Beyond Spherical geometry : Unraveling complex features of objects orbiting around stars from its transit light curve using deep learning

Ushasi Bhowmick 1, Shivam Kumaran 1 · 2025

Read Beyond Spherical geometry : Unraveling complex features of objects orbiting around stars from its transit light curve using deep learning by Ushasi Bhowmick 1, Shivam Kumaran 1 (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-2509-14875.txt.

Introduction

Ushasi Bhowmick and Shivam Kumaran ask how much of an orbiting object's shape can be read from its transit light curve, in a 2025 astronomy paper. When an object crosses its star, the dip in brightness forms the light curve. Many different shapes can produce nearly the same dip, so the problem is ill-posed from the start.

The authors generate a library of random two-dimensional shapes and simulate their transits with a light curve simulator. Each shape is broken into elliptical components described by Fourier coefficients, from broad outlines to fine wiggles. Neural networks then learn to predict those coefficients directly from a light curve. The networks recover the low-order components well, capturing overall shape and orientation. Fine detail remains harder, which maps the limits of shape inference from transits.

Catalog record

CREATORUshasi Bhowmick 1, Shivam Kumaran 1
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-14875

Browse the full library · About the Tower of Babel

Full text

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

Beyond Spherical geometry : Unraveling complex features of objects orbiting around stars from its transit light curve using deep learning Ushasi Bhowmick 1, Shivam Kumaran 1 Abstract Characterizing the geometry of an object orbiting around a star from its transit light curve is a powerful tool to uncover various complex phenomena. This problem is inherently ill-posed, since similar or identical light curves can be produced by multiple different shapes. In this study, we investigate the extent to which the features of a shape can be embedded in a transit light curve. We generate a library of two-dimensional random shapes and simulate their transit light curves with light curve simulator, Yuti . Each shape is decomposed into a series of elliptical components expressed in the form of Fourier coefficients that adds increasingly diminishing perturbations to an ideal ellipse. We train deep neural networks to predict these Fourier coefficients directly from simulated light curves. Our results demonstrate that the neural network can successfully reconstruct the low-order ellipses, which describe overall shape, orientation and large-scale perturbations. For higher order ellipses the scale is successfully determined but the inference of eccentricity and orientation is limited, demonstrating the extent of shape information in the light curve. We explore the impact of non-convex shape features in reconstruction, and show its dependence on shape orientation. The level of reconstruction achieved by the neural network underscores the utility of using light curves as a means to extract geometric information from transiting systems. 1 Introduction The launch of exoplanet survey satellites such as Kepler and TESS has led to the discovery of more than 5000+ exoplanets from a set of more than 15000 project candidates. With increasing photometric precision of space-based telescopes, new possibilities have opened up for the characterization of diverse phenomena with the help of photometric transit light curves. The light curve of the system WASP 103b shows evidence of tidal distortion ( Barros et al. 2022 ) . The transits of \beta -Pictoris b show assymetries corresponding to debris disk and exo-comets ( Lecavelier des Etangs et al. 2022 ; Heller 2024 ) . Williams et al. 2006 theorized the use of eclipse light curves to obtain longitudinal flux variations for exoplanets. In the era of transmission spectroscopy of exoplanets, light curve anomalies are being used to infer assymetric limb features ( von Paris et al. 2016 ; Murphy et al. 2024b ; Murphy et al. 2024a ; Chen et al. 2025 ) . Yip et al. 2020 and Changeat et al. 2024 demonstrate the utility of combining time domain along with spectral domain in atmospheric retrievals. This shows that anomalies in transit light curves can give us information about the geometric distortions in the transiting systems. The use of light curve inversion as a technique to reconstruct aspects of the shape geometry was first conceptualized for asteroid light curves ( Ostro et al. 1988 ; Cellino et al. 1989 ; Kaasalainen & Torppa 2001 ; Kaasalainen et al. 2001 ) . Asteroids are considered as a 3D complex surface, and the light reflected from its surface is monitored over a given time period. The variations in the projected geometry over time gives rise to changes in flux. Ostro et al. 1988 ; Kaasalainen et al. 2001 demonstrate the detectability of convex hulls of asteroid shape. Muinonen et al. 2015 ; Muinonen et al. 2020 ; Muinonen et al. 2022 establish an Inversion model based on ellipsoid assumption and utilize it to perform retrievals of asteroid parameters from observed light curves. These methods have extended to the identification of near-earth objects and space debris ( Bradley & Axelrad 2014 ; Jiang et al. 2023 ; Galeano et al. 2025 ) . Machine learning models are used extensively for optimizing inversion routines ( Furfaro et al. 2019 ; Allworth et al. 2021 ; Tang et al. 2025 ) . Tang et al. 2025 uses deep neural networks to identify regions of non-convexity in reconstructed shapes. The technique has been extended for identification and characterization of starspots ( Harmon & Crews 2000 ; Roettenbacher et al. 2013 ; Santos et al. 2017 ; Luo et al. 2019 ; Luo et al. 2021 ; Luger et al. 2021 ) . These works highlight the ill-posed nature of the inversion problem. In case of starspots it manifests as a degeneracy between the number and location vs the size of the starspots, especially when considering rotational systems ( Luger et al. 2021 ) . Luo et al. 2019 mentions the use of surface brightness ratio as priors to constrain the starspot reconstruction algorithms. These algorithms have been successfully used to constrain starspot properties for various systems, such as KIC 5110407 ( Roettenbacher et al. 2013 ) , Kepler-17b ( Luo et al. 2021 ) etc. Figure 1: Methodology for inverting photometric transit light curves to recover 2-D shapes: (a) generate a comprehensive library of 2D shapes. (b) : using Yuti simulator ( \mathcal{Y} ) to generate corresponding transit light curves(c). (d) train neural-network to approximate the inverse function, \mathcal{Y}^{-1} . (e) Decomposition of 2D shapes into component ellipses and selection of subspace which can be predicted by the NN, effectively identifying invertible subspace (represented with dashed region) The extent of research carried out on the inversion problem indicates that light curve inversion can be potentially extended to transiting systems. In this work, we investigate the extent to which features of a shape are embedded in a transit light curve using machine learning. We generate a large number of arbitrary shapes, and use Yuti 1 1 1 https://github.com/ushasi-bhowmick/Yuti ( Bhowmick & Khaire 2024 ) to generate the transit light curve. Since inversion is an ill-posed problem, instead of a direct reconstruction of shape, we decompose the shapes into a low-dimensional encoding vector. We train deep neural networks to learn the parameter space from the input transit light curve. Based on the ability of the neural network to learn the parameter values, we can make inferences about the feature space of shapes, and identify patterns that are detectable, and those that are ambiguous. The paper is organized as follows. Section 2 describes the methodology of generating arbitrary shapes and their elliptical Fourier decomposition. Section 3 describes the machine-learning model and the training strategies utilized. Section 4 shows the results of training for the first order Fourier coefficients and Section 5 shows the results for higher order coefficients. Section 6 and 7 presents the discussion and conclusions of the work. 2 The Parameter Space of 2D Shapes Photometric transit observations provide a one-dimensional imprint of a two-dimensional projection of a transiting object against the host star. Consequently, the mapping from an arbitrary 2D shape \mathcal{S} to the observed light curve \mathcal{L} is intrinsically many-to-one as many different shapes can produce indistinguishable light curves. For example, shapes flipped along x-axis (the axis parallel to the direction of transit, passing through the center of the circular projection of a star) of the transit will result in an identical light curves. Some skewed features in the shape may not result in any change in the light curve from that of an identical shape without the protrusion. Two visually different shapes may result in similar light curves. Such

Read the complete text