Estimating Orbital Parameters of Direct Imaging Exoplanet Using Neural Network

Bo Liang, Hanlin Song, Chang Liu, Tianyu Zhao, Yuxiang Xu, Zihao Xiao, Manjia Liang, Minghui Du, Wei-Liang Qian, Li-e Qiang, Peng Xu, Ziren Luo · 2025

Read Estimating Orbital Parameters of Direct Imaging Exoplanet Using Neural Network by Bo Liang, Hanlin Song, Chang Liu, Tianyu Zhao, Yuxiang Xu, Zihao Xiao, Manjia Liang, Minghui Du, Wei-Liang Qian, Li-e Qiang, Peng Xu, Ziren Luo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-17459.txt.

Introduction

Working out the orbit of a planet photographed directly, far from its star, usually relies on slow sampling methods that test millions of guesses. Bo Liang, leading a team with eleven co-authors, pairs a modern generative method with the classic sampling approach to speed that work up. Their combined method first narrows the plausible range of orbital values, then samples carefully within that range to pin down the final estimates.

On the well studied planet beta Pictoris b, the combined method ran many tens of times faster than the standard samplers it was compared with, while keeping similar accuracy. Liang and his co-authors also record the best average fit score among the methods tried. They argue the same pairing could serve other fields that face hard inference problems, from cosmology to medical imaging.

Catalog record

CREATORBo Liang, Hanlin Song, Chang Liu, Tianyu Zhao, Yuxiang Xu, Zihao Xiao, Manjia Liang, Minghui Du, Wei-Liang Qian, Li-e Qiang, Peng Xu, Ziren Luo
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-17459

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Estimating Orbital Parameters of Direct Imaging Exoplanet Using Neural Network Bo Liang, Hanlin Song, Chang Liu, Tianyu Zhao, Yuxiang Xu, Zihao Xiao, Manjia Liang, Minghui Du, Wei-Liang Qian, Li-e Qiang, Peng Xu, Ziren Luo Abstract In this work, we propose a new flow-matching Markov chain Monte Carlo (FM-MCMC) algorithm for estimating the orbital parameters of exoplanetary systems, especially for those only one exoplanet is involved. Compared to traditional methods that rely on random sampling within the Bayesian framework, our approach first leverages flow matching posterior estimation (FMPE) to efficiently constrain the prior range of physical parameters, and then employs MCMC to accurately infer the posterior distribution. For example, in the orbital parameter inference of β \beta Pictoris b, our model achieved a substantial speed-up while maintaining comparable accuracy—running 77.8 times faster than Parallel Tempered MCMC (PTMCMC) and 365.4 times faster than nested sampling. Moreover, our FM-MCMC method also attained the highest average log-likelihood among all approaches, demonstrating its superior sampling efficiency and accuracy. This highlights the scalability and efficiency of our approach, making it well-suited for processing the massive datasets expected from future exoplanet surveys. Beyond astrophysics, our methodology establishes a versatile paradigm for synergizing deep generative models with traditional sampling, which can be adopted to tackle complex inference problems in other fields, such as cosmology, biomedical imaging, and particle physics. 1 Introduction Since the first confirmed detection of an exoplanetary system in 1992 Wolszczan and Frail 1992 and the first confirmed detection of an exoplanet in 1995 Mayor and Queloz 1995 , more than 4,615 exoplanetary systems and 5,989 exoplanets have been confirmed as of September 2025 NASA Exoplanet Archive 2025 . These systems exhibit a wide range of characteristics, including variations in the number and composition of exoplanets, the dynamics of their orbital motions, and the types of their host stars. Notably, the majority of these systems are characterized by the detection of only a single planet, although this is partly due to observational biases. For a recent review, see, e.g., Ref. Howe et al. 2025 . Thus, high-precision exoplanet detectors are essential for investigating the mechanisms underlying these characteristics, such as testing theories of planet formation and migration, and understanding the overall evolution of planetary systems Brandt et al. 2021 ; Blunt et al. 2020 . Currently, several observatories remain in operation and are continuously expanding our exoplanet census. Operational projects like Transiting Exoplanet Survey Satellite (TESS) Winn 2024 and CHaracterising ExOPlanets Satellite (CHEOPS) Fortier et al. 2024 conduct high-precision photometric surveys of nearby bright stars, enabling the detection and characterization of transiting exoplanets. In parallel, Gaia Pancino 2020 ultra-precise astrometric data that reveals subtle stellar motions caused by orbiting planets. Additionally, instruments like Spectro-Polarimetric High-contrast Exoplanet REsearch instrument (SPHERE) Beuzit et al. 2019 on the Very Large Telescope (VLT) directly image exoplanets, offering valuable constraints on their orbital architectures and atmospheric properties. Upcoming missions such as PLAnetary Transits and Oscillations of stars (PLATO) Rauer et al. 2024 , Atmospheric Remote-sensing Infrared Exoplanet Large-survey (ARIEL) Changeat and Yip 2023 ; Zingales et al. 2017 , SPHERE+ Boccaletti et al. 2020 , and Nancy Grace Roman Space Telescope Bailey et al. 2023 are expected to significantly advance our understanding of exoplanet populations. By integrating large-scale photometric monitoring, atmospheric spectroscopy, and precise astrometric measurements, these missions are poised to deliver unprecedented insights into the formation, evolution, and diversity of planetary systems. In recent years, a variety of exoplanet discovery techniques have been developed and extensively employed Maire et al. 2023 , including transit photometry, radial velocity (RV), microlensing, and direct imaging. These efforts have led to the detection of a large number of exoplanetary systems, significantly advancing our understanding of the statistical properties of planetary populations and providing critical constraints on models of planet formation and migration. In this work, we focus on exoplanetary systems detected using direct imaging method. Although the number of such detections is smaller than that obtained with the other three techniques Fischer et al. 2014 ; Deeg and Belmonte 2018 , direct imaging provides unique advantages for determining the demographics of exoplanets around nearby young stars, characterizing their atmospheric properties, and exploring their circumstellar disks Boccaletti et al. 2020 . Through synergy with other facilities, such as Gaia, GRAVITY Abuter et al. 2017 , and James Webb Space Telescope (JWST) Lightsey et al. 2012 , more exoplanets will be directly imaged. This growing catalog of directly imaged exoplanets will yield essential observational data, which is vital for testing and validating theoretical models of planet formation and evolution, thereby elucidating the fundamental physical processes that determine planetary system architectures. With the growing accumulation of observational data, a key challenge lies in the rapid and precise estimation of the physical parameters of exoplanetary systems, which are crucial for predicting exoplanet locations, calculating transit probabilities, and assessing the climates and habitability of Earth-like worlds in future missions. In the case of directly imaged exoplanets, several approaches have been proposed for estimating the physical parameters Brandt et al. 2021 , such as linearized least square adjustment Forveille et al. 1999 , the Orbits for the Impatient (OFTI) algorithm Blunt et al. 2017 and the Parallel Tempered Markov Chain Monte Carlo (PTMCMC) algorithm Foreman-Mackey et al. 2013 ; Vousden et al. 2016 . Tools such as orbitize! implement Nested Sampling Speagle 2020 , and have been demonstrated in studies that combine astrometric and radial velocity data to constrain exoplanet orbits ( Brandt et al. 2021 ; Blunt et al. 2020 ) . However, these traditional methods Forveille et al. 1999 ; Blunt et al. 2017 ; Foreman-Mackey et al. 2013 ; Vousden et al. 2016 suffer from limitations: Due to high parameter dimensionality and the potential multimodality of posterior distributions, convergence can be prohibitively slow without an optimal proposal distribution or favorable random initialization. In contrast, while machine learning based methods Sun et al. 2022 ; Ruth 2024 ; Chen et al. 2024 offer a significant advantage in speed for rapid orbital parameter inference, the posterior distributions they generate typically exhibit larger errors than traditional MCMC methods and often fail to accurately capture multimodality. Consequently, achieving rapid inference of exoplanet orbital parameters while maintaining high precision remains a pressing challenge. Recent advances in deep learning have enhanced exoplanet studies by improving the detection and analysis of observational data across various missions Sithajan and Meethong 2023 ; Ofman et al. 2021 ; Du et al. 2024 ; Liang et al. 2024a ; Liang et al. 2024b . Convolutional neural networks effectively identify subtle transit signals in light curves from missions such as Kepler Wang et al. 2024 ; Pan et al. 2024 ; Prithivraj and Kumari 2023 and TESS Ulaş et al. 2025 ; Poleo et al. 2024 , while deep learning approaches applied to radial velocity and direct imaging data help reduce noise and reveal faint planetary features. Generative models like normalizing flows Wald and Steidl 2025 ; Durkan et al. 2019 ; Dinh et al. 2017 ; Dinh et al. 2014 and neural ordinary differential equations Chen et al. 2018 ; Chen et al. 2021 ; Kidger et al. 2021 ; Chen 2018 provide efficient tools for simulation-based inference (SBI) Dax et al. 2023 ; Tejero-Cantero et al.

Read the complete text