What type of study is this?

This is a In Vitro Study study.

August 25, 2025

Deep learning powered numerical relativity surrogate for binary black hole waveforms

Key Points

Fine-tuned neural network models achieve median mismatches of order 10^{-5}, matching top-performing models.
Initial training used four different gravitational-wave approximants before fine-tuning on numerical relativity data.
These surrogate models generate millions of waveforms in under 20 ms using GPU acceleration.
Implemented in the bilby framework, the models aid in parameter estimation tasks for gravitational-wave analysis.

Abstract

Gravitational-wave (GW) approximants are essential for gravitational-wave astronomy, allowing the coverage of the binary black hole parameter space for inference or match filtering without costly numerical relativity (NR) simulations but generally trading some accuracy for computational efficiency. To reduce this trade-off, NR surrogate models can be constructed using interpolation within NR waveform space. We present a two-stage training approach for neural network-based NR surrogate models. We initially train four models on waveforms generated from four different GW approximants and then fine-tune these models on NR data. We show that despite the median mismatches of the pretrained models with NR ranging over two orders of magnitude, the fine-tuned models all reach median mismatches of order 10^-5, on par with top-performing NR surrogates. The dual-stage artificial neural surrogate (DANSur3dq8) models also offer rapid waveform generation, with millions of waveforms being generated in under 20 ms on a GPU. Implemented in the bilby framework, we show DANSur3dq8 can be used for parameter estimation tasks.

Mark Helpful

Bookmark

Relay