© The Authors 2023

1. Introduction

Detailed knowledge of the astrophysical parameters (APs; effective temperatures, radii etc.; see Sect. 2) of stars is fundamental for understanding the structure, formation, and evolution of the astrophysical systems in which they reside. For example, exploring chemical distributions of populations of our Galaxy requires well-constrained stellar effective temperatures (T_eff) and surface gravities (log g) in order to derive precise and accurate abundances; see Nissen & Gustafsson (2018) and Jofré et al. (2019) for reviews. If we want to place our Solar System in the context of exoplanet-system formation and evolution, we need to determine the radius, mass, and age of many exoplanets and their host stars; see for example Kaltenegger & Selsis (2015), Rauer et al. (2014), Rando et al. (2020). Gaia DR3 contains a wealth of new data products. In particular, it provides us with stellar parameters derived from the analysis of the Gaia RVS spectra (Sartoretti et al. 2018), the low-resolution spectra produced by the Blue Photometer (BP) and the Red Photometer (RP; Carrasco et al. 2021; De Angeli et al. 2023), astrometry (Lindegren et al. 2021a), and integrated photometry (Riello et al. 2021) for up to 470 million stars (Andrae et al. 2023; Creevey et al. 2023; Fouesneau et al. 2023; Lanzafame et al. 2023; Recio-Blanco et al. 2023). As expected, the accuracy and precision of these parameters vary with brightness, distance, stellar type and the number of observations. These parameters can be exploited in many ways from detailed studies of individual stars, to statistical studies of large samples of stars or populations. This catalogue, based uniquely on Gaia data, also has long-term legacy value as a rich database for target selection for future follow-up studies and missions.

In this work, we demonstrate the potential of the new data products in Gaia DR3 by producing very high quality samples of astrophysical parameters of stars throughout the HR diagram. We aim to make clean samples of stars based on severe quality cuts. We consider these sources to have the most accurate and precise stellar properties in this catalogue, and advocate their use on a star-by-star basis. However, these quality cuts have an important impact on the selection function and completeness. Our selection criteria will not be optimal for the specific scientific cases of many users, and we fully encourage the exploration of the full catalogue of APs in Gaia DR3.

This paper is laid out as follows. In Sect. 2 we describe the data products that are used in this work. Subsequently, in the first part of this analysis, we focus on three main stellar regimes and produce large high-quality samples of stars covering the hot O-, B-, and A-type stars (OBA, Sect. 3), the cooler F-, G-, K, and M-type stars (FGKM, Sect. 4), and the substellar ultracool dwarfs (UCDs, Sect. 5). We then focus on specific objects of interest: carbon stars (Sect. 6), solar analogues (Sect. 7), and finally the Gaia spectrophotometric standard stars (SPSS, Sect. 8, Pancino et al. 2021). Sections 3–8 are entirely independent sections and readers can choose to focus on their section of choice without missing important information for the rest of the paper. In Sect. 9 we describe the six tables from this work that are made available in Gaia DR3, and then in Sect. 10 we illustrate some applications of the various samples in different astrophysical domains.

2. Data description

To define our samples of stars, we use primarily the astrophysical parameters (APs) from the Gaia DR3 catalogue. These data provide us with a uniformly derived all-sky catalogue of APs: atmospheric properties (T_eff, log g, [M/H], [α/Fe], activity index, emission lines, rotation), abundance estimates for 13 chemical species, evolution characteristics (radius, age, mass, bolometric luminosity), distance, and dust extinction. The APs are found in two tables of the archive: astrophysical_parameters and astrophysical_parameters_supp, and a subset of these are also copied to the gaia_source table for convenience to the user.

These data were produced by the Gaia Data Processing and Analysis Consortium (DPAC) – Coordination Unit 8 (CU8) using the Astrophysical Parameters Inference Software (Apsis, Bailer-Jones et al. 2013; Creevey et al. 2023) and a series of three papers describe the methodologies and content of the APs in Gaia DR3. Creevey et al. (2023) present an overview of the processing, the architecture, and the modules of Apsis, along with a summary of the data products. Fouesneau et al. (2023) focus on the stellar content, its description, and quality assessments, and Delchambre et al. (2023) detail the non-stellar content, in particular object classification, extinction, and extra-galactic objects.

The DPAC data processing chain also uses these APs, for example, to identify the best template spectrum for the extraction of the radial velocities from the RVS spectra, the identification of quasars used to fix the astrometric reference frame, and the optimisation of the BP and RP calibration.

In this work, we focus on the data products provided by six modules of the Apsis chain; the General Stellar Parametrizer from Photometry (GSP-Phot), the General Stellar Parametrizer from Spectroscopy (GSP-Spec), the Extended Stellar Parametrizer for Emission-Line Stars (ESP-ELS), the Extended Stellar Parametrizer for Hot Stars (ESP-HS), the Extended Stellar Parametrizer for Ultra-Cool Dwarfs (ESP-UCD), and the Final Luminosity Age Mass Estimator (FLAME). These are described in detail in Creevey et al. (2023) and in the online documentation. Further details on GSP-Phot and GSP-Spec are also found in the dedicated module papers (Andrae et al. 2023; Recio-Blanco et al. 2023).

Briefly, GSP-Phot processes all sources with mean BP and RP spectra (De Angeli et al. 2023; Montegriffo et al. 2023) to produce spectroscopic parameters and extinction estimates; it also uses parallaxes and photometry. Within the Apsis software, the parallaxes are corrected for the known zero-point biases as a function of latitude, magnitude, and colour; see Lindegren et al. (2021b). It processes the sources considering four stellar libraries, and the individual results for each of these libraries are found in the astrophysical_parameters_supp table. The results from the library responsible for the highest log posterior for that source (see libname_gspphot) are those that appear in the main astrophysical_parameters table. GSP-Spec processes sources with mean RVS spectra (Seabroke et al., in prep.) and produces not only atmospheric parameters but also chemical abundances and the diffuse interstellar band characterisation. These latter products are not the focus of this work, and we refer readers to Gaia Collaboration (2023a,b), respectively, for further information on those. The results from GSP-Spec used in this work are found in the astrophysical_parameters table. ESP-HS processes the BP and RP spectra and the RVS spectra when available, but by default it processes just the BP and RP spectra. It produces stellar parameters for stars hotter than 7500 K along with a spectral type for all stars. The ESP-ELS module analyses emission-line stars and provides class probabilities and labels, along with a measurement of the H-α equivalent width. ESP-UCD is a module dedicated to the analysis of UCDs and it produces a T_eff. All of these results are found in the astrophysical_parameters table. Finally, FLAME processes the output spectroscopic parameters from GSP-Phot and GSP-Spec along with astrometry and photometry to derive evolutionary parameters (R, L, M, age). The FLAME results based on the GSP-Phot input are found in the astrophysical_parameters table, while those based on the GSP-Spec input are found in the astrophysical_parameters_supp table. These six modules (GSP-Phot, GSP-Spec, ESP-ELS, ESP-HS, ESP-UCD, and FLAME) produce the data that are the focus of this paper. For further details on the methods, we refer readers to the above references.

This work also exploits other data products from Gaia DR3; the astrometry (parallaxes errors and proper motions) and properties of the photometry and spectroscopy are found in the main gaia_source table and these were also available in Gaia EDR3; see also Damerdji et al. (in prep.), Lindegren et al. (2021a,b), Riello et al. (2021), Seabroke et al. (in prep.). We additionally exploit the variability analysis performed by the Coordination Unit 7 (Eyer et al. 2023; Clementini et al. 2023; Mowlavi et al. 2023) and the analysis of binary and multiple systems by the Coordination Unit 4 (Gaia Collaboration 2023e; Halbwachs et al. 2023; Holl et al. 2023; Siopis et al., in prep.) to further define our samples.

3. OBA stars

3.1. Scientific motivation

O-, and B-, and A-type (OBA) stars are intermediate- to large-mass stars that evolve rapidly and do not usually migrate very far away from their birth association or cluster. For this reason, they are the best targets to study the structure and dynamics of star forming regions and the Galactic spiral arms (e.g. Gaia Collaboration 2023d). Hot stars also play an important role in the evolution of the Galaxy: they are the main contributors to its enrichment in elements heavier than carbon and their strong ultraviolet radiation is the main source of ionisation of the interstellar medium. The most massive OB stars have strong stellar winds and explode as supernovae at the end of their lives, and are therefore an important contributor to the chemical composition of the interstellar medium. Here we focus on the construction of a sample of OBA-type stars which are part of the Milky Way young disc population. Older stellar populations covering the same effective temperature range (e.g. white dwarfs and blue horizontal branch stars) are therefore excluded from this sample wherever possible. Because young OBA stars are significantly less numerous than cooler stars, their identification can be considered as a key issue in large surveys. Their spectral energy distribution is less sensitive to effective temperature in comparison to later type stars, which hampers the accurate and non-biased determination of T_eff. We fixed the lower T_eff threshold at 7500 K. Moreover, although we have taken care to eliminate the most significant contaminants (e.g. white dwarfs, RR Lyrae, Sect. 3.2), our sample still includes stars that are not young OBA stars, such as blue horizontal branch (HB) stars (Sect. 3.3). In the rest of this section, we continue to refer to the ‘OBA star sample’ as a shorthand for young OBA stars in the disc of the Milky Way.

3.2. Sample selection

GSP-Phot and ESP-HS are the two main Apsis modules that derive the APs of OBA stars. While GSP-Phot processed all targets with G ≤ 19, ESP-HS only processed OBA stars brighter than G = 17.65¹, and additionally it only processed those stars that received a spectraltype_esphs tag of ∈ [‘A’,‘B’,‘O’]. This tag is derived from a Random Forest classification of the BP and RP spectra; see Sect. 11.3.7 of the online documentation for details. In terms of effective temperature, this is equivalent to selecting targets hotter than 7500 K. The same lower T_eff limit is applied to the GSP-Phot stellar sample. Because GSP-Phot processes targets down to G = 19, the corresponding sample initially contains more (11 156 449 stars) candidate-OBA targets than ESP-HS (2 344 484). The GSP-Phot parametrisation partly relies on the use of parallax, and more outliers (e.g. misclassified cool objects, white dwarfs, etc.) are included when the astrometry is less reliable (Fouesneau et al. 2023). To exclude a significant fraction of these, we removed all targets based on the signal-to-noise ratio (S/N) of the parallax ϖ parallax_over_error ≤15, as illustrated in Figs. 1a,c. ESP-HS does not use information that allows to remove white dwarfs for example. Therefore, we applied a lower luminosity threshold to both samples and removed all subluminous objects. The limit was fixed by computing the dispersion around the running median of M_G as a function of T_eff and by using the AP determinations obtained by ESP-HS in its BP/RP+RVS processing mode. The grey shading in Figs. 1b,c shows the area of the HR diagram from which targets were excluded. Ideally, the observed de-reddened (G_BP − G_RP) colour versus T_eff follows the same relation as the one found from synthetic spectra (Figs. 1d,e; blue curve) used to derive the APs. All outliers at more than six standard deviations from the theoretical relation were discarded from the sample, as shown by the grey shading in Figs. 1d,e. The Kiel diagram of each sample is shown in the bottom row of the same figure, with the corresponding number of remaining stars. We noticed that the modules were misclassifiying some RR-Lyrae stars as OBA stars, and therefore the list was cross-matched with the RR-Lyrae table vari_rrlyrae in Gaia DR3 (Clementini et al. 2023). After filtering, 3 023 388 unique sources remained in the list of candidate-OBA stars. Among these, 1 661 459 and 843 324 have ESP-HS or GSP-Phot APs, respectively, while 518 605 have both. Among those targets with GSP-Phot parameters, all but 889 received a spectraltype_esphs tag.

Fig. 1. Selection of the OBA sample. Left and right panels: the OBA samples from ESP-HS and GSP-Phot, respectively. A first filter on the parallax (S/N) is applied to the GSP-Phot targets (from panels a to c). For both samples, subluminous targets are removed (grey shading in panels b and c), and then the outliers at six standard deviations from the expected colour vs. Teff relation (blue line) are filtered out (grey shading in panels d and e). The absolute magnitude MG is computed using the measured parallax and the estimated interstellar extinction AG provided by both modules. The de-reddened colour, (GBP − GRP)0, is derived using the value of E(GBP − GRP). The resulting Kiel diagrams are shown in the bottom rows (panels f and g). The over-densities seen at Teff = 15 000 K, 20 000 K, and 30 000 K are linked to the temperature limits of the adopted synthetic spectra libraries.

The corresponding gold_sample_oba_stars table (this will appear as gaiadr3.gold_sample_oba_stars) has two columns: one lists the source_id and the other a flag that provides information on the kinematics of the targets (Sect. 3.3). We tested the completeness of the GDR3 OBA sample by cross-matching it with the Galactic open cluster members identified by Cantat-Gaudin et al. (2020). The selection of the expected OBA stars in each cluster is based on the (G_BP − G_RP)₀ colour at T_eff = 7500 K, which is estimated by taking into account the published cluster extinction A₀. Their number, $N_{\text{OBA}}^{\text{expected}}$, was used to estimate the completeness fraction as follows:

$$\begin{array}{c}\hfill \mathrm{fraction}=\frac{N_{\mathrm{OBA}}^{\mathrm{GDR}3}}{N_{\mathrm{OBA}}^{\mathrm{expected}}},\end{array}$$(1)

where $N_{\text{OBA}}^{\text{GDR3}}$ is the number of the OBA open cluster targets found in our sample. We expect the fraction to vary with magnitude; we also expect it to vary with interstellar extinction because of degeneracy between extinction and temperature. Figure 2 shows how the completeness varies with A₀. The fraction of targets we have in common with the LAMOST OBA (Xiang et al. 2022) and GOSC (Maíz Apellániz et al. 2013) catalogues are 0.55 and 0.41, respectively. The T_eff distributions provided by both modules confirm that above 10 000 K, the ESP-HS APs should be preferred over the GSP-Phot estimates in the astrophysical_parameters table, whose temperature scale tends to be underestimated in this regime. This is especially true at higher interstellar extinction.

Fig. 2. Completeness of the OBA list in various open clusters (Cantat-Gaudin et al. 2020) as a function of interstellar extinction. The fraction corresponds to the ratio between the number of cluster members present in our list and the number of expected OBA stars. The colour code follows the cluster age provided by Cantat-Gaudin et al. (2020).

A Simbad query of the proposed OBA sample provides 34 055 targets with a confirmed main object type not equal to ‘*’. Among these, 27% have types not compatible with what would be expected for hot young stars, and 79% of these are known HB stars. This high density of hot HB stars can be seen, for example, in the bottom panels of Fig. 1, where their presence produces a significant overdensity of stars with T_eff ranging from 8000 to 10 000 K and log g lower than 3.5. As explained in the following section, a number of these lower mass evolved targets can be flagged by studying their kinematics. Furthermore, 134 498 targets in our list have a spectral type recorded in Simbad, which in 96% of the cases starts with the letter ‘O’, ‘B’, or ‘A’.

3.3. Using kinematics to remove halo contaminants

To further clean the sample of (young) OBA stars from contaminating populations, we propose a simple kinematic filter which removes what are presumably blue HB stars from the halo, which occupy the same colour–brightness space in the colour–magnitude diagram as the OBA stars, as well as the same T_eff–log g space in the Kiel diagram. We filter on the tangential velocity $v_{\text{tan}}=A_v{({\mu }_{\alpha \ast }{}^2+{\mu }_{\delta }{}^2)}^{1/2}/\varpi$, where μ_α* and μ_δ are the proper motions in Right Ascension and Declination, and A_v = 4.74074... km yr s⁻¹, using similar limits for the thin disc, thick disc, and halo as in Gaia Collaboration (2018). Thin disc stars are defined as having v_tan < 40 km s⁻¹, thick disc stars as having 40 ≤ v_tan ≤ 180 km s⁻¹, and halo stars have v_tan > 180 km s⁻¹. We next illustrate the effects of this kinematic selection and thereby focus on stars for which ϖ/σ_ϖ > 10. This parallax quality cut ensures a reliable calculation of the tangential velocities.

Fig. 3. Histogram of tangential velocities of the stars in the OBA sample with ϖ/σϖ > 10. The combined OBA star sample is shown as well as the individual O, B, and A star samples (based on the classifications from the ESP-HS module). The limits in tangential velocity separating the thin disc, thick disc, and halo populations are shown as vertical dashed lines.

Fig. 4. Toomre diagram for the OBA stars for which a radial velocity is available in Gaia DR3. See text for explanations on the diagram. The colour coding indicates the median value of vtan at a given location on this diagram. The half circles indicate limits on the total velocity with respect to the local circular velocity of 50 and 180 km s−1.

Figure 3 shows the distribution of tangential velocities for the OBA star subsample for which ϖ/σ_ϖ > 10. The vertical dashed lines indicate the above limits on v_tan and these correspond well to the inflections in the histograms for the full, B, and A star samples. The O star sample contains almost no sources with v_tan > 180 km s⁻¹. To further explore the tangential velocity selection we show the Toomre diagram in Fig. 4, which shows $\sqrt{V_R^2+V_z^2}$ along the vertical axis and V_ϕ along the horizontal axis, where (V_R, V_ϕ, V_z) are the velocity components of the stars in the Galactocentric cylindrical coordinate system, with R pointing from the Galactic centre to the Sun, z along the axis perpendicular to the Galactic plane, and ϕ along the azimuthal direction in the Milky Way disc plane (where a left-handed coordinate system is used such that the value of V_ϕ is positive for prograde stars in the disc). The values of (V_R, V_ϕ, V_z) are calculated assuming the local circular velocity from the MWPotential2014 Milky Way model (Bovy 2015), which is 219 km s⁻¹ at the distance of the Sun from the Galactic centre (8277 pc, GRAVITY Collaboration 2022). The height of the Sun above the disc plane is assumed to be 20.8 pc (Bennett & Bovy 2019) and the peculiar motion of the Sun is assumed to be (U, V, W) = (11.1, 12.24, 7.25) km s⁻¹ (Schönrich et al. 2010). Figure 4 only contains stars for which the radial velocity is available in Gaia DR3 and the colour coding indicates the value of v_tan. The two half circles indicate the limits on total velocity ${\mathit{v}}_{\mathrm{tot}}=\sqrt{V_R^2+V_{\varphi }^2+V_z^2}$ of 50 and 180 km s⁻¹ which separate thin disc, thick disc, and halo populations (Gaia Collaboration 2018). In this figure a population of stars can be seen at total velocities of more than 180 km s⁻¹ from the local circular velocity and these are most probably halo stars, in particular the population at negative V_ϕ which is associated with merger debris in the halo (e.g. Helmi et al. 2018).

Fig. 5. Distribution of the parameters of the OBA sample with ϖ/σϖ > 10 and vtan > 180 km s−1. Left: observational HR diagram. Right: Kiel diagram. The contours indicate the distribution of the full sample. The colour code indicates the density of sources satisfying the above criteria.

The colour coding in Fig. 4 suggests that the halo contaminants in the OBA sample can be filtered out by demanding v_tan < 180 km s⁻¹, although clearly there will be stars left at low tangential velocities which have a total velocity that puts them in the halo. Figure 5 shows the observational Hertzsprung-Russell (HR) and the Kiel diagrams for the sample of OBA stars with ϖ/σ_ϖ > 10. Extinction corrections using A_G and E(G_BP − G_RP) from the ESP-HS module were applied. The contours show the distribution of the full sample, while the colour coded density images show the distribution of stars selected according to v_tan > 180 km s⁻¹. The velocity-filtered sample mostly occupies the colour–magnitude space where blue HB stars are expected, around (G_BP − G_RP)₀ ∼ 0.05 and M_G, 0 ∼ 0.5 (compare to the rightmost panel of Fig. 21 in Gaia Collaboration 2018). The Kiel diagram shows a prominent feature at log₁₀T_eff ∼ 4, from log g ∼ 4 to log g ∼ 2, corresponding to the known location of the HB stars in this diagram. These same stars are also primarily located at high Galactic latitude, as expected for a halo population. A search in SIMBAD (Wenger et al. 2000) results in 8124 matches for which there is information on stellar type, of which 5770 are incompatible with stellar types corresponding to hot young stars, including 5499 sources classified as HB stars. This further supports using v_tan > 180 km s⁻¹ as a filter to clean the OBA sample from halo star contamination.

Fig. 6. Distribution of stars in the OBA sample projected on the Galactic plane. The Galactic centre is to the right at (X, Y) = (0, 0) and the Sun is at ( − 8, 0). From left to right, the panels show the full sample (with ϖ/σϖ > 10) and the samples selected according to the vtan ranges indicated. The red contours indicate lines of constant vtan, calculated with the simple kinematic disc model as explained in the text.

One might consider further filtering on v_tan, however we caution that because of the large reach of the OBA sample, this can lead to significant spatial selection effects. This is illustrated in Fig. 6. The figure shows the OBA stars with ϖ/σ_ϖ > 10 projected on the Galactic plane. The full sample is shown in the leftmost panel and the other panels show the effects of filtering on v_tan. The star positions in Galactocentric coordinates were calculated using the same Milky Way parameters as listed above. The red contours show the limits of 40 km s⁻¹ and 180 km s⁻¹ on the observed tangential velocities predicted from a simplistic model of the Milky Way disc kinematics. In this model it is assumed that all stars are located in the disc and follow perfectly circular orbits according to the rotation curve from the MWPotential2014 Milky Way model (Bovy 2015). The expected values of v_tan are then calculated over a grid of (X, Y) positions using the method outlined in Brunetti & Pfenniger (2010). The contours indicate the boundaries between smaller v_tan values to the left and larger ones to the right. The contours show that due to the large reach of Gaia, even for stars moving at zero velocity dispersion on circular orbits in the disc, one can still expect to observe tangential velocities out to values normally associated with the thick disc and halo. The rightmost panel in Fig. 6 again confirms that v_tan > 180 km s⁻¹ can be used to clean the OBA sample from halo stars, as the stars are all located to the left of the 180 km s⁻¹ contour, where these would be expected on the right (in the simple model used) if they were disc stars. The middle panels illustrate the spatial selection bias introduced when further restricting the tangential velocity. The second panel from the right shows that limiting the sample to v_tan < 40 km s⁻¹ leads to the exclusion of a significant fraction of young OBA stars which occupy regions of the Galactic disc where the values of v_tan are expected to be larger than 40 km s⁻¹. In addition, there is a lack of stars in the sample along the X = 0 line which roughly follows the shape of the 40 km s⁻¹ contour. The simple disc model predicts zero stars there, and therefore the shape of the gap shows that the model is useful in assessing the spatial selection biases induced by the kinematic selection. In the third panel from the right, the shape of the sample distribution also roughly follows the 40 km s⁻¹ contour.

In conclusion, we provide a table of 3 023 388 young OBA disc stars for exploitation by the community; this sample has been cleaned to remove as many older stellar populations as possible. We recommend to further clean the OBA star sample by applying the kinematic filter v_tan ≤ 180 km s⁻¹. Sources with v_tan > 180 km s⁻¹ have the flag in the table gold_sample_oba_stars set to 1; all other sources have the flag set to 0. We have only used the simple Galactic disc kinematic model to make the point that one should be careful not to introduce spatial biases when selecting on kinematics. Zari et al. (2021) describe a more sophisticated way of employing a simple disc kinematic model to select a clean sample of OBA stars. By assuming the stars follow disc kinematics, the observed proper motions can be used to infer distances. Stars with kinematic distances inconsistent with distances based on the parallaxes and photometric information can then be analysed further to see if they should be removed from the OBA star sample. Further filtering can of course be done on the various data-quality indicators available in Gaia DR3 (see the following section for examples), and one can also use the astrometric fidelity indicator from Rybizki et al. (2022).

4. FGKM stars

4.1. Scientific motivation

F, G, K, and M stars form the majority of the stars of our Milky Way. These stars inform us of how our Galaxy was formed and how it has evolved and are therefore the targets of many Milky Way surveys. These stars are also the targets of the future ESA PLATO mission (Rauer et al. 2014) which promises to help answer questions about the formation and evolution of our own Solar System by studying other exoplanet systems. In this section, we focus on F, G, K, and M star types (FGKM) to provide a clean sample of stars with the following astrophysical parameters: T_eff, log g, [M/H], R, M, age, evolutionary stage, and spectral type. Our final sample contains 3 273 041 stars after vigorous quality cuts based on astrometric, photometric, and astrophysical parameters, along with other Gaia-based criteria.

Our sample selection is described in Sects. 4.2 and 4.3 where we analyse the GSP-Phot-based and GSP-Spec-based atmospheric parameters individually. For both samples, we also report on evolutionary parameters from FLAME and the spectral type from ESP-HS. We then perform some additional filtering by removing variables and binaries. We also further filter on individual parameters from FLAME and ESP-HS for some sources. We validate the target list using open clusters and comparisons with external survey catalogues. In Sect. 10 we illustrate two applications of this sample by analysing known transiting exoplanets and studying unseen UCD-companions in the Gaia data.

Fig. 7. Comparisons between sample fgkm_1 and intermediate samples based on some of the criteria used to define sample fgkm_2. Top panel: illustrates the distance–parallax–error constraint and the lower panel shows the (G − GRP)0 − (GBP − GRP)0 relation after imposing the colour–Teff and colour–colour cuts described in Sect. 4.2. In both panels, the sources in fgkm_1 are shown in the background, while those satisfying the criteria are illustrated in the foreground, colour-coded according to logarithmic count.

4.2. GSP-Phot sample selection

GSP-Phot provides stellar and extinction parameters, distances, radii, and absolute magnitude for 470 million stars with G ≤ 19. We performed our initial query on the full Gaia archive by selecting sources with a parallax (S/N) better than 10, along with a number of other initial quality cuts based on astrometric and photometric parameters. These criteria are based on an analysis of a random set of 2 million sources. This resulted in a total of 70.4 million stars which we refer to as sample fgkm_1, and which is described by the following Astronomical Data Query Language (ADQL) query:

parallax_over_error > 10
ipd_frac_multi_peak < 6
phot_bp_n_blended_transits < 10
teff_gspphot > 2500,

in addition to a quality cut on bprperror < 0.06 (σ_{(BP − RP)}). This latter quantity is calculated from a standard propagation of errors using the parameters photbpmeanfluxovererror $\left(\frac{f_{\mathrm{BP}}}{\sigma f_{\mathrm{BP}}}\right)$ and photrpmeanfluxovererror $\left(\frac{f_{\mathrm{RP}}}{\sigma f_{\mathrm{RP}}}\right)$ from the archive².

We then refined this selection by considering the number of photometric transits, as well as colour–colour and colour–T_eff correlations, ensuring that the source is classified as a star by DSC, along with further constraints based on the GSP-Phot parameters themselves. These are described in the following paragraphs. We retained sources whose parameters are within the FGKM regime: T_eff < 7500 K, M_G < 12, R < 100 R_⊙, and had a log posterior >  − 4000 (goodness-of-fit indicator). We also retained sources with [M/H] > –0.8 which excludes low-metallicity sources with unreliable metallicities (Andrae et al. 2023; Creevey et al. 2023; Fouesneau et al. 2023).

GSP-Phot provides four results for each source based on different stellar libraries, namely MARCS, PHOENIX, A, and OB. Only MARCS and PHOENIX are applicable to the stellar regime considered here. The results for all libraries are found in the astrophysical_parameters_supp table, and we used the difference between teff_gspphot_marcs and teff_gspphot_phoenix (below called ΔT_eff or dteff) as a criterion to further refine the sample. There is a small bias of up to 100 K between results from these two libraries, which is due in part to the different spectral energy distributions (SEDs) of the different models and this bias varies with stellar parameters. We therefore selected those sources where the two values were in agreement in terms of their peak offset, that is, |ΔT_eff + 65|< 150 K. Additionally, we only retained sources when their uncertainties (upper − lower) are < 150 K, and the sources for which the ‘best’ model is the MARCS one (75% of sample), that is, is, libname_gspphot = ‘MARCS’ in the astrophysical_parameters table. These strict criteria based on T_eff removed about 70% of the sources. We also imposed that distance_gspphot was less than the distance corresponding to the parallax decreased by four times parallax_error (and vice versa). The top panel of Fig. 7 illustrates the impact of the cut based on distance. The sources in fgkm_1 are shown in the background, and those with the distance criteria applied (40%) are shown in the foreground. We also show the one-to-one line to guide the eye.

We corrected the G_BP, G_RP, and G observed colours for the interstellar extinction provided by GSP-Phot: G_{BP 0} = G_BP − A_BP, G_{RP 0} = G_RP − A_RP, and G₀ = G − A_G. We then fitted polynomials to the T_eff versus (G_BP − G)₀ (difference between fit and values denoted as dtb) and (G − G_RP)₀ versus (G_BP − G)₀ (difference denoted as dgb), and used these polynomial fits to remove sources further than 3σ (∼7% of fgkm_1). The coefficients of the polynomials are given in Table 1. The bottom panel of Fig. 7 illustrates the (G − G_RP)₀ − (G_BP − G_RP)₀ relation for sample fgkm_1 in the background and the sample with the 3σ constraints on the colour–colour and the colour–T_eff relations in the foreground.

All of the above criteria along with further constraints on DSC class probabilities and number of transits (n_obs below) were used to define the sample fgkm_2 which resulted in a total of 6.3M sources, that is, 12.5% of the fgkm_1 sample. A projection of the retained sources on the Galactic plane is shown in Fig. 8. We note that the criteria on the number of transits were adjusted in order to ensure a full-sky coverage. The full list of constraints for sample fgkm_2 is summarised as follows:

|dgb| < 203,
|dtb| < 0.0267
|dteff + 65| < 150
libname_gspphot = "MARCS"
teff_gspphot_upper-teff_gspphot_lower < 150
teff_gspphot < 7500
mh_gspphot > -0.8
distance_gspphot < 1e3/
       (parallax-4*parallax_error)
distance_gspphot > 1e3/
       (parallax+4*parallax_error)
radius_gspphot < 100
mg_gspphot < 12
logposterior_gspphot > -4000
classprob_dsc_combmod_star > 0.9
phot_bp_n_obs > 19
phot_rp_n_obs > 19
phot_g_n_obs > 150.

Fig. 8. Galactic plane projections illustrating the density of sources of the samples fgkm_2 (top) and fgkm_3 (bottom).

The final selection, fgkm_3, was made by applying different quality cuts based on the position of the star in the HR diagram. Giants were defined as log g < 3.6 and T_eff < 5900 K, and outliers were removed by retaining sources with log g < 0.34M_G + 2.45. Subgiants were defined as 3.6 ≤ log g ≤ 4.0 and T_eff < 5900 K, and outliers were removed by retaining sources with log g < 0.75M_G + 1.13. Main sequence stars were defined as log g > 4.0 and T_eff < 7450 K, and we imposed a further constraint of parallax_over_error > 33.34 in order to select sources with relative errors on R and L with contributions of parallax errors at 3% or less. To further refine the main sequence sample, we applied different criteria in three different colour regimes. For x < 0.98 where x = (G_BP − G_RP)₀, no further selection was done. For 0.98 ≤ x ≤ 1.8, we removed the sequence of young pre-main sequence stars and binaries by retaining sources that satisfied log L < 2.32 − 3.20x + 0.78x² where L is lum_flame. For x > 1.8, we retained sources that satisfied log g < 8.525 − 6.950x + 3.680x² − 0.584x³. This final refinement resulted in a total sample size of 3 530 174 sources.

We illustrate the different selection criteria in the HR diagram in Fig. 9. The top left panel shows the HR diagram using a random sample of data from the Gaia archive and imposing only that T_eff and L exist. The top right panel shows the selection of sources after applying the ADQL search criteria (fgkm_1) which is dominated by the criterion on parallax (S/N). One can see that many outliers and artefacts have already been removed with this cut. The bottom left panel shows the sample fgkm_2 where constraints were based on the GSP-Phot parameters, along with further constraints on the photometry and DSC class probabilities. The HR diagram has not changed drastically, but the quality of the data in fgkm_2 is much higher than in fgkm_1. Finally, the bottom right panel illustrates sample fgkm_3 which was separated into five parts (giants, subgiants, upper, middle, and lower main sequence as described above) before applying different polynomial cuts based on log g, M_G, L, and (G_BP − G_RP)₀.

Fig. 9. HR diagram based on GSP-Phot and FLAME for the definition of the FGKM sample. Top left panel: illustrates the HR diagram before any selection is made using a random sample of 2 Million stars. The rest of the panels show the various quality cuts. Top right: fgkm_1, bottom left is fgkm_2, and bottom right is fgkm_3 before cleaning for variables and binaries.

The Galactic projection of the density of sources is illustrated in the bottom panel of Fig. 8. We also illustrate the distribution of the observable parameters, G, parallax, and (G_BP − G_RP)₀ in Fig. 10. The main sequence stars occupy the dense triangular region and extend to approximately 1900 pc for the hottest stars.

Fig. 10. Distribution of the final sample fgkm_3 of the observed parameters G and parallax, colour-coded by (GBP − GRP)0.

4.3. GSP-Spec sample selection

The selection described in the previous section relies entirely on the BP and RP spectra and their parametrisation, apart from a few criteria on astrometric and photometric parameters. BP and RP spectra have important degeneracies between T_eff and A_G, and by imposing our strict selection criteria, we not only inevitably remove sources with excellent parameters derived from the RVS spectra by GSP-Spec, but we cannot guarantee that they fulfill the ‘gold’ criteria. We therefore made an independent selection by first querying the archive for sources with flags_gspspec LIKE ‘0000000000000%’, that is, sources for which the first 13 characters of the 41-character long quality flag provided by GSP-Spec are equal to ‘0’; see Recio-Blanco et al. (2023). These flag settings indicate low potential biases on T_eff, log g, [M/H], and to some extent [α/Fe] due to rotational velocity, macroturbulence, uncertainties in the radial velocity shift correction and in the RVS flux, and extrapolation, absence of undefined or negative flux values or emission lines, non-zero uncertainties in the parameters, as well as high-quality parameters for KM-type giants (see online documentation). The remaining flag characters are related to element abundances and CN and diffuse interstellar band (DIB) features and were not taken into account for the current selection. This resulted in about 1.9 million sources.

For the further selection, we considered the quality parameters parallax_over_error and rvs_spec_sig_to_noise. The latter contains the S/N in the mean RVS spectrum and is only provided for stars for which the mean RVS spectrum is published in Gaia DR3. We produced HR diagrams (lum_flame_spec teff_gspspec) and Kiel diagrams (logg_gspspec versus teff_gspspec) by imposing different lower limits on rvs_spec_sig_to_noise (S/N ≥ 0, 50, 100, 150, 200, 250, 400, 500). Visual inspection of the HR diagrams showed a group of sources at T_eff ∼ 4000 K clustered around unrealistically high luminosities. Applying the criterion rvs_spec_sig_to_noise ≥ 150 removed 99% of these sources. We combined this with the criterion parallax_over_error > 33.34, similarly to what was applied to main sequence stars in the GSP-Phot-based sample, resulting in 22 143 sources (∼1% of the flag-selected sources), hereafter referred to as ‘fgkm_spec’.

The HR and Kiel diagrams for this selection are shown in Fig. 11. The HR diagram displays a distinct giant branch and red clump as well as a region with turn-off stars and a clear main-sequence. However, as can be seen in the Kiel diagram, the log g values for main sequence stars show a large spread. This is addressed by further filtering, which is described in the following section.

Fig. 11. HR and Kiel diagrams using GSP-Spec-based parameters for the fgkm_spec sample described in Sect. 4.3 colour-coded by the metallicity from GSP-Spec, with parallax_over_error ≥33.34 and rvs_spec_sig_to_noise ≥ 150.

We also compared the distributions of uncertainties in T_eff, log g, [M/H], and [α/Fe] from GSP-Spec for the flag-selected sample and the fgkm_spec sample, where the uncertainty was defined as half of the difference between the upper and lower confidence levels. We found that the distributions for the latter sample have a smaller width by a factor of between 3 and 9 and peak at about half the uncertainty compared with the former sample.

4.4. Final sample and table description

We merged the two samples described in Sects. 4.2 and 4.3 with the objective being to provide one unique FGKM gold sample. As both sample definitions contain criteria that are not applicable to the other sample, we publish an independent table in the Gaia archive, gold_sample_fgkm_stars, which also accounts for additional filtering on specific parameters. The description of the published table is given in Table 2. Further filtering is also done based on other archive products. This is described in this section, and results in a total of 3 273 041 sources.

4.4.1. Filtering of FLAME, GSP-Spec, and ESP-HS parameters in samples fgkm_3 and fgkm_spec

The fgkm_3 GSP-Phot sample includes 3 529 613 sources with FLAME parameters, and 3 313 190 with at least one model-dependent parameter (mass, age, evolutionary stage). Figure 12 shows an HR diagram using T_eff and L, colour-coded by evolutionary stage (ϵ). There is a region on the giant branch that has low evolutionary stages compared to the bulk of the giant branch. These could be red clump stars that have been incorrectly assigned, because the models that were used to produce these parameters only span from the zero age main sequence (ZAMS) to the tip of the giant branch. These targets also have masses of greater than 2  M_⊙. Validation of FLAME parameters has shown that the model values are inaccurate when M > 2  M_⊙ for giants (Babusiaux et al. 2023; Creevey et al. 2023). We therefore only retained mass_flame, age_flame, and evolstage_flame for giants if the following conditions were met: log g < 3.5 and M < 2.0  M_⊙ and age > 1.0 Gyr. For log g > 3.5, no filtering was done. These same criteria were applied to the FLAME parameters in the fgkm_spec sample.

Fig. 12. HR diagram using sample fgkm_3 colour coded according to evolstage_flame. The low values of evolution stage on the giant branch correspond to the FLAME parameters that were removed from the table; see Sect. 4.4.1 for details.

The fgkm_spec sample shown in Fig. 11 shows some problems with log g below a certain threshold. Validation of these values indicates a systematic offset on the order of 0.3 with respect to external catalogues for main sequence stars; see e.g. Creevey et al. (2023), Fouesneau et al. (2023), Recio-Blanco et al. (2023). We therefore removed log g when log g > 4.0 in order to retain a ‘gold’ status, and kept all of the other parameters. As explained in the above references, a calibration of this parameter has been provided and a user can safely use the archive values with or without the calibration, depending on their use case.

We only retained the spectraltype tag from ESP-HS in our table if it had a quality flag of rank 1 or 2 (out of 5). This is given in the flags_esphs field in the astrophysical_parameters table as the second character in that string field.

4.4.2. Further filtering of the merged sample

To ensure that our samples are as clean as possible, we further exploited other Gaia DR3 products. We removed all sources that were considered variable or non-single stars by cross-matching our final source list with the source lists given in the vari_summary table, which removed 249 020 sources, 4873 of which are eclipsing binaries. We also removed the sources appearing in any of the non-single star tables nss_two_body_orbit, nss_acceleration_astro, nss_non_linear_spectro, or nss_vim_fl, which removed a further 28 896 sources. We then used the DPAC-Source Environment Analysis Pipeline (SEAPipe) to further check for any new binary contaminants, and this removed a further 16 sources³.

4.5. Validation of the sample

4.5.1. Validation with clusters

We take advantage of the properties of open clusters to assess the global quality of the FGKM sample. From the FGKM sample, we selected those stars classified as cluster members in the Cantat-Gaudin et al. (2020) catalogue as refined by Gaia Collaboration (2023d). The cross-match between those stars and our sample corresponds to 4132 stars and contains only cross-matches with the GSP-Phot sample. Using the full set of cluster members, we approximated each cluster with an isochrone and derived reference values of T_eff and log g. Using this T_eff, we derived A_G, adopting the literature value of A_V as a proxy of A₀. We made use of the PARSEC isochrone data set (Bressan et al. 2012). Differential extinction was assumed to be negligible inside the clusters for this validation work. This is justified by the fact that our sample excluded clusters younger than 100 Myr.

Fig. 13. Comparison of Teff and AG from GSP-Phot compared to the reference values from isochrones for stars of the FGKM sample in clusters. Left: comparison of Teff, with colour indicating the density of sources. The red line indicates the one-to-one values. Right: ΔAG = AG, GSP − Phot − AG, isochrones versus TeffGSP − Phot.

We compared the T_eff, log g, M, A_G, and [M/H] reference values with those from GSP-Phot and FLAME in our sample. For the cluster [M/H], we adopted the average [M/H] value of all the members. Table 3 and Figs. 13 and 14 present the results, which show good agreement with the reference values. Stars cooler than T_eff ∼ 4500 K have GSP-Phot parameters that show the largest differences with reference values. This overestimation of T_eff at low temperatures often shows increased extinction in this regime.

Fig. 14. Δlog g = log gGSP − Phot − log gisochrones versus TeffGSP − Phot for stars of the FGKM sample in clusters. The colour indicates the distance modulus (m − M) as derived from the GSP-Phot distance.

4.5.2. Validation with other Galactic surveys

We compared the FGKM sample parameters with those of the major spectroscopic surveys using a cross-match computation specifically performed with the Gaia DR2 cross-match software (Marrese et al. 2017, 2019) for the Survey of Surveys project (SoS Tsantaki et al. 2022). The used surveys are APOGEE (DR16, Ahumada et al. 2020), GALAH (DR2, Buder et al. 2018), Gaia-ESO (DR3, hereafter GES, Gilmore et al. 2012), RAVE (DR6, Steinmetz et al. 2020), and LAMOST (DR5, Deng et al. 2012). For each survey, we applied the quality selection criteria suggested in the relevant survey papers and summarised by Tsantaki et al. (2022). The SoS is based on Gaia DR2. Therefore, we used the cross-match between DR2 and EDR3 to find the updated source IDs. We removed all sources with a DR2-DR3 magnitude difference higher than 0.5 mag, angular difference higher than 0.5″, and all sources with more than one ‘neighbour’ or ‘mate’ (see Marrese et al. 2017 for technical definitions of ‘neighbour’ and ‘mate’). We further removed all the confirmed and candidate spectroscopic binaries identified in the surveys (Merle et al. 2017; Birko et al. 2019; Qian et al. 2019; Price-Whelan et al. 2020; Tian et al. 2020; Traven et al. 2020; Kounkel et al. 2021). The summary of the number of FGKM stars from the golden sample found in each survey is given in Table 4, where the median differences in terms of the main parameters – computed as the Gaia value minus the value of the surveys – are reported together with their median absolute deviation (MAD). A graphical comparison for the main parameters can be found in Fig. 15.

Fig. 15. Comparison of atmospheric parameters with the spectroscopic surveys for the FGKM sample. Top panels: comparison of GSP-Spec parameters and the bottom panels GSP-Phot. Left panels: the case of Teff, the middle panels that of log g, and the right panels that of [Fe/H]. The differences on the y-axes are the Gaia values minus the other survey values, where the latter are calculated as the median values in equally populated bins (solid lines, coloured according to the legend in the bottom-left panel). The dotted lines for the GSP-Spec log g are obtained after the corrections recommended by Recio-Blanco et al. (2023).

The T_eff comparison shows agreement with all surveys, both in GSP-Phot and GSP-Spec, within uncertainties. The median offsets for GSP-Spec are generally negative and of the order of –50 K to 100 K, and the same is true for the GSP-Phot offsets. The spreads range from roughly ±70 to ±120 K, in line with expectations. We note that the surveys agree with each other within a few tens of Kelvin, at least in the central portion of the T_eff range. Figure 15 shows some systematic substructures in the comparisons. For GSP-Spec, we find good agreement in T_eff. At the extremes of the T_eff range, some discrepancies occur between GSP-Phot and the comparison with LAMOST, which has the lowest resolution among the surveys.

The GSP-Spec log g comparison shows an offset of about –0.3 dex, which is a known feature, as reported in Sect. 9 of Recio-Blanco et al. (2023, see their Eqs. (1) and (2)), while the GSP-Phot comparison shows excellent agreement with the surveys. When applying the recommended correction to the GSP-Spec log g (dotted lines in Fig. 15), the offsets and main trends are highly mitigated. The spreads in the comparisons are roughly around ±0.1 dex in GSP-Phot and up to ±0.2–0.3 dex in GSP-Spec (before correction). The GSP-Phot estimates show good agreement for the subgiants, and most of the dwarfs, and disagreements at the level of up to 0.3 dex are found for the very high (> 4.5) and low (< 2) log g stars. Again, we note that the surveys agree with each other to approximately 0.1–0.2 dex over most of the log g range.

For metallicity, we use [Fe/H] as an indicator in order to be able to compare with most other surveys, which we computed from [M/H] and [α/Fe] using the formula provided by Salaris et al. (1993). Again, we note a better agreement of the GSP-Phot parameters with the surveys than for the GSP-Spec ones in terms of median offset, which is about zero dex for GSP-Phot and 0.1 dex for GSP-Spec. This was also reported by Recio-Blanco et al. (2023); see their Eqs. (3) and (4). The spreads are of about 0.10–0.15 dex in both cases, which is more than reasonable. We note that the surveys themselves tend to agree with each other to 0.1 dex or better. There is a tendency of both the GSP-Spec and GSP-Phot parameters to overestimate the [Fe/H] of metal-poor stars and to underestimate it for metal-rich ones. This effect has been observed in several other projects where the parameters were derived from low- or medium-resolution spectroscopy or photometry.

In conclusion, the overall agreement with the main spectroscopic surveys is good, but there are substructures in the comparisons that need to be kept in mind. Additionally, depending on the type of stars, we note that the GSP-Spec parameters do not necessarily produce a better agreement with the survey results compared with the GSP-Phot ones, and the use of the GSP-Spec log g and [M/H] corrections (Recio-Blanco et al. 2023) is recommended. This is in part due to the fact that the RVS spectral range extent and resolution are limited, but also to the fact that we are dealing with a high-S/N regime free from major systematic problems, where both the GSP-Spec and GSP-Phot perform close to optimal.

4.5.3. Validation with the PLATO input catalogue

We cross-matched our source list with the PLATO input catalogue (PIC) version 1.1 (Montalto et al. 2021) and obtained 10 828 common sources. In Fig. 16, we compare T_eff, R, and M (in the sense Gaia – PIC) normalised by the combined uncertainties added in quadrature. We also show the ±3σ lines, which show good agreement between the catalogues, but some insignificant artefacts for the comparison of masses. While we show the differences in terms of σ, we report the median (MD) and the MAD of their differences in absolute values in each panel. The agreement with T_eff is similar to that reported in the previous sections, where the GSP-Phot T_eff is on average 50 K lower. There are no matches with the GSP-Spec sources. Radius and mass differences are on the order of 1% and 6%, respectively.

Fig. 16. Difference between Teff (top), R (middle), and M (bottom) from Gaia for the FGKM sample and the PICv1.1 catalogue values normalised to their combined uncertainties for stars in common. We overlay the ±3σ lines. On each panel we also give the median difference (MD) and the MAD in K, R⊙, and M⊙, respectively.

In conclusion, we have compiled a clean sample of 3 273 041 FGKM stars comprising main sequence, subgiant, and giant stars. This sample was selected using many Gaia-based indicators along with GSP-Phot- and GSP-Spec-based astrophysical parameters. The APs of interest are T_eff, log g, [M/H], A_G, E(G_BP − G_RP), L, R, M, age, and spectral type, and we provide a separate table of these parameters in the Gaia archive. We have not applied any calibration or correction to the values in Gaia DR3, but we have filtered some parameters for some sources. We validated our selection by comparing with parameters from clusters and other surveys. These comparisons reveal typical offsets of < 100 K in T_eff from other surveys. In Sect. 10.2 we exploit the T_eff, radius, and mass of this sample in order to analyse known exoplanet systems, and in Sect. 10.4 we analyse the ages of 11 unseen UCD-companions. A user could further filter by selecting in a specific T_eff range, or by excluding distances above a certain threshold, or by providing an upper limit to the amount of extinction between the observer and the star.

5. Ultracool dwarfs

5.1. Scientific motivation

Ultracool dwarfs (UCDs) are objects at the faint end of the main sequence, and defined in Kirkpatrick et al. (1997) as sources with spectral types M7 or later. This definition includes the coolest hydrogen-burning stars and brown dwarfs. Even though brown dwarfs can sustain lithium or deuterium fusion at their cores for a short period of time in the early phases of their evolution (Burrows et al. 2001), the nuclear reactions stop by the time they reach the main sequence and the stars keep cooling and fading thereafter. Despite the fundamental differences in the internal structure across the stellar and substellar regimes, the atmospheric properties overlap at this boundary and it becomes very difficult to distinguish between the two regimes based on photometric or spectroscopic properties. In this section, we define a high-quality sample of UCDs which we propose as excellent candidates with which to advance our knowledge of such low-mass objects. To complement the T_eff of UCDs in Gaia DR3 we provide a catalogue of radii and luminosities by complementing the Gaia data with infrared photometry and we explore the existence of a minimum in the mass–radius relation slope (e.g. Dieterich et al. 2014; Smart et al. 2018; Cifuentes et al. 2020).

5.2. Sample selection

Our initial sample of UCD candidates is from the astrophysical_parameters table where a total of 94 158 sources have been processed as UCD candidates and we estimate teff_espucd. We remind readers that the main difference with respect to existing compilations of UCD candidates (for example Reylé 2018) is the use of the Gaia DR3 RP spectra to produce T_eff and to help define the selection criteria as described in the Gaia DR3 online documentation. We also imposed that the first digit of flags_espucd = 0 or 1 (the most reliable categories), which gives a total of 67 428 candidates. We then required that the Gaia astrometric flags fulfil the following conditions: ruwe < 1.4, ipd_frac_multi_peak = 0 and ipd_gof_harmonic_amplitude < 0.1 to reduce contamination by unresolved binaries. We then selected sources with a cross-match (as provided in the Gaia archive) in the 2MASS (Skrutskie et al. 2006) and AllWISE (Wright et al. 2010; Mainzer et al. 2011) catalogues, with available measurements in the J, H, and K_s 2MASS bands, and the W1 and W2 AllWISE bands, all with quality A flags. The W3 band was not included as a requirement because the lack of measurement uncertainties reduces the number of sources drastically. Finally, we remove sources above the G + 5log₁₀(ϖ)+5 = 3 + 2.5 (G − J) line to avoid including suspected low-gravity UCDs that have not yet contracted and reached equilibrium. This gives a total of 31 822 candidates for this study.

We used the virtual observatory VOSA (Bayo et al. 2008) to calculate the minimum reduced χ² fits between the constructed SEDs from the Gaia G and G_RP bands and the infrared photometry, and the CIFIST 2011_2015 BT Settl models (Allard et al. 2012). We retain the sources whose reduced ${\chi }_r^2<100$. We allow for rather large values of ${\chi }_r^2$ in order to account for the known discrepancies between the models and observations, and the discrete nature of the model library. The distribution of log₁₀(${\chi }_r^2$) is approximately normal and 96.5% of all the values are below the imposed threshold which therefore only removes obvious pathological fits. The final sample has a total of 21 068 sources.

5.3. Combining Gaia with external data to derive R and L

R and L are parameters that are also calculated by the FLAME module and available in the astrophysical_parameters table. However, these are only available for sources with T_eff > 2500 K. A comparison of the values for the sources in common is discussed in the following section. We computed bolometric fluxes using Gaia and IR photometry. To account for the unobserved flux outside the observed wavelength bands, we needed to calculate bolometric corrections. We used the CIFIST 2011_2015 BT Settl models again in order to calculate the ratio of observed to total flux for the aforementioned set of photometric bands⁴. This produces a theoretical flux correction factor f_obs/f_total for the T_eff range between 1200 and 2700 K in steps of 100 K. For each of the UCD candidates with full photometry, we obtain the correction factor by interpolating the T_eff value derived by the ESP-UCD module in this grid. The resulting corrections are in the range between 0.48 and 0.54 mag with a median value of 0.53 mag. We use this ratio to infer the total flux that would be observed at the Earth and derive the bolometric luminosity using the Gaia parallax measurement. Finally, using the ESP-UCD T_eff estimate and the bolometric luminosity, we inferred radii for the UCD candidates using the Stefan-Boltzmann law. Figures 17 and 18 show the scatter plot of the inferred radii and luminosities as a function of the ESP-UCD T_eff. The uncertainties (only shown for sources cooler than 1900 K to aid readability) were calculated using a simple Taylor expansion and neglecting correlations amongst the intervening variables.

Fig. 17. Radii of candidate UCDs in the Gaia golden sample. The colour code indicates the logarithm of the VOSA fit χ2 values, squares represent the data points in Table 1 of Dieterich et al. (2014), and black asterisks denote unresolved binaries therein. The box plots are calculated within bins of 100 K.

Fig. 18. Bolometric luminosities of candidate UCDs in the Gaia golden sample. The colour code indicates the logarithm of the VOSA fit χ2 values, squares represent the data points in Table 1 of Dieterich et al. (2014), and black asterisks denote unresolved binaries therein.

To fully exploit this UCD golden sample, we provide an accompanying table in the Gaia archive, gold_sample_ucd, which lists source_id, the correction factor to calculate the bolometric flux, radius, luminosity, and uncertainties, along with the ${\chi }_r^2$ value. This table can be used with the teff_espucd provided in the astrophysical_parameters table.

Fig. 19. Comparison of the radii estimated for the UCD sample by the ESP-UCD (x axis) and FLAME (y axis) modules for the sources in common. The colour code reflects the effective temperature used by the FLAME module to estimate the radii.

5.4. Validation

In Fig. 19 we compare the radii values of sources with estimates from the FLAME and ESP-UCD modules. This comparison reveals remarkable agreement for the lowest temperature regime (T_eff < 2600 K) but also shows evidence for a systematic difference in the larger FLAME radii above. This is due in part to a difference of approximately 85 K in the temperatures used for the derivation of radii (i.e. the T_eff used by FLAME – from GSP-Phot – are hotter than the ones produced by the ESP-UCD module).

Figure 17 shows the expected decrease in radius as the temperature decreases down to temperatures of the order of ≈2200 − 2000 K. The radius then increases for even cooler temperatures until T_eff ≈ 1400 K where the trend reverses and the slope becomes positive again.

In Fig. 18 we can see a systematic difference between the luminosities estimated by Dieterich et al. (2014; represented by the black squares) and the ones from this work in the range T_eff > 2000 K. This difference translates into an offset in radius in Fig. 17. The offset in luminosity could be due to (1) a difference in the T_eff estimates if our temperatures were systematically cooler than those of Dieterich et al. (2014) in that regime and/or (2) a difference in the calculation of the bolometric correction (BC) if BCs derived by Dieterich et al. (2014) produce bolometric luminosities systematically fainter than the ones derived here. We examine the two alternatives more closely in the following paragraphs.

Fig. 20. Comparison of the effective temperatures used to derive radii in this work (x-axis) and those used in the literature (y-axis) for the UCD sample. Black filled circles denote sources from Cifuentes et al. (2020) and orange filled circles denote those from Dieterich et al. (2014).

Figure 20 shows a comparison of the temperatures used by Dieterich et al. (2014) and Cifuentes et al. (2020) to infer radii with those estimated by the ESP-UCD module. This comparison suggests a systematic difference of approximately 65 K above T_eff ≈ 2200 K. This is different from but consistent with the difference encountered in the comparison with the FLAME outputs.

The ESP-UCD T_eff are based on an empirical training set built from the Gaia ultracool dwarf sample (GUCDS; Smart et al. 2017, 2019) and the spectral type–T_eff relation by Stephens et al. (2009). The values derived by the ESP-UCD regression module were calibrated as described in the Gaia DR3 online documentation to account for a discrepancy that was found with respect to the regression module trained on BT-Settl models. The RP spectra, which were simulated from the BT-Settl library of synthetic spectra, were found to nicely reproduce the observed RP spectra in this T_eff regime. Also, the calibrated temperatures were found to produce relatively good agreement with the SIMBAD spectral types where available (again, using the relations by Stephens et al. 2009) as illustrated in the validation of the ESP-UCD module in Fouesneau et al. (2023). However, in view of the comparisons described above, it is not implausible that the correction applied in the calibration of the results from the empirical training set was overestimated by an amount of the order of 65 K. In any case, the systematic difference in effective temperatures explains part but not all of the discrepancy in the luminosities and radii. Hence, we suspect that this discrepancy may also be caused by differences in the corrections applied to the observed fluxes to derive bolometric luminosities. Our procedure to estimate the bolometric luminosity is different from that used by Dieterich et al. (2014) and this could be the source of the systematic difference in the luminosities above 2000 K apparent in Fig. 18. While we directly interpolate the fraction of the total flux emitted in the photometric bands on a grid of BT Settl models, Dieterich et al. (2014) apply a wavelength-dependent correction to the BT Settl models such that they agree with the observed photometric magnitudes before that fraction is estimated. As a direct comparison is not possible because of the unavailability of their correction factors, we cannot discard this different procedure as a potential explanation of the difference.

The overall trends of decreasing radii down to ∼2000 K and slowly increasing radii for even cooler temperatures are confirmed with the Gaia data, although the associated uncertainties are large. The final positive slope in the regime T_eff < 1400 K is also compatible with that shown in Cifuentes et al. (2020) but not predicted (to the best of our knowledge) by theoretical studies. The sample of UCDs used here can be expected to be a combination of different ages, masses, and metallicities (all of them with an impact on the effective temperatures and radii), and therefore it is not straightforward to draw any direct conclusions about these fundamental parameters from Fig. 17.

In summary, we provide a catalogue of 21 068 UCDs that we consider to be of very high quality from the available sources in the astrophysical_parameters. We derive their luminosities and radii by calculating bolometric corrections and make these new parameters available in the accompanying gold_sample_ucd table.

6. Carbon stars

6.1. Scientific motivation

A high number of asymptotic giant branch (AGB) stars have carbon-enriched atmospheres and show C₂ and CN molecular bands that are stronger than usual in stars cooler than 3800 K (i.e. G_BP − G_RP ≥ 2). The origin of the enrichment can be due to mass transfer in binary systems or to pollution by nuclear He fusion products from the inner to the outer layers. Because they belong to a late stage of stellar evolution where mass loss occurs, and which precedes the formation of the planetary nebula, carbon stars are important contributors to the interstellar medium and provide good reference cases with which to study the physical processes affecting the end of the life of low-mass stars. During the Gaia DR3 development and processing, no synthetic spectra showing such high carbon abundances were included in the simulations that are used in the Apsis software to produce APs (Creevey et al. 2023). The spectral libraries used as templates to derive the astrophysical parameters from BP and RP, as well as those adopted to measure the radial velocities, are therefore not fully adapted to analysing the data for carbon stars. Therefore, an attempt was made by the ESP-ELS module to flag suspected carbon stars.

6.2. Sample selection

The identification of candidate carbon stars by ESP-ELS is based on a Random Forest classifier trained on the synthetic BP and RP spectra as well as on the observed Gaia data obtained for a sample of Galactic carbon stars (Abia et al. 2020). This identification is saved in the spectraltype_esphs field of the astrophysical_parameters table. In total, 386 936 targets received the ‘CSTAR’ tag. While most of these stars are M stars, only a smaller fraction of the sample exhibit significant C₂ and CN molecular bands. To identify these cases, we measured the band head strength as follows:

$$\begin{array}{c}\hfill R_{{\lambda }_2}=\frac{f({\lambda }_2)}{g_{{\lambda }_1,{\lambda }_3}({\lambda }_2)},\end{array}$$(2)

where f(λ₂) is the flux measured at the top of the band head of the molecular band, and g_λ1, λ3 the value linearly interpolated between wavelengths λ₁ and λ₃. The four band heads we considered are described in Table 5. These were computed for a random sample of 27 528 stars with G_BP − G_RP (not dereddened) colours uniformly distributed between 1 and 5 in order to locate the range of R_773.3 and R_895.0 values occupied by non-carbon stars. The upper limit of the interquantile dispersion (2.7% and 97.3%) is the threshold below which the targets providing the weakest values are excluded (i.e. it provides one lower threshold on R_773.3, and one on R_895.0).

Fig. 21. Band head strengths (see Eq. (2) and Table 5) measured in the BP and RP spectra of known Galactic (MW, orange points, Alksnis et al. 2001), Large Magellenic Cloud (LMC, black points, Kontizas et al. 2001), and Small Magellenic Cloud (SMC, green points, Morgan & Hatzidimitriou 1995) carbon stars. Only targets within 1 arcsec of a Gaia DR3 source_id are taken into account. Upper panels: locus occupied by non-carbon stars represented by the blue shaded area. Middle and lower panels: targets with weaker or non-existing CN features shown with blue points (i.e. they fall in the shaded areas of the upper panels). The pink broken and full lines delimit the domain occupied by 87% and 98% of the carbon stars with strong CN features, respectively.

Fig. 22. Same as Fig. 21 but for the 386 936 candidate carbon stars flagged by ESP-ELS. Pink curves represent the domain occupied by the carbon stars found in the literature (Fig. 21, and Sect. 6.2).

Figures 21 and 22 show the results obtained for known carbon stars and candidate carbon stars flagged by the ESP-ELS module, respectively. Most of the 386 936 candidate carbon stars (upper panels of Fig. 22) flagged by the algorithm have G_BP − G_RP > 2 mag, and have colours consistent with M stars. However, the known carbon stars, especially those in the Magellanic clouds, have colours down to ∼1 mag. A significant fraction of these have therefore not been detected and are not part of the golden sample. Our proposed sample of carbon stars is obtained after applying the lower thresholds on both R_773.3 and R_895.0 ratios.

6.3. Validation of sample

Fig. 23. Mollweide view in Galactic coordinates of the carbon stars sample described in this work. The locations of the Magellanic Clouds and the Sagittarius stream are shown in blue.

Fig. 24. Magnitude and colour distribution of carbon stars. Left panels: vertical black dashed line shows the upper magnitude limit of the data processed by ESP-ELS. Upper panels: all the targets belonging to the golden sample of carbon stars are taken into account. Other panels: distributions obtained for the known MW (Alksnis et al. 2001), LMC (Kontizas et al. 2001), and SMC (Morgan & Hatzidimitriou 1995) carbon stars are shown in blue. In orange, we show the distribution of the targets in common with the sample we propose in this work.

The sample we propose includes 15 740 stars exhibiting the strongest CN molecular bands. Their spatial distribution is shown in Fig. 23. As previously noted, most of the remaining carbon stars have G_BP − G_RP > 2 mag, which is consistent with what is expected from M-type stars (Fig. 24). From a cross-match with the three main catalogues of carbon stars, about two-thirds are known cases. The magnitude and colour distributions of the targets found in the literature and in common with the proposed sample are shown in Fig. 24. Most of the carbon stars that have not been identified correspond to targets bluer than G_BP − G_RP = 2 or/and fainter than G = 17.65 mag. Taking magnitude and colour/T_eff constraints into account, the fractions of detected known carbon stars are shown in Table 6.

Carbon stars are located at the very cool edge of the GSP-Phot T_eff domain (T_eff > 2500 K). In addition, no synthetic spectra adapted for the accurate AP determination of carbon stars were available, and only a fraction of the carbon stars have their astrophysical parameters published in GDR3. Therefore, it is not surprising that the obtained T_eff tends to be overestimated (by 500 to 1500 K) and should be considered with caution. However, the Kiel diagrams obtained for the known carbon stars (Fig. 25, left panel) and those from our list (same figure, right panel) are consistent with each other. Nevertheless, the estimated T_eff and their location in the diagram are also consistent with AGB stars. We note that a few targets (254) have T_eff hotter than 6000 K, while the corresponding SEDs are typical of AGB carbon stars (showing typical CN bands in the RP) as shown in Fig. 26.

Fig. 25. Kiel diagram of carbon stars with published GSP-Phot parameters. Left panel: density plot for known MW, SMC, and LMC C stars. Right panel: density plot for the carbon stars in our sample.

Fig. 26. BP and RP spectra of the 20 randomly chosen (amongst 254) carbon stars (this work) with Teff GSP − Phot > 6000 K. The ordinate axis provides the flux normalised to the total flux, and shifted by k × 0.003 (where k is an integer that varies from 0 to 9 from the bottom to the top spectrum).

To exploit this sample, the list of source_id is made available as a separate table in the archive gold_sample_carbon_stars for the 15 740 bone fide carbon stars, which were also flagged in the main astrophysical_parameters table (see flags_esphs for details). The initial set of 386 936 carbon-candidate stars can still be found in the same table, as these remain tagged as ‘CSTAR’ in the https://gea.esac.esa.int/archive/documentation/GDR3/Gaia_archive/chap_datamodel/sec_dm_astrophysical_parameter_tables/ssec_dm_astrophysical_parameters.html#astrophysical_parameters-spectraltype_esphs field.

7. Solar analogues

7.1. Scientific motivation

The Sun is the reference point in much of stellar astronomy and astrophysics. Solar analogues are stars that resemble the Sun in terms of a restricted set of parameters. In contrast to the Sun, they can be observed in the night sky and with the very same instruments used to study stars in the Milky Way. There is no strict definition of what constitutes a solar analogue; both the set of parameters and allowed parameter ranges vary in the literature. For astrophysical purposes, one often aims to constrain the photometrically and spectroscopically accessible parameters T_eff, log g, and overall metallicity [M/H] to within typical measurement uncertainties. Depending on data quality and analysis technique applied, uncertainties as small as 10 K in T_eff, 0.03 in log g, and 0.01 in [M/H] are achievable⁵ (Yana Galarza et al. 2021), but 50 K, 0.15, and 0.05 are more typical values. These small errors are the result of line-by-line differential analyses relative to the Sun, a technique which cancels many of the systematic sources of error that stellar analyses otherwise often suffer from.

The most accurate analyses have revealed systematic differences in the chemical composition of the Sun relative to solar analogues in the solar neighbourhood: When selected to be good matches in [Fe/H] (iron abundance), the Sun is among the 10%–15% of stars that are rich in volatile elements (Meléndez et al. 2009). A tight (broken) trend of abundance with condensation temperature of the various elements is found with an amplitude of 0.08 dex (20% in linear abundance). The reason for this effect is still unknown, but has been proposed to be related to selective accretion of gas over dust because of the presence of planets. This finding potentially opens up new avenues for systematic evolutionary studies of solar-type stars and their planets.

Solar analogues have also been used to identify the abundance ratios that depend most sensitively on stellar age and can therefore serve as precise spectroscopic clocks. One such study identifies the [Y/Mg] abundance ratio as particularly age sensitive (Nissen 2015). Working with ages rather than metallicity as a proxy for age puts chemical-evolution studies on a much firmer footing. Loosening constraints on the stellar parameters, one can also study ‘the Sun as a star’ and its evolution.

Finally, solar analogues also serve a purpose in the study of minor bodies of the Solar System. In this context, they are used to subtract the solar spectrum (and earth-atmospheric contributions in the case of ground-based observations) from reflectance spectroscopy of asteroids, for example, with the aim being to obtain a more uniform classification; see for example the sso_reflectance_spectrum table in Gaia DR3 and Gaia Collaboration (2023c). We note that this type of science case requires stars whose SED resembles that of the Sun as closely as possible; however, a perfect match in stellar parameters is not necessarily needed, especially if one considers fainter G dwarfs which may suffer from extinction and associated reddening. A scientific application of solar analogues is presented in Sect. 10.

7.2. Candidate selection

In order to identify candidate solar analogues from the full Gaia sample, we need to define selection criteria. We apply two general criteria: (i) Apparent magnitude brighter than G < 16 since fainter sources would be difficult to follow up efficiently with ground-based spectroscopy. (ii) Excellent parallax quality, ϖ/σ_ϖ > 20, in order to reliably place sources in the HR diagram. From these basic criteria, we continue to select candidates from GSP-Spec results. On a known sample of solar analogues and twins (Porto de Mello et al. 2014; Ramírez et al. 2014; Nissen 2015; Mahdi et al. 2016; Tucci Maia et al. 2016; Lorenzo-Oliveira et al. 2018; Giribaldi et al. 2019; Casali et al. 2020; Yana Galarza et al. 2021), GSP-Spec on average deviates from solar values by 14.4 K in T_eff, by −0.071 in log g, and by −0.05 in [M/H] (Fouesneau et al. 2023, Sect. 4.1 therein). Taking those average differences into account, we require that GSP-Spec results agree with 5772 K to within 100 K, to log g = 4.44 to within 0.25, and to [M/H] = 0 to within 0.2. Furthermore, we require good GSP-Spec flags⁶. Finally, we combine GSP-Spec results with FLAME estimates to further clean the sample of possible contamination: First, we require that mass_flame_spec is between 0.95 M_⊙ and 1.05 M_⊙. Second, we require that radius_flame_spec is between 0.8 R_⊙ and 1.2 R_⊙. This results in a total of 5863 GSP-Spec candidates for solar analogues, of which 916 have RVS spectra published in Gaia DR3.

GSP-Phot results can also be used to select candidates. Here, we only consider PHOENIX and MARCS in the context of solar analogues. We require that GSP-Phot results for both libraries agree with 5772 K within 100 K, to log g = 4.44 within 0.25, and to [M/H] = 0 within 0.2, where we correct results from each library for its mean differences from known solar analogues (Fouesneau et al. 2023, Sect. 4.1 therein). This results in a total of 234 779 GSP-Phot candidates for solar analogues, 7884 of which have RVS spectra. However, we do not publish this candidate list. Interested readers may contact the authors. The list of Gaia DR3 source IDs for the 5863 solar-analogue candidates from GSP-Spec is provided in the Gaia archive as a separate table gold_sample_solar_analogues.

Because of the selection on very high parallax quality (ϖ/σ_ϖ > 20), the candidates tend to be nearby and thus scatter more or less uniformly over the whole sky. Nevertheless, the sky distribution shows the imprint of the Gaia scanning law, because high parallax quality is easiest to achieve for sources with many transits.

In Fig. 27, we check and verify that the solar-analogue candidates have [α/Fe] abundances that are statistically consistent with the solar value of zero. The standard deviation of [α/Fe] for this particular subset of solar-like stars is 0.056, which is lower than the global uncertainty reported for all stellar types in Recio-Blanco et al. (2023).

Fig. 27. Distribution of [α/Fe] abundances from GSP-Spec for solar-analogue candidates. Grey shows the raw alphafe_gspspec values and black shows the calibrated values (Recio-Blanco et al. 2023). The dashed red line shows a Gaussian distribution with a mean of –0.028 and a standard deviation of 0.056.

7.3. RVS spectra of candidates

For visual confirmation of the candidate selection, we inspect the published RVS spectra of the candidates. For comparison, we also take the RVS spectra of 13 known solar analogues which have RVS spectra published in Gaia DR3. Figure 28a shows that the 916 GSP-Spec candidates with RVS spectra are in excellent agreement with the mean RVS spectrum of known solar analogues. Most of these 916 GSP-Spec candidates are brighter than G < 11.7. The 1985 candidates with RVS spectra one would obtain from GSP-Phot and that are similarly bright (G < 11.7) are shown in Fig. 28b, and show equally good agreement with the known solar analogues shown in Fig. 28a. This demonstrates that GSP-Phot results are also very reliable under these selection criteria. For orientation, Fig. 28c shows RVS spectra of 7589 random stars with G < 11.7 and here we see clear differences; for example the Ca lines vary in depth, where for hot stars in particular the Ca lines are usually weak and instead Paschen lines start to appear. In Fig. 28c, we can also see the DIB around 860 nm (Gaia Collaboration 2023b).

Fig. 28. RVS spectra of 916 solar-analogue candidates from GSP-Spec (panel a) where 95% of GSP-Spec candidates satisfy G < 11.7. We also show the solar-analogue candidates obtained from a possible selection from GSP-Phot results in panel b, but we only show 1985 GSP-Phot candidates with G < 11.7. For comparison, panel c shows RVS spectra of 7589 randomly selected stars (i.e. no solar-analogue candidates) also with G < 11.7. In all panels, the red line shows the median in each pixel and the shaded red contours show the pixel-wise central 68% and 90% intervals. The solid blue line is identical in all three panels and shows the mean RVS spectrum of 13 solar analogues known from the literature.

7.4. Candidates with extinction

Solar analogues with notable extinction would be of particular scientific interest; for example for inferring the extinction law. In Fig. 29, we show colours of GSP-Spec candidates including photometry from Gaia and AllWISE (Cutri et al. 2021) as a function of the GSP-Phot A₀ estimate. The G − W₁ colour clearly reddens with increasing A₀ in Fig. 29a whereas the W₁ − W₂ colour remains virtually constant in Fig. 29b⁷.

Fig. 29. Colours of GSP-Spec solar-analogue candidates as a function of GSP-Phot extinction estimates. W1 and W2 denote AllWISE photometry (Cutri et al. 2021). We restrict the comparison to candidates with W1 > 8 mag, because AllWISE photometry suffers from saturation for brighter sources. Panel c: the red line is a linear increase with abp_gspphot offset by the mean GBP − W2 colour of 589 stars where A0 < 0.001 mag according to GSP-Phot. The red interval marks the uncertainty from the standard deviation of the mean. The quoted root-mean-square (RMS) difference is between the GBP − W2 colour and abp_gspphot plus the mean.

In Fig. 29c, we further investigate the reddening of the G_BP − W₂ colour, which has the largest wavelength coverage from the near-UV (320–670 nm for G_BP) to the near-infrared (4.6 μm for W₂). In particular, G_BP will be much more affected by extinction than W₂; indeed A_BP ≫ A_W2, such that we can take the GSP-Phot A_BP estimate as an approximation for the reddening of the G_BP − W₂ colour. Indeed, Fig. 29c shows a linear relation with a low RMS deviation of 0.087 mag across an A_BP range of 1.75 mag. This attests to the quality of the A_BP estimate from GSP-Phot (at least for bright sources with high-quality parallax measurements).

Having established in Fig. 29 that the GSP-Phot extinction estimates agree with the reddening of colours of the candidates, we inspect how the low-resolution BP and RP spectra themselves vary with GSP-Phot extinction. For the 5863 GSP-Spec candidates, Fig. 30 shows that the BP and RP spectra clearly redden and dim as the GSP-Phot estimate of A₀ increases. While BP and RP spectra at low extinction show much more flux in BP than in RP, BP and RP spectra at A₀ ∼ 1.5 mag already show equally high peak fluxes in both BP and RP while their overall flux is reduced by a factor of approximately five in BP and about three in RP with respect to a zero-extinction solar-like BP/RP spectrum.

Fig. 30. Variation of low-resolution BP and RP spectra of GSP-Spec solar-analogue candidates with the GSP-Phot A0 estimate. In order to make the BP and RP spectra comparable, they have been rescaled to an apparent magnitude of G = 15 + AG with AG taken from GSP-Phot.

8. SPSS

The Gaia SPSS⁸ are a grid of flux tables specifically designed to calibrate Gaia photometry and BP and RP spectra. They are the result of a dedicated set of ground-based observing campaigns to collect spectrophotometry (Altavilla et al. 2015), light curves for constancy monitoring (Marinoni et al. 2016), and absolute photometry for validation (Altavilla et al. 2021) over more than ten years. The latest version of the grid, SPSS V2, was used to calibrate the Gaia photometry in EDR3 and the BP and RP spectra in DR3 and contains 111 stars⁹ based on ≃1500 spectra, and is calibrated on the 2013 version of the CALSPEC¹⁰ grid (Bohlin et al. 1995, 2019; Bohlin 2014) with a zero-point accuracy of better than 1%. The SPSS grid is designed to cover those areas of the stellar parameter space that are not well sampled by CALSPEC, in particular the FGKM star types, and to cover the entire Gaia wavelength range (330–1050 nm). The final release, SPSS V3, will be used to calibrate Gaia DR4, it will contain about 200 stars and will make full use of all the ≃6500 spectra collected in the observing campaigns. It will be calibrated on the latest version of the CALSPEC grid (Bohlin 2014; Bohlin et al. 2019), which differs by about < 0.5% from the 2013 one in the grid zero point. The S/N of the ground-based SPSS spectra is generally well above 100, with the exception of the blue and red extremes of the Gaia wavelength range. The SPSS flux tables were therefore extended with theoretical spectra and adjusted to match the central high-S/N region of the observed spectra (see Pancino et al. 2021, for details). It is therefore of the utmost importance for the next SPSS release to have a robust estimate of the spectral type, atmospheric parameters (T_eff, log g, [Fe/H], and [α/Fe]), and of the interstellar absorption for as many of the SPSS as possible.

To this aim, we explored and selected relevant information from the astrophysical_parameters table of Gaia DR3 for the SPSS V2 stars. In particular, whenever available, we selected the GSP-Spec parameters over the GSP-Phot ones for T_eff, log g, and [Fe/H]. To obtain [Fe/H] from the GSP-Phot [M/H] estimates, we used the formula by Salaris et al. (1993) and assumed an α-enhancement of +0.35 for metal-poor stars, +0.15 for intermediate metallicities, and zero for solar or higher metallicity. We note that we did not apply any re-calibration to the log g and [Fe/H] GSP-Phot and GSP-Spec values. Similarly, for the choice of the FLAME parameters, that is, mass, age, luminosity, and radius, we always selected the corresponding FLAME-spec determinations when available (in the astrophysical_parameters_supp table). Parameters were available for all the SPSS in the sample, except for the 56 white dwarfs. For hot stars, a handful of parameters from ESP-HS were available that were not parametrised by GSP-Phot or GSP-Spec. The two binarity estimators available (specmod and combmod) agreed on indicating SPSS 028 (SA105-663) as a binary, while 15 different SPSS were indicated as photometrically variable (phot_variable_flag) and will be carefully re-evaluated in the preparation of the SPSS V3 release.

To explore the quality of the results, we compared them with the two sets of parameters presented by Pancino et al. (2021): (1) a collection of literature estimates and (2) the best-fit parameters obtained by extending the SPSS V2 flux tables with model libraries. First, we compared the spectral type determinations and find that only 16 SPSS out of 111 have discrepant spectral types, and in all cases the discrepancies never span more than one spectral class (e.g. an F star classified as G). For one star, SPSS 313 (M5–S1490), discordant previous literature spectral type determination (from A to F) was available, and we find it to be a K giant, about 500 K cooler than the coolest literature determination. We then compared the three main atmospheric parameters (Fig. 31) with both reference sets. As can be seen, apart from a very small number of outliers, the agreement with the two sets of reference parameters appears good, especially when considering the heterogeneity of the literature estimates. There is an indication that a few stars with A₀ ≳ 1 mag have problems in some of the parameters. However, for the majority of stars, the agreement for T_eff and log g is excellent, with median differences of ΔT_eff = –4 ± 322 K and Δlog g = –0.04 ± 0.59 dex. The comparison of [Fe/H] is still good if one includes metal-poor stars, with Δ[Fe/H] = 0.15 ± 0.61 dex. When excluding stars below [Fe/H] ≃ –2 dex, which appear to have an overestimated iron metallicity, the comparison improves, with Δ[Fe/H] = –0.09 ± 0.44 dex. We note that an overestimate for metal-poor stars is a common problem when metallicities or iron abundances are derived from photometric data or low-resolution spectra (see also Miller 2015; Anders et al. 2022; Xu et al. 2022).

Fig. 31. Comparison of the SPSS sample main parameters derived here with the two reference sets by Pancino et al. (2021): the best-fit parameters to the SPSS flux tables are shown in grey, while a collection of literature spectroscopic estimates is coloured according to the interstellar absorption A0 obtained here. Left, middle, and right panels: the cases of Teff, log g, and [Fe/H], respectively. The 1:1 line is shown in green in all panels.

In Gaia DR3, we present a table gold_sample_spss, which contains the 111 SPSS stars, and for each source, their Gaia DR3 source_id, name, spectral type, binary and variability flags, along with the stellar parameters, extinction, distance, radial velocity, and v sin i (where available) for the 52 non-subdwarf and white dwarf stars of the sample.

9. Summary of golden samples

In Sects. 3– 8, we define several samples of stars that are carefully selected to be homogeneous and of the highest quality, and can be used in many different astrophysical contexts. Complementary data tables are made available in the Gaia DR3 archive to help in exploiting these samples; see here in the online documentation. Table 7 summarises the names, sizes, and contents of these tables, and here we provide an overview.

The six tables are all entitled gold_sample_name where name is specific to the sample; that is, oba_stars, fgkm_stars, carbon_stars, solar_analogues, spss, and ucd. These can be called in an ADQL query in Gaia DR3 as gaiadr3.gold_sample_name.

The tables for the solar analogues and the carbon stars contain the source_id only. The OBA table also includes a flag that allows one to apply a kinematic filter. As well as source_id, the table for the UCDs contains the newly derived radii and luminosities from the analysis of the Gaia and infrared data, and the bolometric flux correction. The SPSS sample table contains all 111 SPSS sources along with information such as binary and variability flags, radial velocity, and v sin i. The stellar parameters and extinction are given for the non-white dwarf stars, some of which are based on GSP-Spec parameters and others on GSP-Phot or on ESP-HS; this is indicated by the notes in that table. Finally, for the FGKM sample, a table with source_id, atmospheric parameters, evolutionary parameters, and spectral type is provided, where specific parameters for some sources have been removed (compared to the astrophysical_parameters table).

10. Exploitation of the golden samples

In this section, we demonstrate four applications of the golden samples presented in this paper. For the first application we exploit the OBA sample to derive the parameters of the Milky Way rotation curve and the peculiar motion of the Sun. We then use the FGKM sample to characterise known transiting exoplanets. This is followed by an exploitation of the solar analogue sample to derive the colours of the Sun, and finally we use the stellar companions of unseen UCDs to explore the ages of these substellar systems.

10.1. Milky Way rotation curve

A classical application for the OBA star sample is to infer the parameters of the Milky Way rotation curve near the Sun. Young disc stars have often been used for this purpose because of the low dispersion of their velocities around the overall differential rotation of stars in the thin disc (for a recent example based on Gaia EDR3 data, see Bobylev & Bajkova 2022). We illustrate this application with a very simple modelling of the proper motions in terms of the Milky Way disc rotation curve. The rotation curve is described with the circular velocity and the slope of the circular velocity as a function of Galactocentric cylindrical distance R, both evaluated at the position of the Sun (or equivalently, the Oort constants A and B for an axisymmetric Milky Way, see e.g. Olling & Dehnen 2003). We use a subsample of the OBA stars, namely those with spectraltype_esphs equal to ‘B’, with ϖ/σ_ϖ > 10 and v_tan < 180 km s⁻¹. The sample is further restricted to (1000/ϖ)× sin b < 250 pc and 6.5 < R < 15 kpc. Over this range in R, the above approximation to the rotation curve is reasonable (see e.g. Eilers et al. 2019, their Fig. 3). Figure 32 shows the proper motions in ℓ and b as a function of Galactic longitude for the 385 423 B-stars in this sample. The figure beautifully reveals the variation of μ_ℓ* with cos(2ℓ), a consequence of Galactic differential rotation, and shows the slight offset of the proper motions in latitude from zero, reflecting the Sun’s motion perpendicular to the Galactic plane. The width of the proper motion distributions mainly reflects the range of distances to the stars in the sample.

Fig. 32. Proper motions in Galactic longitude (top) and latitude (bottom) as a function of Galactic longitude for the sample of 385 423 B-stars described in the text. The lines show the proper motions predicted from the rotation curve model parameters resulting from the fit to the data for stars at 500 pc (dashed line) and 2000 pc from the Sun (solid line, close to the median distance of stars in the sample).

To derive the rotation curve parameters, we use a Bayesian model for the proper motions in Fig. 32. The model has parameters similar to the simple kinematic model described in Sect. 3: the circular velocity at the position of the Sun V_circ, ⊙, the slope of the circular velocity curve dV_circ/dR, the peculiar motion vector of the Sun v_⊙, pec = (U_⊙, V_⊙, W_⊙), and the velocity dispersions in the plane and perpendicular to the plane, σ_xy and σ_z. The position of the Sun is fixed at a Galactocentric distance of 8277 pc (GRAVITY Collaboration 2022), while the height above the Galactic plane is taken as the median of −(1000/ϖ)× sin b for the B-star sample, which is 16 pc. The model velocities v of the stars are calculated from the azimuthal component of the velocity in Galactocentric cylindrical coordinates, V_ϕ = −(V_circ, ⊙ + dV_circ/dR × (R − R_⊙)), as v = ( − V_ϕ sin ϕ, V_ϕ cos ϕ, 0). Here we use a right-handed coordinate system, and so V_ϕ is negative for the stars in the disc of the Milky Way. The model proper motions μ_pred are then calculated from v − v_⊙, the parallaxes, and celestial positions of the stars (with v_⊙ = (U_⊙, V_⊙ + V_circ, ⊙, W_⊙)). The parallaxes were used as error-free observables. The velocity of the local standard of rest with respect to the rotational standard of rest is assumed to be 0 km s⁻¹ (Bland-Hawthorn & Gerhard 2016) and not included in the model.

The seven model parameters are optimised through a Markov chain Monte Carlo sampling of the posterior. The likelihood for the observed proper motions is a normal distribution centred on μ_pred with a covariance matrix that accounts for the covariance matrix of the observed proper motions and the velocity dispersions using the appropriate form of Eq. (16) in Lindegren et al. (2000). The priors on the model parameters are broad normal distributions centred on 220, 11, 12, and 7 km s⁻¹ for V_circ, ⊙, U_⊙, V_⊙, and W_⊙, respectively. The prior on dV_circ/dR is a normal distribution centred on 0 km s⁻¹ kpc⁻¹. The priors on the velocity dispersions are Gamma distributions with parameters α = 2 and β = 0.1. The model was implemented in Stan (Stan Development Team 2022) using the CmdStanPy interface. The posterior was sampled with four Markov chain Monte Carlo (MCMC) chains for 1500 steps each, and the first 500 steps were discarded as ‘burn-in’. To keep the required computational resources within bounds, the Stan model was run using a random subset of 20 000 stars chosen from the B-star sample above.

The resulting model parameters are: V_circ, ⊙ = 234 ± 0.5 km s⁻¹, dV_circ/dR = −3.6 ± 0.1 km s⁻¹ kpc⁻¹, U_⊙ = 8.1 ± 0.1 km s⁻¹, V_⊙ = 11.2 ± 0.2 km s⁻¹, W_⊙ = 8.1 ± 0.1 km s⁻¹, σ_xy = 14.2 ± 0.1 km s⁻¹, and σ_z = 7.3 ± 0.1 km s⁻¹. These numbers are consistent with results from the literature (e.g. as compiled by Bland-Hawthorn & Gerhard 2016). The corresponding Oort parameters are A = 16 km s⁻¹ kpc⁻¹, B = −12 km s⁻¹ kpc⁻¹, and A − B = 28 km s⁻¹ kpc⁻¹. The total velocity of the Sun translates to an apparent proper motion at the position of Sgr A^* of −6.25 mas yr⁻¹ along the plane and −0.21 mas yr⁻¹ perpendicular to the plane. This is consistent with the most recent evaluation of the proper motion of Sgr A^* by Reid & Brunthaler (2020).

Fig. 33. Distributions of the observed and model proper motions for the sample of 385 423 B-stars described in the text, with proper motions in Galactic longitude and latitude in the upper and lower panels, respectively. The black lines show the observed proper motion distributions. The thin orange lines are the predicted proper motion distributions for 200 randomly sampled MCMC model parameters. The mean of all such sample distributions is indicated by the thick dashed line.

The uncertainties quoted for the above results should be interpreted as the precision achieved in the context of the model and the subsample used. The uncertainties are underestimated. They do not account for the variance due to the choice of the specific random subsample of 20 000 B stars. More importantly, the obvious model deficiencies are not accounted for, such as ignoring the effects of the Milky Way disc warp, the motions induced by spiral arms (Olling & Dehnen 2003), and deviations of the true rotation curve from the simple model. The ‘mode-mixing’ effect discussed in Olling & Dehnen (2003) is not an issue here because of the precise knowledge of the parallaxes of the stars in the sample. The model deficiencies are apparent in Fig. 33 which shows a comparison between the observed and model proper motion distributions. As noted above, the modelling here is a mere illustration of the possibilities offered in analysing the proper motions for a sample of young disc stars covering a large range in R. For a much more in-depth look at a sample of young disc stars, selected slightly differently from what we presented in Sect. 3, we refer to the Gaia DR3 paper on mapping the asymmetric disc of the Milky Way (Gaia Collaboration 2023d). This latter paper presents maps showing rich structure in the velocity field of OB stars, which can be traced to the spiral arms, something which the above model does not capture. On the other hand, the average V_ϕ curve shown in that paper for the OB stars (calculated from the proper motions, parallaxes, and radial velocities) shows that the description of the rotation curve used above is accurate in an average sense.

10.2. Exoplanet characterisation

The search for and characterisation of exoplanet systems is at the forefront of scientific research, with many current and future ground- and space-based projects dedicated to this quest; see for example Gardner et al. (2006), Borucki et al. (2008), Rauer et al. (2014), Ricker (2014), Tinetti et al. (2018). Characterisation of the planet itself relies on the knowledge of the planet host. In particular, the planet’s radius and mass depends directly on the stars’s radius and mass through the following equations:

$$\begin{array}{c}\hfill M_{p}sin(i)=\frac{M_{\star }^{2/3}{P}^{1/3}K{(1-e^2)}^{0.5}}{{(2\pi G)}^{1/3}}\end{array}$$(3)

and

$$\begin{array}{c}\hfill d_{\mathrm{tr}}={\left(\frac{R_p}{R_{\star }}\right)}^2,\end{array}$$(4)

where M_p, M_⋆ are the mass of the planet and star, respectively, i and e are the inclination and eccentricity of the orbital system, P is the orbital period, K is the semi-amplitude of the radial velocity curve, and d_tr is the transit depth due to the planet with radius R_p passing in front of the star with radius R_⋆ and blocking a part of its light. In reality, the relationship between the transit depth and relative radii is a little more complicated than Eq. (4) (see e.g. Heller 2019), but we keep things simple here for the purpose of illustration. Additionally, we consider only transiting systems, and so the inclination of the system is very close to 90° and sin(i)∼1.

We obtained a list of the known transiting planets and light-curve parameters from exoplanets.org. For four of the planets, we adopted the values from the reference paper because of errors or inconsistencies: XO-6b from Crouzet et al. (2017), KELT-8b from Fulton et al. (2015), Kepler-407b from Marcy et al. (2014), and Kepler-68b Gilliland et al. (2013). This catalogue contains (as of March 2022) 2651 confirmed transiting exoplanets. We cross-matched these sources with the FGKM sample and obtained 593 planet matches. Of these, 354 contain transiting parameters with which to estimate the planetary radius while 108 entries contain parameters that can be used to estimate both the planet mass and radius; but only 95 have a valid stellar mass in our sample.

We calculated the radii of the exoplanets using radius_flame, along with the available transit depth parameter. To evaluate the uncertainties, we used a bootstrap method where we perturbed each of the input observations 1000 times and used the resulting standard distribution of the evaluated parameters to estimate the uncertainties. We show the distribution of the planetary radii as a function of orbital separation of the planet–star system in Fig. 34. We colour-coded the planet symbols according to teff_gspphot and the symbol size indicates the orbital period of the system, which ranges from 0.57 days to just under 365 days. We also show the position of the Earth and Jupiter as grey squares, which highlights the difference between other planetary systems and our own. In particular, many of these planets are well inside the inner limit of the habitable zone and the Jupiter-sized planets are equally close to their host star.

Fig. 34. Distribution of planetary radii compared to the separation from their host star (orbital semi-major axis a) for planetary systems in the FGKM sample. The colour code indicates the Teff of the host star, while the symbol size indicates the orbital period in log10 scale (range = 0.57–364.8 days). The dotted lines indicate 1 REarth and 1 RJup and the Earth and Jupiter are denoted by the square symbols.

We furthermore calculated the mass of the planets for the 95 sources with radial velocity parameters and stellar masses. Of these, four did not have a reported eccentricity, and of the other 91, only 24 have non-zero values. For the planets with no reported eccentricity, we assumed circular orbits. Ten of the planets also did not have a reported inclination and so we assumed i = 90° (sin i = 1) which is reasonable for a transiting system. The median value of the inclinations of the other 85 planetary systems is 87.2° (sin i = 0.9988 ∼ 1).

We show our results in the planet radius–planet mass diagram in Fig. 35. We also show some models corresponding to model mass–radius relationships for different Earth-like planet compositions from Zeng et al. (2016) and Jupiter-like planet compositions from Guillot et al. (2015). The black lines represent Earth-like planet mass–radius relations assuming an ice-like (dashed), rocky Earth-like (dashed-dotted), and iron (dashed) composition. The coloured lines represent models of an isolated planet of solar composition at 5 Gyr (like for Jupiter, blue), a heavily irradiated planet with an equilibrium temperature of 1000 K with no core (green) and one with a 100 M_Earth central core (red).

Fig. 35. Planet mass and radius (in Jupiter units) of 95 planets with radial velocity and stellar parameters in the FGKM sample. The colour coding is as in Fig. 34 and symbol size corresponds to the semi-major axis. Some radius–mass models of planets are also shown; see text for details.

The precision on our results (the error bars are shown although they are not always visible) does allow one to distinguish between different bulk compositions of these planets provided we have full control of the potential systematic errors. We provide the mass, radius, and age properties of the planet and their hosts in Table 8.

As these figures highlight, there is a dearth of knowledge and accurate characterisations of Earth-size exoplanets in the habitable zone. The upcoming ESA PLATO mission promises to populate the habitable zone by observing (at least) one large field over a two-to-three year period, which will allow us to detect and confirm planets in Earth-like orbits around Sun-like stars.

10.3. The colours of the Sun

The colours of the Sun are not as well known as one might have imagined, either observationally or from modelling. Solar analogues offer the possibility to validate, and if necessary calibrate, our understanding of the solar flux as a function of wavelength (Holmberg et al. 2006; Casagrande & VandenBerg 2018). They have also been used to estimate the solar bolometric correction in Gaia’s photometric system. Below we make a new attempt to determine precise and accurate solar colours.

We use the sample of solar-analogue candidates selected from GSP-Spec from Sect. 7 in order to estimate the colours of the Sun. As we demonstrate in Sect. 7.4, these stars have reliable extinction estimates from GSP-Phot. Consequently, the BP/RP spectra with very low extinction (according to GSP-Phot) can be used to indirectly estimate the intrinsic continuum shape of the Sun. Among the GSP-Spec solar-analogue candidates, there are 682 with a GSP-Phot A₀ < 0.001 mag. Given these, we obtain an absolute magnitude of

$$\begin{array}{c}\hfill M_{G,\odot }=(4.614\pm 0.179)\mathrm{mag}.\end{array}$$(5)

For this, we adopt the inverse parallax as a distance estimator because our candidate selection requires very high parallax quality ($\frac{\varpi }{{\sigma }_{\varpi }}>20$). For comparison, a value of M_G, ⊙ = 4.66 is adopted for FLAME (Creevey et al. 2023, Sect. 4.3 therein). Given the 682 candidates with A₀ < 0.001 mag, we also obtain mean colours and standard deviations of

$$\begin{array}{cc}\hfill {(G_{\mathrm{BP}}-G_{\mathrm{RP}})}_{\odot }& =(0.818\pm 0.029)\mathrm{mag}\hfill \end{array}$$(6)

$$\begin{array}{cc}\hfill {(G_{\mathrm{BP}}-G)}_{\odot }& =(0.324\pm 0.016)\mathrm{mag}\hfill \end{array}$$(7)

$$\begin{array}{cc}\hfill {(G-G_{\mathrm{RP}})}_{\odot }& =(0.494\pm 0.020)\mathrm{mag}\hfill \end{array}$$(8)

$$\begin{array}{cc}\hfill {(G-J)}_{\odot }& =(0.969\pm 0.578)\mathrm{mag}\hfill \end{array}$$(9)

$$\begin{array}{cc}\hfill {(G-H)}_{\odot }& =(1.292\pm 0.401)\mathrm{mag}\hfill \end{array}$$(10)

$$\begin{array}{cc}\hfill {(G-K_s)}_{\odot }& =(1.371\pm 0.351)\mathrm{mag}\hfill \end{array}$$(11)

$$\begin{array}{cc}\hfill {(G-W_1)}_{\odot }& =(1.449\pm 0.066)\mathrm{mag}\hfill \end{array}$$(12)

$$\begin{array}{cc}\hfill {(G-W_2)}_{\odot }& =(1.405\pm 0.065)\mathrm{mag},\hfill \end{array}$$(13)

where we restrict the AllWISE comparison to cases with W₁ > 8 mag in order to avoid saturation. These colours are in excellent agreement with the values (G_BP − G_RP)_⊙ = 0.82 mag, (G_BP − G)_⊙ = 0.33 mag, and (G − G_RP)_⊙ = 0.49 mag obtained by Casagrande & VandenBerg (2018) from Gaia DR2 passbands and synthetic as well as observed spectra for the Sun. The absolute magnitude of M_G, ⊙ = 4.67 mag obtained by these latter authors is also consistent with our estimate. In order to make this comparison with Gaia DR3 passbands and also include near-infrared photometry, we take the Kurucz model sun_mod_001.fits from the CALSPEC library¹¹ (Bohlin et al. 1995, 2014, 2020) and simulate its photometry using the pyphot package¹². We obtain synthetic colours of (G_BP − G_RP)_⊙ = 0.813 mag, (G_BP − G)_⊙ = 0.324 mag, and (G − G_RP)_⊙ = 0.490 mag, which are again in excellent agreement with our estimated colours. For colour combinations with 2MASS, we obtain (G − J)_⊙ = 0.992 mag, (G − H)_⊙ = 1.320 mag, and (G − Ks)_⊙ = 1.360 mag, which are again in excellent agreement with our candidates. Concerning AllWISE (Cutri et al. 2021), we obtain (G − W₁)_⊙ = 1.380 mag and (G − W₂)_⊙ = 1.301 mag. These values are slightly bluer than the values we estimate from the GSP-Spec candidates, but are still within 1σ and 1.6σ, respectively.

10.4. Ages of UCDs not seen by Gaia

Another application of the Gaia astrophysical parameters is to constrain the characteristics of faint UCDs that are beyond the mission magnitude limit but are in binary systems with brighter objects that are observed by Gaia. Once we have identified a multiple system we assume that the UCD has the same chemical composition, age, distance, and, after allowing for orbital motion, proper motions. In addition, if the movement due to the orbital motion is detected by Gaia, this will provide a constraint on the mass of the various components. Brown dwarfs evolve and cool over time and their observational properties are degenerate with age, mass, and metallicity; binary systems are therefore benchmarks for understanding these processes. Gaia will provide a large homogeneous multi-parametric sample with intersecting constraints that will tie down the UCD regime. For this illustrative discussion, we concentrate on the age parameter¹³. We note however that more precise ages can be obtained by combining Gaia with other observational data such as asteroseismology.

To identify a potential list of objects with a high probability to be in a binary system, we used the positional and kinematical criteria given by Eq. (2) in Smart et al. (2019) and the list of known UCDs from that study. When the faint UCD did not have a measured parallax we used its spectro-photometric distance. We found eight UCDs without Gaia DR3 five-parameter solutions that are in binary systems in the FGKM sample, while also in the regime of reliable ages (see Fouesneau et al. 2023). We added a further three interesting targets with reliable ages here because they were rejected from the FGKM sample for failing on only one of the criteria: A ipd_frac_multi_peak = 22; and B and F classprob_dsc_combmod_binarystar > 0.99. These 11 UCDs are listed in Table 9 with name, adopted parallax, spectral type, and mass along with the companion Gaia source_id, age, and the median published ages with 16% and 84% percentiles.

The number of literature age estimates vary from 6 to 46 for each target and are from varied sources: model comparisons (Holmberg et al. 2009; Casagrande et al. 2011), chromospheric activity (Pace 2013; Metchev & Hillenbrand 2004), or Galactic kinematics (Gontcharov 2012). The published age percentiles often indicate uncertainties of a factor of 2 or a large portion of the age of the Galaxy indicating the current difficulty in determining ages for stars. Figure 36 shows the Gaia versus the median published values from Table 9. When available, we used the values based on the GSP-Spec T_eff: age_flame_spec; these are denoted by the filled circles. The open circles are age_flame which are based on GSP-Phot T_eff. For most of the stars, we find general agreement with the literature, with the worst agreements for systems E, I, and K. For E, the T_eff from both GSP-Phot and GSP-Spec agree to within 25 K and we would therefore trust its age if the star were within the regime of models that were used. For I and K, we find significant disagreements between the GSP-Phot and GSP-Spec T_eff, and this could indicate a possible issue with age_flame. We discuss each of the systems individually in the following section.

Fig. 36. Literature ages versus FLAME ages for companions of UCDs not seen by Gaia. Filled circles correspond to age_flame_spec (i.e. using the Teff from GSP-Spec) and open circles are age_flame. The error bars represent the 16% and 84% percentiles.

The interpolated masses of the UCDs are estimated from a comparison to the illustrated tracks in Fig. 37 taken from Baraffe et al. (2015) for stars and Phillips et al. (2020) for brown dwarfs assuming the age of the companion star from this work, and these are reported in Table 9.

Fig. 37. Evolutionary tracks and UCD locations in the H-band absolute magnitude versus age diagram, adopting the companion age. The tracks are colour coded by mass. The dashed lines indicate the stellar to substellar transition zone (from 0.072 to 0.075 M⊙).

Notes on individual systems 2MASSI J0025036+475919 (A) is an L4+L4 binary in a multiple system with the spectroscopic binary HD 2057 (Reid et al. 2006, and references therein) and another component 11″ from the primary (Gaia EDR3 392562179817297536). Lithium absorption has been detected in the combined spectra of the secondary indicating it has an age less than 1.0 Ga (Cruz et al. 2007); 2010AJ....139..176F; 2015APJ...810..158F which is much lower than the primary age indicated here. This is one of the widest binary systems (∼10 000 AU) with an ultracool component but in the range of other systems of similar total mass. The difference in age estimates of the primary and secondary is not easily reconcilable. One possible solution is that it is not a binary system; the agreement of high proper motions is a strong constraint, but the spectroscopic distance is very uncertain as the binary nature of the secondary requires an assumption of the component flux contributions. Another possibility is that the primary age estimate is high because of its binary nature.

2MASS J02233667+5240066 (B) was first noted to be in a common proper motion system with HIP 11161 in Deacon et al. (2014). The primary has been shown to have acceleration terms (Kervella et al. 2022; Brandt 2021) but the separation with the UCD is large (41″) and the primary has now been resolved by Gaia into two components and is listed as a spectroscopic binary in the non-single stars orbital solution results. It also has a very high classprob_dsc_combmod_binarystar (> 0.99). The observed acceleration is therefore due to the primary binarity and not the UCD. Using age_flame_spec we find a mass of 80 M_♃ which defines the end of the stellar main sequence.

2MASS J06462756+7935045 (C) was indicated as being in a binary system with HD 46588 based on a high common proper motion (Loutrel et al. 2011). It is an L9 brown dwarf, one of the few known at the L/T transition in wide binary systems. These allow constraints on their astrophysical properties. The age_flame_spec is lower by 1σ than the primary literature age. As this is one of the few L9s where an independent age is known, it is important to clarify this discrepancy. Assuming the literature age and distance from the primary, Loutrel et al. (2011) find a $T_{\mathrm{eff}}={1360}_{1280}^{1410}K$, which is an important constraint for the temperature at the L/T boundary. If we assume the lower age_flame_spec, this will increase the temperature estimate at this boundary.

HD 49197 B (D) has been studied extensively since its first discovery by Metchev & Hillenbrand (2004) using high-resolution observing techniques. It is at a separation of 0.95″ from the primary. There are ongoing adaptive optics projects to try to determine a binary solution (Bowler et al. 2020; Tokovinin 2014). With a magnitude difference of greater than ten, Gaia will not be able to resolve the system. If we adopt the low end of the literature age range, HD 49197 B is a brown dwarf; if we adopt the high end –for example that indicated by age_flame_spec– the object becomes a star. As there is also a possibility of finding the mass of this companion either through high-resolution imaging or the detection of acceleration terms in the Gaia primary solution (proper motion anomalies between the HIPPARCOS and Gaia results have already been detected in Kervella et al. 2022), knowing its age will be crucial for constraining the stellar–substellar boundary.

2MASS J12173646+1427119 (E) was first discovered in the Pan-STARRS survey as a companion to HIP 59933 at 40″. The secondary is detected by Gaia (EDR3 3921177219942653696) but with only a two-parameter solution. The primary, EDR3 3921176983720146560, has a non-single star solution which indicates a companion of 0.09 M_⊙ with a period of 1 yr and a corresponding separation of 1 AU; given the small separation there must be a third component in the system. Any age above 0.5 Ga would indicate that this latter component is a stellar object but very close to the stellar–substellar boundary as indicated by our 82 M_♃.

HD 118865B (F) is a T5 in a system with an F4 spectral type first noted in Burningham et al. (2013) where these authors find an age range of 1.5–4.9 Ga and mass of 45–60 M_♃. We find a primary age that is at the top end of their range and hence a slightly larger mass. If confirmed it will provide a high mass for this T5 compared to other brown dwarfs of a similar type.

2MASS J14165987+5006258 (G) is noted as a binary system in Faherty et al. (2010) with a very large separation of ∼26 000 AU. The primary age_flame_spec estimate is consistent with published values and near-IR colours in Faherty et al. (2010) where they also re-evaluate its spectral type from L5.5 to L4. The estimated mass indicates that this object is of stellar and not sub-stellar type and further characterisation will contribute to our understanding of very old borderline stellar objects.

ULAS J142320.79+011638.2 (H) is the coolest object in this sample in a system with an early-G dwarf, HIP 70319. There are a significant number of age estimates from very young to very old and the age_flame_spec is in agreement with the median. This age is consistent with a low-metallicity primary and also with a broader Y-band peak and more depressed K-band peak than other T8s (Kirkpatrick et al. 2021). This is an important benchmark for metal-poor T dwarfs.

Gl 564 B/C (I) is an L4+L4 binary in a triple system with Gl 564, a G2 V star. The majority of the published age ranges are very young because Gl 564 is chromospherically active with a high lithium abundance and fast rotation (Potter et al. 2002). The space motion also puts the object in the Ursa Major moving group from the Banyan Σ tool (Gagné et al. 2018), which has an age of around 500 Myr (King et al. 2003). This is in contrast to the high age_flame which is difficult to reconcile given the high lithium abundance and space motion. A possible explanation for this discrepancy is in the limitations of the models that were used; for example, they do not include rotation. If the system is in the first 0.5 Ga, they will be contracting brown dwarfs. The orbital period of the UCD binary system is around 10 yr (Potter et al. 2002) and we will therefore soon have dynamical masses with Gaia. These objects will provide a well-constrained calibration point for the theoretical models describing low-mass, ultracool objects.

Gl779B (J) is an L4.5 UCD at 0.7″ from GJ 779, a G0 star. High levels of chromospheric activity suggest a young age, whereas lithium abundance indicates a slightly older age than the Hyades and kinematics indicate an old disc star. The age_flame_spec is consistent with the published estimates. The orbit is such that it should be visible in the future Gaia observations which will lead to a dynamical mass estimate (Crepp et al. 2014). A comparison of the accelerations found from comparisons of HIPPARCOS and Gaia DR2 results indicate a mass of around 0.07 M_⊙ (Brandt et al. 2019) and this is therefore on the stellar–substellar boundary currently defining the end of the main sequence, and in agreement with our estimated mass.

Eps Ind C (K) is the second closest brown dwarf binary T1+T6 system in a triple system with the K5V star eps Ind. One of the brown dwarfs has a Gaia solution (Gaia EDR3 6412596012146801152) which we assume is the T1. Later releases should provide a dynamical solution for the component masses. There is a significant history of publications for both the primary and the secondary system. With a period of around 11 yr and an observed separation that varies from 0.6 to 2″, it is a defining system for parameters of early T dwarfs. The https://gea.esac.esa.int/archive/documentation/GDR3/Gaia_archive/chap_datamodel/sec_dm_astrophysical_parameter_tables/ssec_dm_astrophysical_parameters_supp.html#astrophysical_parameters_supp-age_flame_spec is at the low end of the published age range for the primary, and the masses of the secondaries from Dieterich et al. (2018) also imply an inconsistency with such a young age. A dynamical mass determination from the Gaia observations should resolve this inconsistency.

We have seen that the results of Gaia can be brought to bear on our understanding of objects fainter than its magnitude limit. Indeed there will probably be less than 1000 brown dwarfs brighter than the Gaia magnitude limit (Smart et al. 2019) while we expect there to be tens of thousands in binary systems or detected from astrometric and radial velocity perturbations. Therefore, the contribution of Gaia to brown dwarf studies will be predominantly due to indirectly detected objects rather than direct detections.

11. Conclusion

In this work, we define homogeneous samples of high-quality astrophysical parameters by exploiting many Gaia data products that appear in Gaia DR3, while focusing on the sources and data products in the astrophysical_parameters and the astrophysical_parameters_supp tables which were produced by the Apsis software (Creevey et al. 2023; Delchambre et al. 2023; Fouesneau et al. 2023). We consider different regimes of stars all across the HR diagram. In the first part of this work, we consider large samples of young massive disc OBA stars (Sect. 3), FGKM spectral-type stars (Sect. 4), and faint ultracool dwarfs (UCDs, Sect. 5). We then focus on smaller samples of specific object types; carbon stars (Sect. 6), solar analogues (Sect. 7), and the Gaia spectrophotometric standard stars (SPSS; Pancino et al. 2021, Sect. 8). Concerning the latter, this paper provides the first homogeneous determination of the SPSS dataset to date. We validate each of the samples using the Gaia data themselves and external catalogues, and our results are published in six tables that appear alongside Gaia DR3; see Sect. 9 and Table 7.

In Sect. 10, we demonstrate some use cases of these samples of stars. We use a subset of the OBA sample to illustrate its usefulness in analysing the Milky Way rotation curve (Sect. 10.1). We then use the properties of the FGKM stars to analyse known exoplanet systems including the determination of planet radii and masses (Sect. 10.2). We then predict the colours of the Sun in various passbands using the solar analogue sample (Sect. 10.3). Finally, we analyse the ages of some unseen UCD-companions to the FGKM stars (Sect. 10.4).

The aim of this work is to highlight the science that can be done with Gaia DR3. We focus on specific types of stars using strict quality criteria on many of the data products, which sometimes includes some ad hoc filtering criteria tuned with a particular science case in mind. We emphasise that our strict personal selections may not be applicable to some specific science cases, and users should acknowledge this before exploiting these samples. We fully encourage all users to exploit all of the astrophysical parameters in Gaia DR3 independent of our specific selection criteria highlighted in this work. Indeed, there are up to 470 million stars with stellar parameters derived using the mean BP and RP spectra, up to 6 million stellar parameters and abundances derived from the mean RVS spectra, up to 130 million masses and ages, and many other new stellar products that have not been mentioned in this work, such as DIB estimates, activity index of active stars, and Hα emission. As illustrated in this work, many science cases can be explored with these data.

Acknowledgments

We thank the referee for their constructive comments on the manuscript. This work has made use of data from the European Space Agency (ESA) mission Gaia (https://www.cosmos.esa.int/gaia), processed by the Gaia Data Processing and Analysis Consortium (DPAC, https://www.cosmos.esa.int/web/gaia/dpac/consortium). Funding for the DPAC has been provided by national institutions, in particular, the institutions participating in the Gaia Multilateral Agreement. The full list of funding agencies and grants is given in Appendix A. This research has used NASA’s Astrophysics Data System, and the VizieR catalogue access tool (CDS, Strasbourg, France). The data processing and analysis made use of matplotlib (Hunter 2007), NumPy (Harris et al. 2020), the IPython package (Pérez & Granger 2007), Vaex (Breddels & Veljanoski 2018), TOPCAT (Taylor 2005), pyphot (http://github.com/mfouesneau/pyphot), R (R Core Team 2013), Astropy (Astropy Collaboration 2013); astropy:2018, CmdStanPy (https://github.com/stan-dev/cmdstanpy), and ArviZ (Kumar et al. 2019). In case of errors or omissions, please contact the Gaia Helpdesk. The full list of acknowledgements can also be found in the official online documentation for Gaia DR3.

References

Appendix A: Gaia funding institutions

This work presents results from the European Space Agency (ESA) space mission Gaia. Gaia data are processed by the Gaia Data Processing and Analysis Consortium (DPAC). Funding for the DPAC is provided by national institutions, in particular the institutions participating in the Gaia MultiLateral Agreement (MLA). The Gaia mission website is https://www.cosmos.esa.int/gaia. The Gaia archive website is https://archives.esac.esa.int/gaia.

The Gaia mission and data processing have financially been supported by, in alphabetical order by country:

– the Algerian Centre de Recherche en Astronomie, Astrophysique et Géophysique of Bouzareah Observatory;

– the Austrian Fonds zur Förderung der wissenschaftlichen Forschung (FWF) Hertha Firnberg Programme through grants T359, P20046, and P23737;

– the BELgian federal Science Policy Office (BELSPO) through various PROgramme de Développement d’Expériences scientifiques (PRODEX) grants, the Research Foundation Flanders (Fonds Wetenschappelijk Onderzoek) through grant VS.091.16N, the Fonds de la Recherche Scientifique (FNRS), and the Research Council of Katholieke Universiteit (KU) Leuven through grant C16/18/005 (Pushing AsteRoseismology to the next level with TESS, GaiA, and the Sloan DIgital Sky SurvEy – PARADISE);

– the Brazil-France exchange programmes Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) and Coordenação de Aperfeicoamento de Pessoal de Nível Superior (CAPES) - Comité Français d’Evaluation de la Coopération Universitaire et Scientifique avec le Brésil (COFECUB);

– the Chilean Agencia Nacional de Investigación y Desarrollo (ANID) through Fondo Nacional de Desarrollo Científico y Tecnológico (FONDECYT) Regular Project 1210992 (L. Chemin);

– the National Natural Science Foundation of China (NSFC) through grants 11573054, 11703065, and 12173069, the China Scholarship Council through grant 201806040200, and the Natural Science Foundation of Shanghai through grant 21ZR1474100;

– the Tenure Track Pilot Programme of the Croatian Science Foundation and the École Polytechnique Fédérale de Lausanne and the project TTP-2018-07-1171 ‘Mining the Variable Sky’, with the funds of the Croatian-Swiss Research Programme;

– the Czech-Republic Ministry of Education, Youth, and Sports through grant LG 15010 and INTER-EXCELLENCE grant LTAUSA18093, and the Czech Space Office through ESA PECS contract 98058;

– the Danish Ministry of Science;

– the Estonian Ministry of Education and Research through grant IUT40-1;

– the European Commission’s Sixth Framework Programme through the European Leadership in Space Astrometry (ELSA) Marie Curie Research Training Network (MRTN-CT-2006-033481), through Marie Curie project PIOF-GA-2009-255267 (Space AsteroSeismology & RR Lyrae stars, SAS-RRL), and through a Marie Curie Transfer-of-Knowledge (ToK) fellowship (MTKD-CT-2004-014188); the European Commission’s Seventh Framework Programme through grant FP7-606740 (FP7-SPACE-2013-1) for the Gaia European Network for Improved data User Services (GENIUS) and through grant 264895 for the Gaia Research for European Astronomy Training (GREAT-ITN) network;

– the European Cooperation in Science and Technology (COST) through COST Action CA18104 ‘Revealing the Milky Way with Gaia (MW-Gaia)’;

– the European Research Council (ERC) through grants 320360, 647208, and 834148 and through the European Union’s Horizon 2020 research and innovation and excellent science programmes through Marie Skłodowska-Curie grant 745617 (Our Galaxy at full HD – Gal-HD) and 895174 (The build-up and fate of self-gravitating systems in the Universe) as well as grants 687378 (Small Bodies: Near and Far), 682115 (Using the Magellanic Clouds to Understand the Interaction of Galaxies), 695099 (A sub-percent distance scale from binaries and Cepheids – CepBin), 716155 (Structured ACCREtion Disks – SACCRED), 951549 (Sub-percent calibration of the extragalactic distance scale in the era of big surveys – UniverScale), and 101004214 (Innovative Scientific Data Exploration and Exploitation Applications for Space Sciences – EXPLORE);

– the European Science Foundation (ESF), in the framework of the Gaia Research for European Astronomy Training Research Network Programme (GREAT-ESF);

– the European Space Agency (ESA) in the framework of the Gaia project, through the Plan for European Cooperating States (PECS) programme through contracts C98090 and 4000106398/12/NL/KML for Hungary, through contract 4000115263/15/NL/IB for Germany, and through PROgramme de Développement d’Expériences scientifiques (PRODEX) grant 4000127986 for Slovenia;

– the Academy of Finland through grants 299543, 307157, 325805, 328654, 336546, and 345115 and the Magnus Ehrnrooth Foundation;

– the French Centre National d’Études Spatiales (CNES), the Agence Nationale de la Recherche (ANR) through grant ANR-10-IDEX-0001-02 for the ‘Investissements d’avenir’ programme, through grant ANR-15-CE31-0007 for project ‘Modelling the Milky Way in the Gaia era’ (MOD4Gaia), through grant ANR-14-CE33-0014-01 for project ‘The Milky Way disc formation in the Gaia era’ (ARCHEOGAL), through grant ANR-15-CE31-0012-01 for project ‘Unlocking the potential of Cepheids as primary distance calibrators’ (UnlockCepheids), through grant ANR-19-CE31-0017 for project ‘Secular evolution of galxies’ (SEGAL), and through grant ANR-18-CE31-0006 for project ‘Galactic Dark Matter’ (GaDaMa), the Centre National de la Recherche Scientifique (CNRS) and its SNO Gaia of the Institut des Sciences de l’Univers (INSU), its Programmes Nationaux: Cosmologie et Galaxies (PNCG), Gravitation Références Astronomie Métrologie (PNGRAM), Planétologie (PNP), Physique et Chimie du Milieu Interstellaire (PCMI), and Physique Stellaire (PNPS), the ‘Action Fédératrice Gaia’ of the Observatoire de Paris, the Région de Franche-Comté, the Institut National Polytechnique (INP) and the Institut National de Physique nucléaire et de Physique des Particules (IN2P3) co-funded by CNES;

– the German Aerospace Agency (Deutsches Zentrum für Luft- und Raumfahrt e.V., DLR) through grants 50QG0501, 50QG0601, 50QG0602, 50QG0701, 50QG0901, 50QG1001, 50QG1101, 50QG1401, 50QG1402, 50QG1403, 50QG1404, 50QG1904, 50QG2101, 50QG2102, and 50QG2202, and the Centre for Information Services and High Performance Computing (ZIH) at the Technische Universität Dresden for generous allocations of computer time;

– the Hungarian Academy of Sciences through the Lendület Programme grants LP2014-17 and LP2018-7 and the Hungarian National Research, Development, and Innovation Office (NKFIH) through grant KKP-137523 (‘SeismoLab’);

– the Science Foundation Ireland (SFI) through a Royal Society - SFI University Research Fellowship (M. Fraser);

– the Israel Ministry of Science and Technology through grant 3-18143 and the Tel Aviv University Center for Artificial Intelligence and Data Science (TAD) through a grant;

– the Agenzia Spaziale Italiana (ASI) through contracts I/037/08/0, I/058/10/0, 2014-025-R.0, 2014-025-R.1.2015, and 2018-24-HH.0 to the Italian Istituto Nazionale di Astrofisica (INAF), contract 2014-049-R.0/1/2 to INAF for the Space Science Data Centre (SSDC, formerly known as the ASI Science Data Center, ASDC), contracts I/008/10/0, 2013/030/I.0, 2013-030-I.0.1-2015, and 2016-17-I.0 to the Aerospace Logistics Technology Engineering Company (ALTEC S.p.A.), INAF, and the Italian Ministry of Education, University, and Research (Ministero dell’Istruzione, dell’Università e della Ricerca) through the Premiale project ‘MIning The Cosmos Big Data and Innovative Italian Technology for Frontier Astrophysics and Cosmology’ (MITiC);

– the Netherlands Organisation for Scientific Research (NWO) through grant NWO-M-614.061.414, through a VICI grant (A. Helmi), and through a Spinoza prize (A. Helmi), and the Netherlands Research School for Astronomy (NOVA);

– the Polish National Science Centre through HARMONIA grant 2018/30/M/ST9/00311 and DAINA grant 2017/27/L/ST9/03221 and the Ministry of Science and Higher Education (MNiSW) through grant DIR/WK/2018/12;

– the Portuguese Fundação para a Ciência e a Tecnologia (FCT) through national funds, grants SFRH/BD/128840/2017 and PTDC/FIS-AST/30389/2017, and work contract DL 57/2016/CP1364/CT0006, the Fundo Europeu de Desenvolvimento Regional (FEDER) through grant POCI-01-0145-FEDER-030389 and its Programa Operacional Competitividade e Internacionalização (COMPETE2020) through grants UIDB/04434/2020 and UIDP/04434/2020, and the Strategic Programme UIDB/00099/2020 for the Centro de Astrofísica e Gravitação (CENTRA);

– the Slovenian Research Agency through grant P1-0188;

– the Spanish Ministry of Economy (MINECO/FEDER, UE), the Spanish Ministry of Science and Innovation (MICIN), the Spanish Ministry of Education, Culture, and Sports, and the Spanish Government through grants BES-2016-078499, BES-2017-083126, BES-C-2017-0085, ESP2016-80079-C2-1-R, ESP2016-80079-C2-2-R, FPU16/03827, PDC2021-121059-C22, RTI2018-095076-B-C22, and TIN2015-65316-P (‘Computación de Altas Prestaciones VII’), the Juan de la Cierva Incorporación Programme (FJCI-2015-2671 and IJC2019-04862-I for F. Anders), the Severo Ochoa Centre of Excellence Programme (SEV2015-0493), and MICIN/AEI/10.13039/501100011033 (and the European Union through European Regional Development Fund ‘A way of making Europe’) through grant RTI2018-095076-B-C21, the Institute of Cosmos Sciences University of Barcelona (ICCUB, Unidad de Excelencia ‘María de Maeztu’) through grant CEX2019-000918-M, the University of Barcelona’s official doctoral programme for the development of an R+D+i project through an Ajuts de Personal Investigador en Formació (APIF) grant, the Spanish Virtual Observatory through project AyA2017-84089, the Galician Regional Government, Xunta de Galicia, through grants ED431B-2021/36, ED481A-2019/155, and ED481A-2021/296, the Centro de Investigación en Tecnologías de la Información y las Comunicaciones (CITIC), funded by the Xunta de Galicia and the European Union (European Regional Development Fund – Galicia 2014-2020 Programme), through grant ED431G-2019/01, the Red Española de Supercomputación (RES) computer resources at MareNostrum, the Barcelona Supercomputing Centre - Centro Nacional de Supercomputación (BSC-CNS) through activities AECT-2017-2-0002, AECT-2017-3-0006, AECT-2018-1-0017, AECT-2018-2-0013, AECT-2018-3-0011, AECT-2019-1-0010, AECT-2019-2-0014, AECT-2019-3-0003, AECT-2020-1-0004, and DATA-2020-1-0010, the Departament d’Innovació, Universitats i Empresa de la Generalitat de Catalunya through grant 2014-SGR-1051 for project ‘Models de Programació i Entorns d’Execució Parallels’ (MPEXPAR), and Ramon y Cajal Fellowship RYC2018-025968-I funded by MICIN/AEI/10.13039/501100011033 and the European Science Foundation (‘Investing in your future’);

– the Swedish National Space Agency (SNSA/Rymdstyrelsen);

– the Swiss State Secretariat for Education, Research, and Innovation through the Swiss Activités Nationales Complémentaires and the Swiss National Science Foundation through an Eccellenza Professorial Fellowship (award PCEFP2_194638 for R. Anderson);

– the United Kingdom Particle Physics and Astronomy Research Council (PPARC), the United Kingdom Science and Technology Facilities Council (STFC), and the United Kingdom Space Agency (UKSA) through the following grants to the University of Bristol, the University of Cambridge, the University of Edinburgh, the University of Leicester, the Mullard Space Sciences Laboratory of University College London, and the United Kingdom Rutherford Appleton Laboratory (RAL): PP/D006511/1, PP/D006546/1, PP/D006570/1, ST/I000852/1, ST/J005045/1, ST/K00056X/1, ST/K000209/1, ST/K000756/1, ST/L006561/1, ST/N000595/1, ST/N000641/1, ST/N000978/1, ST/N001117/1, ST/S000089/1, ST/S000976/1, ST/S000984/1, ST/S001123/1, ST/S001948/1, ST/S001980/1, ST/S002103/1, ST/V000969/1, ST/W002469/1, ST/W002493/1, ST/W002671/1, ST/W002809/1, and EP/V520342/1.

The GBOT programme uses observations collected at (i) the European Organisation for Astronomical Research in the Southern Hemisphere (ESO) with the VLT Survey Telescope (VST), under ESO programmes 092.B-0165, 093.B-0236, 094.B-0181, 095.B-0046, 096.B-0162, 097.B-0304, 098.B-0030, 099.B-0034, 0100.B-0131, 0101.B-0156, 0102.B-0174, and 0103.B-0165; and (ii) the Liverpool Telescope, which is operated on the island of La Palma by Liverpool John Moores University in the Spanish Observatorio del Roque de los Muchachos of the Instituto de Astrofísica de Canarias with financial support from the United Kingdom Science and Technology Facilities Council, and (iii) telescopes of the Las Cumbres Observatory Global Telescope Network.

All Tables

All Figures

Fig. 1. Selection of the OBA sample. Left and right panels: the OBA samples from ESP-HS and GSP-Phot, respectively. A first filter on the parallax (S/N) is applied to the GSP-Phot targets (from panels a to c). For both samples, subluminous targets are removed (grey shading in panels b and c), and then the outliers at six standard deviations from the expected colour vs. Teff relation (blue line) are filtered out (grey shading in panels d and e). The absolute magnitude MG is computed using the measured parallax and the estimated interstellar extinction AG provided by both modules. The de-reddened colour, (GBP − GRP)0, is derived using the value of E(GBP − GRP). The resulting Kiel diagrams are shown in the bottom rows (panels f and g). The over-densities seen at Teff = 15 000 K, 20 000 K, and 30 000 K are linked to the temperature limits of the adopted synthetic spectra libraries.

In the text

	Fig. 2. Completeness of the OBA list in various open clusters (Cantat-Gaudin et al. 2020) as a function of interstellar extinction. The fraction corresponds to the ratio between the number of cluster members present in our list and the number of expected OBA stars. The colour code follows the cluster age provided by Cantat-Gaudin et al. (2020).
In the text

	Fig. 3. Histogram of tangential velocities of the stars in the OBA sample with ϖ/σϖ > 10. The combined OBA star sample is shown as well as the individual O, B, and A star samples (based on the classifications from the ESP-HS module). The limits in tangential velocity separating the thin disc, thick disc, and halo populations are shown as vertical dashed lines.
In the text

	Fig. 4. Toomre diagram for the OBA stars for which a radial velocity is available in Gaia DR3. See text for explanations on the diagram. The colour coding indicates the median value of vtan at a given location on this diagram. The half circles indicate limits on the total velocity with respect to the local circular velocity of 50 and 180 km s−1.
In the text

	Fig. 5. Distribution of the parameters of the OBA sample with ϖ/σϖ > 10 and vtan > 180 km s−1. Left: observational HR diagram. Right: Kiel diagram. The contours indicate the distribution of the full sample. The colour code indicates the density of sources satisfying the above criteria.
In the text

Fig. 6. Distribution of stars in the OBA sample projected on the Galactic plane. The Galactic centre is to the right at (X, Y) = (0, 0) and the Sun is at ( − 8, 0). From left to right, the panels show the full sample (with ϖ/σϖ > 10) and the samples selected according to the vtan ranges indicated. The red contours indicate lines of constant vtan, calculated with the simple kinematic disc model as explained in the text.

In the text

Fig. 7. Comparisons between sample fgkm_1 and intermediate samples based on some of the criteria used to define sample fgkm_2. Top panel: illustrates the distance–parallax–error constraint and the lower panel shows the (G − GRP)0 − (GBP − GRP)0 relation after imposing the colour–Teff and colour–colour cuts described in Sect. 4.2. In both panels, the sources in fgkm_1 are shown in the background, while those satisfying the criteria are illustrated in the foreground, colour-coded according to logarithmic count.

In the text

	Fig. 8. Galactic plane projections illustrating the density of sources of the samples fgkm_2 (top) and fgkm_3 (bottom).
In the text

	Fig. 9. HR diagram based on GSP-Phot and FLAME for the definition of the FGKM sample. Top left panel: illustrates the HR diagram before any selection is made using a random sample of 2 Million stars. The rest of the panels show the various quality cuts. Top right: fgkm_1, bottom left is fgkm_2, and bottom right is fgkm_3 before cleaning for variables and binaries.
In the text

	Fig. 10. Distribution of the final sample fgkm_3 of the observed parameters G and parallax, colour-coded by (GBP − GRP)0.
In the text

	Fig. 11. HR and Kiel diagrams using GSP-Spec-based parameters for the fgkm_spec sample described in Sect. 4.3 colour-coded by the metallicity from GSP-Spec, with parallax_over_error ≥33.34 and rvs_spec_sig_to_noise ≥ 150.
In the text

	Fig. 12. HR diagram using sample fgkm_3 colour coded according to evolstage_flame. The low values of evolution stage on the giant branch correspond to the FLAME parameters that were removed from the table; see Sect. 4.4.1 for details.
In the text

	Fig. 13. Comparison of Teff and AG from GSP-Phot compared to the reference values from isochrones for stars of the FGKM sample in clusters. Left: comparison of Teff, with colour indicating the density of sources. The red line indicates the one-to-one values. Right: ΔAG = AG, GSP − Phot − AG, isochrones versus TeffGSP − Phot.
In the text

	Fig. 14. Δlog g = log gGSP − Phot − log gisochrones versus TeffGSP − Phot for stars of the FGKM sample in clusters. The colour indicates the distance modulus (m − M) as derived from the GSP-Phot distance.
In the text

Fig. 15. Comparison of atmospheric parameters with the spectroscopic surveys for the FGKM sample. Top panels: comparison of GSP-Spec parameters and the bottom panels GSP-Phot. Left panels: the case of Teff, the middle panels that of log g, and the right panels that of [Fe/H]. The differences on the y-axes are the Gaia values minus the other survey values, where the latter are calculated as the median values in equally populated bins (solid lines, coloured according to the legend in the bottom-left panel). The dotted lines for the GSP-Spec log g are obtained after the corrections recommended by Recio-Blanco et al. (2023).

In the text

	Fig. 16. Difference between Teff (top), R (middle), and M (bottom) from Gaia for the FGKM sample and the PICv1.1 catalogue values normalised to their combined uncertainties for stars in common. We overlay the ±3σ lines. On each panel we also give the median difference (MD) and the MAD in K, R⊙, and M⊙, respectively.
In the text

	Fig. 17. Radii of candidate UCDs in the Gaia golden sample. The colour code indicates the logarithm of the VOSA fit χ2 values, squares represent the data points in Table 1 of Dieterich et al. (2014), and black asterisks denote unresolved binaries therein. The box plots are calculated within bins of 100 K.
In the text

	Fig. 18. Bolometric luminosities of candidate UCDs in the Gaia golden sample. The colour code indicates the logarithm of the VOSA fit χ2 values, squares represent the data points in Table 1 of Dieterich et al. (2014), and black asterisks denote unresolved binaries therein.
In the text

	Fig. 19. Comparison of the radii estimated for the UCD sample by the ESP-UCD (x axis) and FLAME (y axis) modules for the sources in common. The colour code reflects the effective temperature used by the FLAME module to estimate the radii.
In the text

	Fig. 20. Comparison of the effective temperatures used to derive radii in this work (x-axis) and those used in the literature (y-axis) for the UCD sample. Black filled circles denote sources from Cifuentes et al. (2020) and orange filled circles denote those from Dieterich et al. (2014).
In the text

Fig. 21. Band head strengths (see Eq. (2) and Table 5) measured in the BP and RP spectra of known Galactic (MW, orange points, Alksnis et al. 2001), Large Magellenic Cloud (LMC, black points, Kontizas et al. 2001), and Small Magellenic Cloud (SMC, green points, Morgan & Hatzidimitriou 1995) carbon stars. Only targets within 1 arcsec of a Gaia DR3 source_id are taken into account. Upper panels: locus occupied by non-carbon stars represented by the blue shaded area. Middle and lower panels: targets with weaker or non-existing CN features shown with blue points (i.e. they fall in the shaded areas of the upper panels). The pink broken and full lines delimit the domain occupied by 87% and 98% of the carbon stars with strong CN features, respectively.

In the text

	Fig. 22. Same as Fig. 21 but for the 386 936 candidate carbon stars flagged by ESP-ELS. Pink curves represent the domain occupied by the carbon stars found in the literature (Fig. 21, and Sect. 6.2).
In the text

	Fig. 23. Mollweide view in Galactic coordinates of the carbon stars sample described in this work. The locations of the Magellanic Clouds and the Sagittarius stream are shown in blue.
In the text

Fig. 24. Magnitude and colour distribution of carbon stars. Left panels: vertical black dashed line shows the upper magnitude limit of the data processed by ESP-ELS. Upper panels: all the targets belonging to the golden sample of carbon stars are taken into account. Other panels: distributions obtained for the known MW (Alksnis et al. 2001), LMC (Kontizas et al. 2001), and SMC (Morgan & Hatzidimitriou 1995) carbon stars are shown in blue. In orange, we show the distribution of the targets in common with the sample we propose in this work.

In the text

	Fig. 25. Kiel diagram of carbon stars with published GSP-Phot parameters. Left panel: density plot for known MW, SMC, and LMC C stars. Right panel: density plot for the carbon stars in our sample.
In the text

	Fig. 26. BP and RP spectra of the 20 randomly chosen (amongst 254) carbon stars (this work) with Teff GSP − Phot > 6000 K. The ordinate axis provides the flux normalised to the total flux, and shifted by k × 0.003 (where k is an integer that varies from 0 to 9 from the bottom to the top spectrum).
In the text

	Fig. 27. Distribution of [α/Fe] abundances from GSP-Spec for solar-analogue candidates. Grey shows the raw alphafe_gspspec values and black shows the calibrated values (Recio-Blanco et al. 2023). The dashed red line shows a Gaussian distribution with a mean of –0.028 and a standard deviation of 0.056.
In the text

Fig. 28. RVS spectra of 916 solar-analogue candidates from GSP-Spec (panel a) where 95% of GSP-Spec candidates satisfy G < 11.7. We also show the solar-analogue candidates obtained from a possible selection from GSP-Phot results in panel b, but we only show 1985 GSP-Phot candidates with G < 11.7. For comparison, panel c shows RVS spectra of 7589 randomly selected stars (i.e. no solar-analogue candidates) also with G < 11.7. In all panels, the red line shows the median in each pixel and the shaded red contours show the pixel-wise central 68% and 90% intervals. The solid blue line is identical in all three panels and shows the mean RVS spectrum of 13 solar analogues known from the literature.

In the text

Fig. 29. Colours of GSP-Spec solar-analogue candidates as a function of GSP-Phot extinction estimates. W1 and W2 denote AllWISE photometry (Cutri et al. 2021). We restrict the comparison to candidates with W1 > 8 mag, because AllWISE photometry suffers from saturation for brighter sources. Panel c: the red line is a linear increase with abp_gspphot offset by the mean GBP − W2 colour of 589 stars where A0 < 0.001 mag according to GSP-Phot. The red interval marks the uncertainty from the standard deviation of the mean. The quoted root-mean-square (RMS) difference is between the GBP − W2 colour and abp_gspphot plus the mean.

In the text

	Fig. 30. Variation of low-resolution BP and RP spectra of GSP-Spec solar-analogue candidates with the GSP-Phot A0 estimate. In order to make the BP and RP spectra comparable, they have been rescaled to an apparent magnitude of G = 15 + AG with AG taken from GSP-Phot.
In the text

Fig. 31. Comparison of the SPSS sample main parameters derived here with the two reference sets by Pancino et al. (2021): the best-fit parameters to the SPSS flux tables are shown in grey, while a collection of literature spectroscopic estimates is coloured according to the interstellar absorption A0 obtained here. Left, middle, and right panels: the cases of Teff, log g, and [Fe/H], respectively. The 1:1 line is shown in green in all panels.

In the text

Fig. 32. Proper motions in Galactic longitude (top) and latitude (bottom) as a function of Galactic longitude for the sample of 385 423 B-stars described in the text. The lines show the proper motions predicted from the rotation curve model parameters resulting from the fit to the data for stars at 500 pc (dashed line) and 2000 pc from the Sun (solid line, close to the median distance of stars in the sample).

In the text

Fig. 33. Distributions of the observed and model proper motions for the sample of 385 423 B-stars described in the text, with proper motions in Galactic longitude and latitude in the upper and lower panels, respectively. The black lines show the observed proper motion distributions. The thin orange lines are the predicted proper motion distributions for 200 randomly sampled MCMC model parameters. The mean of all such sample distributions is indicated by the thick dashed line.

In the text

Fig. 34. Distribution of planetary radii compared to the separation from their host star (orbital semi-major axis a) for planetary systems in the FGKM sample. The colour code indicates the Teff of the host star, while the symbol size indicates the orbital period in log10 scale (range = 0.57–364.8 days). The dotted lines indicate 1 REarth and 1 RJup and the Earth and Jupiter are denoted by the square symbols.

In the text

	Fig. 35. Planet mass and radius (in Jupiter units) of 95 planets with radial velocity and stellar parameters in the FGKM sample. The colour coding is as in Fig. 34 and symbol size corresponds to the semi-major axis. Some radius–mass models of planets are also shown; see text for details.
In the text

	Fig. 36. Literature ages versus FLAME ages for companions of UCDs not seen by Gaia. Filled circles correspond to age_flame_spec (i.e. using the Teff from GSP-Spec) and open circles are age_flame. The error bars represent the 16% and 84% percentiles.
In the text

	Fig. 37. Evolutionary tracks and UCD locations in the H-band absolute magnitude versus age diagram, adopting the companion age. The tracks are colour coded by mass. The dashed lines indicate the stellar to substellar transition zone (from 0.072 to 0.075 M⊙).
In the text