Abstract Details
| Name: Ushasi Bhowmick Affiliation: Space Applications Centre, ISRO Conference ID: ASI2025_389 Title : Lightcurve Inversion: Analyzing the extent of feature embedding of a two-dimensional projected shape on its transit lightcurve. Authors and Co-Authors : Ushasi Bhowmick 1, Shivam Kumaran 1 Abstract Type : Oral Abstract Category : Sun, Solar System, Exoplanets, and Astrobiology Abstract : The increased sensitivity and resolution of space-based telescopes has led to observations of deviations from spherical exoplanet transits, caused by tidal distortions, disintegrating planets etc. Therefore, a proper understanding of geometrical anomalies in photometric lightcurves is of key importance. The lightcurve inversion problem is considered to be ill-posed, however, it has been addressed in great detail especially in the context of asteroid lightcurves. Since a number of different shapes may give rise to identical lightcurves, we need to identify what features of a two-dimensional (2D) projected shape can be successfully embedded in its transit lightcurve. We generate a large number of arbitrary shapes and their transit lightcurves using Yuti (Bhowmick and Khaire 2024). As a demonstration, we use the complexity parameter defined in Chen and Sundaram 2005 as a scalar metric for characterizing a 2-D shape. We design a Deep Neural Network (DNN) and show that it retrieves the complexity parameter with error <15% from lightcurve alone. The error in retrieval increases for shapes with larger complexity (C>0.3) depicting regimes where shapes are degenerate to transit lightcurves. To capture more complex features, we create a low-dimension latent-space representation (Λ) of the shapes using autoencoders. We train a separate DNN model to retrieve this latent-space from the lightcurves. The capability of the DNN to retrieve Λ from the lightcurves itself reflects the successful embeddings of shape complexity in the lightcurves. The latent-space can be segregated into clusters depicting different ‘classes’ of shapes, based on their detectability from transit. We aim to study the intra-class similarities and inter-class differences in shape features. The lightcurve inversion problem can be reduced to identification of such ‘classes’ which will enable us to narrow-down the geometrical properties of the shape based on transit lightcurves. |