A new framework for lightning-fast gravitational wave analysis of pulsar timing data
Abstract
Pulsar timing array data analysis is computationally expensive, limiting the complexity of models which can be studied. As pulsar timing datasets and their respective models grow in size and sophistication, faster and scalable inference methods are essential. In this paper, we accelerate pulsar timing analyses by sampling in the space of Fourier coefficients instead of analytically marginalizing over them. Previous studies have shown the Fourier space induces a complex, high-dimensional posterior geometry, from which it is generally difficult to sample. We show that under an appropriate coordinate transformation the Fourier coefficients approximately follow a standard normal distribution, and may be efficiently sampled using a Hamiltonian Monte Carlo scheme. Under this coordinate transformation, for datasets of size and complexity comparable to the NANOGrav 15-year release, the new method produces converged posterior distributions for a range of models which include inter-pulsar correlations, stochastic, and deterministic signals in approximately 15 minutes on an NVIDIA GeForce RTX 3090 GPU. By comparison, the legacy pulsar timing analysis software \texttt{ENTERPRISE} would require months of computation on a CPU cluster to analyze comparable datasets under the same joint stochastic and deterministic models.
Versions
Related
- Optimal Strategies for Gravitational-Wave Memory Detection with Noise Marginalization in Pulsar Timing Arrays
- Accurate pulsar timing array likelihoods: better low-rank covariance for gravitational wave background
- Gravitational-Wave Sky Mapping with Pulsar Timing Arrays: The Full Earth-Pulsar Response and Resolution Operational Regimes
- Full-Stokes response to a stochastic gravitational-wave background in pulsar timing arrays: Fully analytic overlap reduction functions at finite distance and coherence of close pulsar pairs
- Prior volume effect inflates the gravitational-wave background in pulsar timing arrays
BibTeX
@misc{gundersen2026framework,
title = {{A new framework for lightning-fast gravitational wave analysis of pulsar timing data}},
author = {Aiden Gundersen and Rutger van Haasteren and Neil J. Cornish and Michele Vallisneri and Patrick M. Meyers},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2607.06834},
eprint = {2607.06834},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.06834},
}
RIS
TY - UNPB
TI - A new framework for lightning-fast gravitational wave analysis of pulsar timing data
AU - Aiden Gundersen
AU - Rutger van Haasteren
AU - Neil J. Cornish
AU - Michele Vallisneri
AU - Patrick M. Meyers
PY - 2026
DA - 2026/10/30
PB - arXiv
DO - 10.48550/arxiv.2607.06834
UR - https://arxiv.org/abs/2607.06834
AB - Pulsar timing array data analysis is computationally expensive, limiting the complexity of models which can be studied. As pulsar timing datasets and their respective models grow in size and sophistication, faster and scalable inference methods are essential. In this paper, we accelerate pulsar timing analyses by sampling in the space of Fourier coefficients instead of analytically marginalizing over them. Previous studies have shown the Fourier space induces a complex, high-dimensional posterior geometry, from which it is generally difficult to sample. We show that under an appropriate coordinate transformation the Fourier coefficients approximately follow a standard normal distribution, and may be efficiently sampled using a Hamiltonian Monte Carlo scheme. Under this coordinate transformation, for datasets of size and complexity comparable to the NANOGrav 15-year release, the new method produces converged posterior distributions for a range of models which include inter-pulsar correlations, stochastic, and deterministic signals in approximately 15 minutes on an NVIDIA GeForce RTX 3090 GPU. By comparison, the legacy pulsar timing analysis software \texttt{ENTERPRISE} would require months of computation on a CPU cluster to analyze comparable datasets under the same joint stochastic and deterministic models.
ER -