiral, merger, and ringdown of binary black holes (i.e. IMRPhenomD) can be written in JAX and demonstrate that the serial evaluation speed of the waveform (and its derivative) is similar to the lalsuite implementation in C. I will then focus on three applications where efficient and differentiable waveforms are essential. Firstly, I will demonstrate how gradient descent can be used to optimize the ∼ 200 coefficients that are used to calibrate the waveform model. Secondly, I will show that Fisher forecasting calculations can be sped up by ∼ 100× (on a CPU) with no loss in accuracy. This increased speed makes population forecasting substantially simpler. Finally, I will show that by combining likelihood heterodyning, automatically-differentiable and accelerator-compatible waveforms, and gradient-based Markov chain Monte Carlo (MCMC) sampling enhanced by normalizing flows, we achieve full Bayesian parameter estimation for real events like GW150914 and GW170817 within a minute of sampling time.