Modeling Dependence Structures in Astronomical Multi-Band Time Series Data via Multi-Output Gaussian Processes
Abstract
Modern astronomical time-domain surveys routinely collect multi-band light curves that provide complementary information about the physical processes governing source variability. Although various multi-output Gaussian process (GP) models have been applied in different astronomical contexts, the statistical and astrophysical implications of their dependence structures have received comparatively limited attention. In this work, we examine three existing approaches to modeling cross-band dependence within a common multi-output GP framework and investigate how their underlying assumptions affect statistical inference and astrophysical interpretation. These include covariance-based, stochastic differential equation (SDE)-based, and latent-process approaches, which differ mainly in how cross-band dependence is introduced. The covariance-based approach directly specifies cross-band covariance functions to characterize dependence, whereas the SDE-based approach specifies joint temporal dynamics, with cross-band dependence arising from coupled dynamics or correlated driving processes. The latent-process approach represents dependence through shared stochastic processes and their transformations, which can reflect physical relationships among multiple light curves. We demonstrate the practical implications of these modeling choices through applications to multi-band active galactic nucleus light curves and emission-line reverberation mapping. Rather than advocating a universally preferred approach, this work provides guidance for selecting dependence structures according to scientific objectives and clarifies how these choices influence the statistical characterization and scientific interpretation of multi-band stochastic variability.
Related
- Improving the Physical Interpretability of Gaussian Processes in Stellar Activity Modeling: A Study Case on Photometric Variability among Stellar Clusters
- Multivariate Continuous-Time Autoregressive Moving Average Processes for Astronomical Multiband Time Series
- Nuclear-physics-guided Gaussian Processes
- Characterizing Binary Black Hole Subpopulations in GWTC-4 with Binned Gaussian Processes: On the Origins of the 35 M ⊙ Peak
- Gaussian processes on ray-guided transformed uniform grids for fast, flexible, and auto-differentiable adaptive source reconstruction in lens modelling
BibTeX
@misc{das2026modeling,
title = {{Modeling Dependence Structures in Astronomical Multi-Band Time Series Data via Multi-Output Gaussian Processes}},
author = {Samata Das and Lishan Shi and Yasaman Hamayouni and Hyungsuk Tak and Jong-Hak Woo},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2607.21431},
eprint = {2607.21431},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.21431},
}
RIS
TY - UNPB
TI - Modeling Dependence Structures in Astronomical Multi-Band Time Series Data via Multi-Output Gaussian Processes
AU - Samata Das
AU - Lishan Shi
AU - Yasaman Hamayouni
AU - Hyungsuk Tak
AU - Jong-Hak Woo
PY - 2026
DA - 2026/10/27
PB - arXiv
DO - 10.48550/arxiv.2607.21431
UR - https://arxiv.org/abs/2607.21431
AB - Modern astronomical time-domain surveys routinely collect multi-band light curves that provide complementary information about the physical processes governing source variability. Although various multi-output Gaussian process (GP) models have been applied in different astronomical contexts, the statistical and astrophysical implications of their dependence structures have received comparatively limited attention. In this work, we examine three existing approaches to modeling cross-band dependence within a common multi-output GP framework and investigate how their underlying assumptions affect statistical inference and astrophysical interpretation. These include covariance-based, stochastic differential equation (SDE)-based, and latent-process approaches, which differ mainly in how cross-band dependence is introduced. The covariance-based approach directly specifies cross-band covariance functions to characterize dependence, whereas the SDE-based approach specifies joint temporal dynamics, with cross-band dependence arising from coupled dynamics or correlated driving processes. The latent-process approach represents dependence through shared stochastic processes and their transformations, which can reflect physical relationships among multiple light curves. We demonstrate the practical implications of these modeling choices through applications to multi-band active galactic nucleus light curves and emission-line reverberation mapping. Rather than advocating a universally preferred approach, this work provides guidance for selecting dependence structures according to scientific objectives and clarifies how these choices influence the statistical characterization and scientific interpretation of multi-band stochastic variability.
ER -