Conditional Autoencoder for Generating Binary Neutron Star Waveforms with Tidal and Precession Effects

TL;DR

This work tackles the computational bottleneck of generating accurate gravitational-wave waveforms for binary neutron star mergers that exhibit spin precession and tidal effects. It introduces a conditional autoencoder (cAE) that learns amplitude $A(t)$ and phase $\Phi(t)$ representations conditioned on system parameters $\Theta$, achieving high fidelity with a mismatch around $2.13\times10^{-3}$ while delivering substantial speedups (roughly 5x for single waveforms and ~10x for batches) compared with traditional models. The model leverages a dual-branch architecture, phase- and amplitude-specific encoders, and a latent space alignment constraint to enable real-time parameter estimation and rapid GW searches, validated against IMRPhenomXP_NRTidalv2. Latent-space analyses reveal physically meaningful structures corresponding to tidal effects, mass, and spin variations, supporting interpretable generative behavior. The approach promises significant practical impact for low-latency GW detection and large-scale multi-messenger analyses, with future directions including diffusion-based extensions and inference-optimization techniques to push latency and scalability further.

Abstract

Gravitational waves from binary neutron star mergers provide critical insights into dense matter physics and strong-field gravity, yet accurate waveform modeling remains computationally intensive. We present a deep generative model for gravitational waveforms from binary neutron star mergers that captures the late inspiral, merger, and ringdown phases while incorporating spin precession and tidal effects. Using a conditional autoencoder architecture, the model efficiently produces high-fidelity waveforms across a broad parameter space, including component masses (m1, m2), spin components (S1x, S1y, S1z, S2x, S2y, S2z), and tidal deformabilities (Lambda1, Lambda2). Trained on 1*10^6 waveforms generated by the IMRPhenomXP_NRTidalv2 model, our network achieves a mean mismatch of 2.13*10^-3. The generation time for a single waveform is 0.12 s, compared to 0.66 s for IMRPhenomXP_NRTidalv2, representing a speedup of about fivefold. When generating 1000 waveforms, the model completes the task in 0.75 s, roughly ten times faster than the baseline. This significant acceleration facilitates rapid parameter estimation and real-time gravitational-wave searches. With improved precision and efficiency, the model can support low-latency detection and broader applications in multi-messenger astrophysics.

Figures (19)

• Figure 1: Input samples in time domain. Top: amplitude curve $A(t)$; middle: phase curve $\Phi(t)$; bottom: waveform strain $h(t)$.
• Figure 2: (a) Structure of an AE. (b) Structure of a variational VAE.
• Figure 3: Training and testing architecture of our model, along with detailed structures of individual components, the structure of the Transformer and ResNet block are shown in Appendix \ref{['sec:appendix']} (FIG \ref{['fig:resnet_transformer']}).
• Figure 4: Training and validation loss curves for the amplitude and phase models.
• Figure 5: Mismatch distributions of $h_+$ and $h_\times$ for both the cAE and Residual-MLP-CNN models over test set.