Weights for neural network surrogate model of NRSur7dq4

Trained neural-network weights and reference implementation for jax-nrsur7dq4-nn, a neural-network surrogate model that emulates the NRSur7dq4 gravitational-waveform model for precessing binary black hole mergers.

The surrogate decomposes the waveform into its constituent dynamical pieces (orbital frequency, co-precessing quaternion, spin trajectories, and co-orbital frame modes) and models each with an independent multilayer-perceptron network. A parallelized model bank reassembles these into inertial-frame strain polarizations using the same rotation and mode-summation pipeline as NRSur7dq4. The model covers the NRSur7dq4 calibration region: mass ratios 1 ≤ q ≤ 4 and dimensionless spin magnitudes |χ₁|, |χ₂| ≤ 0.8.

The full waveform-to-likelihood pipeline is implemented in JAX; it is differentiable and GPU-accelerated, enabling gradient-based and GPU-accelerated inference including Fisher information matrices, nested sampling, gradient-based MCMC, and importance sampling.

This record contains the trained weights distributed as an HDF5 file. For the model construction, accuracy validation, and performance benchmarks, see the accompanying paper: Fast, accurate, and differentiable: a neural-network surrogate for NRSur7dq4 precessing binary black hole waveforms [arXiv:2607.24960]. Please cite that paper when using these weights.