Astrostatistics and Data Science
We present a summary of the posters of this astronomical topic. If you want to see the full poster, click on the document:
Title: Searching Eccentric Eclipsing Binary Stars with TESS
Fontirroig, V (ULS). Monsalves Gonzalez, N. (ULS), Segura-VandePerre, J. (ULS), Jaque Arancibia (ULS), M. , Damke G. (ULS, AURA)
Abstract: With new space telescopes such as the Transiting Exoplanet Survey Satellite (TESS), better data quality is in our hands and, as time goes by, the quantity of this data is increasing almost exponentially, making it almost impossible to analyze them in a traditional way. However, astroinformatics techniques with machine learning and deep learning algorithms could solve this problem. The aim of this work is to find eccentric eclipsing binary stars using machine learning algorithms with the help of the TESS database. A sample of 94947 detached eclipsing binary (EB) stars obtained from The International Variable Star Index (VSX) were searched in the Mikulski Archive for Space Telescopes (MAST) using the lightkurve package. Then, a Light Curve (LC) analysis is done by a Fourier transform and a double gaussian curve fit. This is an important step that gives critical information such as the difference between eclipses and their depth. Finally, Principal Component Analysis and the K-means methods are used to check if this fitting is correct, removing those that were not (i.e. bad quality LC). A quantitative analysis of the remaining EB stars is made based on the separation of both eclipses.
Title: Identification and determination of metallicities of red giant stars using Machine Learning techniques applied to the narrow and broadband photometry of the S PLUS survey
Molina Jorquera, Francisca
Molina Jorquera,Francisca (1), Damke, Guillermo (1,2,3),Jaque Arancibia, Marcelo (1,2)
1: Departamento de Astronomía, Universidad de La Serena
2: AURA Observatory in Chile
3: Instituto de Investigación Multidisciplinar en Ciencias y Tecnología, ULS
Abstract: Red giant stars are characterized for their high luminosity and relatively large number compared to other giants. As a consequence, they are im-portant tracers of substructures in the Milky Way and the Local Group. However, the study of these substructures, including their formation and evolution, usually requires obtaining chemical information of their stars. The latter is achieved though spectroscopy, which requires a substantial amount of telescope time.
In this work, we present the derivation of metallicities of red giant stars from the Southern Photometric Local Universe Survey (S-PLUS) , and perform giant/dwarf discrimination, using Machine Learning techniques applied to the S-PLUS photometry. The combination of the five broadband and the seven narrowband filters of S-PLUS—the latter especially centered on prominent stellar spectral lines—provides sufficient photometric information to train Machine Learning algorithms. We created the training catalog by cross-matching the photometric catalogs of stars of S-PLUS and their respective spectroscopic data from the APOGEE survey .
Finally, we will use the trained algorithm to generate a catalog of red giant stars from S-PLUS, which allow us to study stellar populations and the structure of the Milky Way.
Title: Massive Stars Synthetic Spectral Line Classification
F. Ortiz¹, R. Pezoa², I. Araya³ & M. Curé¹
¹ Instituto de Física y Astronomía, Universidad de Valparaíso.
² Escuela de informática FACING Universidad de Valparaíso, CCTVal-UTFSM.
³ Centro DAiTA Lab, Universidad Mayor.
Abstract: Spectral lines of two massive stars have been classified in the Hα lines models of the database ISOSCELES (GrId of Stellar AtmOSphere and HydrodynamiC ModELs for MassivE Stars). To do this, first two sub-grids are made, one (grid 1) being representative of the full grid, but with a reduce number of models, and the other (grid 2) being selected by the range of temperatures of the star HD 115842. In second place, the sub-grids are segmented using the clustering algorithm Gaussian Mixture Models (GMM), which classes are selected by a search using the Bayesian Information Criterion (BIC) estimator. Then, the classes generated, within the data of the sub-grids, are used to train deep neural networks. The last part of the method consist in using the respective neural network to predict the class of an observed line, and then calculate the χ² between the line and each model in the predicted class.
Title: New cosmic rays propagation scheme
Gonzalo Jaque, Universidad de Concepción, Stefano Bovino, Universidad de Concepción, Alessandro Lupi, Institut d’Astrophysique de Paris, Tommaso Grassi, Ludwig Maximilian University of Munich.
Abstract: Low-energy cosmic rays (CRs) below 10e9 eV are known as the particles able to make through to the very cold dense core within molecular clouds where star formation is taking place. They trigger the chemistry on those regions by interacting with H2 and other species driving ions-neutral reactions. To understand the chemistry of cold cores is then of fundamental importance to accurately know the Cosmic Rays Ionization Rate (CRIR). Unfortunately, it is very challenging to retrieve this quantity from observations and strong approximation are needed in astrochemical models. For instance, attenuated models or constant CRIR value as 1e-17[s−1] on astrochemical or galactic simulations are quite often employed.
As a way to solve this we present an scheme algorithm which uses part of the Smoothed-Particle Hydrodynamics from the meshless code GIZMO, modified to act as an interpolator of points spread over the simulation domains and to follow CRs path while coupled and interact with the MHD system by giving them the physical properties of it, which allows us to retrieve the CRIR of it on-the-fly at the current simulation. The methodology and the current state of development of the scheme is presented in this poster.