Cross-matching the General Catalogue of Variable Stars with the Gaia DR3 source catalogue Open Access

ABSTRACT

The most recent release of the General Catalogue of Variable Stars (GCVS) (Samus et al. 2017) contains 518 different variable types, in eight different variable categories. The catalogue has now reached its 5th version available via the VizieR service and can be considered one of the primary catalogues on the subject. The Gaia Data Release 3 (DR3) contains the most extensive catalogue of variable stars over the entire sky, still it is an intermediate step towards a better understanding of the quality of data and the automated algorithms being put in place to achieve a more concise classification as the work progresses. Ongoing work to identify all variability types in Gaia requires that a complete set of variable classes is represented. We investigated the most recent variability types listed therein, and compared them to the literature used to classify variable stars in Gaia. We have come across close to 10 000 individual variables in the GCVS that are not classified as variable in Gaia DR3, which include 56 variability types – some of which are bright stars. In this investigation, we demonstrate that there are still a large number of those bright stars missing from the Gaia variable classification. Clear indications show that variables with very short (<1 d), and very long periods, were missed by Gaia DR3 Gaia (Prusti et al. 2016 and Vallenari et al. 2022). Moreover, variables with large amplitudes were also missing. We discuss our findings in some detail.

1 INTRODUCTION

Since decades, the General Catalogue of Variable Stars (GCVS; Samus et al. 2017) is considered a primary source of variable stars. Being one of the first catalogues of its kind, the initial version was published in 1948 by Kukarkin and Parango, it has grown significantly since then. In its most recent version (version 5.1), 58 200 variable stars and over 500 variable types are listed. Those are divided into eight variable classes (Table A1).

Gaia (Prusti et al. 2016) is the most ambitious all-sky multi-epoch census of stars in our Milky Way, providing time domain data for astrometry, photometry, spectrophotometry, and spectroscopy. The Gaia Data Processing and Analysis Consortium (DPAC) is performing a systematic analysis of the variable sources iteratively delivered through data releases.

The emergence of numerous large-scale multi-epoch surveys, their profound impact on astronomy research, and the growing adoption of machine learning techniques have made it crucial to perform meticulous validation of the survey results for the variability type classification. This is necessary to ensure that the data can serve as effective training sets for future surveys, by so creating a virtuous spiral. Even within the iterative processing of Gaia, such efforts are required to improve the completeness and purity of the results from one data release to the next.

The DPAC embarked on the enormous task of compiling the most extensive collection of published literature on variability types utilizing the Gaia’s astrometry and photometry (Gavras et al. 2022). This vast compilation encompassed 4.9 million variable sources, and the consortium employed cleaned subsamples of this compilation as the training data for Gaia Data Release 3 (DR3). It is worth noting that although large-scale efforts are highly valuable, they should not overshadow the significance of conducting more targeted analyses on specific catalogues. Such efforts can provide a more focused understanding of particular aspects of the data and uncover novel insights that may have gone unnoticed in larger compilations.

In order to contribute to the reliability and completeness of identifying variability types in Gaia DR3, and to identify possible missing ones, we cross-matched the GCVS found in VizieR with the Gaia DR3 source catalogue (Eyer et al. 2022; Rimoldini et al. 2022; Vallenari et al. 2022). Out of the 58 200 stars listed in the GCVS, 9881 were found missing (as variables) in the Gaia DR3 catalogue, as they have a photometric variable flag set to the value NOT AVAILABLE.

In Fig. 1, we plot the sample of 9881 variables sources missing in Gaia DR3 and indicate in which category they fall. In this paper, we analyse those variability types that are missing and share our analysis.

2 METHOD

We extracted the entire GCVS from VizieR, which we then uploaded into the Gaia Archive at the European Space Astronomy Centre (ESAC). We then used the Gaia DR3 source catalogue to cross-match the two by using a cone search with a radius of 5 arcsec. This was found to be sufficiently small to make a positive identification of the stars in both catalogues. By applying angular distance and magnitude as described in Gavras et al. (2022), as well as the GCVS id running number, we filtered out the duplicates. To further ensure, and compare our results to other cross-matches, we ran spot checks on identifications found in SIMBAD astronomical data base. We also ran spot checks on the given G magnitude versus the GCVS magnitude for a number of variable types to gain confidence in the approach. As a further check on the physical parameters of the missing variables, we extracted the known periods and amplitudes for all variables in the GCVS. We then used a method applied by Kerschbaum & Hron (1992) to bin all stars in approximated intervals of periods and intervals, and to compare them to the missing set in order to determine trends and possible causes of not being included in the Gaia DR3 variable list.

3 RESULTS

In Appendix  A, we summarize the number of variables that were cross-matched and provide a list of types that were flagged as not classified in Gaia DR3. 9881 stars listed in GCVS have not been classified as variables in Gaia DR3. Of those, 56 variable types have not yet been positively identified. That is about 17 per cent of all GCVS variables, and equally, the percentile of missing variable types. Some of the GCVS stars have multiple components with differing magnitudes, which may be due to Gaia’s capability of detecting multiple stars that were not listed as such in the GCVS. There are also several bright stars that appear to fall in this category. To investigate the scope of missing bright stars, we show Hertzsprung–Russell (HR) diagrams in Fig. 2 to demonstrate this. We purposefully did not include the ‘colon’, ‘cst’, ‘*’, and ‘blank’ variable types that appear in the GCVS, as those have uncertainties in their variability.

The missing types in the colour versus absolute magnitude diagram (CaMD) are very instructive. We note that the Gaia DR3 catalogue of variables is missing a population of long-period variable (LPV) stars that forms a clump centred about G_BP − G_RP = 2–5, M_G = −2. Binaries and also many eruptive stars are missing, for example, in the probable pre-main-sequence phase.

Misidentification could very well be due to the eclipsing binaries being missed due to the temporal sampling of Gaia.

In Fig. 3, we show a histogram of the missing variable classes according to the GCVS variability categories. A large number of those stars that were missed fall in the eruptive category. Only 5 per cent make up the pulsating class of variables. We also make a representation of the CaMD of all the categories for those stars that are part of our overall missing list (see Fig. 4). This diagram confirms that the clump identified belongs to pulsating stars (LPVs), it also shows that Gaia DR3 is missing a significant population of Ap stars (grey points).

In order to further investigate the causes of having missed such a large number of variables, we looked at the periods of the GCVS variables that appeared to be missing in Gaia DR3. As Fig. 5 shows very short period variables (<1 d) and very long period variables have not yet been detected by Gaia. This could be due to the following.

• The current algorithms in place have not yet been fine-tuned to detect very short periods.

• In the case of long periods, 70 per cent of all stars are of variability type M or SR, and their periods (>1000 d) have not yet been sufficiently covered to allow a secure detection and classification.

Furthermore, and supporting our findings on missing GCVS variability classes as discussed above, we find that large-amplitude stars are often not included as variables in Gaia DR3. We plot the distribution of amplitudes for all missing stars in Fig. 6. The main physical reason for this low detection rates at large amplitudes is that those stars are very often of the type cataclysmic variables, being novae and supernovae (types N, NA, and NB) showing no relevant variability during the Gaia observations. The second quite incomplete variability detection in Gaia is at very low amplitudes with objects like Ap stars or small-amplitude δ Scuties.

Of course, periods and amplitudes are highly correlated in many variability types and frequently low amplitudes go hand in hand with short periods and large amplitudes with long periods.

4 CONCLUSIONS

We can summarize our findings as follows.

• Most of the unidentified variable types listed in GCVS contain a single star, which had not been detected variable in the Gaia DR3. See Table A2 for a list of all those types that have not yet been identified.

• Gaia’s photometric resolution should be used with more confidence to reclassify or unify some of the GCVS variables, which may allow to better understand the variability types and regroup some. This is in particular true for eclipsing binaries and phenomena, such as planetary nebulae (see Gavras et al. 2022 for more details).

• Various photometric systems must be compared with caution to ensure reliable results. This is exemplified in Fig. 1, which displays a set of pulsating stars exhibiting a lower magnitude in the GCVS (with probable heterogeneous photometric systems, but mostly V band) compared to the Gaia photometric broad-band. The shift occurs for the G band from 2 to 8. They are red objects mostly LPVs. Thanks to the Gaia BP and RP spectrophotometry, Montegriffo et al. (2022) were able be transformed the observed Gaia data into several well-known photometric systems. It is noteworthy that the systematic shift observed between the GCVS ‘magnitude’ and G band disappears once the GCVS magnitude is compared to the V-band-derived photometry as shown in Fig. 7.

• Discrepancies in magnitude can be due to the different phases of the variables being observed, and will require examining the light curves for those stars or by biases introduced for large amplitude variables when weighting procedures are used (see Eyer et al. 2022).

• Some variables that were originally classified in GCVS may have significantly diminished in brightness or ceased significant variability. This is true for at least 50 variables that were cross-checked in both sources, where the magnitude difference is greater than 2 as Fig. 1 shows.

• There were many duplicates in spite of the use of a very tight cone to search within the given radius. For some stars with high proper motion the propagation of this motion should be used. In this study it was not taken into account.

• In Gaia DR3, there can be overlapping types, i.e. one star is associated with several variability types, but there are no identifications of composite types such as hybrid pulsators, or a pulsator in a binary system. In the GCVS such designations, such as E/PSR, exist and form 56 variability types missed. It is hoped that in future Gaia data releases such identification of some of these variability composite types will be possible and published.

• Among the type missed, all the transients, the X-ray variability, eclipses are easily explained as missing. Similar can be said for very short and very long period variables, both not sampled properly with Gaia at the moment.

• The small amplitude variability in G can be also missed, such as in Ap stars a problem already identified in Eyer (1998) with the broad Hp band.

It is hoped that with the initial cross-match, a better understanding of those variables that are currently unaccounted for will be better identified in Gaia Data Release 4 (DR4), in order to complete the list of variable stars. With the introduction of new types of variability that can go into the process of identification, the sample will certainly grow bigger. It is beyond the scope of this paper to delve deeper into the intricacies of the individual variable classes, or try to differentiate why some variables have multiple types. This can only become evident, once we have a clearer picture of the samples available to us in Gaia DR4.

Much work will go into better understanding variability phenomena, with the help of larger samples of the individual classes, which will include a larger number of variables having similar characteristics, to better explain their physical phenomena, and thereby significantly reducing the uncertainties in the current classification. The process of data releases in Gaia is an iterative one that began with Data Release 1 (DR1) as a showcase, followed by Data Release 2 (DR2), which is considered a first step, now we are at DR3, an intermediate one, and ongoing towards a complete analysis in DR4 and Data Release 5 (DR5). As we gain more confidence in the artificial intelligence (AI) algorithms being developed, by introducing more details in the intricacies of these variables, we hope to optimize the analysis as much as possible and contribute to a clearer set of the major variability classes.

ACKNOWLEDGEMENTS

LE would like to thank Nicholas Chornay for discussions.

The majority of diagrams and plots were made using the topcat application (Taylor, 2005).

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.

DATA AVAILABILITY

The data underlying this paper were accessed from VizieR (https://vizier.unistra.fr). In particular, the General Catalogue of Variable Stars (GCVS): B/gcvs/gcvs_cat.

References

APPENDIX A: