Evidence for the Third Stellar Population in the Milky Way's Disk

One of the most actively explored areas of contemporary astronomy is the formation and evolution of Milky Way-like galaxies and their main structural components. It is now possible to perform accurate chemodynamical studies of stellar populations in our Galaxy, thanks to the advent of large-scale surveys such as the Sloan Digital Sky Survey (SDSS; York et al. 2000), LAMOST (Cui et al. 2012), and the Gaia mission (Brown et al. 2018). The photometry and medium-resolution spectroscopy from SDSS or LAMOST provides estimates of stellar atmospheric parameters (effective temperature, surface gravity, and metallicity), and radial velocities, while the Gaia satellite delivers accurate positions, trigonometric parallaxes, and proper motions for many millions of stars. The combination of chemical information and kinematics is a powerful tool to explore the structures of the Milky Way and to derive the properties of the associated stellar populations, defined as samples of stars that exhibit common spatial distributions, kinematics, and chemical composition.

In the case of old stellar populations in the Milky Way (the bulge, thick disk, and halo system), spatial motions and chemical abundances can reveal crucial information on the assembly history of the Galaxy, because these stars conserve the kinematic and chemical properties of the ancient Galactic building blocks and assembly mechanisms (Bullock & Johnston 2005; Moore et al. 2006; Brook et al. 2007; Tissera et al. 2010). Previous works have successfully employed kinematics and chemical abundances to analyze the complexity of the Galaxy's halo, and have revealed the presence of (at least) two diffuse stellar components (Carollo et al. 2007, 2010; Beers et al. 2012; Tian et al. 2019) and numerous structures and over-densities (Grillmair & Carlin 2016).

Recent works based on the second Gaia data release (DR2; Brown et al. 2018) have shown the presence of additional structures or debris in the halo system, likely relics of past merging events (Helmi et al. 2018; Myeong et al. 2019).

The thick disk (TD), one of the main structural components of the Milky Way, has been extensively studied since its discovery (Yoshii 1982; Gilmore & Reid 1983). Stars in the TD possess different kinematics and chemical composition from the younger thin-disk stellar population. The metallicity distribution of the TD exhibits a peak at [Fe/H] = −0.6 (Chiba & Beers 2000; Ivezic et al. 2008; Carollo et al. 2010; Lee et al. 2011b); however, stars with disk-like kinematics have been observed at even lower metallicity, down to [Fe/H] = −1.7 (Chiba & Beers 2000; Carollo et al. 2010), and possibly lower. This low-metallicity tail of the TD is known as the metal-weak thick disk (MWTD; Morrison et al. 1990; Beers et al. 2014).

The properties of the MWTD, such as its velocity components, metallicity range, and stellar orbital parameters have been challenging to derive, due to the strong overlap with the TD and inner stellar halo, as well as the lack of accurate stellar distances and intrinsic motions. Previous attempts to model the MWTD as an independent stellar population from the TD (Carollo et al. 2010; Ivezic et al. 2008) have revealed that its mean Galactocentric rotational velocity could be in the range 100–150 km s⁻¹, while its metallicity spans values from [Fe/H] = −0.8 to −1.7.

In this article, we explore the nature of the MWTD and present evidence that it is an independent stellar population, using a sample of SDSS stars combined with accurate astrometric parameters provided by the second Gaia data release.

The stellar disk and halo populations are defined by their fundamental properties in kinematics and chemical space. For the stellar halo, the quantity V_rot/σ (where V_rot is the mean rotational velocity around the Galaxy center, and σ is the total velocity dispersion) is smaller than 1, indicating that halo stars have random motions around the Galactic center, with high energy, and can achieve large distances during their orbits. Also, their kinematic parameters and the spatial density do not exhibit strong dependence on the Galactocentric distance (R) or distance from the Galaxy's plane ($| Z|$), in the solar neighborhood. On the contrary, the disk components exhibit a ratio V_rot/σ larger than 1, suggesting a large mean rotation, and their kinematic and spatial density depend on the R and Z distances. Galaxy disks show a highly flattened distribution along the $| Z|$ direction as well. In chemical space, the disk and halo components differ in the content of metals ([Fe/H]) and their α-element ratios ([α/Fe]). Halo stars are metal poor ([Fe/H] < −1.0) and α-element rich. The two previously recognized disk components, the thin disk and the TD, have different chemical properties—the thin disk is metal rich $(\langle [\mathrm{Fe}/{\rm{H}}]\rangle \sim 0)$ and exhibits a low $\langle [\alpha /\mathrm{Fe}]\rangle$ ratio, while the TD is more metal poor $(\langle [\mathrm{Fe}/{\rm{H}}]\rangle \sim -0.6)$ and exhibits a higher [α/Fe] ratio. These definitions and properties should be considered when performing kinematics and chemical analysis of Galactic stellar populations to avoid a mixture of definitions.

2.1. Selection of the Data

2.2. Derivation of Kinematic and Dynamic Parameters

Figure 1 shows the distribution of stars (with relative errors on parallax less than 20%, and heliocentric distances less than 4 kpc) in the plane defined by L_Z and L_P. Stars that rotate in the same direction as the Galactic rotation (prograde) have positive values of ${L}_{{\rm{Z}}}$. In this diagram, the stars that come from the same progenitor(s) might be expected to cluster together, as they have similar values in the (L_Z, L_P) plane.

Zoom In Zoom Out Reset image size

The full sample of stars shown in Figure 1(a) exhibits two groups, with prograde rotation at L_Z ∼ 1800 kpc km s⁻¹ and L_Z ∼ 1200 kpc km s⁻¹, respectively. The group with larger vertical angular momentum, corresponding to a larger Galactocentric rotational velocity, represents the superposition of the thin-disk and TD components (Carollo et al. 2010). The less-prograde group of stars has rotational properties not previously associated with a separate stellar population in the Milky Way: this is the MWTD, which up to now appeared to be consistent with the low-metallicity tail of the TD, but as we demonstrate below, it is a distinct population.

Figure 1(b) represents the metallicity, as a function of L_Z, for the entire stellar sample, color coded on the relative abundance ratio between the α-elements (e.g., Mg, Ti) and metallicity ([α/Fe]). Typical values of [α/Fe] for the thin disk are between 0.0 and +0.15 (Lee et al. 2011b; Haywood et al. 2013; Recio-Blanco et al. 2014; Hayden et al. 2015), with metallicities −0.3 < [Fe/H] < +0.5. However, some investigations report evidence of the existence of a subpopulation of thin-disk stars with metallicities −0.6 < [Fe/H] < −0.2 and 0.0 < [α/Fe] < +0.2, located at Galactocentric distance R > 9 kpc, and possessing large mean rotational velocity (Lee et al. 2011b; Bovy et al. 2012b). These lower metallicity outer thin-disk stars are partially present in our sample (we estimate a fraction of ∼14% among the stars at R > 9 kpc; see Section 4), however they do not affect the properties of the less-prograde stellar population, and for the rest of the analysis we only consider their influence on the parameters derived for the TD. In Figure 1(b), stars with +0.15 < [α/Fe] < +0.2 are represented by a cyan color, while stars with [α/Fe] > +0.22 are shown in red. From inspection of this figure, the more-prograde stellar population comprising the TD and some thin-disk contamination, and the MWTD-like populations separate very well in the (L_Z, [Fe/H]) plane. The rest of the stars, with large values of [α/Fe] and lower metallicity, belong to the halo system. The average error on L_Z and L_P is 100 kpc km s⁻¹ and 70 kpc km s⁻¹, respectively; they are represented by error bars legend in Figure 1(a).

The peaks seen in Figure 1(c) clearly show the separation of the more-prograde and the less-prograde stellar components in the vertical angular momentum distribution. Rotational properties similar to those exhibited by the MWTD-like component were obtained in a previous work, assuming that the MWTD could be treated as an independent stellar population ($100\,\mathrm{km}\,{{\rm{s}}}^{-1}\lt {V}_{\phi }\lt 150\,\mathrm{km}$ s⁻¹, Carollo et al. 2010). Average rotational velocities for the thin disk and TD are 220 km s⁻¹ (Rix & Bovy 2013) and 180 km s⁻¹ (Carollo et al. 2010), respectively.

Figure 1(d) shows the normalized density distribution for the full sample (gray), the more-prograde stellar populations (cyan) and the high-α components (red), which are dominated by the MWTD-like and halo populations (the TD is also present in this range of α-element abundances). It is worth noting that the peak in L_Z for the cyan cluster (L_Z ∼ 1800–2000 kpc km s⁻¹) is larger than the known average value for the TD (L_Z ∼ 1600 kpc km s⁻¹, assuming the Sun's location at 8.5 kpc), due to thin-disk contamination. The panels of Figure 1 show, for the first time, clear evidence that the MWTD-like stellar population has different kinematics and chemical composition than the TD and halo, suggesting a distinct astrophysical origin.

Figure 2 shows [α/Fe], as a function of the metallicity, for the full sample of stars (panel a), and a subsample obtained by selecting stars with vertical angular momentum L_Z > 500 kpc km s⁻¹ and distance from the plane $| Z| \gt 1$ kpc (panel b). The cut in angular momentum reduces the contamination from halo stars, while the cut in vertical distance removes most of the thin-disk stars.

Zoom In Zoom Out Reset image size

In Figure 2(a), stars with metallicity −0.9 < [Fe/H] < −0.2 and +0.1 < [α/Fe] < +0.25 represent the abundance sequence of the TD (Hayden et al. 2015) with some thin-disk contamination, including the outer thin-disk stars that can reach large distances from the Galactic plane (Lee et al. 2011b; Bovy et al. 2012b; Savino & Posti 2019). The sequence delimited by the abundance ranges −1.2 < [Fe/H] < −0.6 and +0.22 < [α/Fe] < +0.35 is dominated by the prograde component with lower vertical angular momentum identified in Figure 1 (MWTD-like). The rest of the distribution, with a range in abundances [Fe/H] < −1.2 and +0.2 < [α/Fe] < +0.32, belongs to the halo system of the Galaxy.

The MWTD-like sequence is well represented in the subsample shown in Figure 2(b), although some stars in this stellar population were removed by the adopted cuts in L_Z and [Fe/H]. It is worth noting that in Figure 2(b), the two sequences in the range −1.2 < [Fe/H] < −0.3 are neither pure TD or pure MWTD. The TD (including outer thin-disk contaminants) and the MWTD-like stellar populations are strongly overlapped in the abundance space, and they dominate in the range of metallicity −1.2 < [Fe/H] < −0.3 with α-element ratios +0.1 < [α/Fe] < +0.37.

Figure 2(c) shows the metallicity distribution for the same sample of stars as in Figure 2(b), but divided into two subsamples, the α-poor ([α/Fe] < +0.22) and α-rich ([α/Fe] > +0.22) stars. This distribution exhibits two metallicity peaks at [Fe/H] ∼ −0.6 (cyan; right edge of the maxima of the distribution) and −1.0 (red; left edge of the maxima of the distribution), that correspond to the TD and MWTD-like stellar populations, with some contamination from lower metallicity halo stars ([Fe/H] ∼ −1.6) and outer thin-disk stars, which are identified in the abundance space with [Fe/H] up to −0.7 and +0.1 < [α/Fe] < +0.2 (Lee et al. 2011b; Bovy et al. 2012b). Figure 2(d) shows the [α/Fe] abundance distribution for the sample of stars from Figure 2(b), with an additional cut in metallicity to minimize the halo-star contamination. A mixture model analysis employing two independent Gaussian components applied to this distribution provides a very good fit, with mean values of [α/Fe] = +0.18 and [α/Fe] = +0.28, and standard deviations σ₁ = 0.035 dex, and σ₂ = 0.036 dex, respectively. The low-α component (TD dominated) accounts for 34% of the total number of stars in the subsample, while the high-α component (MWTD dominated) accounts for 66%. In Figure 2(d), the low-α and high-α Gaussian fits are displayed in cyan and red, respectively.

Figure 3 shows the distribution of the logarithmic number density, overplotted with equidensity contours for the entire sample in the (L_Z, L_P) and ([Fe/H], L_Z) planes (panels (a) and (b), respectively). The two prograde components can be clearly identified with the high-density reddish clusters with equidensity contour of 1.2 (panel a) and 1.0 (panel b). In panels (c) and (d), the density distributions are represented in the ([α/Fe], [Fe/H]) plane for the entire sample, and for a subsample obtained by selecting stars with vertical angular momentum L_Z > 500 kpc km s⁻¹, and distance from the plane of $| Z| \gt 1$ kpc, respectively, as in Figure 2. The density contour plots provide better visualization of the various stellar populations described in Figures 1 and 2.

Zoom In Zoom Out Reset image size

The rotational properties and abundances are key parameters for the separation of overlapping stellar populations in the Milky Way (Chiba & Beers 2000; Carollo et al. 2007, 2010). For the remaining analysis, we adopted a subsample of stars with metallicity −1.2 < [Fe/H] < −0.6 and α-element abundances [α/Fe] > +0.2 (Sample A). These cuts select stars belonging mainly to the MWTD-like and TD components, with some contamination from the inner halo (IH), while the previously noted contamination from the outer thin-disk stars is not present in this range of metallicity and α-element abundances.

Figure 4 shows the distribution of the vertical angular momentum (left column of panels) and (signed) orbital eccentricity (right column of panels) of Sample A for different cuts in the absolute values of vertical distance, $| Z|$, where the plus (minus) signs in eccentricity denote prograde (retrograde) motions. As a comparison, an additional subsample (Sample B), comprising mainly TD stars, obtained by adopting the cuts −0.9 < [Fe/H] < −0.6 and +0.1 < [α/Fe] < +0.2, is overplotted in red. This sample also has some thin-disk contamination.

Zoom In Zoom Out Reset image size

Visual inspection of the Sample A distributions reveals that, up to a vertical distance of $| Z| =1$ kpc, the TD and MWTD-like stellar populations are strongly overlapped, with some contamination from the IH, while Sample B is completely dominated by the overlapping thin disk and TD (L_Z ∼ 2000 kpc km s⁻¹). As expected, close to the Galactic plane, most of the stars have low orbital eccentricities (ecc ∼ +0.1). At larger distances, 1 kpc $\lt | Z| \lt 2$ kpc, the dominant components in Sample A are the MWTD-like and TD populations, with some contribution from the IH (L_Z ∼ 1200 kpc km s⁻¹, L_Z ∼ 1800 kpc km s⁻¹, and L_Z ∼ 0 kpc km s⁻¹, respectively).

The orbital eccentricity distributions shown in the right column of panels has stars with larger values associated with the MWTD-like component (up to ecc ∼ +0.5) and the IH ($| \mathrm{ecc}| \gt 0.5$; Carollo et al. 2010). In this range of distances, Sample B is still dominated by the overlapping thin disk and TD, with L_Z ∼ 1800 kpc km s⁻¹ and ecc ∼ +0.1. In the range 2 kpc $\lt | Z| \lt 3$ kpc, the MWTD is the dominant component in Sample A, with a larger contribution from the IH (L_Z ∼ 1200 kpc km s⁻¹ and L_Z ∼ 0 kpc km s⁻¹), while Sample B is TD dominated, although the number of stars at these vertical distances is rather low (the histograms are normalized to 1 to allow a comparison between the two samples). This is also evident in the orbital eccentricity distribution, where most of the stars possess +0.2 < ecc < +0.6 (and ecc < −0.5, corresponding to IH stars). At distances $| Z| \gt 3$ kpc, the MWTD-like stellar population is still present, but it is very weak. In all of the panels, the distributions are normalized to facilitate a visual comparison between samples, in particular, in those cases where Sample B contains a much lower number of stars than Sample A (at larger vertical distances). The peaks of the L_Z distributions include 95, 308, 92, and six stars, respectively, for Sample A, and 60, 88, 13, and one stars, respectively, for Sample B.

A mixture model analysis employing three independent (Gaussian) components applied to the distributions of stars in Sample A, up to $| Z| =3$ kpc, provides estimates for the mean vertical angular momentum and dispersion reported in Table 1. The mixture model includes three independent components identified in the table as TD, MWTD-like, and IH. Close to the Galactic plane, the TD and MWTD dominate the distribution, with 43% and 38% stellar fractions. At a larger vertical distance, the main component (60% of the stars) is MWTD-like, with a mean angular momentum of ∼1100 to ∼1200 kpc km s⁻¹. The inner-halo stellar population accounts for 19%, 14%, and 30% of the total number of stars in Sample A, respectively, while the TD-dominated component contributes 43%, 26%, and 8%, at $0\,\mathrm{kpc}\lt | Z| \lt 1\,\mathrm{kpc}$, $1\,\mathrm{kpc}\,\lt | Z| \lt 2\,\mathrm{kpc}$, and $2\,\mathrm{kpc}\lt | Z| \lt 3\,\mathrm{kpc}$, respectively.

Table 1.  Parameters from Mixture Modeling Analysis for Sample A

$| Z|$	N	$\langle {L}_{{\rm{Z}}}\rangle$	${\sigma }_{{L}_{{\rm{Z}}}}$	P
(kpc)		(kpc km s−1)	(kpc km s−1)	(%)
0–1	TD	1904 ± 89	144 ± 133	43
	MWTD	1267 ± 72	353 ± 77	38
	IH	−8 ± 199	443 ± 123	19
1–2	TD	1744 ± 231	210 ± 34	26
	MWTD	1173 ± 51	363 ± 31	60
	IH	134 ± 200	656 ± 85	14
2–3	TD	1774 ± 231	236 ± 111	8
	MWTD	1076 ± 96	416 ± 63	62
	IH	−198 ± 179	473 ± 92	30

Download table as:  ASCIITypeset image

It is well known that the thick disk exhibits a negative gradient in the mean rotational velocity as a function of the vertical distance (Chiba & Beers 2000; Carollo et al. 2010 and references therein). This is an intrinsic property of this stellar population, which is likely due to its mechanism of formation. By using a subsample of stars in the range −1.1 < [Fe/H] < −0.6 and [α/Fe] > +0.22, we show that the MWTD-like population exhibits a dependence on the vertical distance (Figure 5), with a gradient of ${\rm{\Delta }}\langle {V}_{\phi }\rangle /{\rm{\Delta }}\langle | Z| \rangle =-21$ km s⁻¹ kpc⁻¹, lower than that determined for the canonical TD (${\rm{\Delta }}\langle {V}_{\phi }\rangle /{\rm{\Delta }}\langle | Z| \rangle =-36$ km s⁻¹ kpc⁻¹, Carollo et al. 2010). Determination of the mean rotational velocity and its dispersion for the MWTD is accomplished by assuming a fiducial sample of stars selected in the range $1\,\mathrm{kpc}\lt | Z| \lt 2\,\mathrm{kpc}$, −1.1 < [Fe/H] < −0.6, and [α/Fe] > +0.25. The results are $\langle {V}_{\phi }\rangle \,=147\pm 2$ km s⁻¹ and ${\sigma }_{{V}_{\phi }}=60\pm 2$ km s⁻¹. Note that a similar velocity dispersion was obtained in a recent work (Robin et al. 2017) for the old thick-disk stellar population, corresponding to the oldest episode of star formation in the disk (12 Gyr), according to the Besancon population synthesis model (Robin et al. 2014). Tian et al. (2019) identify a disk-like component with mean rotation and velocity dispersion which they claimed could be identified with the MWTD, but it appears that their values are more consistent with the canonical TD.

Zoom In Zoom Out Reset image size

The mean orbital parameters, such as the apo-Galactic and peri-Galactic distances ($\langle {r}_{\mathrm{apo}}\rangle$, $\langle {r}_{\mathrm{peri}}\rangle$, maximum and minimum distance from the Galaxy's center achieved by a star during its orbit) and $\langle {Z}_{\max }\rangle$ (maximum distance from the Galaxy's plane achieved by a star during its orbit) are $\langle {r}_{\mathrm{apo}}\rangle \sim 10\,\mathrm{kpc}$, $\langle {r}_{\mathrm{peri}}\rangle \sim 5$ kpc, and $\langle {Z}_{\max }\rangle \sim 3$ kpc, respectively.

By using the fiducial sample of stars employed to determine the mean rotational velocity and dispersion for the MWTD-like stellar population ($1\,\mathrm{kpc}\lt | Z| \lt 2\,\mathrm{kpc}$, −1.1 > [Fe/H] < −0.6 and [α/Fe] > +0.25), we found that the mean Galactocentric distance for this component is R = 8.6 ± 0.2 kpc.

The ratio $\langle {V}_{\phi }\rangle /\sigma$ (where σ is the total velocity dispersion) for stellar populations in the Galaxy's disk is typically larger than 1. In the case of the MWTD-like component, this parameter has a value of ∼2.5. Moreover, its rotational velocity shows a clear dependence on the vertical distance. Such properties unequivocally identify the MWTD as a disk component.

4.1. On the Thin-disk Contamination

The thin-disk component is not well represented among the SDSS DR7 calibration stars (Carollo et al. 2010) because these stars were selected at Galactic latitudes that would avoid thin-disk stars. However, some contamination may still be present in our data, which can be largely removed by selecting stars at $| Z| \gt 1$ kpc. Previous analyses have shown that a portion of the thin-disk stellar population might possess higher [α/Fe] ratios (+0.1 < [α/Fe] < +0.2), lower metallicities ([Fe/H] ∼ −0.5), and located at larger Galactocentric distance than the low-α metal-rich thin disk (the so-called outer-disk stars, R > 9 kpc; Bovy et al. 2012b). The α-rich portion of this stellar disk can reach larger vertical distances than the α-poor counterpart. To test if the α-rich thin-disk stars can affect our results, we constructed a new version of Figure 1 (labeled as Figure 6) by selecting stars at Galactocentric distance of 7 kpc < R < 9 kpc. Such a cut would exclude the outer thin-disk stars, as well as a fraction of the TD, MWTD, and halo stars.

Zoom In Zoom Out Reset image size

Figure 6 shows that the two prograde groups identified in the (${L}_{{\rm{Z}}},{L}_{{\rm{P}}}$) diagram (panel a) are still present in this new sample, although the more-prograde cluster contains a lower number of stars than the sample represented in Figure 1. In the (L_Z, [Fe/H]) plane (panel b), stars with +0.1 < [α/Fe] < +0.2 (low-α) are shown in cyan, while stars with [α/Fe] > +0.22 (high-α) are shown in red. From an inspection of this figure, the low-α prograde component, dominated by TD stars, is clearly separated from the high-α prograde component, which is dominated by the MWTD-like population.

It is important to remark that the above low-α and high-α separation reflects the range of [α/Fe] ratios exhibited by the two clusters at [Fe/H] > −1.2 in panels 1(a) and (b) of Figure 2, and that the high-α (as defined above) prograde stellar population also includes TD stars. In panels 1(c) and (d), the peak in L_Z associated with the less-prograde component is well represented, while the peak related to the more-prograde component is somewhat flattened. This is mainly due to the fact that the cut in Galactocentric distance (R < 9 kpc) also removes a significant number of TD stars.

We applied the mixture modeling analysis to the subsamples of stars obtained for different cuts of the vertical distance, $| Z|$, starting from the sample with −1.2 < [Fe/H] < −0.6, α-element ratios [α/Fe] > +0.2, and 7 kpc < R < 9 kpc (Sample C). The results are reported in Table 2. The kinematic parameters of the MWTD-like stellar population remain unchanged; it is the dominant component in all three distributions, with a mean angular momentum of 1100–1200 kpc km s⁻¹. The main differences between these results and those reported for Sample A in Table 1 are (1) a lower mean vertical angular momentum for the TD close to the Galactic plane, $\langle {L}_{{\rm{Z}}}\rangle \,\sim 1700$ kpc km s⁻¹ (∼1900 kpc km s⁻¹ when 7 kpc < R < 10 kpc is used) and (2) the TD component is not present at vertical distances $2\,\mathrm{kpc}\lt | Z| \lt 3\,\mathrm{kpc}$ (the best fit was obtained with two components only). We suspect that by removing stars with R > 9 kpc, the TD stellar population is significantly reduced, affecting the presence of such populations at these distances.

Table 2.  Parameters from Mixture Modeling Analysis for Sample C

$| Z|$	N	$\langle {L}_{{\rm{Z}}}\rangle$	${\sigma }_{{L}_{{\rm{Z}}}}$	P
(kpc)		(kpc km s−1)	(kpc km s−1)	(%)
0–1	TD	1698 ± 166	200 ± 78	31
	MWTD	1193 ± 93	157 ± 100	41
	IH	427 ± 224	490 ± 124	28
1–2	TD	1691 ± 62	151 ± 49	10
	MWTD	1176 ± 38	327 ± 25	74
	IH	134 ± 200	656 ± 85	15
2–3	MWTD	1076 ± 96	416 ± 63	82
	IH	−198 ± 179	473 ± 92	18

Download table as:  ASCIITypeset image

In previous works (Lee et al. 2011b; Bovy et al. 2012b), it was found that the outer thin disk is located in the metallicity range −0.7 < [Fe/H] < −0.2, with 0.0 < [α/Fe] < +0.2, and resides at Galactocentric distance R > 9 kpc. In this abundance range, both the thin-disk and TD populations are present, and it is difficult to accurately quantify the thin-disk contamination because this component is not well represented in our sample.

Based on the results reported in previous works, we considered outer thin-disk stars to be those in the metallicity range −0.7 < [Fe/H] < −0.2 with α-abundance ratios 0.0 < [α/Fe] < +0.2, and located at R > 9 kpc and $0\,\mathrm{kpc}\lt | Z| \,\lt 2\,\mathrm{kpc}$. The total number of stars located at R > 9 kpc in our sample is N = 4527, while the number of likely outer thin-disk stars is N = 637, corresponding to a contamination of 14%.

4.2. Rapidly Rotating TD Stars

In Figure 4, the peak values of L_Z for the subsamples in different cuts of $| Z|$, obtained from Sample B, are larger (∼2000 kpc km s⁻¹) than those expected for a distribution dominated by TD stars (∼1600 kpc kms⁻¹, assuming the Sun's position at R = 8.5 kpc). Sample B is selected in a range of metallicity and α-elements abundance (−0.9 < [Fe/H] < − 0.6 and +0.1 < [α/Fe] < +0.2) that excludes the majority of thin-disk stars (see, for example, Figure 4 in Hayden et al. (2015) and Figure 2 in Cheng et al. (2012)). In this range of abundances, however, the outer thin-disk stars have been identified, −0.7 < [Fe/H] < −0.2, and 0.0 < [α/Fe] < +0.2 (Lee et al. 2011b; Bovy et al. 2012b). The total number of stars in Sample B is 415, while the number of likely outer thin-disk stars, selected according to the above criteria, is 77. This means that Sample B has a contamination from "likely" outer thin-disk stars on the order of ∼18%. This would explain the large values of L_Z obtained for the TD. To investigate further, we explored the trend of the mean rotational velocity, $\langle {V}_{\phi }\rangle$, for Sample B (−0.9 < [Fe/H] < −0.6; +0.1 < [α/Fe] < +0.2) as a function of metallicity. It is known that thin-disk stars show a negative rotational velocity gradient ($\langle {V}_{\phi }\rangle /{\rm{\Delta }}\langle | Z| \rangle$) as the metallicity increases, while TD stars exhibit a positive or zero velocity gradient (Lee et al. 2011b). Figure 7 shows the mean Galactocentric rotational velocity, as a function of the metallicity, for Sample B, in two different intervals of vertical distance, $0\,\mathrm{kpc}\lt | Z| \lt 1\,\mathrm{kpc}$ (black), and $1\,\mathrm{kpc}\,\lt | Z| \lt 2\,\mathrm{kpc}$ (red). Visual inspection of this figure reveals that the mean rotational velocity has a zero or slightly positive gradient as the metallicity increases. Therefore, it is unlikely that Sample B has significant contamination from outer thin-disk stars. We conclude that in the range of abundances where Sample B is selected, the majority of the stars belong to the TD, while other works claim that in such an interval of metallicity and α-abundance, the outer thin-disk stars are still present (Lee et al. 2011b; Bovy et al. 2012b). It remains unclear the reason why these TD stars possess such large rotational velocities, requiring further investigation.

Zoom In Zoom Out Reset image size

How could a second stellar population in the TD have formed, and what are the implications for the assembly history of the Galaxy?

In a previous work (Carollo et al. 2010), it was suggested that there may exist a connection between the MWTD-like component and the Monoceros Ring (an overdensity of stars in a large area of the sky approximately parallel to the Galactic plane toward the anticenter of the Galaxy, in the Galactic latitude range $10^\circ \lt | b| \lt 35^\circ$; this spans a large area in Galactic longitude, covering most of the second and third quadrants (Newberg et al. 2002; Ibata et al. 2003) on the basis of its mean metallicity, [Fe/H] ∼ −1.0 and its rotational velocity, ∼110 km s⁻¹ (Ivezic et al. 2008). The origin of the Monoceros Ring has been a matter of debate, and recent evidence tends to support the hypothesis that it is not the debris of an accreted ancient satellite, but probably a part of the wobbly Galactic disk (de Boer et al. 2018; Sheffield et al. 2018) induced by the tidal interaction with massive satellites such as the Sagittarius dwarf and/or the Large Magellanic Cloud (Laporte et al. 2018). If this is the case, then the MWTD observed near the Sun may simply be a gravitationally heated/warped ensemble of TD stars, exhibiting a slower mean rotation than the TD, due to asymmetric drift. However, this simple scenario cannot explain the systematically lower metalliticites and higher [α/Fe] ratios of the MWTD stars compared with TD stars, as found in this work.

Instead, the MWTD stars might be more ancient TD stars, heated and raised to higher vertical distances, caused by a Gaia-Enceladus-like merger about 10 billion years ago, in the course of the formation of the Milky Way's IH (Helmi et al. 2018; Bignone et al. 2019); the main body of the currently observed TD may be formed by the more recent merging of a dwarf satellite and associated disk heating (Gallart et al. 2019).

Another possibility for explaining the MWTD stars is that they are originally formed at lower Galactocentric radii, R ∼ 5 kpc, than the Sun, before the main body of the TD formed, accounting for their lower L_Z, lower [Fe/H], higher [α/Fe], and probably older ages than the TD. Then, internal dynamical process in this stellar disk, such as bar instability and/or spiral-arm formation may have induced radial migration of these stars to the solar neighborhood. Alternatively, the transport of stars might be caused by the merging of Gaia-Enceladus, inducing a nonaxisymmetric structure such as a bar and/or spiral in this early disk, before the formation of the main body of the TD.

As an alternative scenario, the MWTD could result from partially circularized debris of a relatively massive accreted satellite. The circularization occurs through dynamical friction, which favors a satellite of mass comparable to the SMC. The observed metallicity of the MWTD would also be consistent with a satellite of such mass, through the mass–metallicity relation for dwarf galaxies (Kirby et al. 2013). Deposition of satellite stars into the Galactic disk in this way would suggest the possibility that dark matter from the satellite is also deposited into the disk. To assess the validity of these formation scenarios, it is worth further exploring state-of-the-art high-resolution numerical simulations for the formation of the Milky Way's disk and halo system in a consistent manner.

D.C. acknowledges support from the ESO Scientific Visitor Programme and support by Sonderforschungsbereich SFB 881 "The Milky Way System" (subproject A9) of the German Research Foundation (DFG). M.C. and M.N.I. acknowledge partial support by MEXT Grant-in-Aid for Scientific Research (No. 17H01101 and 18H04334 for MC, 18H05437 for M.C. and MNI). K.C.F. acknowledges support from Australian Research Council grant DP160103747. T.C.B. acknowledges partial support from grant PHY 14-30152; Physics Frontier Center/JINA Center for the Evolution of the Elements (JINA-CEE), awarded by the US National Science Foundation. Y.S.L. acknowledges support from the National Research Foundation (NRF) of Korea grant funded by the Ministry of Science and ICT (No.2017R1A5A1070354 and NRF-2018R1A2B6003961). P.B.T. acknowledges partial support from Fondecyt 1150334 and UNAB Grant 10/19. C.B. acknowledges support by Sonderforschungsbereich SFB 881 "The Milky Way System" (subproject A9) of the German Research Foundation (DFG).

There are two potential problems to take into account when using Gaia DR2 parallaxes: the issues that could arise by calculating the distance as the inverse of parallax, and the well-known parallax zero-point offset.

In this analysis, the stellar distances were estimated by taking the inverse of parallax for stars with relative error σ_π/π < 0.2. However, distance determination from parallax measurements is an inference problem and was extensively discussed by Bailer–Jones et al. (2018). The top-right panel of Figure 8 shows the comparison between the distances obtained by taking the inverse of parallax (d_g) with those derived by Bailer–Jones et al. (2018) (d_BJ) for the sample of stars satisfying the condition σ_π/π < 0.2. As can be appreciated from this figure, there is very good agreement up to ∼2 kpc, then d_g tends to be somewhat overestimated with respect to d_BJ, with deviations on the order of ∼(0.2–0.6) kpc that increases with distance. To assess if such deviations may influence the results presented in this paper, the kinematic parameters are evaluated by adopting the Bailer–Jones distances. The (L_Z, L_P) diagram obtained with these distances is shown in the bottom-left panel of Figure 8. This new angular momenta diagram is identical to the top-left panel of Figure 1, and exhibit two groups of stars with prograde rotation at L_Z ∼ 1800 kpc km s⁻¹ and L_Z ∼ 1200 kpc km s⁻¹. This exercise shows that the distance obtained by the inverse of parallax for the sample of stars with σ_π/π < 0.2 is a robust determination.

Zoom In Zoom Out Reset image size

Lindegren et al. (2018) showed that parallax measurements are affected by an average zero-point offset of δ_π = −0.029 mas. This bias appears to be dependent on multiple variables, such as stellar magnitudes, colors, sky position, and the employed stellar sample, and it affects primarily distant stars (small parallaxes). Recent works report a slightly higher value of the parallax zero-point offset, on the order of δ_π = −0.05 mas (Graczyk et al. 2019; Leung & Bovy 2019; Schönrich et al. 2019; Zinn et al. 2019). In order to establish the important of this offset on the results, we performed the bulk of the analysis by adding a constant value of 0.054 mas to the parallaxes. The assumption of a constant parallax zero-point offset can be justified by following the results reported in Leung & Bovy (2019), where it is shown that the constant offset model is in very good agreement with the multivariate offset model up to 10 kpc for a sample of ∼265,000 stars in common between APOGEE DR14 (Holtzman et al. 2015; Garcia Perez et al. 2016; Abolfathi et al. 2018) and Gaia DR2.

The top-right panel of Figure 8 shows the distance obtained with the inverse of parallax corrected for the bias, as a function of the distance inferred with the Bailer–Jones's method. The agreement between the two distances is improved, and the maximum deviation is only 0.2 kpc. Again, to assess the importance of the parallax correction for the zero-point offset on the results reported in this paper, we derived the kinematic parameters with these new distances. The bottom-right panel of Figure 8 shows the (L_Z, L_P) diagram for the sample of stars with relative error on parallax < 20%, which now contains a larger number of stars with respect to the sample obtained with no zero-point offset. This can be explained by the fact that many stars are closer and are included in the local volume of 4 kpc. The bottom-right panel clearly shows that the two prograde groups of stars at L_Z ∼ 1800 kpc km s⁻¹ and L_Z ∼ 1200 kpc km s⁻¹ are now even better defined than before, thanks also to the increased number of stars in the local volume, below 4 kpc.

The vertical angular momentum distribution for the samples A and B obtained with the parallax zero-point offset correction is shown in Figure 9. A comparison with Figure 4 shows that, at $0\,\mathrm{kpc}\lt | Z| \lt 3\,\mathrm{kpc}$, both L_Z and the eccentricity distribution for the two samples are almost identical to those reported in the three top panels of this figure. Farther from the Galactic plane, at $3\,\mathrm{kpc}\lt | Z| \lt 4\,\mathrm{kpc}$, there are only a few stars due to the slightly shorter distances obtained with the addition of the parallax zero-point offset. The results of the mixture model analysis applied to these new distributions are reported in Table 3. A comparison with Table 1 shows that the values of mean angular momentum and dispersion for the TD, MWTD, and IH are very similar. Close to the Galactic plane ($0\,\mathrm{kpc}\lt | Z| \lt 1\,\mathrm{kpc}$), the distribution is dominated by the TD and MWTD-like components, while at larger vertical distances the MWTD-like component is dominant, with mean angular momentum of ∼1100 to ∼1200 kpc km s⁻¹. The inner-halo stellar population accounts for 15%, 17%, and 25% of the total number of stars in Sample A, respectively, while the TD-dominated component contributes 45%, 40%, and 18%, at $0\,\mathrm{kpc}\lt | Z| \lt 1\,\mathrm{kpc}$, $1\,\mathrm{kpc}\lt | Z| \lt 2\,\mathrm{kpc}$, and $2\,\mathrm{kpc}\lt | Z| \,\lt 3\,\mathrm{kpc}$, respectively.

Zoom In Zoom Out Reset image size

Table 3.  Parameters from Mixture Modeling Analysis for Sample A (With Parallax Zero-point Correction)

$| Z|$	N	$\langle {L}_{{\rm{Z}}}\rangle$	${\sigma }_{{L}_{{\rm{Z}}}}$	P
(kpc)		(kpc km s−1)	(kpc km s−1)	(%)
0–1	TD	1917 ± 35	187 ± 19	45
	MWTD	1318 ± 56	251 ± 61	40
	IH	372 ± 226	550 ± 114	15
1–2	TD	1819 ± 40	217 ± 19	30
	MWTD	1210 ± 51	320 ± 46	53
	IH	184 ± 150	546 ± 70	17
2–3	TD	1696 ± 124	230 ± 60	18
	MWTD	1108 ± 125	359 ± 89	57
	IH	−2.5 ± 198	465 ± 100	25

Download table as:  ASCIITypeset image

The mean Galactocentric rotational velocity and dispersion for the MWTD-like stellar population was obtained by adopting the fiducial sample of stars described in Section 3 ($1\,\mathrm{kpc}\lt | Z| \lt 2\,\mathrm{kpc}$, −1.1 < [Fe/H] < −0.6, and [α/Fe] > +0.25). The results are $\langle {V}_{\phi }\rangle =150\pm 2$ km s⁻¹ and ${\sigma }_{{V}_{\phi }}\,=58\pm 2$ km s⁻¹ which are totally in agreement with the values reported in Section 3, obtained without adding the zero-point offset bias to the parallaxes.

An additional sample of stars, selected by cross-matching the SDSS DR15 (Aguado et al. 2019) with Gaia DR2, was employed to obtain a separate confirmation of the MWTD component in phase space and chemical space. The sample contains only stars with available spectroscopy, and for which the radial velocities and the chemical abundances are determined through the SSPP pipeline. The [α/Fe] ratios were taken from SDSS DR7 (Abazajian et al. 2009).

We selected stars with a relative error on parallax σ_π/π < 0.2, and, calculated distances using the inverse of parallax, as for the calibration-star sample (with no parallax zero-point offset correction). The resulting number of stars is ∼68,000, and the majority have heliocentric distances d < 4 kpc. Kinematical parameters are obtained by assuming the Sun's location at 8.5 kpc (Ghez et al. 2008) from the Galaxy's center and correction for Sun's motion of (U, V, W) = (−9, 12, 7) km s⁻¹. The circular velocity at the position of the Sun is 220 km s⁻¹ (Bovy et al. 2012a). Orbital angular momenta are computed by using the procedures described above.

Figure 10 shows the distribution of the selected stars in the (${L}_{{\rm{Z}}},{L}_{{\rm{P}}}$) plane. It is worth noting that in this additional sample that the thin-disk component is better represented than in the calibration-star sample used in our main analysis. Therefore, to reduce the contamination from this stellar population, we only plot stars with metallicity [Fe/H] < −0.6 and α-element abundance ratio [α/Fe] > +0.1. We also included an additional cut in metallicity ([Fe/H] > −1.6) to reduce the contamination from halo stars. In panel (a), the more-prograde (TD-dominated) and less-prograde components are clearly recognizable as the two clusters, with angular momentum L_Z ∼ 1800 kpc km s⁻¹ and ${L}_{{\rm{Z}}}\sim 1200\,\mathrm{kpc}$ km s⁻¹, that we associate with the TD-dominated and the MWTD-like stellar populations, respectively. The density distribution shown in panel (b) exhibits two peaks in L_Z, as in the calibration-star sample (Figure 1), confirming the detection of the MWTD-like component in SDSS DR15 as well.

Zoom In Zoom Out Reset image size