Fast Parameter Inference on Pulsar Timing Arrays with Normalizing Flows.

Area of Science:

• Astronomy and Astrophysics
• Computational Physics
• Data Science

Background:

• Pulsar timing arrays (PTAs) are crucial for detecting stochastic gravitational wave backgrounds (SGWB).
• Current Bayesian inference methods, like Markov chain Monte Carlo (MCMC), are computationally intensive, requiring days to weeks for analysis.
• The high dimensionality of the search space in MCMC leads to significant computational bottlenecks, primarily due to costly likelihood evaluations.

Purpose of the Study:

• To develop a faster and more accurate method for estimating SGWB posteriors using PTAs.
• To leverage simulation-based inference techniques to overcome the computational limitations of traditional MCMC methods.
• To demonstrate the efficacy of conditional normalizing flows in accelerating PTA data analysis.

Main Methods:

• Utilized conditional normalizing flows, a type of simulation-based inference, trained on simulated PTA data.
• Applied these flows to estimate posterior distributions for the SGWB.
• Compared the computational time and accuracy against traditional MCMC methods.

Main Results:

• Conditional normalizing flows achieved extremely fast and accurate estimation of SGWB posteriors.
• The sampling time was reduced from days or weeks to mere seconds.
• This represents a significant acceleration in the analysis of PTA data for gravitational wave detection.

Conclusions:

• Conditional normalizing flows offer a powerful and efficient alternative to MCMC for PTA data analysis.
• This approach dramatically speeds up the process of searching for the stochastic gravitational wave background.
• The findings pave the way for more rapid discoveries and characterizations of gravitational wave sources.