Determination of the relative inclination and the viewing angle of an interacting pair of galaxies using Convolutional Neural Networks. Issue 3 (20th July 2020)

• Record Type: Journal Article
• Authors: Prakash, Prem; Banerjee, Arunima; Perepu, Pavan Kumar
• Is Part Of: Monthly notices of the Royal Astronomical Society. Volume 497:Issue 3(2020)
• Page Start: 3323
• Page End: 3334
• Publication Date: 2020-07-20
• Subjects: methods: data analysis -- methods: statistical -- virtual observatory tools -- galaxies: evolution -- galaxies: interactions -- galaxies: kinematics and dynamics; Astronomy -- Periodicals; Periodicals; 520.5
• DOI: 10.1093/mnras/staa2109
• Languages: English
• ISSNs: 0035-8711

Abstract

ABSTRACT: Constructing dynamical models for interacting galaxies constrained by their observed structure and kinematics crucially depends on the correct choice of the values of their relative inclination ( i ) and viewing angle (θ) (the angle between the line of sight and the normal to the plane of their orbital motion). We construct Deep Convolutional Neural Network (DCNN) models to determine the i and θ of interacting galaxy pairs, using N -body + smoothed particle hydrodynamics (SPH) simulation data from the GalMer data base for training. GalMer simulates only a discrete set of i values (0°, 45°, 75°, and 90°) and almost all possible values of θ values in the range, [−90°, 90°]. Therefore, we have used classification for i parameter and regression for θ. In order to classify galaxy pairs based on their i values only, we first construct DCNN models for (i) 2-class ( i = 0 °, 45°) (ii) 3-class ( i = 0°, 45°, 90°) classification, obtaining F 1 scores of 99 per cent and 98 per cent respectively. Further, for a classification based on both i and θ values, we develop a DCNN model for a 9-class classification using different possible combinations of i and θ, and the F 1 score was 97${{\ \rm per\ cent}}$ . To estimate θ alone, we have used regression, and obtained a mean-squared error value of 0.12. Finally, we also tested our DCNN model on real data from Sloan Digital Sky Survey. Our DCNN models could be extended to determine additional dynamical parameters, currently determined by trial and error method.