Image of the Week

MK classification of stars from BP/RP spectrophotometry across the Hertzsprung-Russell diagram

Animation showing the evolution of BP/RP spectra for stellar spectral types covered in a low-extinction Hertzsprung-Russell diagram. Credits: ESA/Gaia/DPAC - CC BY-SA 3.0 IGO. Acknowledgements: ESA/Gaia/DPAC/CU8/OA, Lara Pallas-Quintela.

The animation shows a Hertzsprung-Russell Diagram built with Gaia EDR3 G, G_RP, G_BP magnitudes and parallaxes for a sample of 14,351,682 low-extinction stars in the Milky Way, together with two different representations of the composite BP/RP spectra of a selection of stars along the Main Sequence, the Red Giant Branch, and the White Dwarf sequence. Such a representation illustrates the power of BP/RP spectra to give detailed information about the evolutionary state and physical properties of stars.

Our goal is to obtain a set of templates covering typical stellar types to improve, in forthcoming data releases, the usefulness and performance of the unsupervised classification algorithm based on the sampled mean spectra generated by SMSGen from the BP/RP spectra and used by the Outlier Analysis (OA) package within Coordination Unit 8 (CU8). OA groups similar spectra using Self-Organizing Maps (SOMs), a technique that does employ tagged data during the training process. Hence, once the model is built, such templates are crucial to both validate the algorithm’s performance and to determine the spectral type of each neuron (group of objects). This clustering is publicly available in the GUASOM web app, a novel tool to be officially released with Gaia DR3 on 13 June 2022.

We selected low-extinction stars by setting a limit for their galactic latitudes avoiding the galactic plane (b >= 30 deg or b <= -30 deg) and adding the stars in the solar neighbourhood (r_med_geo <= 60 pc). Additionally, stars in Gaia DR2 with a very low A_G value derived by CU8, a_g_val <= 0.03 mag, were included. The next step was to obtain the fiducial curves corresponding to the Main Sequence and the Red Giant Branch by fitting polynomials in the maximum density zones of the Colour-Magnitude Diagram (CMD).

The stars were selected with a relative parallax uncertainty better than 10% and the astrometric and photometric quality filters as explained in A&A 616, A10, 2018.

We then used the relationships between G_BP-G_RP and T_eff for dwarfs and giants in Mucciarelli and Bellazzini 2020 and sampled both the Main Sequence and the Red Giant Branch at different T_eff intervals covering the standard spectral subtypes, which are shown in the animation.

In the case of white dwarfs (WDs), we cross-matched our sample with the WD EDR3 catalogue by Gentile-Fusillo et al. 2021, which provides a classification in the different WD classes and T_eff values for those stars. Such classification was refined using the scheme in Sion et al. 1983, which allows us to assign spectral subtypes to our white dwarfs according to the value of their effective temperatures. For the sake of simplicity, only a sequence of DA stars is shown in the animation, as well as high-luminosity WD stars, populated by Planetary Nebulae central stars and hot subdwarfs.

In the animation, Gaia spectrophotometry is shown in two formats: internally calibrated BP/RP spectra (De Angeli et al. 2022) on the left, and externally calibrated BP/RP spectra (Montegriffo et al. 2022) on the right, both normalised. While internally calibrated spectra are those processed by the OA (Outlier Analysis) module of Coordination Unit 8 and contain the instrumental profile of both spectrophotometers, externally calibrated spectra can be directly generated by Gaia users from Coordination Unit 5 BP/RP coefficients once Gaia DR3 is published. For these purposes, the GaiaXPy Python library has been used, developed by Coordination Unit 5 and available with more details from the Gaia DR3 preview page for GaiaXPy.

Credits: ESA/Gaia/DPAC/CU8/OA; Lara Pallas-Quintela, Daniel Garabato, Minia Manteiga, and Carlos Dafonte from the University of A Coruña, Spain. Our thanks to the entire CU8 team, especially Orlagh Creevey, Morgan Fouesneau, and Rosanna Sordo.

[Published: 10/06/2022]