TY - JOUR
AU - Marinoni,, S.
AB - Abstract The treatment of crowded fields in Gaia data will only be a reality in a few years from now. In particular, for globular clusters, only the end-of-mission data (public in 2022–2023) will have the necessary full crowding treatment and will reach sufficient quality for the faintest stars. As a consequence, the work on the deblending and decontamination pipelines is still ongoing. We describe the present status of the pipelines for different Gaia instruments, and we model the end-of-mission crowding errors on the basis of available information. We then apply the nominal post-launch Gaia performances, appropriately worsened by the estimated crowding errors, to a set of 18 simulated globular clusters with different concentration, distance and field contamination. We conclude that there will be 103–104 stars with astrometric performances virtually untouched by crowding (contaminated by <1 mmag) in the majority of clusters. The most limiting factor will be field crowding, not cluster crowding: the most contaminated clusters will only contain 10–100 clean stars. We also conclude that (i) the systemic proper motions and parallaxes will be determined to 1 per cent or better up to ≃15 kpc, and the nearby clusters will have radial velocities to a few km s−1; (ii) internal kinematics will be of unprecedented quality, cluster masses will be determined to ≃10 per cent up to 15 kpc and beyond, and it will be possible to identify differences of a few km s−1 or less in the kinematics (if any) of cluster sub-populations up to 10 kpc and beyond; (iii) the brightest stars (V ≃ 17 mag) will have space-quality, wide-field photometry (mmag errors), and all Gaia photometry will have 1–3 per cent errors on the absolute photometric calibration. astrometry, parallaxes, globular clusters: general 1 INTRODUCTION The ESA1 (European Space Agency) space mission Gaia2 (Perryman et al. 2001; Mignard 2005; Gaia Collaboration 2016a,b) is the successor of Hipparcos (Perryman et al. 1997), with the goal of providing astrometry for billions of point-like sources across the whole sky, with an error of 24 μas-level for stars of G ≃ 15 mag. Gaia will also provide broad-band magnitudes and colours for all sources, down to the Gaia white-light magnitude of G = 20.7 mag (V ≃ 21 mag). Low-dispersion spectra will also be obtained in two broad-bands with the red and blue spectrophotometers (BP and RP). Finally, Gaia will produce spectra in the calcium triplet region with the RVS (the radial velocity spectrometer) down to G ≃ 17 mag, from which radial velocities (RVs). Object classification and parametrization will be possible for all sources. Gaia was launched in 2013 December (de Bruijne, Rygl & Antoja 2014), and the first data release was in 2016 September (Gaia Collaboration 2016b), containing positions and white-light magnitudes for the best behaved stars, and additional information like parallaxes and proper motions for ≃2 million stars observed previously with Tycho-2 (Høg et al. 2000; Michalik, Lindegren & Hobbs 2015; Lindegren et al. 2016). Gaia will observe not only stars, but also tens of thousands of quasars, unresolved galaxies, Solar system objects, many transient and variable objects like supernovae, and finally the interstellar medium (Altavilla et al. 2012; Ducourant et al. 2014; Eyer et al. 2014; de Bruijne et al. 2015; Proft & Wambsganss 2015; Zwitter & Kos 2015; Bachchan, Hobbs & Lindegren 2016; Tanga et al. 2016). Gaia will also pose a challenge because of its data amount and complexity, pushing the astrophysical community farther into the path of big data and data mining (Gaia Collaboration 2016a). Gaia is limited in dense stellar fields, owing to the onboard and downstream telemetry bandwidth. For spectroscopy, an additional limitation is provided by the large physical size of the dispersed images and spectra on the focal plane. This has particular relevance for studies of the Galactic plane, the bulge and globular clusters (hereafter, GCs). In the first few Gaia data releases, disturbed sources like binaries, multiple stars and stars in crowded fields will likely not be part of the released material (Gaia Collaboration 2016b). In any case, the inclusion of a sufficient sample of stars at the main-sequence turn-off point or fainter – with good-quality measurements – is very important for GC research. Therefore, only the latest few Gaia releases (2020–2023) are expected to provide a significant breakthrough in GC research. To prepare the work, we explore in this paper the expected behaviour of Gaia data in several simulated Galactic GCs, adopting the official post-launch Gaia science performances and some simplified recipes to describe additional deblending error components, based on the Gaia deblending pipelines. A very preliminary – and now outdated – version of this work was presented by Pancino, Bellazzini & Marinoni (2013). This paper is organized as follows. In Section 2, we describe the current status of crowding treatment in Gaia; in Section 3, we present our crowding errors modelling; in Section 4, we describe our simulated clusters and the computation of Gaia observed quantities and final errors; in Section 5, we explore the simulations and show the potential of Gaia data for GC studies; in Section 6, we summarize the main results and draw our conclusions. 2 CROWDING IN GAIA DATA As mentioned in the previous section, Gaia is a complex space mission, with two different telescopes projecting their light on a large common focal plane, captured by 106 different CCDs (charge-coupled devices), and passing through different instruments. Gaia scans the whole sky by precessing its spin axis, and describing great circles on the sky that slowly drift with its precession. Each region is scanned from a minimum of ≃40 times, to a maximum of ≃250 times, with an average number of ≃70 passages for the AF, BP and RP, and ≃40 passages for RVS. All the CCDs in the focal plane are read in TDI (time-delayed integration) mode to closely follow Gaia's movement across the sky. On the focal plane, stars ‘move’ along the scanning direction, encountering different instruments: The first two columns of the CCD array are called Sky Mappers (SM), and they are used for the onboard detection of point-like sources; each of the SM columns sees only light from one of the two telescopes. The AF (astrometric field) provides astrometry and photometry of point-like sources in the G band, a white-light passband defined by the telescope and instrument transmission and by the CCD quantum efficiency (Fig. 1). The BP and RP provide low-resolution spectra (R = λ/δλ = 20–100) in the ranges shown in Fig. 1, and the integrated GBP and GRP magnitudes; the spectra are necessary for the chromaticity displacement correction in astrometric measurements. The RVS provides R ≃ 11 700 spectra in the calcium triplet region, for stars down to G ≃ 17 mag, depending on the object. Figure 1. Open in new tabDownload slide The Gaia nominal passbands for the astrometric field (G band), the blue and red spectrophotometers (BP and RP) and the radial velocity spectrometer (RVS). Figure courtesy of C. Jordi. Figure 1. Open in new tabDownload slide The Gaia nominal passbands for the astrometric field (G band), the blue and red spectrophotometers (BP and RP) and the radial velocity spectrometer (RVS). Figure courtesy of C. Jordi. To save telemetry bandwidth, given the enormous amount of data produced daily by Gaia, observations in each instrument are only transmitted for pixels contained in rectangular windows, that follow the detected point-like objects along the focal plane. For the faint stars, data in the allocated windows are binned in the AC direction by the onboard processing software. The adopted window sizes and relevant quantities for our treatment of crowding are listed in Table 1. More information can be found in the Gaia mission paper (Gaia Collaboration 2016a). Table 1. Adopted Gaia relevant quantities. We note that the AF window sizes are relevant for the astrometry; the BP/RP sizes for the photometry, chromaticity correction, and object classification and parametrization; the RVS window sizes are relevant for RV, object parametrization and abundance analysis. Size Size Description (arcsec) (pix) 0.176 789 1 (AC) Across scan (AC) pixel size 0.058 933 1 (AL) Along scan (AL) pixel size 0.176 789 3 (AL) Gaia PSF 2.121 468 12 (AC) AC window size (AF, BP and RP) 0.070 720 12 (AL) AL window size (AF)a 1.767 890 10 (AC) AC window size (RVS) 3.535 980 60 (AL) AL window size (BP and RP) 74.785 977 1269 (AL) AL window size (RVS) Size Size Description (arcsec) (pix) 0.176 789 1 (AC) Across scan (AC) pixel size 0.058 933 1 (AL) Along scan (AL) pixel size 0.176 789 3 (AL) Gaia PSF 2.121 468 12 (AC) AC window size (AF, BP and RP) 0.070 720 12 (AL) AL window size (AF)a 1.767 890 10 (AC) AC window size (RVS) 3.535 980 60 (AL) AL window size (BP and RP) 74.785 977 1269 (AL) AL window size (RVS) aIn some CCD columns in AF, as well as in the SM CCD columns, the windows are longer to allow for background measurements around the sources. The longer wings of these windows can in principle be used also to check the source profile behaviour outside normal window limits. Open in new tab Table 1. Adopted Gaia relevant quantities. We note that the AF window sizes are relevant for the astrometry; the BP/RP sizes for the photometry, chromaticity correction, and object classification and parametrization; the RVS window sizes are relevant for RV, object parametrization and abundance analysis. Size Size Description (arcsec) (pix) 0.176 789 1 (AC) Across scan (AC) pixel size 0.058 933 1 (AL) Along scan (AL) pixel size 0.176 789 3 (AL) Gaia PSF 2.121 468 12 (AC) AC window size (AF, BP and RP) 0.070 720 12 (AL) AL window size (AF)a 1.767 890 10 (AC) AC window size (RVS) 3.535 980 60 (AL) AL window size (BP and RP) 74.785 977 1269 (AL) AL window size (RVS) Size Size Description (arcsec) (pix) 0.176 789 1 (AC) Across scan (AC) pixel size 0.058 933 1 (AL) Along scan (AL) pixel size 0.176 789 3 (AL) Gaia PSF 2.121 468 12 (AC) AC window size (AF, BP and RP) 0.070 720 12 (AL) AL window size (AF)a 1.767 890 10 (AC) AC window size (RVS) 3.535 980 60 (AL) AL window size (BP and RP) 74.785 977 1269 (AL) AL window size (RVS) aIn some CCD columns in AF, as well as in the SM CCD columns, the windows are longer to allow for background measurements around the sources. The longer wings of these windows can in principle be used also to check the source profile behaviour outside normal window limits. Open in new tab 2.1 Gaia deblending and decontamination pipelines Crowding treatment ideally requires a preliminary evaluation of the crowding conditions of each source and transit, based on knowledge of the scene, i.e. the distribution and characteristics of all the neighbouring sources, as collected before the current observation. Different pipelines are employed for different instruments and to treat different cases. Below we describe the current status of the ones that are relevant for this study. 2.1.1 AF deblending Stars closer than the Gaia PSF width (which is assumed here3 to be 0.177 arcsec, see also Table 1) can be recognized as blends already in the astrometric processing of AF data. These blends can be detected, for example, from the high errors in the centroid determination and its wobbling from one transit to another, or by photometric variability, RV variability, or in general because the fits of the PSF or LSF to the data show large residuals or require more than one component. Assessing whether a star is isolated or has detected or suspected companions is a crucial task. Multiple or blended objects are redirected to the NSS (non-single stars) pipeline where an attempt to model them as binary systems is carried out (Pourbaix 2011; Gaia Collaboration 2016a). If none of the available binary models or configurations produces a good fit to the data, then a stochastic model is employed to derive preliminary parameters of the secondary (or tertiary and so on) source. Therefore we can assume that – considering also the small Gaia PSF – the vast majority of NSS will be known. Stars that are further apart than the tiny Gaia PSF are easily deblended by the PSF fitting algorithms, with results much more similar to HST (Hubble Space Telescope) than to typical ground-based telescopes. This is the main reason why – as we will see in the following – astrometric measurement uncertainties (and G magnitudes) are in general less affected by crowding, unlike BP/RP and RVS measurements. 2.1.2 BP and RP deblending and decontamination pipelines BP and RP transits of sources that are not isolated will be called here either blended or contaminated. The idea is that when the sources are so close that they occupy the same window or interfering windows in the vast majority of the transits (blends, with D < 2.12 arcsec), they will require a different treatment than objects that will often be assigned well-separated windows, or transits that are just altered by the flux of a bright source that is well outside of the window (contaminants, with D < 3.54 arcsec). This is illustrated in Fig. 2, where different cases are shown together with the background evaluation regions, and in Fig. 3, showing how the differing orientation and AL projected distance between sources affects window assignment. Figure 2. Open in new tabDownload slide An illustration of the crowding effects on BP and RP dispersed images, with their assigned windows (black rectangles). Different cases are represented, some going beyond the scope of this paper. Stars are represented by their surface density profiles (solid contours), with the outmost contours at the background level. The brightest star (G = 15 mag) is in window (2), while the faintest one (G = 20 mag) is in window (1). Stars fainter than the Gaia magnitude limit (G = 20.7 mag) are assigned no window, like for star (3). The coloured portions of the windows are the background samples, in green when they are free from contamination, and in red when they are contaminated and cannot be used. When stars are too close, they can be assigned to the same widow or two (or more) truncated windows, like in cases (9), (10) and (11). There are also empty windows (virtual objects) like in case (6). Figure courtesy of A. Brown. Figure 2. Open in new tabDownload slide An illustration of the crowding effects on BP and RP dispersed images, with their assigned windows (black rectangles). Different cases are represented, some going beyond the scope of this paper. Stars are represented by their surface density profiles (solid contours), with the outmost contours at the background level. The brightest star (G = 15 mag) is in window (2), while the faintest one (G = 20 mag) is in window (1). Stars fainter than the Gaia magnitude limit (G = 20.7 mag) are assigned no window, like for star (3). The coloured portions of the windows are the background samples, in green when they are free from contamination, and in red when they are contaminated and cannot be used. When stars are too close, they can be assigned to the same widow or two (or more) truncated windows, like in cases (9), (10) and (11). There are also empty windows (virtual objects) like in case (6). Figure courtesy of A. Brown. Figure 3. Open in new tabDownload slide Simplified description of transits in BP/RP. In case (a), two stars are closer than half the AC window size, and no matter the orientation of the satellite, they will always be assigned the same window (or truncated windows). In case (b), two stars have a distance that is in between the AC and AL half window sizes, and depending on the orientation, they are sometimes assigned the same window (or truncated window), sometimes different windows. Depending on their brightness, they can still contaminate each other's window. In case (c), the stars are farther apart than the AL window size, and they are always assigned two different windows but, if one of the two stars is very bright, it can still contaminate the other significantly. Figure 3. Open in new tabDownload slide Simplified description of transits in BP/RP. In case (a), two stars are closer than half the AC window size, and no matter the orientation of the satellite, they will always be assigned the same window (or truncated windows). In case (b), two stars have a distance that is in between the AC and AL half window sizes, and depending on the orientation, they are sometimes assigned the same window (or truncated window), sometimes different windows. Depending on their brightness, they can still contaminate each other's window. In case (c), the stars are farther apart than the AL window size, and they are always assigned two different windows but, if one of the two stars is very bright, it can still contaminate the other significantly. For blended sources, there are different pipelines that can be used in different phases of the mission. Blind pipelines (without knowledge of the scene) can be applied in the initial phases, when the history of each source in different instruments has not built up to a sufficient level. The two (or more) blended sources can be roughly modelled without a priori knowledge of their exact positions, astrophysical parameters and fluxes, just by modelling them with two overlapping spectral energy distributions. This approach was successfully applied to Gaia commissioning data4 of bright stars, recovering the vast majority of the Ticho-2 binary and double stars. Once a few transits are accumulated, they can be better disentangled if they are modelled simultaneously (per source rather than per transit), even if there is still not enough information on the scene, improving the quality of the reconstruction. Finally, once the scene is well characterized and the history of the source is well developed, other parameters of the modelling like the spectral type, the projected distance of the sources along scan and the relative fluxes, can be used to further improve the involved sources reconstruction. For contaminated sources, it is necessary to know well the flux of the contaminating sources around, thus knowledge of the scene is necessary. Each known source is modelled to reconstruct the flux even at large distances from the window (especially for bright stars). The amount of reconstructed contaminating flux from neighbouring sources is computed in each pixel of the contaminated source window, and subtracted. For all these reasons, decontamination will have to be performed contextually with the scene reconstruction and crowding evaluation. 2.1.3 RVS deblending and decontamination pipelines The deblending philosophy adopted by RVS is slightly different from the one adopted in AF, or BP and RP. The deblending pipelines are being adapted and rewritten to mitigate the stray light issues found after launch5 (Mora et al. 2016) that impact mostly on RVS spectra, with an expected loss of ≃1.4 mag in sensitivity (Seabroke et al. 2016). We therefore based our modelling of deblending errors on a previous algorithm that was based on general geometrical considerations (Allende Prieto 2008). The treatment of contamination is instead part of the background treatment, which is a vital part of RVS processing, because of the length of the spectra (see Table 1). The background is divided in diffused and point-like, and treated separately. In both cases, a modelling takes place based on knowledge of the scene and all available data on stray light and nearby objects. The model produces an estimate of the total background in the RVS windows that is subtracted from the source signal. Clearly, the model becomes more and more accurate as Gaia data are progressively accumulated and therefore the best results will be obtained towards the end of the mission (Janssen, Huckle & Seabroke, private communication). 2.1.4 Image reconstruction and future possibilities To facilitate the study of crowded areas, two paths are being followed. On the one hand, some crowded regions like the centre of ω Centauri or NGC 1818 in the Magellanic Cloud were imaged and transmitted to the ground in 2D mode in the SM,6 during commissioning, and further regions might be observed in 2D during the mission lifetime. The goal of these 2D images is to fully reconstruct the scene around each Gaia source to help improving and testing the deblending algorithms in crowded areas. On the other hand, pipelines for the 2D image reconstruction from individual Gaia transits are being developed. One such pipeline, called Source Environment Analysis (sea) pipeline, is based on the FastStack image reconstruction (Harrison 2011; Gaia Collaboration 2016a). The initial tests are promising and the reconstructed images could be useful, among other things, to test and improve the deblending and decontamination pipelines as well. Therefore, the simulations presented in this paper, being based on the current status of the deblending and decontamination pipelines, have to be seen as generally pessimistic. 2.2 A note about completeness in Gaia data A particular limitation of the Gaia design is related to the number of simultaneous object and background (virtual object) windows that can be allocated by the onboard detection algorithm in each CCD and in any given moment. This number varies from ≃35 000 to 1050 000 objects per square degree, depending on the instrument (see Gaia Collaboration 2016a, for more details). Areas of the sky that can suffer from this limitation are clearly the Galactic bulge and plane, and all other fields of view that happen to overlap them on the Gaia focal plane.7 More relevant to this study, the limitation applies to the central regions of GCs, where the local star density can be very high. At the single transit level, completeness is also influenced by the onboard detection algorithm. Every source that enters the SM is assigned a window, prioritizing brighter objects down to the limiting magnitude. The windows follow the stars along the focal plane as Gaia scans the sky, and when the stars exit the field, the window slots are ‘freed’ and can be assigned to new sources again. For this reason, the central dense areas of the clusters will – statistically speaking – contain more (bright) stars, and thus obtain more windows with respect to the periphery of the cluster, in the AC direction. Different scans, oriented in the sky with different angles, will lose different stars and thus at the end of mission, each star that is not entirely lost will have lost part of its transits. This is illustrated in the simulated cluster in Fig. 4, and is also visible in the ω Centauri actual Gaia data shown in a press release.8 Figure 4. Open in new tabDownload slide Effect of the onboard detection on completeness. The BP/RP image of a GC was simulated with GIBIS (Babusiaux 2005). The green rectangles are the assigned windows. There is an apparent loss of stars above and below the cluster centre, where the AC stellar density is lower and windows have a lower probability of being assigned to a source. The percentage of stars lost is roughly constant with stellar density in each transit, but each transit has a different orientation. As a results, the end-of-mission incompleteness will tend to be flatter with distance from the centre than in usual GC photometries. Figure 4. Open in new tabDownload slide Effect of the onboard detection on completeness. The BP/RP image of a GC was simulated with GIBIS (Babusiaux 2005). The green rectangles are the assigned windows. There is an apparent loss of stars above and below the cluster centre, where the AC stellar density is lower and windows have a lower probability of being assigned to a source. The percentage of stars lost is roughly constant with stellar density in each transit, but each transit has a different orientation. As a results, the end-of-mission incompleteness will tend to be flatter with distance from the centre than in usual GC photometries. We do not attempt to simulate these completeness effects in this paper, as they would require a full end-of-mission simulation of different lines of sight in the sky, including the detailed behaviour of the onboard detection algorithms. 3 MODELLING OF CROWDING ERRORS As mentioned above, our goal is to illustrate what Gaia can do for GC studies, rather than attempting a rigorous simulation of deblending errors. Therefore, we will model available simulations of crowded Gaia data, to derive simple formulas with as few free parameters as possible. The crowding error models derivation for different Gaia instruments and different types of blends are described in the following sections. The models describe the expected percentage errors on flux caused by crowding as a function of relevant parameters like the contaminating flux, contaminating colour, and distance of the contaminating object(s). Later, in Section 4 we will transform our flux errors into errors on other parameters like positions, parallaxes, proper motions and RVs. We will then use these errors to worsen the nominal Gaia post-launch science performances, depending on the crowding level suffered from each of our simulated GC stars. 3.1 ‘Classic’ blends We term in this paper classic blends those stars that are closer than the AF Gaia PSF (0.177 arcsec, see Table 1), and that will remain so along the five years of Gaia observations. Stars that will move apart thanks to their parallax or proper motion differences, or that have RV differences that allow to deblend them, will not be considered as classic blends, but treated as normal Gaia blends (see below). The precession of Gaia has an important implication for crowding treatment: each time Gaia scans a particular region of the sky, it is oriented differently, and thus the projected distance of two stars in the SM columns, on which the onboard object detection and window assignment are based, are different (see Fig. 3). Therefore, as described in Section 2.1.1, even the closest classic blends have a chance to be deblended, although it is difficult to simulate realistically the errors caused by deblending when so many observables enter iteratively the deblending procedure along the whole mission. In this paper, we will assume that all blends of stars closer than the Gaia PSF will be recognized as such, thanks to the mission long lifetime, its tiny astrometric and photometric errors, and sophisticated data analysis methods. Because of the limited number of classic blends,9 this assumption is not going to have a significant impact on the following analysis. If the relatively small number of classic blends comes as a surprise, it has to be noted that Gaia observations only reach V ≃ 21 mag in GCs, and thus the actual crowding levels are much lower than those of the typically deeper photometry from the ground or with HST. We will also assume that the actual deblending errors follow the relation adopted for the BP and RP blends (see below), extrapolating the present simulations towards smaller source distances. While highly uncertain, this is our only available estimate of the Gaia deblending capabilities at the present stage. For RVS, there currently are no plans to attempt deblending of sources closer than 0.17 arcsec, and therefore we will not attempt to simulate the effect of crowding in these RVS sources. 3.2 ‘Gaia’ blends As discussed in Section 2.1.2, Gaia blends are all those cases in which two stars are closer than half the AC window size of the relevant instrument (Table 1, these all fall into case (a) of Fig. 3). Because the AF image deblending will be simpler (the AL profile is sharper) than that of BP, RP and RVS spectra, we will conservatively base our estimate of crowding errors for deblending on a set of BP and RVS simulations. For the BP and RP blending simulations, we used a per-transit, partially informed deblending pipeline10 on the GIBIS (Gaia Instrument and Basic Image Simulator; Babusiaux 2005) simulated BP spectra of a typical GC red giant, with Teff = 4000 K and G = 15 mag. The star was blended with other similar stars, having either Teff = 4000 or 6000 K, G = 15.75 or 17.00 mag, and at different distances equal to 0.25, 0.5 and 1.0 arcsec (see Fig. 5). Figure 5. Open in new tabDownload slide The simulated BP deblending errors as a function of distance (circles), based on a G = 15 mag star with Teff = 4000 K, and our derived error models (curves). The dotted vertical lines mark the applicability range, from the Gaia PSF size (0.177 arcsec) to half the length of the BP/RP windows (1.76 arcsec). Blends with different characteristics are shown: cooler by 2000 K (small circles and solid curves), with the same temperature (medium circles and dashed curves), and hotter by 2000 K (large circles and dotted curves). The colours refer to the relative contaminating flux: 17 per cent (blue), 50 per cent (green), 200 per cent (yellow) and 630 per cent (red). Figure 5. Open in new tabDownload slide The simulated BP deblending errors as a function of distance (circles), based on a G = 15 mag star with Teff = 4000 K, and our derived error models (curves). The dotted vertical lines mark the applicability range, from the Gaia PSF size (0.177 arcsec) to half the length of the BP/RP windows (1.76 arcsec). Blends with different characteristics are shown: cooler by 2000 K (small circles and solid curves), with the same temperature (medium circles and dashed curves), and hotter by 2000 K (large circles and dotted curves). The colours refer to the relative contaminating flux: 17 per cent (blue), 50 per cent (green), 200 per cent (yellow) and 630 per cent (red). We then modelled the deblending (relative) errors on the recovered flux as an exponential αeβ D, where D is the distance between the two blended stars. The coefficients α and β vary with the temperature difference and the relative contaminating flux, respectively, as illustrated in Fig. 5. For RP, the deblending errors were arbitrarily multiplied by 1.5, owing to the generally wider AL shape of the spectra that produces worse deblending results than in BP. These simplified recipes are intentionally pessimistic, to accommodate for unexpected sources of error. The approximate range of validity of the model (highlighted in Fig. 5) goes from the Gaia PSF size to half the AL size of the BP and RP windows. As discussed in the previous section, we applied an extrapolation of this relation to D < 0.17 arcsec to classic blends, for lack of a more detailed study at this stage. For RVS, the only available deblending simulations were computed before launch (Allende Prieto 2008, see also Fig. 6), when the windowing scheme and background treatment were different (see Section 2.1.3). Thus, the error modelling presented here will be much more uncertain for RVS than for BP and RP, or for AF. The temperature or colour difference between two sources has little impact on the deblending ability of the RVS pipelines, because the spectral shape is approximately flat along the RVS windows. What mostly counts is the relative contaminating flux. As a result, the best-fitting exponential laws had a constant β, and α varying with the relative contaminating flux. As mentioned above, we did not extend the model below the PSF size for RVS (shaded region in Fig. 6). Figure 6. Open in new tabDownload slide The simulated pre-launch RVS deblending errors, for different contamination levels, and our simplified exponential fits. The shaded area (D < 0.17 arcsec) is not yet treated in the current RVS deblending pipelines, and our fits there are extremely uncertain, therefore we will not provide any simulation for classic blends in RVS. Figure 6. Open in new tabDownload slide The simulated pre-launch RVS deblending errors, for different contamination levels, and our simplified exponential fits. The shaded area (D < 0.17 arcsec) is not yet treated in the current RVS deblending pipelines, and our fits there are extremely uncertain, therefore we will not provide any simulation for classic blends in RVS. 3.3 Contaminants For contaminated stars [cases (b) and (c) in Fig. 3], the reconstruction of star fluxes at distances larger than the window size is subject to a variety of uncertainty sources, and the relevant pipelines will be effective only when a sufficient history is accumulated to build a scene. Not only the stars positions are needed, but also their spectral energy distributions. Additionally, a complete characterization of the PSF profiles AC and AL, at different wavelengths and positions on the focal plane will be fundamental. Therefore no detailed simulations are available at the moment (Section 2.1.2). However, we do know that a step-like behaviour is expected in the errors at 1.76 arcsec, where the processing switches from decontamination to deblending, in the sense that decontamination pipelines perform slightly worse than the deblending ones. There will be, on purpose, a ‘grey area’ in projected distance between sources, where both pipelines will be applied, for cross-validation purposes. Some tests were performed to develop the current decontamination algorithms (Piersimoni, De Luise & Busso 2011, and De Luise, private communication). According to those tests, we decided to roughly approximate the decontamination errors with a factor of 2 worsening with respect to deblending errors. This is a pessimistic assumption. Our modelling of decontamination residual errors is applied to contaminant stars with distances ranging from half the AC window size to the full AL window size or to the contaminating star's ‘size’, whichever is larger. We modelled the contaminating stars sizes as a linear function of G magnitude (Pulone & Marrese, private communication, see also Fig. 2). Stars fainter than G ≃ 15 mag are generally smaller than the BP and RP AL window size.11 For RVS, the discrete background pipelines are being presently integrated and tested, so no simulations are available at the moment and we just extended our simplified treatment of the RVS blends (previous section) to RVS contaminants, as done for BP and RP, using the appropriate window sizes. 4 CLUSTER SIMULATIONS We simulated 18 different GCs, as summarized in Table 2, with different concentration, background contamination and distance. The final simulated clusters, after the Gaia science performances and the crowding error simulations are included, are presented in Table 3. Table 2. Simulated cluster properties and crowding evaluation results. For each of the 18 simulated GCs we list (see the text for more details): (1) the GC ID number; (2) the GC concentration parameter, c = log(rt/rc); (3) the GC projected distance; (4) the type of simulated background; (5) the total number of simulated stars (GC and background) above Gaia detection limit; (6) the number of GC members; (7) the number of classic blends; (8) the number of Gaia blends; (9) the number of contaminated stars; (10) the number of clean stars, i.e. contaminated by less than 0.1 per cent in flux (roughly 1 mmag); (11) the quick GC designation used in the paper and (12) some examples of GCs with similar distance, concentration and background field contamination. Cluster c Distance Background ntot nGC nclassic nblends ncontam nclean Designation Similar to (dex) (kpc) # 1 1.0 5 halo 73 385 72 510 4621 45 101 52 142 16 093 Easy case M13, M92 # 2 1.0 10 halo 30 200 29 325 3521 21 550 23 754 4275 M92 # 3 1.0 15 halo 14 027 13 152 2203 10 312 11 203 1495 NGC 5053 # 4 1.0 5 disc 54 923 26 816 1028 14 115 17 289 7379 M71 # 5 1.0 10 disc 33 435 5328 326 3302 3812 1164 Intermediate case M56, NGC 2298 # 6 1.0 15 disc 29 737 1630 123 1048 1208 311 M79 # 7 1.0 5 bulge 1537 592 71 996 9833 69 887 71 600 218 M22, NGC 6553 # 8 1.0 10 bulge 1494 601 29 005 5620 28 497 28 919 36 M9, NGC 6638 # 9 1.0 15 bulge 1478 538 12 942 2998 12 762 12 911 15 Pal 11 # 10 2.5 5 halo 73 385 72 510 11 902 47 043 53 029 15 623 M5 # 11 2.5 10 halo 30 200 29 325 6914 21 977 23 944 4117 M3 # 12 2.5 15 halo 14 027 13 152 3663 10 407 11 176 1525 NGC 5466 # 13 2.5 5 disc 54 923 26 816 3236 15 372 18 085 6946 M92, Pal 10 # 14 2.5 10 disc 33 435 5328 906 3491 3968 1055 M79, NGC 1851 # 15 2.5 15 disc 29 737 1630 292 1097 1250 300 M15 # 16 2.5 5 bulge 1537 592 71 996 16 564 70 125 71 648 190 NGC 6540, NGC 6558 # 17 2.5 10 bulge 1494 601 29 005 8726 28 556 28 924 40 NGC 6325, NGC 6342 # 18 2.5 15 bulge 1478 538 12 942 4378 12 770 12 914 13 Difficult case M54, NGC 6517 Cluster c Distance Background ntot nGC nclassic nblends ncontam nclean Designation Similar to (dex) (kpc) # 1 1.0 5 halo 73 385 72 510 4621 45 101 52 142 16 093 Easy case M13, M92 # 2 1.0 10 halo 30 200 29 325 3521 21 550 23 754 4275 M92 # 3 1.0 15 halo 14 027 13 152 2203 10 312 11 203 1495 NGC 5053 # 4 1.0 5 disc 54 923 26 816 1028 14 115 17 289 7379 M71 # 5 1.0 10 disc 33 435 5328 326 3302 3812 1164 Intermediate case M56, NGC 2298 # 6 1.0 15 disc 29 737 1630 123 1048 1208 311 M79 # 7 1.0 5 bulge 1537 592 71 996 9833 69 887 71 600 218 M22, NGC 6553 # 8 1.0 10 bulge 1494 601 29 005 5620 28 497 28 919 36 M9, NGC 6638 # 9 1.0 15 bulge 1478 538 12 942 2998 12 762 12 911 15 Pal 11 # 10 2.5 5 halo 73 385 72 510 11 902 47 043 53 029 15 623 M5 # 11 2.5 10 halo 30 200 29 325 6914 21 977 23 944 4117 M3 # 12 2.5 15 halo 14 027 13 152 3663 10 407 11 176 1525 NGC 5466 # 13 2.5 5 disc 54 923 26 816 3236 15 372 18 085 6946 M92, Pal 10 # 14 2.5 10 disc 33 435 5328 906 3491 3968 1055 M79, NGC 1851 # 15 2.5 15 disc 29 737 1630 292 1097 1250 300 M15 # 16 2.5 5 bulge 1537 592 71 996 16 564 70 125 71 648 190 NGC 6540, NGC 6558 # 17 2.5 10 bulge 1494 601 29 005 8726 28 556 28 924 40 NGC 6325, NGC 6342 # 18 2.5 15 bulge 1478 538 12 942 4378 12 770 12 914 13 Difficult case M54, NGC 6517 Open in new tab Table 2. Simulated cluster properties and crowding evaluation results. For each of the 18 simulated GCs we list (see the text for more details): (1) the GC ID number; (2) the GC concentration parameter, c = log(rt/rc); (3) the GC projected distance; (4) the type of simulated background; (5) the total number of simulated stars (GC and background) above Gaia detection limit; (6) the number of GC members; (7) the number of classic blends; (8) the number of Gaia blends; (9) the number of contaminated stars; (10) the number of clean stars, i.e. contaminated by less than 0.1 per cent in flux (roughly 1 mmag); (11) the quick GC designation used in the paper and (12) some examples of GCs with similar distance, concentration and background field contamination. Cluster c Distance Background ntot nGC nclassic nblends ncontam nclean Designation Similar to (dex) (kpc) # 1 1.0 5 halo 73 385 72 510 4621 45 101 52 142 16 093 Easy case M13, M92 # 2 1.0 10 halo 30 200 29 325 3521 21 550 23 754 4275 M92 # 3 1.0 15 halo 14 027 13 152 2203 10 312 11 203 1495 NGC 5053 # 4 1.0 5 disc 54 923 26 816 1028 14 115 17 289 7379 M71 # 5 1.0 10 disc 33 435 5328 326 3302 3812 1164 Intermediate case M56, NGC 2298 # 6 1.0 15 disc 29 737 1630 123 1048 1208 311 M79 # 7 1.0 5 bulge 1537 592 71 996 9833 69 887 71 600 218 M22, NGC 6553 # 8 1.0 10 bulge 1494 601 29 005 5620 28 497 28 919 36 M9, NGC 6638 # 9 1.0 15 bulge 1478 538 12 942 2998 12 762 12 911 15 Pal 11 # 10 2.5 5 halo 73 385 72 510 11 902 47 043 53 029 15 623 M5 # 11 2.5 10 halo 30 200 29 325 6914 21 977 23 944 4117 M3 # 12 2.5 15 halo 14 027 13 152 3663 10 407 11 176 1525 NGC 5466 # 13 2.5 5 disc 54 923 26 816 3236 15 372 18 085 6946 M92, Pal 10 # 14 2.5 10 disc 33 435 5328 906 3491 3968 1055 M79, NGC 1851 # 15 2.5 15 disc 29 737 1630 292 1097 1250 300 M15 # 16 2.5 5 bulge 1537 592 71 996 16 564 70 125 71 648 190 NGC 6540, NGC 6558 # 17 2.5 10 bulge 1494 601 29 005 8726 28 556 28 924 40 NGC 6325, NGC 6342 # 18 2.5 15 bulge 1478 538 12 942 4378 12 770 12 914 13 Difficult case M54, NGC 6517 Cluster c Distance Background ntot nGC nclassic nblends ncontam nclean Designation Similar to (dex) (kpc) # 1 1.0 5 halo 73 385 72 510 4621 45 101 52 142 16 093 Easy case M13, M92 # 2 1.0 10 halo 30 200 29 325 3521 21 550 23 754 4275 M92 # 3 1.0 15 halo 14 027 13 152 2203 10 312 11 203 1495 NGC 5053 # 4 1.0 5 disc 54 923 26 816 1028 14 115 17 289 7379 M71 # 5 1.0 10 disc 33 435 5328 326 3302 3812 1164 Intermediate case M56, NGC 2298 # 6 1.0 15 disc 29 737 1630 123 1048 1208 311 M79 # 7 1.0 5 bulge 1537 592 71 996 9833 69 887 71 600 218 M22, NGC 6553 # 8 1.0 10 bulge 1494 601 29 005 5620 28 497 28 919 36 M9, NGC 6638 # 9 1.0 15 bulge 1478 538 12 942 2998 12 762 12 911 15 Pal 11 # 10 2.5 5 halo 73 385 72 510 11 902 47 043 53 029 15 623 M5 # 11 2.5 10 halo 30 200 29 325 6914 21 977 23 944 4117 M3 # 12 2.5 15 halo 14 027 13 152 3663 10 407 11 176 1525 NGC 5466 # 13 2.5 5 disc 54 923 26 816 3236 15 372 18 085 6946 M92, Pal 10 # 14 2.5 10 disc 33 435 5328 906 3491 3968 1055 M79, NGC 1851 # 15 2.5 15 disc 29 737 1630 292 1097 1250 300 M15 # 16 2.5 5 bulge 1537 592 71 996 16 564 70 125 71 648 190 NGC 6540, NGC 6558 # 17 2.5 10 bulge 1494 601 29 005 8726 28 556 28 924 40 NGC 6325, NGC 6342 # 18 2.5 15 bulge 1478 538 12 942 4378 12 770 12 914 13 Difficult case M54, NGC 6517 Open in new tab Table 3. Column-by-column description of the final simulated stars that will only be available in the electronic edition of the journal, and at CDS. Content Column Units Description Cluster (1) Cluster number Star (2) Star ID RA΄ (3) (deg) Position along the RA direction Dec΄ (4) (deg) Position along the Dec direction δcoord (5) (deg) Error on RA΄ and Dec΄ |$\mu _{\rm {RA^{{\prime}}}}$| (6) (mas yr−1) RA΄ proper motion μDec΄ (7) (mas yr−1) Dec΄ proper motion δμ (8) (mas yr−1) Error on proper motion ϖ (9) (mas) Parallax δϖ (10) (mas) Parallax error G (11) (mag) G-band integrated magnitude δG (12) (mag) G magnitude error GBP (13) (mag) GBP integrated magnitude δGBP (14) (mag) GBP magnitude error GRP (15) (mag) GRP integrated magnitude δGRP (16) (mag) GRP magnitude error Membership (17) True membership Content Column Units Description Cluster (1) Cluster number Star (2) Star ID RA΄ (3) (deg) Position along the RA direction Dec΄ (4) (deg) Position along the Dec direction δcoord (5) (deg) Error on RA΄ and Dec΄ |$\mu _{\rm {RA^{{\prime}}}}$| (6) (mas yr−1) RA΄ proper motion μDec΄ (7) (mas yr−1) Dec΄ proper motion δμ (8) (mas yr−1) Error on proper motion ϖ (9) (mas) Parallax δϖ (10) (mas) Parallax error G (11) (mag) G-band integrated magnitude δG (12) (mag) G magnitude error GBP (13) (mag) GBP integrated magnitude δGBP (14) (mag) GBP magnitude error GRP (15) (mag) GRP integrated magnitude δGRP (16) (mag) GRP magnitude error Membership (17) True membership Note. The pseudo-coordinates RA΄ and Dec΄ are just distances from the centre in deg, not true sky coordinates. The true membership, in the case of classic blends, relates to the brightest star of the blend. It is 1 for cluster members and 0 for field stars. Open in new tab Table 3. Column-by-column description of the final simulated stars that will only be available in the electronic edition of the journal, and at CDS. Content Column Units Description Cluster (1) Cluster number Star (2) Star ID RA΄ (3) (deg) Position along the RA direction Dec΄ (4) (deg) Position along the Dec direction δcoord (5) (deg) Error on RA΄ and Dec΄ |$\mu _{\rm {RA^{{\prime}}}}$| (6) (mas yr−1) RA΄ proper motion μDec΄ (7) (mas yr−1) Dec΄ proper motion δμ (8) (mas yr−1) Error on proper motion ϖ (9) (mas) Parallax δϖ (10) (mas) Parallax error G (11) (mag) G-band integrated magnitude δG (12) (mag) G magnitude error GBP (13) (mag) GBP integrated magnitude δGBP (14) (mag) GBP magnitude error GRP (15) (mag) GRP integrated magnitude δGRP (16) (mag) GRP magnitude error Membership (17) True membership Content Column Units Description Cluster (1) Cluster number Star (2) Star ID RA΄ (3) (deg) Position along the RA direction Dec΄ (4) (deg) Position along the Dec direction δcoord (5) (deg) Error on RA΄ and Dec΄ |$\mu _{\rm {RA^{{\prime}}}}$| (6) (mas yr−1) RA΄ proper motion μDec΄ (7) (mas yr−1) Dec΄ proper motion δμ (8) (mas yr−1) Error on proper motion ϖ (9) (mas) Parallax δϖ (10) (mas) Parallax error G (11) (mag) G-band integrated magnitude δG (12) (mag) G magnitude error GBP (13) (mag) GBP integrated magnitude δGBP (14) (mag) GBP magnitude error GRP (15) (mag) GRP integrated magnitude δGRP (16) (mag) GRP magnitude error Membership (17) True membership Note. The pseudo-coordinates RA΄ and Dec΄ are just distances from the centre in deg, not true sky coordinates. The true membership, in the case of classic blends, relates to the brightest star of the blend. It is 1 for cluster members and 0 for field stars. Open in new tab 4.1 3D simulated clusters The simulated clusters were computed with the mcluster code12 (Küpper et al. 2011), that is designed to produce initial conditions for N-body simulations. We used the code version (mcluster_sse) that implements the stellar evolution recipes by Hurley, Pols & Tout (2000). We produced two clusters containing 800 000 stars, with positions and velocities drawn from equilibrium King (1966) models. The two clusters differed only in their concentration parameter, c = log(rt/rc), one having c = 1.0 and the other c = 2.5 (see also Fig. 7). Among the key input parameters for the simulation, the half-mass radius was chosen to produce a (projected) half-light final radius similar to what typically observed for Galactic GCs: 4 pc for the c = 1.0 cluster and 3 pc for c = 2.5 one (see also Harris 1996, and following updates). Among the key input stellar population parameters, we adopted an age of 12 Gyr, a metallicity of Z = 0.0003, corresponding to [Fe/H] = −1.79 dex, and a Kroupa (2001) IMF (initial mass function). The simulations were performed without including binaries, for simplicity, and the resulting clusters are spherical and non-rotating. Figure 7. Open in new tabDownload slide The logarithmic density profiles (left-hand panels) and RV dispersion profiles (right-hand panels) for the two synthetic clusters with concentration parameter c = 1.0 (top panels) and c = 2.5 (bottom panels). The red lines are the corresponding theoretical density profiles (left-hand panels) and ±3 σ contours about the systemic velocity for King (1966) models with the same parameters. Figure 7. Open in new tabDownload slide The logarithmic density profiles (left-hand panels) and RV dispersion profiles (right-hand panels) for the two synthetic clusters with concentration parameter c = 1.0 (top panels) and c = 2.5 (bottom panels). The red lines are the corresponding theoretical density profiles (left-hand panels) and ±3 σ contours about the systemic velocity for King (1966) models with the same parameters. The main implications of our choices of input parameters are: (i) a blue HB (horizontal branch) morphology, which may be useful to test the performance of Gaia on relatively hot stars and (ii) a stellar M/LV ratio larger (≥3 instead of ≤2) than what typically observed in Galactic GCs having a present-day mass function compatible with a low-exponent power law (Paust et al. 2010). The high M/LV ratio is also due to a relatively large fraction of dark remnants: ≃25 per cent of the cluster mass is contributed by objects with M > M⊙.13 Independently of the recipes giving raise to such a large fraction of dark remnants and their actual nature, this is not a concern for our purpose: the only effect is that the potential well may be slightly deeper than for most real clusters and, consequently, the velocity dispersion is slightly larger, at fixed total luminosity and scale radius. What is relevant for us is that the central velocity dispersion is still in the right range and that the model is self-consistent. The resulting simulated clusters have absolute integrated magnitudes of MV = −7.6 mag, i.e. just above the peak of the clusters luminosity function (MV = −7.4 mag, according to Brodie & Strader 2006). The clusters projected surface density distributions are well fitted by a King (1966) profile, when obtaining rc (the core radius) from the fit. Finally, the derived central velocity dispersions agree well with the predictions by King (1966), where the total mass is obtained by summing the masses of all individual stars. 4.2 Simulated field contamination We simulated the field Galactic population in three directions using the Besanco̧n models14 (Robin et al. 2003, and references therein), including kinematics and without observational errors. The models produced catalogues of magnitudes (in the Johnson–Cousins system), 3D positions, and 3D motions. We adopted the ‘standard’ extinction law offered by the Besanco̧n simulator, a mean and diffused absorption of 0.7 mag kpc−1, neglecting small-scale variations, and decreasing away from the Galactic plane with a smooth Einasto profile (for more details, see Robin et al. 2003). We selected all available spectral types in a magnitude range of 0 < V< 25 mag. The simulations were computed in a 0.7 × 0.7 deg2 centred on each of the following three directions (see also Table 2): a rather empty ‘halo’ field at l = 150 and b = 80 deg, containing ≃2300 stars in total; this is meant to represent the best cases of low background contamination in Gaia observations; an extremely dense ‘bulge’ field at l = 5 and b = 5 deg, containing almost 6 millions of stars, which should cover even the most crowded Gaia observations, when two relatively crowded lines of sight overlap on the focal plane (see also Section 2); a crowded ‘disc’ field in the anticentre direction, at l = 180 and b = 0, containing roughly 140 000 stars, with a relatively high extinction of up to ≃2 mag in the V band (even if these values are only reached in ≃25 per cent of the known GCs in Harris 1996); while not as extreme as the bulge line of sight, this direction provides still severe crowding levels – on the Galactic plane – coupled with significant reddening. 4.3 Simulated Gaia measurements Each of the two clusters (see Section 4.1) was projected at three different distances of 5, 10 and 15 kpc and combined with each of the three backgrounds, to produce the 18 clusters in Table 2. A typical systemic RV of 100 km s−1 was assigned to each cluster (broadly compatible with the values listed in the Harris 1996, catalogue, in its most recent version), while a proper motion of −5 mas yr−1 in both the RA and Dec was assigned to all clusters (broadly compatible with the values measured by, e.g. Dinescu et al. 1999). Distances were also converted into parallaxes. Integrated Gaia magnitudes G, GBP and GRP (see also Section 2 and Fig. 1) were obtained for each star using the formulae provided by Jordi et al. (2010, using their coefficients from Table 3), starting from the B−V colour. For extinction, we adopted the pessimistic hypothesis that the cluster reddening was equal to the corresponding background field's highest value (i.e. as if the cluster was entirely behind the field population). While this assumption is not completely realistic, it gave us the possibility of exploring higher reddening conditions, at least for the disc projections. This yielded an AV of 0.062 mag for the halo projections, 0.080 mag for the bulge projections and of 2.149 mag for the disc projections. 4.4 Crowding evaluation For each simulated star in each GC, we counted all neighbours (with flux above 0.1 per cent of the star) within a specified distance, depending on the considered instrument and on the types of blends (Section 3). We adopted an exhaustive algorithm for neighbours search.15 The relative number of each type of neighbours (classic blends, Gaia blends, contaminants) for the 18 simulated clusters is shown in Fig. 8 as a function of distance from the GC centre. As can be seen, the parameter that dominates the fraction of classic blends is the GC concentration: core-collapsed clusters have more than twice the number of classically blended stars in their central parts compared to normal GCs. The vast majority of classic blends are in the GC central regions, and are blends of GC members with other GC members. Distance and reddening have a curious effect: because GCs become fainter with increasing distance and reddening, their luminosity function crosses the detection limit of Gaia where it is less populous. As a consequence, not only the number of stars detected by Gaia decreases, but also the probability of blending, and so the blends and contaminant fractions. On the other hand, the effect of field contaminants is to increase the fraction of Gaia blends and contaminants at all distances from the centre. Extreme cases are the bulge background GCs, where the vast majority of stars are blended and/or contaminated to some degree at all radii, mostly by field stars. Figure 8. Open in new tabDownload slide Results of the crowding evaluation on simulated GCs. Each panel displays the relative fraction of classic blends, Gaia blends and contaminants as a function of distance from the GC centre. Only stars that are blended or contaminated by at least 0.01 mag in total (i.e. ≃1 per cent of their flux) are shown in the figure. Only classic blends have a significant effect on the astrometric performances, while BP/RP and RVS are significantly affected also by Gaia blends and contaminants. Figure 8. Open in new tabDownload slide Results of the crowding evaluation on simulated GCs. Each panel displays the relative fraction of classic blends, Gaia blends and contaminants as a function of distance from the GC centre. Only stars that are blended or contaminated by at least 0.01 mag in total (i.e. ≃1 per cent of their flux) are shown in the figure. Only classic blends have a significant effect on the astrometric performances, while BP/RP and RVS are significantly affected also by Gaia blends and contaminants. We listed in Table 2 some key quantities resulting from the crowding evaluation. As can be seen, the number of stars that are clean (i.e. contaminated by less than 1 mmag) can be substantial when the background is not extreme. The vast majority of Gaia blends and contaminants will be known, as a large fraction of the classic blends (see Section 2.1.1). Therefore, it will always be possible to select a reliable sample of clean stars from Gaia data. Additionally, we report here that already in the first Gaia release, which does not contain stars in truncated windows or stars with low-quality measurements (Gaia Collaboration 2016b), more than 200 000 stars were actually detected in the ω Cen field. This compares well with our #1 cluster, which is reasonable considering the higher mass of ω Cen, and gives support to the figures discussed above and reported in Table 2. 4.5 Simulated Gaia errors To compute the Gaia errors, we combined the post-launch Gaia science performances (Gaia Collaboration 2016a) with our crowding error estimates. For the magnitudes we combined the two errors in quadrature, because they are fully independent. For the astrometric and spectroscopic measurements, the crowding errors are not fully independent from the Gaia science performances, because they both depend on the star magnitude. Therefore, we conservatively summed the errors in modulus. 4.5.1 Crowding errors We used the modelling described in Section 3, which provided a relative error on the flux of each star, caused by the various type of blends and contaminants affecting it. These modelled errors represent well the crowding errors for G, GBP and GRP, i.e. on the photometry. To evaluate the crowding errors on astrometry, we considered two components: the error on centroiding and a chromatic shift. For the centroiding errors, only classic blends were considered, because only stars closer than the PSF are expected to have a significant impact on the final centroid measurement of a star. We expect that the PSF of a star will be perturbed by the residual flux left by the close companions after the deblending procedure, and these residuals will depend on the contaminating flux and distance of the blending stars, as modelled in Section 3. We therefore computed the percentual flux in the residuals as an average of the blended star and the blending ones, weighted on distance and flux of the blending stars. We then assumed that the percentual centroiding error caused by blending was equal to this residual contaminating flux, a pessimistic assumption. The resulting centroiding errors caused by crowding are typically of the order of ∼10 μas yr−1 (with maximum values of ∼100), but they affect a relatively low number of stars in each GC (see Table 2). For the chromaticity correction, we know that chromatic effects for Gaia are relevant, because of its high accuracy, and they are corrected using the BP/RP information. If an error is made on that colour determination because of blends, the error is transported into the astrometric measurements as well. It was estimated that the shift on the single Gaia measurement could amount to 500–800 μas, depending on the spectral type and on the position along the focal plane. By averaging out transits and by correcting the effect of spectral type with GBP–GRP, the residual end-of-mission errors would amount to 0.4–1.4 μas (Jordi et al. 2006)16 in most of the well-behaved stars. However, for objects with non-stellar or peculiar colours, such as a quasar or a (multiple) stellar blend, the residual error could be still be roughly 30 μas, as computed from the typical residual centroid shift difference between a B and an M star, after chromaticity correction (Jordi et al. 2006). We thus approximated linearly17 the additional chromatic error as 0.002 μas for each K of temperature difference between each blended stars pair (both classic and Gaia blends), multiplied by the relative contaminating flux. The typical residuals from chromatic errors caused by crowding are of a few μas, but can be substantially higher for companions that are very bright or have a very different colour (up to ≃30–50 μas in our simulations). For RV error computations, the first consideration is that every time a star is contaminated by another star with a net RV difference above ≃30 km s−1 (roughly the Gaia resolution element FWHM), there is a chance of separating the two RVs, depending on the relative flux contamination. The percentage of flux contamination (obtained as in Section 3) was used to degrade the signal of the contaminated (or blended) star. We thus recomputed the magnitude of each blended or contaminated star, taking into account the residual errors after deblending or decontamination. We similarly recomputed the colours taking into account the residual colour errors. The recomputed colours and magnitudes were fed into the RV error equation (Gaia Collaboration 2016a), to derive the error on RV implied by the magnitude and colour deblending errors. The second consideration is that an artificial RV shift of roughly 30 km s−1 can also be obtained if two blended stars are offset, in the AL direction, by at least 4 AL pixels (or 0.235 arcsec), therefore we treated these stars as in the above case. We finally did not attempt to simulate any RV measurement for stars closer than the Gaia PSF (i.e. one AC pixel or 0.177 arcsec) as explained previously. Moreover, we did not compute RVs or RV errors for stars fainter than G = 16 mag. 5 RESULTS We illustrate in the following sections the type of data quality we expect from Gaia at the end of the mission, based on our simulations. As mentioned already, given all the conservative assumptions and the use of preliminary deblending pipelines, the presented results have to be considered pessimistic, in the sense that the pipelines will be more sophisticated in a few years from now. We will show in all figures three simulated GCs: an easy case with D = 5 kpc, c = 1.0 and a halo background (cluster #1, plotted in green in all figures); an intermediate case with 10 kpc, c = 1.0, and a disc background, with a relatively high reddening for GCs (cluster #5, plotted in yellow); and a difficult case with D = 15 kpc, c = 2.5, and a bulge background, which is an extreme condition for Gaia (cluster #18, plotted in red). 5.1 Proper motions Because proper motions are determined with AF, the impact of crowding is less severe than on BP/RP spectrophotometry, or on RVS spectra. Fig. 9 shows the histogram of the differences between the final proper motion – including all error sources – and the initial error-free simulated one, for the known cluster members. The median systemic proper motion can be recovered with a bias of 5–10 μas yr−1, i.e. with a systematic error of roughly 0.1–0.2 per cent. The intrinsic spread of the simulated GCs is of the order of 100 μas yr−1, depending on the GC. The observed spread is around 400–500 μas yr−1, if one includes all the simulated stars down to the magnitude limit. It is almost entirely explained with the nominal post-launch performances applied on top of the intrinsic spread, especially when including faint stars. In fact (see Section 4.5.1) the crowding errors on the affected stars are in the worst cases a few 100 μas yr−1 for the centroiding determination, and ≃30 μas yr−1 for chromaticity errors caused by GBP/RP crowding errors. Figure 9. Open in new tabDownload slide Histograms of differences between the final and error-free simulated proper motions of the easy (green histogram), intermediate (yellow histogram) and difficult cases (red histogram). Only cluster members are shown. The resulting absolute displacements in the systemic proper motion determination is of about 1–10 μas yr−1, corresponding to errors of <0.1 per cent. We note that the intrinsic dispersion is of the order of 100 μas yr−1, and the final, simulated one is 300–600 μas yr−1, when including all stars down to G ≃ 20.7 mag. Figure 9. Open in new tabDownload slide Histograms of differences between the final and error-free simulated proper motions of the easy (green histogram), intermediate (yellow histogram) and difficult cases (red histogram). Only cluster members are shown. The resulting absolute displacements in the systemic proper motion determination is of about 1–10 μas yr−1, corresponding to errors of <0.1 per cent. We note that the intrinsic dispersion is of the order of 100 μas yr−1, and the final, simulated one is 300–600 μas yr−1, when including all stars down to G ≃ 20.7 mag. Concerning individual stars, Fig. 10 shows the behaviour of the final simulated errors as a function of G magnitude. It can be useful to compare these simulations with the first results from the HST PROMO project for M15 (Hubble Space Telescope PROper MOtion project; Bellini et al. 2014). Their typical error around the G = 15 mag is 5–10 times higher than the error expected from Gaia, about 0.1 mas yr−1, but at G = 21 mag they are about half the Gaia error, at 0.5 mas yr−1. Moreover, HST proper motions can reach several magnitudes below the Gaia limit. On the other hand, the advantage of Gaia's proper motions is that they cover a larger area around each GC – actually, the whole sky – and that they are absolute proper motions. Figure 10. Open in new tabDownload slide Final simulated proper motion errors, including the effect of crowding for the easy (green points), intermediate (yellow points) and difficult cases (red points), as a function of final simulated G magnitude. Only cluster members are shown. Figure 10. Open in new tabDownload slide Final simulated proper motion errors, including the effect of crowding for the easy (green points), intermediate (yellow points) and difficult cases (red points), as a function of final simulated G magnitude. Only cluster members are shown. We show in Fig. 11 the typical vector diagrams of all simulated stars, including faint stars and background contaminants. The clusters are always easily separated from the field population, except in the case of the extreme bulge background. In the following sections, we will select the probable members with a loose criterium, i.e. stars within 1 mas yr−1 from the systemic GC motion, and with a strict criterium, i.e. within 0.3 mas yr−1 from the systemic GC motion. Figure 11. Open in new tabDownload slide Proper motion diagrams for all the stars, cluster and background members, in the easy (green points), intermediate (yellow points) and difficult cases (red points). The colour scale refers to the density of points, with darker toner referring to higher densities. The cluster lies at (−5, −5) in all diagrams, and it is always easily recognized, but it is overwhelmed by field stars in the last case. Figure 11. Open in new tabDownload slide Proper motion diagrams for all the stars, cluster and background members, in the easy (green points), intermediate (yellow points) and difficult cases (red points). The colour scale refers to the density of points, with darker toner referring to higher densities. The cluster lies at (−5, −5) in all diagrams, and it is always easily recognized, but it is overwhelmed by field stars in the last case. 5.1.1 Measurement of derived quantities To assess the derived quantities that will be measurable from Gaia data, we performed a basic test on the easy case GC. We excluded all stars that were clearly unrelated to the GC with a very loose proper motion selection (±3 mas yr−1) and all the classic blends. We then converted proper motions in the RA and Dec directions into velocities, using the GC distance and propagating its error (see Section 5.2). We then estimated the RV spreads σRA and σDec as a function of distance from the GC centre with their errors. We used the maximum likelihood estimation (MLE) method with the likelihood formulation described by Pryor & Meylan (1993), Walker et al. (2006) and Martin et al. (2007), which takes into account the errors on measurements as well. The actual errors on the RV dispersions in each radial bin are of about 0.4–1.4 km s−1. The result is displayed in Fig. 12, showing that Gaia will be able to determine the radial profile of the RV dispersion of nearby GCs with errors of about 1 km s−1 in the nearest GCs. Figure 12. Open in new tabDownload slide RV dispersion profile in RA and Dec for the easy case cluster. Figure 12. Open in new tabDownload slide RV dispersion profile in RA and Dec for the easy case cluster. More in general, we can say that: Proper motions obtained by Gaia in GCs will allow the study of dynamical relaxation in the 5–10 closest GCs (see also Section 5.3 and Fig. 14). This is because at least a few magnitudes below the GC turn-off point are required to adequately sample a range of stellar masses. Fig. 14 clearly shows that this is possible only for clusters at ≃5 kpc or less, because of the Gaia detection limit. Therefore, these studies will have to rely on deeper proper motion data sets for more distant GCs.18 Two past studies dealt with the determination of masses in GCs with Gaia data (An, Evans & Deason 2012; Sollima et al. 2015) yielding somewhat conflicting results. The main problem is to break the mass-anisotropy degeneracy, with the anisotropy signal being stronger outside the central regions where relaxation has the highest effect. Therefore, blends by cluster members are irrelevant there, as discussed in Section 4.4 and shown in Fig. 8. And even in the case of heavy background contamination, proper motions will be only mildly affected (Fig. 10). The assumption by Sollima et al. (2015) that one can only use stars contaminated by less than 10 per cent of the flux by objects more distant than 3.54 arcsec, is far too pessimistic. The simulations by An et al. (2012), prove that 10 per cent mass estimates can be provided for GCs if one can count on at least 100 tracers (we can use red giants) with proper motion errors <100 μas. As shown in Table 2 and Fig. 10, the number of stars per GC with these errors will be more like a few 103–104, except maybe for the most extreme cases of bulge clusters. Moreover, our test with the easy GC above shows that RV spreads can be measured as a function of distance from the centre with 1 km s−1 errors at 5 kpc. Therefore, as concluded by An et al. (2012), it will be possible to determine GC masses with an error of 10 per cent or less at least up to 15 kpc. This requires ground-based RVs to complement Gaia proper motions.19 Gaia will be able to provide the necessary quantities for computing accurate GC orbits within the Galaxy, because, as we saw, it will provide high-quality systemic proper motions for each GC, with a bias of a few μas yr−1 and errors of the order of 1 per cent. Distances will be obtained with similar errors, as discussed below (Section 5.2). Once the orbits are obtained they could in turn be used to look for extra tidal stars and tidal tails. Even if the G < 20.7 mag limit reduces the density of targets that will be actually measured by Gaia around GCs, the proper motion signal of escaped GC stars should be very sharp, and in most cases significantly different – within the uncertainties – from that of the surrounding field population (see also Fig. 11). Gaia could be able to help in the open and hotly debated problem of multiple populations (Kraft 1994; Gratton, Sneden & Carretta 2004), in particular by revealing kinematical differences among different GC sub-populations, if they are present (Hénault-Brunet et al. 2015). For example, Bellazzini et al. (2012) reported the possible detection of a different rotation pattern between the Na-rich and Na-poor stars in some of the GCs in their sample, with rotation amplitude differences not larger than ≃5 km s−1. Depending on the distance and the inclination of the rotation axis, this would correspond, in terms of proper motions, to a difference in rotation of the order of 1 mas yr−1, which would be detectable for those GCs having at least 100 stars with measurement errors of a few 10 μas. 5.2 Parallaxes The major problem in deriving distances from parallaxes, especially when the error is comparable to the measurement, is that the errors on distances are not symmetric: by simply inverting the parallax one could obtain a very wrong distance estimate (Bailer-Jones 2015). Statistically speaking, it is also possible to obtain negative parallaxes, as illustrated in Fig. 13, which would result in meaningless distance estimates if not treated properly. Figure 13. Open in new tabDownload slide Histograms of the parallaxes for the easy (top panel), intermediate (middle panel) and difficult (lower panel) clusters. The vertical dotted line in all panels marks the zero parallax limit, while the solid line marks the true parallax of each simulated GC. Lightly shaded histograms show the member stars selected with the loose criterium and heavily shaded ones with the strict criterium (see Section 5.1). Figure 13. Open in new tabDownload slide Histograms of the parallaxes for the easy (top panel), intermediate (middle panel) and difficult (lower panel) clusters. The vertical dotted line in all panels marks the zero parallax limit, while the solid line marks the true parallax of each simulated GC. Lightly shaded histograms show the member stars selected with the loose criterium and heavily shaded ones with the strict criterium (see Section 5.1). When one is interested in the systemic distance of a GC, however, negative parallaxes are useful to assess the uncertainty on the measurements and their distribution. Fig. 13 shows the distribution of parallaxes for probable members, as defined in Section 5.1. As can be seen, not only the average parallax of the clusters is very well recovered, but also its uncertainty, provided that one makes use also of the negative parallaxes. The exception is the case of the difficult cluster, which has a background contamination that overwhelms the signal from the cluster members. As can be seen, the recovered parallax is strongly biased by the background contaminants. A more strict members selection apparently produces better results (but see below). However, the resulting distribution of strict members is still asymmetric, because of the blends between cluster and field stars. This type of problem is not specific of Gaia data, of course, but it is extremely important to employ a reliable membership selection and robust statistical modelling of the parallax distribution, which can be skewed and is heteroscedastic. We employed an MLE analysis of the 18 simulated GCs using the strict and loose membership criteria described above. We used the likelihood estimator described by Pryor & Meylan (1993), Walker et al. (2006) and Martin et al. (2007), which takes into account the highly variable errors, and we computed the most probable parallax ϖ and its error δϖ. The result is displayed in Table 4. We first note that the depth of a GC, as measured with parallaxes, can correspond at most to a few μas for the nearest GCs that have large radii, even if one considers the tidal radius, and in the majority of cases is well below 1 μas. Thus the observed spread is caused almost entirely by measurement errors. The bias itself, i.e. the difference between the true input parallax and the recovered one, is always 1 per cent or smaller, even for the difficult case GC. The formal error on the recovered ϖ is of the same order. We also note that choosing a more restrictive membership selection can often increase the formal errors without resulting in a better ϖ determination. On the contrary, the bias is slightly increased when applying a more restrictive membership selection (see also Bailer-Jones 2015). Table 4. Systemic parallax for the simulated clusters, based on members selected with the strict and loose criteria (see text for more details). The columns contain: (1) the cluster number; (2) concentration parameter; (3) projected cluster distance and (4) parallax; (5) background type; (6) systemic parallax and error for members selected with the strict criterium and (7) with the loose criterium. Cluster c D0 ϖ0 Field ϖstrict ϖloose (kpc) (μas) (μas) (μas) # 1 1.0 5 200.0 Halo 200.2 ± 2.6 200.2 ± 0.6 # 2 1.0 10 100.0 Halo 101.6 ± 2.1 99.3 ± 0.8 # 3 1.0 15 66.7 Halo 69.2 ± 2.6 69.1 ± 1.0 # 4 1.0 5 200.0 Disc 203.9 ± 6.1 197.4 ± 1.3 # 5 1.0 10 100.0 Disc 95.6 ± 4.4 98.5 ± 1.5 # 6 1.0 15 66.7 Disc 74.0 ± 5.6 65.8 ± 1.9 # 7 1.0 5 200.0 Bulge 206.8 ± 2.3 200.8 ± 0.6 # 8 1.0 10 100.0 Bulge 100.1 ± 1.9 99.1 ± 0.6 # 9 1.0 15 66.7 Bulge 63.9 ± 2.3 67.9 ± 0.7 # 10 2.5 5 200.0 Halo 197.5 ± 0.7 198.4 ± 0.4 # 11 2.5 10 100.0 Halo 101.4 ± 1.3 101.2 ± 0.6 # 12 2.5 15 66.7 Halo 62.2 ± 1.5 66.4 ± 0.9 # 13 2.5 5 200.0 Disc 200.0 ± 1.6 200.2 ± 0.8 # 14 2.5 10 100.0 Disc 100.0 ± 3.1 102.8 ± 1.3 # 15 2.5 15 66.7 Disc 74.3 ± 3.2 70.5 ± 1.8 # 16 2.5 5 200.0 Bulge 199.3 ± 0.7 198.6 ± 0.4 # 17 2.5 10 100.0 Bulge 101.6 ± 1.3 99.5 ± 0.5 # 18 2.5 15 66.7 Bulge 65.8 ± 1.4 67.0 ± 0.7 Cluster c D0 ϖ0 Field ϖstrict ϖloose (kpc) (μas) (μas) (μas) # 1 1.0 5 200.0 Halo 200.2 ± 2.6 200.2 ± 0.6 # 2 1.0 10 100.0 Halo 101.6 ± 2.1 99.3 ± 0.8 # 3 1.0 15 66.7 Halo 69.2 ± 2.6 69.1 ± 1.0 # 4 1.0 5 200.0 Disc 203.9 ± 6.1 197.4 ± 1.3 # 5 1.0 10 100.0 Disc 95.6 ± 4.4 98.5 ± 1.5 # 6 1.0 15 66.7 Disc 74.0 ± 5.6 65.8 ± 1.9 # 7 1.0 5 200.0 Bulge 206.8 ± 2.3 200.8 ± 0.6 # 8 1.0 10 100.0 Bulge 100.1 ± 1.9 99.1 ± 0.6 # 9 1.0 15 66.7 Bulge 63.9 ± 2.3 67.9 ± 0.7 # 10 2.5 5 200.0 Halo 197.5 ± 0.7 198.4 ± 0.4 # 11 2.5 10 100.0 Halo 101.4 ± 1.3 101.2 ± 0.6 # 12 2.5 15 66.7 Halo 62.2 ± 1.5 66.4 ± 0.9 # 13 2.5 5 200.0 Disc 200.0 ± 1.6 200.2 ± 0.8 # 14 2.5 10 100.0 Disc 100.0 ± 3.1 102.8 ± 1.3 # 15 2.5 15 66.7 Disc 74.3 ± 3.2 70.5 ± 1.8 # 16 2.5 5 200.0 Bulge 199.3 ± 0.7 198.6 ± 0.4 # 17 2.5 10 100.0 Bulge 101.6 ± 1.3 99.5 ± 0.5 # 18 2.5 15 66.7 Bulge 65.8 ± 1.4 67.0 ± 0.7 Open in new tab Table 4. Systemic parallax for the simulated clusters, based on members selected with the strict and loose criteria (see text for more details). The columns contain: (1) the cluster number; (2) concentration parameter; (3) projected cluster distance and (4) parallax; (5) background type; (6) systemic parallax and error for members selected with the strict criterium and (7) with the loose criterium. Cluster c D0 ϖ0 Field ϖstrict ϖloose (kpc) (μas) (μas) (μas) # 1 1.0 5 200.0 Halo 200.2 ± 2.6 200.2 ± 0.6 # 2 1.0 10 100.0 Halo 101.6 ± 2.1 99.3 ± 0.8 # 3 1.0 15 66.7 Halo 69.2 ± 2.6 69.1 ± 1.0 # 4 1.0 5 200.0 Disc 203.9 ± 6.1 197.4 ± 1.3 # 5 1.0 10 100.0 Disc 95.6 ± 4.4 98.5 ± 1.5 # 6 1.0 15 66.7 Disc 74.0 ± 5.6 65.8 ± 1.9 # 7 1.0 5 200.0 Bulge 206.8 ± 2.3 200.8 ± 0.6 # 8 1.0 10 100.0 Bulge 100.1 ± 1.9 99.1 ± 0.6 # 9 1.0 15 66.7 Bulge 63.9 ± 2.3 67.9 ± 0.7 # 10 2.5 5 200.0 Halo 197.5 ± 0.7 198.4 ± 0.4 # 11 2.5 10 100.0 Halo 101.4 ± 1.3 101.2 ± 0.6 # 12 2.5 15 66.7 Halo 62.2 ± 1.5 66.4 ± 0.9 # 13 2.5 5 200.0 Disc 200.0 ± 1.6 200.2 ± 0.8 # 14 2.5 10 100.0 Disc 100.0 ± 3.1 102.8 ± 1.3 # 15 2.5 15 66.7 Disc 74.3 ± 3.2 70.5 ± 1.8 # 16 2.5 5 200.0 Bulge 199.3 ± 0.7 198.6 ± 0.4 # 17 2.5 10 100.0 Bulge 101.6 ± 1.3 99.5 ± 0.5 # 18 2.5 15 66.7 Bulge 65.8 ± 1.4 67.0 ± 0.7 Cluster c D0 ϖ0 Field ϖstrict ϖloose (kpc) (μas) (μas) (μas) # 1 1.0 5 200.0 Halo 200.2 ± 2.6 200.2 ± 0.6 # 2 1.0 10 100.0 Halo 101.6 ± 2.1 99.3 ± 0.8 # 3 1.0 15 66.7 Halo 69.2 ± 2.6 69.1 ± 1.0 # 4 1.0 5 200.0 Disc 203.9 ± 6.1 197.4 ± 1.3 # 5 1.0 10 100.0 Disc 95.6 ± 4.4 98.5 ± 1.5 # 6 1.0 15 66.7 Disc 74.0 ± 5.6 65.8 ± 1.9 # 7 1.0 5 200.0 Bulge 206.8 ± 2.3 200.8 ± 0.6 # 8 1.0 10 100.0 Bulge 100.1 ± 1.9 99.1 ± 0.6 # 9 1.0 15 66.7 Bulge 63.9 ± 2.3 67.9 ± 0.7 # 10 2.5 5 200.0 Halo 197.5 ± 0.7 198.4 ± 0.4 # 11 2.5 10 100.0 Halo 101.4 ± 1.3 101.2 ± 0.6 # 12 2.5 15 66.7 Halo 62.2 ± 1.5 66.4 ± 0.9 # 13 2.5 5 200.0 Disc 200.0 ± 1.6 200.2 ± 0.8 # 14 2.5 10 100.0 Disc 100.0 ± 3.1 102.8 ± 1.3 # 15 2.5 15 66.7 Disc 74.3 ± 3.2 70.5 ± 1.8 # 16 2.5 5 200.0 Bulge 199.3 ± 0.7 198.6 ± 0.4 # 17 2.5 10 100.0 Bulge 101.6 ± 1.3 99.5 ± 0.5 # 18 2.5 15 66.7 Bulge 65.8 ± 1.4 67.0 ± 0.7 Open in new tab The scientific implications of ≃1 per cent distances for most of the GCs in the Galaxy are far reaching. They directly help in the determination and modelling of GC orbits, which can then be used for dynamical studies of the MW, of the GC themselves, and of the interactions between the two. For example, they would help simulations that use the GC orbits and tidal tails and streams to constrain the Galactic potential (Peñarrubia, Koposov & Walker 2012; Price-Whelan et al. 2014), or they can help discerning the origin of the GC themselves through dynamical modelling. But the area in which significant breakthrough is expected lies in the determinations of stellar ages (Gratton et al. 1997) and masses (Feuillet et al. 2016). Distance has in fact a similar effect as age when using high-quality CMDs (from HST or ground-based observations) to estimate the absolute ages of GCs. By reducing the distance determination errors by more than a factor of 10, we can expect a significant improvement on the age determinations, obtaining absolute ages with errors below 10 per cent. On the other hand, ≃1 per cent distances can help in constraining the stellar masses and more importantly, the surface gravities. When trying to determine relative ages in GCs with multiple stellar populations, one needs to disentangle the effects of chemical composition – mainly helium and C+N+O abundance – from age effects. A very accurate surface gravity determination would remove one of the major sources of uncertainty in the determination of stellar chemistry from spectroscopy. 5.3 Photometry The Gaia colour–magnitude diagrams (CMDs) of the three example clusters are shown in Fig. 14. The top panels show all the simulated stars, including the background, while the bottom panels show only the probable members, selected with the loose criterium described in the previous section. As can be seen, background contamination is the first obvious cause of crowding errors for Gaia photometry. A large reddening (like in the case of the disc cluster) or a distance larger than 10 kpc also impact the quality of the CMD, because the photometric errors, especially those on the GBP–GRP colour, increase quite rapidly with magnitude. Figure 14. Open in new tabDownload slide The simulated Gaia photometry for the three example GCs. Top panels show all the simulated stars, regardless of membership. Bottom panels show only the loosely selected probable members (see text for more details). The left-hand panels refer to the easy case, the centre ones to the intermediate case and the right ones to the difficult case. The colour scale refers to the density of points, with darker toner referring to higher densities. Figure 14. Open in new tabDownload slide The simulated Gaia photometry for the three example GCs. Top panels show all the simulated stars, regardless of membership. Bottom panels show only the loosely selected probable members (see text for more details). The left-hand panels refer to the easy case, the centre ones to the intermediate case and the right ones to the difficult case. The colour scale refers to the density of points, with darker toner referring to higher densities. It will therefore not be possible to reach the nominal photometric errors promised by Gaia in those clusters that lie on a bulge or disc background. This is mainly caused by the extended shape of the BP and RP dispersed images, while the G magnitudes will not suffer significantly from crowding. However, for clusters that are relatively free from background contamination, the BP/RP photometry of stars brighter than G ≃ 15 mag will have an extremely good quality, comparable to HST photometry. Fainter stars down to G ≃ 18–19 mag will still have a BP/RP quality that is comparable with the best ground-based catalogues. While HST photometry will certainly be preferable for some applications, Gaia photometry will have a few advantages: (1) the field of view of Gaia is not limited by any field size: it covers the whole sky; (2) each star that has Gaia astrometry from AF, also has BP/RP spectra from which rough stellar parameters and a reddening estimate can be obtained (Gaia Collaboration 2016a); (3) brighter stars, generally red giants in the case of GCs, will also have RVs and more accurate estimates of parameters and reddening from RVS spectra20 (Kordopatis et al. 2012) and (4) the absolute photometric calibration of Gaia will be based on one the largest, most homogeneous, and most accurate set of spectrophotometric standard stars to date (Pancino et al. 2012; Altavilla et al. 2015; Marinoni et al. 2016), that will grant an accuracy of ≃1–3 per cent with respect to Vega (Bohlin & Gilliland 2004). 5.4 Radial velocities Stars brighter than G ≃ 17 mag will have RVs measurements. Gaia has already produced billions of RVS spectra and by the end of the mission, each star will be observed on average 40 times. This can be compared with the extremely successful RAVE survey (Radial VElocity Experiment; Kordopatis et al. 2013), which measured RVs for half a million stars with V < 12 mag. The Gaia end-of-mission errors will vary with the star's colour and will be of the order of 1 km s−1 for the bright red stars (G < 12.5 mag and cooler than F types) and will be about 15–20 km s−1 or more for fainter and bluer stars. Crowding will affect RVS more than any of the other instruments on board. Even if the magnitude limit is brighter, the AL size of the spectra is more than an arcminute on the sky (see Table 1). Fig. 15 shows the simulated RV measurements, as a function of the G magnitude, for the three example cases. Given the limitations of our simulations, we immediately note that only a handful of reliable members in the intermediate case have meaningful RV determinations, because of the combined effect of high reddening – for a GC – and background contamination. We also note the large scatter in RV for the faintest stars in the difficult case, where background contamination takes its toll. However, for the easy case, we notice that a sample of more than 100 stars with magnitudes above ≃13–14 mag is available, with errors around 1 or few km s−1. This will happen for the closest 10–20 GCs. The stars with large errors between 13.5 and 14.5 mag are hotter HB stars. To conclude, we note that a mission extension would increase the quality of the RVS spectra. Figure 15. Open in new tabDownload slide RV as a function of magnitude for the easy (top panel), medium (middle panel) and difficult (bottom panel) cases. The entire samples of available RV measurements are plotted in grey. The dotted lines mark zero and the true systemic velocity of the simulated clusters, 100 km s−1. Only true cluster members are coloured in green, yellow and red, respectively. Figure 15. Open in new tabDownload slide RV as a function of magnitude for the easy (top panel), medium (middle panel) and difficult (bottom panel) cases. The entire samples of available RV measurements are plotted in grey. The dotted lines mark zero and the true systemic velocity of the simulated clusters, 100 km s−1. Only true cluster members are coloured in green, yellow and red, respectively. 5.5 The central arcminute As discussed in Section 2.2, it is not feasible to simulate the degree of completeness of Gaia data, because it varies across the sky, based on the number of different passages and on their respective orientation. As discussed in that section, nearby and relatively sparse GCs like ω Cen should be 100 per cent complete down to V ≃ 16 mag, even in the very core. This might appear surprising, but it is a beneficial byproduct of the relatively shallow Gaia magnitude limit. However, still the question remains: how far into the GC core can we obtain reliable measurements, at least for the stars that we will be able to measure? Fig. 16 shows the end-of-mission error distributions for all member stars, down to G = 20.7 mag, for the central arcminute of the simulated clusters. As can be seen, astrometry is not too badly affected by crowding: the maximum errors are of 1 mas yr−1 for proper motions and 0.7 mas for parallaxes (to compare with Fig. 10). The majority of stars have performances of the order of a few hundred μas yr−1 or μas, and this including also the faintest stars. If this appears surprising, we recall that what mostly governs the end-of-mission astrometric errors caused by crowding in Gaia is the PSF size, which is comparable to that of HST. Figure 16. Open in new tabDownload slide End-of-mission error distribution for all member stars (defined as in Fig. 15) down to G = 20.7 mag, within the central arcminute of the simulated GCs. The top panel shows proper motion errors, the middle one parallax errors and the bottom ones G magnitude errors. The easy cluster is represented in green in all panels, the intermediate one in yellow and the difficult one in red. Figure 16. Open in new tabDownload slide End-of-mission error distribution for all member stars (defined as in Fig. 15) down to G = 20.7 mag, within the central arcminute of the simulated GCs. The top panel shows proper motion errors, the middle one parallax errors and the bottom ones G magnitude errors. The easy cluster is represented in green in all panels, the intermediate one in yellow and the difficult one in red. The situation is different for BP/RP photometry or RVS, because even if the PSF in the AC direction is still small, and thus allows to detect blends quite efficiently, the AL size of the window is larger, and aligned with the dispersion axis of the spectra. The performances in the central arcminute are in fact significantly worse for these instruments than for the AF and the G-band magnitude in the case of crowded fields. 6 SUMMARY AND CONCLUSIONS We used all the available information on the presently available Gaia deblending pipelines, along with their results on simulated data, to model the behaviour of Gaia in crowded areas. We computed additional crowding errors that were combined with the post-launch science performances of Gaia and applied to a set of simulated GCs with different concentration (c = 1.0 and 2.5), distance (5, 10 and 15 kpc), and field population (halo, disc and bulge). We showed that: Gaia will provide estimates of the parallax and systemic motion of GCs with unprecedented accuracies: ≃1 per cent and <<1 per cent, respectively, for all GCs as far as 15 kpc at least. Also, systemic RVs with errors of a few km s−1 will be obtained for the 10–20 closest GCs, although our RVS simulations are not as accurate as for the other Gaia instruments. This will allow for a very accurate modelling of GC orbits. The astrometry, obtained through the AF together with G magnitudes, will be only marginally affected by crowding, having performances not too dissimilar from the nominal Gaia performances even for the most field-contaminated bulge GCs. This is an effect of the tiny Gaia PSF (0.17 arcsec) in AF. The G magnitudes will have mmag performances and space quality to G ≃ 17–18 mag. The proper motions of individual stars within GCs have sufficient quality to obtain mass estimates with 10 per cent errors for GCs as far as 15 kpc at least, and to identify small variations of the properties (rotations, spreads) of a few km s−1 for GCs as far as 10 kpc. The distances of GCs will be obtained with errors of ≃1 per cent, and for GCs heavily contaminated by field stars to a few per cent. The impact on the determinations of stellar masses and ages will be significant. It is expected that GC absolute ages with errors below 10 per cent will be obtained. Also, better estimates of surface gravities for GC stars with known distances will remove one of the major uncertainty sources from abundance determinations with high-resolution spectroscopy. While it is difficult to simulate the exact completeness level of Gaia in GCs, we have shown that the astrometric performances are still exceptional in the central arcminute of the simulated GCs: most of the stars have errors around a few 100 μas or μas yr−1. The BP/RP photometry and the RVS spectra, on the other hand, have larger AL sizes and therefore suffer more from crowding effects. The most important factor for these instruments is crowding by field stars, especially in the most extreme cases like the bulge field, which acts at all distances from the GC centre and combines with the crowding effects from GC stars in the central regions. The imminent decision of whether to extend the Gaia mission lifetime will certainly have a beneficial impact on all the above measurements. However, the simulations presented here are already pessimistic and the pipelines are expected to evolve significantly in the next few years. Therefore, we conclude that Gaia measurements will revolutionize our knowledge of GCs. Acknowledgments We warmly thank A. Brown, M. Castellani, A. Di Cecco, F. De Luise, D. Harrison, H. E. Huckle, K. Janssen, C. Jordi, P. M. Marrese, D. Pourbaix, L. Pulone and G. Seabroke, for providing useful information about Gaia BP, RP and RVS deblending and decontamination, and for enlightening discussions about Gaia and crowded fields treatment in general. This work used simulated data provided by the Simulation Unit (CU2) of the Gaia Data Processing and Analysis Consortium (DPAC), run with GIBIS at CNES (Centre national d’études spatiales). This research has made use of the r programming language (https://www.r-project.org/) and more specifically of its ‘data.table’ package, for the treatment of very large data sets. Some of the figures were prepared with topcat (Taylor 2005). 1 A list of the acronyms used in this paper (Table A1) can be found in Appendix A. 2 http://www.cosmos.esa.int/web/gaia 3 The value we adopted is a conservative estimate. The Gaiaeffective PSF varies across the field of view and has a median value of 0.103 arcsec (Fabricius et al. 2016). 4 http://www.cosmos.esa.int/web/gaia/iow_20150226 5 http://www.cosmos.esa.int/web/gaia/news_20141217 6 http://www.cosmos.esa.int/web/gaia/iow_20140206 7 We recall that to obtain absolute astrometric measurements, Gaia projects two different lines of sight on a common focal plane. To disentangle sources coming from the two projected lines of sight, each of the SM CCD columns sees only one of the lines of sight. Therefore, there will be some fields that – even with low stellar background – will happen to overlap crowded areas in some fraction of their transits. 8 http://www.cosmos.esa.int/web/gaia/iow_20141113 9 From 5 to 8 per cent of the total number of stars in our simulations, including field stars, are classic blends, depending on the particular cluster. The classic blends are mostly in the inner parts of GCs, within 1 arcmin approximately. For more details see Section 4.4. 10 For a detailed description of the various BP/RP deblending pipelines, see Section 2.1.2. What we mean here by ‘partially informed’ is that we used information on the blended stars positions in AF, but we did not use any other information, such as astrophysical parameters (Teff, logg, AV, [Fe/H]) and therefore we assumed we did not know the spectral energy distribution of the involved sources. This is still a pessimistic assumption, because towards the end of the mission this kind of information will become available and iteratively improved with every data processing cycle. 11 With our linear relation, a G = 2 mag star would be able to significantly contaminate stars up to 34.4 arcsec. These bright stars are rare: in our 18 simulated GCs there is only one field star brighter than that, appearing in the six disc GCs. 12 https://ahwkuepper.wordpress.com/mcluster/ 13 Dark remnants are white dwarf stars, neutron stars and black holes. Recently, comparable fractions of dark remnants were obtained with state-of-the-art cluster modelling by Sollima et al. (2015, 2016). 14 http://model.obs-besancon.fr/ 15 In other words, we looped over the list of (relevant) stars and computed distances with all other (relevant) stars, one object at a time. This was necessary because smarter neighbour searches pre-compute a distance matrix to increase computation speed, and therefore tend to saturate computer memory for data sets containing more than 105 objects (Hendra Gunadi 2011). Our bulge simulated GCs contain a few million stars, and several tens of millions of relevant pairs. 16 These estimates were computed using the Gaia filter system, now abandoned, but there are no other estimates available of the Gaia residual chromatic displacement in the literature at the moment. 17 The entity of the chromatic correction errors caused by crowding is small for the simulated GCs (a few μas or μas yr−1) compared to the Gaia science performances (see also Fig. 10), and therefore a linear approximation was considered adequate. 18 Deeper measurements could be obtained if the mission lifetime is extended beyond its nominal 5-yr duration. The sea pipelines also have the potential of recovering faint stars. Finally, the use of the 2D images obtained for some crowded fields will help by directly recovering some faint stars and by improving the deblending pipelines, thus recovering more faint stars also in other regions. 19 For the closest GCs, Gaia will provide some good RVs as well, see also Section 5.4, for more details. 20 Accurate reddening estimates can also be obtained from diffuse interstellar bands, that are included in the Gaia RVS wavelength range (see, for example Puspitarini et al. 2015, and references therein). REFERENCES Allende Prieto C. , 2008 , Gaia Technical Note number GAIA-C6-SP-MSSL-CAP-003 Google Preview WorldCat COPAC Altavilla G. , Botticella M. T. , Cappellaro E. , Turatto M. , 2012 , Ap&SS , 341 , 163 Crossref Search ADS Altavilla G. et al. , 2015 , Astron. Nachr. , 336 , 515 Crossref Search ADS An J. , Evans N. W. , Deason A. J. , 2012 , MNRAS , 420 , 2562 Crossref Search ADS Babusiaux C. , 2005 , Turon C. , O'Flaherty K. S. , Perryman M. A. C. , ESA SP-576: The Three-Dimensional Universe with Gaia , ESA , Noordwijk , 417 Google Preview WorldCat COPAC Bachchan R. K. , Hobbs D. , Lindegren L. , 2016 , A&A , 589 , A71 Crossref Search ADS Bailer-Jones C. A. L. , 2015 , PASP , 127 , 994 Crossref Search ADS Bellazzini M. , Bragaglia A. , Carretta E. , Gratton R. G. , Lucatello S. , Catanzaro G. , Leone F. , 2012 , A&A , 538 , A18 Crossref Search ADS Bellini A. et al. , 2014 , ApJ , 797 , 115 Crossref Search ADS Bohlin R. C. , Gilliland R. L. , 2004 , AJ , 127 , 3508 Crossref Search ADS Brodie J. P. , Strader J. , 2006 , ARA&A , 44 , 193 Crossref Search ADS de Bruijne J. H. J. , Rygl K. L. J. , Antoja T. , 2014 , EAS Publ. Ser. Vol. 67, Gaia Astrometric Science Performance – Post-Launch Predictions , EDP Sciences , Les Ullis , 23 Google Preview WorldCat COPAC de Bruijne J. H. J. , Allen M. , Azaz S. , Krone-Martins A. , Prod'homme T. , Hestroffer D. , 2015 , A&A , 576 , A74 Crossref Search ADS Dinescu D. I. , van Altena W. F. , Girard T. M. , López C. E. , 1999 , AJ , 117 , 277 Crossref Search ADS Ducourant C. , Krone-Martins A. , Galluccio L. , Teixeira R. , 2014 , Ballet J. , Martins F. , Bournaud F. , Monier R. , Reylé C. , Proc. Ann. French Soc. Astron. Astrophys., Gaia and the Extragalactic 421 Google Preview WorldCat COPAC Eyer L. et al. , 2014 , Walton N. A. , Figueras F. , Balaguer-Núñez L. , Soubiran C. , EAS Publ. Ser. Vol. 67 , The Milky Way Unravelled by Gaia: GREAT Science from the Gaia Data Releases. p 75 Google Preview WorldCat COPAC Fabricius C. et al. , 2016 , A&A , 595 , A3 Crossref Search ADS Feuillet D. K. , Bovy J. , Holtzman J. , Girardi L. , MacDonald N. , Majewski S. R. , Nidever D. L. , 2016 , ApJ , 817 , 40 Crossref Search ADS Gaia Collaboration , 2016a , A&A , 595 , A1 Crossref Search ADS Gaia Collaboration , 2016b , A&A , 595 , A2 Crossref Search ADS Gratton R. G. , Fusi Pecci F. , Carretta E. , Clementini G. , Corsi C. E. , Lattanzi M. , 1997 , ApJ , 491 , 749 Crossref Search ADS Gratton R. , Sneden C. , Carretta E. , 2004 , ARA&A , 42 , 385 Crossref Search ADS Harris W. E. , 1996 , AJ , 112 , 1487 Crossref Search ADS Harrison D. L. , 2011 , Exp. Astron. , 31 , 157 Crossref Search ADS Hendra Gunadi A. , 2011 , Master thesis , Australian National University Google Preview WorldCat COPAC Hénault-Brunet V. , Gieles M. , Agertz O. , Read J. I. , 2015 , MNRAS , 450 , 1164 Crossref Search ADS Hurley J. R. , Pols O. R. , Tout C. A. , 2000 , MNRAS , 315 , 543 Crossref Search ADS Høg E. et al. , 2000 , A&A , 355 , L27 Jordi C. et al. , 2006 , MNRAS , 367 , 290 Crossref Search ADS Jordi C. et al. , 2010 , A&A , 523 , A48 Crossref Search ADS King I. R. , 1966 , AJ , 71 , 64 Crossref Search ADS Kordopatis G. , Recio-Blanco A. , de Laverny P. , Bijaoui A. , Hill V. , Gilmore G. , Wyse R. F. G. , Ordenovic C. , 2012 , EPJ Web Conf. , 19 , 09010 Crossref Search ADS Kordopatis G. et al. , 2013 , AJ , 146 , 134 Crossref Search ADS Kraft R. P. , 1994 , PASP , 106 , 553 Crossref Search ADS Kroupa P. , 2001 , MNRAS , 322 , 231 Crossref Search ADS Küpper A. H. W. , Maschberger T. , Kroupa P. , Baumgardt H. , 2011 , MNRAS , 417 , 2300 Crossref Search ADS Lindegren L. et al. , 2016 , A&A , 595 , A4 Crossref Search ADS Marinoni S. et al. , 2016 , MNRAS , 462 , 3616 Crossref Search ADS Martin N. F. , Ibata R. A. , Chapman S. C. , Irwin M. , Lewis G. F. , 2007 , MNRAS , 380 , 281 Crossref Search ADS Michalik D. , Lindegren L. , Hobbs D. , 2015 , A&A , 574 , A115 Crossref Search ADS Mignard F. , 2005 , Seidelmann P. K. , Monet A. K. B. , ASP Conf. Ser. Vol. 338, Astrometry in the Age of the Next Generation of Large Telescopes , Astron. Soc. Pac. , San Francisco , 15 Google Preview WorldCat COPAC Mora A. et al. , 2016 , MacEwen H. A. , Fazio G. G. , Lystrupeds M. , Batalha N. , Siegler N. , Tong E. C. , Proc. SPIE Conf. Ser. Vol. 9904, Space Telescopes and Instrumentation 2016: Optical, Infrared, and Millimeter Wave , SPIE , Bellingham , 99042D Google Preview WorldCat COPAC Pancino E. et al. , 2012 , MNRAS , 426 , 1767 Crossref Search ADS Pancino E. , Bellazzini M. , Marinoni S. , 2013 , Mem. Soc. Astron. Ital. , 84 , 83 Paust N. E. Q. et al. , 2010 , AJ , 139 , 476 Crossref Search ADS Peñarrubia J. , Koposov S. E. , Walker M. G. , 2012 , ApJ , 760 , 2 Crossref Search ADS Perryman M. A. C. et al. , 1997 , A&A , 323 , 49 Perryman M. A. C. et al. , 2001 , A&A , 369 , 339 Crossref Search ADS Piersimoni A. , De Luise F. , Busso G. , 2011 , Gaia Technical Note number GAIA-C5-TN-OATE-AP-001 Google Preview WorldCat COPAC Pourbaix D. , 2011 , Docobo J. A. , Tamazian V. S. , Balega Y. Y. , AIP Conf. Proc. Vol. 1346, International Workshop on Double and Multiple Stars: Dynamics, Physics, and Instrumentation , Am. Inst. Phys. , New York , 122 Google Preview WorldCat COPAC Price-Whelan A. M. , Hogg D. W. , Johnston K. V. , Hendel D. , 2014 , ApJ , 794 , 4 Crossref Search ADS Proft S. , Wambsganss J. , 2015 , A&A , 574 , A46 Crossref Search ADS Pryor C. , Meylan G. , 1993 , Djorgovski S. G. , Meylan G. , Proc. ASP Conf. Ser. Vol. 50, Structure and Dynamics of Globular Clusters , Astron. Soc. Pac. , San Francisco , 357 Google Preview WorldCat COPAC Puspitarini L. et al. , 2015 , A&A , 573 , A35 Crossref Search ADS Robin A. C. , Reylé C. , Derrière S. , Picaud S. , 2003 , A&A , 409 , 523 Crossref Search ADS Seabroke G. et al. , 2016 , Bragaglia A. , Arnaboldi M. , Rejkuba M. , Romano D. , Proc. IAU Symp. 317, The General Assembly of Galaxy Halos: Structure, Origin and Evolution , Cambridge Univ. Press , Cambridge , 346 Google Preview WorldCat COPAC Sollima A. , Baumgardt H. , Zocchi A. , Balbinot E. , Gieles M. , Hénault-Brunet V. , Varri A. L. , 2015 , MNRAS , 451 , 2185 Crossref Search ADS Sollima A. et al. , 2016 , MNRAS , 462 , 1937 Crossref Search ADS Tanga P. et al. , 2016 , Planet. Space Sci. , 123 , 87 Crossref Search ADS Taylor M. B. , 2005 , Shopbell P. , Britton M. , Ebert R. , ASP Conf. Ser. Vol. 347, Astronomical Data Analysis Software and Systems XIV , Astron. Soc. Pac. , San Francisco , 29 Google Preview WorldCat COPAC Walker M. G. , Mateo M. , Olszewski E. W. , Bernstein R. , Wang X. , Woodroofe M. , 2006 , AJ , 131 , 2114 Crossref Search ADS Zwitter T. , Kos J. , 2015 , Mem. Soc. Astron. Ital. , 86 , 541 SUPPORTING INFORMATION Supplementary data are available at MNRAS online. Table 3. Column-by-column description of the final simulated stars. Please note: Oxford University Press is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article. APPENDIX A: LIST OF ACRONYMS Table A1 lists all the acronyms used. Table A1. List of acronyms used in this paper. Acronym Description AC ACross scan AF Astrometric field AL ALong scan BP Blue spectro-Photometer CCD Charge-coupled device CDS Centre de Données astronomiques de Strasbourg CMD Colour–magnitude diagram CNES Centre National d’Études Spatiales DPAC Data Processing and Analysis Consortium ESA European Space Agency FWHM Full width at half-maximum GIBIS Gaia Instrument and Basic Image Simulator GC Globular cluster HB Horizontal branch HST Hubble Space Telescope IMF Initial mass function LSF Line spread function MLE Maximum likelihood estimator MW Milky Way NSS Non-single stars PSF Point spread function RAVE RAdial Velocity Experiment RP Red spectro-Photometer RV Radial velocity RVS Radial velocity spectrometer sea Source Environment Analysis SM Sky Mapper TDI Time-delayed integration Acronym Description AC ACross scan AF Astrometric field AL ALong scan BP Blue spectro-Photometer CCD Charge-coupled device CDS Centre de Données astronomiques de Strasbourg CMD Colour–magnitude diagram CNES Centre National d’Études Spatiales DPAC Data Processing and Analysis Consortium ESA European Space Agency FWHM Full width at half-maximum GIBIS Gaia Instrument and Basic Image Simulator GC Globular cluster HB Horizontal branch HST Hubble Space Telescope IMF Initial mass function LSF Line spread function MLE Maximum likelihood estimator MW Milky Way NSS Non-single stars PSF Point spread function RAVE RAdial Velocity Experiment RP Red spectro-Photometer RV Radial velocity RVS Radial velocity spectrometer sea Source Environment Analysis SM Sky Mapper TDI Time-delayed integration Open in new tab Table A1. List of acronyms used in this paper. Acronym Description AC ACross scan AF Astrometric field AL ALong scan BP Blue spectro-Photometer CCD Charge-coupled device CDS Centre de Données astronomiques de Strasbourg CMD Colour–magnitude diagram CNES Centre National d’Études Spatiales DPAC Data Processing and Analysis Consortium ESA European Space Agency FWHM Full width at half-maximum GIBIS Gaia Instrument and Basic Image Simulator GC Globular cluster HB Horizontal branch HST Hubble Space Telescope IMF Initial mass function LSF Line spread function MLE Maximum likelihood estimator MW Milky Way NSS Non-single stars PSF Point spread function RAVE RAdial Velocity Experiment RP Red spectro-Photometer RV Radial velocity RVS Radial velocity spectrometer sea Source Environment Analysis SM Sky Mapper TDI Time-delayed integration Acronym Description AC ACross scan AF Astrometric field AL ALong scan BP Blue spectro-Photometer CCD Charge-coupled device CDS Centre de Données astronomiques de Strasbourg CMD Colour–magnitude diagram CNES Centre National d’Études Spatiales DPAC Data Processing and Analysis Consortium ESA European Space Agency FWHM Full width at half-maximum GIBIS Gaia Instrument and Basic Image Simulator GC Globular cluster HB Horizontal branch HST Hubble Space Telescope IMF Initial mass function LSF Line spread function MLE Maximum likelihood estimator MW Milky Way NSS Non-single stars PSF Point spread function RAVE RAdial Velocity Experiment RP Red spectro-Photometer RV Radial velocity RVS Radial velocity spectrometer sea Source Environment Analysis SM Sky Mapper TDI Time-delayed integration Open in new tab © 2017 The Authors Published by Oxford University Press on behalf of the Royal Astronomical Society
TI - Globular clusters with Gaia
JF - Monthly Notices of the Royal Astronomical Society
DO - 10.1093/mnras/stx079
DA - 2017-05-11
UR - https://www.deepdyve.com/lp/oxford-university-press/globular-clusters-with-gaia-FSRJmS7mSB
SP - 412
VL - 467
IS - 1
DP - DeepDyve
ER -