[Paper Review] An efficient GPU-accelerated multi-source global fit pipeline for LISA data analysis

Michael L. Katz, Nikolaos Karnesis+3 more|arXiv (Cornell University)|May 7, 2024

Pulsars and Gravitational Waves Research4 citations

TL;DR

This paper presents Erebor, a GPU-accelerated, fully automated global fit pipeline for LISA data analysis that simultaneously models Massive Black Hole Binaries (MBHBs), Galactic Binaries (GBs), and instrumental noise. It achieves high-precision recovery of 15 injected MBHBs and catalogs ~12,000 GBs with high confidence, using reversible-jump MCMC, ensemble sampling, and real-time RJMCMC refitting on GPU-accelerated infrastructure.

ABSTRACT

The large-scale analysis task of deciphering gravitational wave signals in the LISA data stream will be difficult, requiring a large amount of computational resources and extensive development of computational methods. Its high dimensionality, multiple model types, and complicated noise profile require a global fit to all parameters and input models simultaneously. In this work, we detail our global fit algorithm, called ``Erebor,'' designed to accomplish this challenging task. It is capable of analysing current state-of-the-art datasets and then growing into the future as more pieces of the pipeline are completed and added. We describe our pipeline strategy, the algorithmic setup, and the results from our analysis of the LDC2A Sangria dataset, which contains Massive Black Hole Binaries, compact Galactic Binaries, and a parameterized noise spectrum whose parameters are unknown to the user. The Erebor algorithm includes three unique and very useful contributions: GPU acceleration for enhanced computational efficiency; ensemble MCMC sampling with multiple MCMC walkers per temperature for better mixing and parallelized sample creation; and special online updates to reversible-jump (or trans-dimensional) sampling distributions to ensure sampler mixing and accurate initial estimates for detectable sources in the data. We recover posterior distributions for all 15 (6) of the injected MBHBs in the LDC2A training (hidden) dataset. We catalog $\sim12000$ Galactic Binaries ($\sim8000$ as high confidence detections) for both the training and hidden datasets. All of the sources and their posterior distributions are provided in publicly available catalogs.

Motivation & Objective

• To develop a fully automated, scalable global fit pipeline for LISA data that handles multiple source types and complex noise simultaneously.
• To address the computational challenge of analyzing high-dimensional, multi-source LISA datasets using GPU acceleration and advanced sampling techniques.
• To produce accurate posterior distributions for MBHBs, GBs, and noise parameters in both training and hidden LDC2A datasets.
• To enable future extension to realistic orbital dynamics, non-stationary noise, and advanced source models through modular, extensible design.
• To release open-source code and public catalogs to advance community-wide LISA data analysis and benchmarking.

Proposed method

• The pipeline uses a global fit framework based on Reversible-Jump Markov Chain Monte Carlo (RJMCMC) to infer the uncertain number of Galactic Binaries (GBs) and their parameters.
• GPU-accelerated sampling operations are implemented using CuPy and custom kernels to significantly speed up likelihood evaluations and MCMC steps.
• Ensemble sampling and tempering are employed to improve mixing and convergence across high-dimensional parameter spaces during global fitting.
• Online refitting of RJMCMC proposal distributions is performed via single-source MCMC runs on residuals, enhancing sampling efficiency and accuracy.
• The pipeline integrates modular components for MBHBs, GBs, and noise power spectral density (PSD), communicating via a shared residual model.
• A parameterized noise model includes both instrumental noise and a confusion foreground from unresolved GBs, fitted jointly with astrophysical sources.

Experimental results

Research questions

• RQ1 Can a fully automated, GPU-accelerated global fit pipeline successfully recover known MBHB and GB signals in the LDC2A Sangria dataset without human intervention?
• RQ2 How effective is ensemble sampling and online RJMCMC refitting in improving mixing and convergence for high-dimensional, multi-source LISA data analysis?
• RQ3 To what extent can GPU acceleration reduce computational cost and energy usage compared to CPU-based pipelines in LISA data fitting?
• RQ4 How well does the pipeline perform on the hidden LDC2A dataset, where source parameters are unknown, and can it produce reliable posterior distributions?
• RQ5 What are the limitations of current template models in capturing complex GB populations, and how do they affect RJMCMC sampling efficiency?

Key findings

• The pipeline successfully recovered all 15 injected MBHBs in the LDC2A training dataset with no false alarms, achieving high-precision posterior estimation.
• The Galaxy sampler detected approximately 12,000 GBs across both training and hidden datasets, with over 8,000 identified as high-confidence detections.
• The pipeline achieved a match rate of >90% between detected GBs and the input population, indicating strong fidelity in source characterization.
• GPU acceleration reduced computational cost and energy usage compared to CPU-based alternatives, with performance expected to scale with future hardware.
• Online refitting of RJMCMC proposals significantly improved sampling efficiency and convergence, especially in high-dimensional GB parameter spaces.
• The noise PSD and confusion foreground were accurately fitted, though the assumption of stationary noise led to potential biases in local sensitivity near MBHB mergers.