112 Symbiotic Star Candidates in Gaia Data Release 3

Abstract

Symbiotic Stars are some of the most elusive and exotic star systems in the universe. Their rarity and peculiarity are of key importance to our understanding of stellar evolution and the origins of elements that make life possible. They give crucial information in determining how binary systems evolve and a peek into fundamental cosmological events. Since the list of known Symbiotics is relatively small, new releases from Gaia provide an opportunity to expand upon this list. Using machine learning techniques to train and test on a multitude of stars found in Gaia Data Release 3 (DR3), the aim is to discover a new list containing potential Symbiotic Stars.

Introduction

Symbiotic Stars are a part of a small set of binaries that usually contain a white dwarf and a red giant star. Both stars were similar to the Sun but are now at an older stage of evolution. The white dwarf is a compact, Earth-size object, while the red giant is cooler and roughly a hundred times the size of the Sun. In a Symbiotic Star, the white dwarf accretes material from its red giant companion resulting in an extremely hot temperature and super bright luminosity. This then allows for ionization of a large fraction of the cool giant’s wind. In combination, this unique light pattern given at a specific wavelength allows for Symbiotics to stand out from other stars.

In order to find these systems, it’s crucial to look at a number of key factors:

1. Spectroscopic data involves an interaction of matter with electromagnetic radiation over a range of wavelengths. This data is used to find the spectral lines, doppler shift, temperature, stellar classifications and compositions of celestial objects.
2. Photometric data includes light measured over a broad range of colors which allows for an accurate determination of the brightness of a celestial object.
3. Astrometric data refers to the measurements of the positions and motions of stars.

We focus on observations from Gaia, a space-based instrument that measured over 1 billion stars. Gaia data provides accurate astrometry, photometry, and spectra in blue and red wavebands. The spectra are provided in the form of coefficients of spectral templates derived by the Gaia team. These factors help create the parameters that are needed to search for the appropriate stars in a machine learning algorithm.

Machine Learning Techniques 

To begin the machine learning process there are some steps to ensure a thorough and well-rounded classifier

1. Gather a training set from Gaia DR3 of known non-Symbiotics.
2. Set-up training parameters.
3. Retrieve a test set of known Symbiotics.
4. Create labels to represent if a star is Symbiotic or not.
5. Create a list of stars in both datasets that have spectroscopic data.
6. Use the first 25 coefficients of the blue spectra data and add them to the training parameters.
7. Create classifier parameters to fit the datasets and make a matrix regarding the accuracy of the predicted versus misclassified stars.
8. Run the classifier onto potential red giant stars gathered from the Gaia DR3 database.

In creating a training set, we took two distance cuts in kiloparsecs, [-1 – 10], [10 – 1000] and then queued Gaia to pull 875 random stars from each of these cuts. This dataset was then run through SIMBAD, a database of astronomical objects, to confirm it contained no known symbiotic stars. The dataset was then combined to form a 1750-star list which was then assigned labels equal to zero for non-symbiotic stars. The known Symbiotics, which we determined from SIMBAD, were assigned labels equal to one. These sources constitute the preliminary analysis reported here.

For machine learning we set up a list of Gaia attributes on which to train the data sets. For Symbiotic stars specifically, bp_rp, phot_g_mean_mag, parallax, parallax_over_error, astrometric_gof_al, pm, ruwe, and astrometric_excess_noise were the essential parameters in initial testing.

Toward including spectroscopic information in our machine learning search, we sought stars with Gaia’s spectral data (has_xp_continuous == True). Our data was then filtered to get a list of total stars that contained spectroscopic information. For our Symbiotic Star search, the first 25 coefficients of the blue spectra list were used. Those 25 coefficients were then added into the initial parameters for testing.

We set up our machine learning classifier with a parameter grid to fine tune the RandomForestClassifier for better accuracy. The classifier is then run on the datasets and should produce a matrix by which it correctly predicts and/or misclassifies them. Our next step will be to run the classifier on promising candidates in the Gaia database and see what turns out.

Summary 

This is a work in progress. While much has been done, there is still much left to do. By leveraging machine learning techniques on Gaia’s extensive dataset, we aim to identify these elusive Symbiotic binaries and discover new members. In finding these new potential candidates, and studying them, we hope to learn more about how they work and what they can tell us about our universe. After all, the universe is a space with infinite possibilities and a vast unexplored frontier waiting for us to discover its mysteries.

References