The Bulge Radial Velocity Assay for RR Lyrae Stars (BRAVA-RR) DR2: A Bimodal Bulge?

The OGLE-IV catalog of bulge RRLs (Pietrukowicz et al. 2015) was used to select the RRLs. As in BRR-DR1, the wavelength covered was ∼8300 Å to 8800 Å at a resolution of R ∼ 10,000 using the AAOmega multifiber spectrograph on the Anglo-Australian Telescope (AAT). Exposure times were between 0.33 and 3 hours, and in general, we collected between one and four epochs for each RRL, the multiple observations required to measure systemic velocities for each star. Observations were taken from 2017 June 16—June 21. ¹³ Reductions were carried out in an identical manner to BRR-DR1, and radial velocities were found using IRAF's xcsao routine. The xcsao cross-correlation was carried out using spectra taken of the APOGEE star 2M18264551–1747096 and adopting the velocity of this star to be −69.2 km s⁻¹ (Alam et al. 2015). The spectrum of this star was acquired using the same setup with AAOmega that was used with our RRLs. The individual uncertainties from xcsao typically ranged from ∼1 km s⁻¹ to ∼4 km s⁻¹.

The radial velocity measurements were phased by the stars' known OGLE period and overplotted with the radial velocity template from Liu (1991). The template is scaled using

$$\begin{eqnarray}&&{A}_{{rv}}=\displaystyle \frac{40.5\times {V}_{\mathrm{amp}}+42.7}{1.37},\end{eqnarray} \tag{ 1 }$$

where V_amp is the V-band amplitude, to account for the correlation between the amplitudes of the velocity curves and light curve (Liu 1991). To be consistent with BRR-DR1, we adopt p = 1.37, the so-called "projection factor," to relate our observed radial velocities to the pulsation radial velocities (see also Sesar 2012). Because OGLE samples the I band much more frequently than the V band, we used the following relation for the V amplitude:

$$\begin{eqnarray}&&{V}_{\mathrm{amp}}={I}_{\mathrm{amp}}\times 1.6\end{eqnarray} \tag{ 2 }$$

(see, e.g., Table 3 in Kunder et al. 2013).

In BRR-DR1, we found that the uncertainty in the center-of-mass radial velocity measurement due to having a small (between one and four) number of epochs to be between 2 and 5 km s⁻¹. Adding in quadrature both the individual uncertainty on each velocity measurement as reported from xcsao as well as that from having a small number of epochs for center-of-mass radial velocity determination, the total center-of-mass radial velocity uncertainties are ∼5–10 km s⁻¹.

Although we made every effort to observe RRLs that were free of a companion within a 2'' radius, we did notice evidence of blending in a small fraction of our RRL stellar spectra (∼1%). For some spectra, not one, but two distinct dips in intensity around the calcium triplet (CaT) was seen. These two distinct CaT lines represent the CaT lines of two different stars: the RRL and a close neighbor. The most likely reason for this blending is that the typical seeing during our observing runs was between 15 and 25. We discarded the spectra with two distinct CaT lines, as we could not be sure which CaT line belonged to the RRL and which one belonged to the neighbor. There may be other stars that have blends but that we were not able to identify due to low signal-to-noise ratio, a small difference in radial velocity between the RRL and the neighboring star, and/or too many spectral lines near the CaT lines from, e.g., shock waves (Chadid et al. 2008; Yang et al. 2014), or the star having a hot temperature where hydrogen Paschen lines dominate as compared to the CaT signature.

For 98% of our sample, we had more than one epoch of observations. Therefore, we visually checked each radial velocity curve and made sure it matched that from Liu (1991) using the zero point in phase fixed using the OGLE-IV time of maximum brightness. Stars with a large scatter in their radial velocity curves are flagged with flag=1, where large scatter means that at least one observation deviates by ∼20 km s⁻¹ or more from the Liu (1991) template. Two typical stars with flag=1 are shown in Figure 1.

Zoom In Zoom Out Reset image size

Some of the stars with scatter may be attributed to stellar blending that went undetected by our visual examinations of stellar spectra. Blends are less likely to be detected if a particular observation is taken when an RRL is at a high temperature during its pulsation cycle, as this is where the hydrogen Paschen lines become significant and the CaT lines are less prominent. We do not believe that stars with flag=0 are blends, as otherwise the individual radial velocities would not follow the Liu (1991) template (see Figure 1). Note that the phase of the observation is fixed using the times of minimum brightness given by OGLE, and the Liu (1991) template is shifted only up or down to fit the individual radial velocity measurements. We cannot be certain that stars with one epoch of observations are not affected by undetected blends.

Figure 2 shows a comparison between the radial velocities presented here and those from both BRR-DR1 and Kunder et al. (2018a). A striking ∼10 km s⁻¹ offset is found from the radial velocities here as compared to BRR-DR1, and no meaningful offset is found from the radial velocities here as compared to Kunder et al. (2018a). The difference in the BRR-DR1 velocities stems from the standard star used for the radial velocity zero point. In contrast to BRR-DR1, where a star from the BRAVA survey was used as a radial velocity standard, here we set the radial velocity zero point to the APOGEE star 2M18264551–1747096. The radial velocity of this star is better determined than the one used by BRR-DR1, and therefore, the velocities presented here should supersede those from BRR-DR1. In Kunder et al. (2018a), the center of the three CaT lines was measured directly by hand, so the agreement between these radial velocities and the ones presented here indicates that our zero point is much improved from BRR-DR1.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Examples of RRL radial velocity curves are shown in Figure 1, where we show those stars with the largest space velocities, which are discussed further in Section 5. All radial velocity measurements are presented in Table 1. Table 1 gives the OGLE ID (1), the R.A. (2), and decl. (3) as provided by OGLE, the star's time-average velocity (4), the number of epochs used for the star's time-average velocity (5), the radial velocity flag (where stars with well-fit radial velocity curves have flag=0 and stars with a large scatter around their radial velocity curves are flagged with flag=1) (6), the period of the star (7), V-band magnitude (8), the I-band magnitude (9) and the I-band amplitude (10) as calculated by OGLE, and lastly, the distance (see Section 2.3) adopted for the orbital integration (11).

Table 1. DR2 Radial Velocities of the BRAVA-RR stars

OGLE ID	R.A. (J2000.0)	Decl. (J2000.0)	${\mathrm{HRV}}_{\phi =0.38}$ (km s−1)	# Epochs	Flag	Period (days)	${(V)}_{\mathrm{mag}}$	${(I)}_{\mathrm{mag}}$	Iamp	Distance (kpc)
01743	17 43 38.31	−34 28 23.30	85	3	0	0.5314405	18.19	16.55	0.687	12.4
01773	17 43 49.26	−34 24 13.40	−147	3	0	0.4375252	18.35	16.78	0.781	9.9
01806	17 44 0.89	−34 27 30.40	−24	1	0	0.5615170	17.93	16.26	0.535	8.3
01860	17 44 18.85	−34 28 13.70	62	1	0	0.5062415	18.13	16.71	0.765	10.7
01862	17 44 19.49	−34 19 38.60	171	1	0	0.4645364	18.72	17.08	0.761	11.2
01873	17 44 22.58	−34 19 6.10	−61	3	0	0.4239042	18.56	16.84	0.827	9.2
01938	17 44 46.95	−34 23 58.00	−22	3	0	0.5459384	18.58	16.97	0.573	11.6
01983	17 45 1.83	−34 11 27.70	32	2	0	0.4759508	18.05	16.37	0.544	7.9
02000	17 45 5.33	−35 50 53.20	12	2	0	0.4803552	17.98	16.53	0.859	9.8
02016	17 45 7.56	−35 21 22.90	−65	2	0	0.5553001	17.35	15.81	0.472	7.2

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

With the Gaia DR2 data release (Gaia Collaboration et al. 2018a), we now have in hand millions of proper motions. These have (for example) the precision to separate bulge RRLs from those RRLs residing in the Sagittarius dwarf galaxy (Sgr; e.g., Kunder et al. 2019) that otherwise complicate the study of the bulge.

Proper motions of the RRLs are obtained from cross-matching with the Gaia DR2 catalog. To ensure that the cross-match for each RRL is correct in the crowded bulge region of the sky, only the matches with Gaia flag phot_variable_flag=VARIABLE were kept. We further limited our cross-matches to those stars with the recommended astrometric quality indicators such as the unit weight error (UWE) and the renormalized unit weight error (RUWE), as described by Lindegren et al. (2018). Specifically, only stars with a RUWE < 1.4 were kept, where the RUWE of a star was calculated using

$$\begin{eqnarray}&&\mathrm{RUWE}=\displaystyle \frac{\mathrm{UWE}}{{\mu }_{0}(G,C)}.\end{eqnarray} \tag{ 3 }$$

Here, µ₀(G, C) is a normalization factor obtained from interpolating the ${phot}\_g\_{mean}\_{mag}$ (G) and rp − bp color of the stars in the table on the ESA Gaia DR2 known issues page. The UWE can be calculated using the equation

$$\begin{eqnarray}&&\mathrm{UWE}=\sqrt{\displaystyle \frac{{\chi }^{2}}{(N-5)}},\end{eqnarray} \tag{ 4 }$$

where χ² is given by the ${astrometric}\_{chi}2\_{al}$ parameter and N is given by the ${astrometric}\_n\_{good}\_{obs}\_{al}$ parameter. Our final sample of RRLs with Gaia DR2 astrometry contains 1389 stars out of the 2768 RRLs presented here.

One advantage of using RRLs as stellar tracers is that they are relatively well-understood standard candles. Distances for each RRL are calculated using the identical procedure as in Pietrukowicz et al. (2015). First, the photometric metallicities of the stars were found using the following relation from Smolec (2005):

$$\begin{eqnarray}&&[\mathrm{Fe}/{\rm{H}}]=-3.142-4.902P+0.824({\phi }_{31}+\pi )),\end{eqnarray} \tag{ 5 }$$

where P is the pulsation periods of the stars and ϕ₃₁ is a Fourier-phase combination for the cosine decomposition for the I-band light curves of the stars. The photometric metallicities indicate a sharply peaked RRL metallicity distribution at [Fe/H] = −1.02 dex, with a dispersion of 0.25 dex (Pietrukowicz et al. 2012, 2015). These photometric metallicities were then used to find their absolute magnitudes, M_V and M_I , using theoretical relations from Catelan et al. (2004):

$$\begin{eqnarray}&&{M}_{V}=-2.288-0.882\,\mathrm{log}({\rm{Z}})+0.108\,\mathrm{log}{\left(Z\right)}^{2},\end{eqnarray} \tag{ 6 }$$

$$\begin{eqnarray}&&{M}_{I}=-0.471-1.132\,{\rm{log}}(P)+0.205\,{\rm{log}}({\rm{Z}}),\end{eqnarray} \tag{ 7 }$$

where

$$\begin{eqnarray}&&\mathrm{log}({\rm{Z}})=[{Fe}/H]-1.765.\end{eqnarray} \tag{ 8 }$$

Because we use [Fe/H] instead of [M/H], we are not taking into account any effect of an enhancement in α-capture elements with respect to a solar-scaled mixture. This is because there are no studies that have been carried out concerning the α abundances of bulge field RRLs (note that the study by Hansen et al. 2016 concerned one RRL in the direction of the bulge that is thought to be a halo interloper and not a bona fide bulge field RRL star). Also, even assuming that the bulge RRLs are considerably enhanced with [α/Fe] = 0.6, this would change our log Z by ∼0.5, resulting in a distance change of ∼2%. A more modest enhancement of [α/Fe] = 0.2 would change our log Z by ∼0.15, resulting in a distance change of ∼0.5%. The reddening along the line of sight of the star is obtained using the following relation from Nataf et al. (2013):

$$\begin{eqnarray}&&{A}_{I}=0.7465[({m}_{V}-{m}_{I})-({M}_{V}-{M}_{I})]+1.37E(J-K),\end{eqnarray} \tag{ 9 }$$

where m_V and m_I are the OGLE mean magnitudes of the RRL in the V and I passbands, respectively, and E(J − K) is the reddening from Gonzalez et al. (2012). Lastly, the distance can be found using

$$\begin{eqnarray}&&d={10}^{0.2({m}_{I}-{M}_{I}-{A}_{I}+5)}.\end{eqnarray} \tag{ 10 }$$

Assuming [Fe/H] = −1 for all RRLs changes the individual distances by ∼1.5%. Assuming [Fe/H] = 0 for all RRLs changes the individual distances by ∼5% and the distances to smaller/closer values. Assuming [Fe/H] = −2 for all RRLs changes the individual distances by ∼8%, shifting the distances to farther/larger values. The 10% uncertainty in distance, as assumed throughout this paper, is not dominated by the uncertainties in metallicity, but rather by the uncertainties in reddening and in the theoretical M_V and M_I relations. In this paper, the RRL distances are used mainly to compare the RRLs to each other (e.g., those RRLs with distances closest to the Galactic center and those with distances farther from the Galactic center). Relative distances are not as independent of the exact zero point used in M_V .

From the RRLs studied here, we find a median distance of 8.48 kpc, with a standard deviation of 1.6 kpc. It has been shown by Pietrukowicz et al. (2015) that when projecting the individual distances using this method onto the Galactic plane (by cos b) and scaling the distribution by d⁻² to compensate for the "cone effect" (i.e., where more objects are seen at larger distances inside a solid angle), a distance to the Galactic center of R₀= 8.27 kpc ± 0.008(stat) ± 0.40(sys) kpc is obtained. They also show that using other absolute magnitude relations, e.g., Marconi et al. (2015) and Bono et al. (2007), do not affect the distance determinations significantly.

As was shown by Kunder et al. (2015), not all RRLs in the direction of the Galactic bulge are actually confined to the bulge. This is the case for other metal-poor stars located toward the Galactic bulge as well (e.g., Howes et al. 2015). The stellar density of the bulge, halo, and disk populations are all considerable and overlap in the vicinity of the Galactic center. Therefore, obtaining a "pure" sample of bulge stars based on photometry/distance and spatial position alone, which is the usual approach when selecting bulge targets, is not statistically probable. With Gaia DR2 and taking advantage of RRLs being standard candles, we are in a position to use orbital analysis to probe which RRLs are confined to the inner Galaxy and which ones are interlopers.

Figure 6 (left panel) shows the galactocentric distances of the RRLs in our sample, overlaid on particles from the R1 Milky Way simulation, ¹⁴ where a bar/bulge forms via the disk undergoing the usual buckling instability (Gardner et al. 2014). This model was designed to understand bulge formation mechanisms by tracing the kinematic and spatial characteristics of individual particles/stars (e.g., Debattista et al. 2005), and so is therefore not fully cosmological; it is a collisionless simulation with no gas, and the dark matter halo is a rigid potential. This model developed a strong peanut structure (Debattista et al. 2005) and can be used to explore the nature of the X shape of the Milky Way bulge (e.g., Gardner et al. 2014). We scale the model with respect to the local standard of rest, and for the RRLs, we assume a Sun–Galactic Center distance of 8.2 kpc (Bland-Hawthorn & Gerhard 2016). Stars designated as halo stars have apocenter distances greater than 3.5 kpc, and stars that belong to the bulge have apocenter distances smaller than 3.5 kpc (see Section 3).

Zoom In Zoom Out Reset image size

It has already been shown that a break in the mean density distribution at a distance of ∼0.5 kpc from the center is evident in the bulge RRLs (Pietrukowicz et al. 2012), and this suggests there may be a difference between the RRLs closest to the Galactic center and those farther out. We separate the RRLs into those spatially closest to the center of the Galaxy (black circles; ${R}_{{\rm{GC}}}\lt 0.9\,{\rm{kpc}})$ and those farther out (crosses; ${R}_{{\rm{GC}}}\gt 0.9\,{\rm{kpc}})$. The division between centrally concentrated RRLs and bar/bulge RRLs was made in order to obtain a sample of RRLs that is as centrally confined as possible but with still enough stars to carry out a statistical analysis of spatial and kinematic properties. Ideally, we would use a division at ${R}_{\mathrm{GC}}=0.5\,\mathrm{kpc}$, but our BRAVA-RR sample does not have a statistical sample of stars with ${R}_{\mathrm{GC}}\lt 0.5\,\mathrm{kpc}$. OGLE-IV avoids the plane of the Galaxy, so the OGLE-IV RRL sample contains essentially no stars with ${R}_{\mathrm{GC}}\lt 0.4\,\mathrm{kpc}$, and peaks at ∼1 kpc (Pietrukowicz et al. 2015).

The RRLs that are more centrally concentrated (${R}_{\mathrm{GC}}\,\lt 0.9$ kpc) have different orbital properties than those that are not (${R}_{\mathrm{GC}}\,\gt 0.9$ kpc): the more centrally concentrated RRL have smaller vertical excursions, $| Z{| }_{\max }$ distances, and have apo- and pericenter distances that indicate their orbits are tightly bound. Both populations of RRLs have very few RRLs with eccentricities smaller than 0.5, As we show below, these two populations, both of which are of the bulge in the sense that they are confined within the central ∼3.5 kpc of the Milky Way, represent distinct Galactic components that overlap in the inner Galaxy.

Our division between bar/bulge RRLs and central/classical bulge RRLs almost certainly does not cleanly separate these two components; some RRLs at greater distances from the Galactic center could still belong to the classical bulge component. A cleaner separation may be possible with the inclusion of metallicity information.

The spatial distribution of the two bulge stellar populations are shown in Figure 7, and the bar as traced from VVV red clump stars (taken from Gonzalez et al. 2012) is also shown. The top panel shows the spatial distribution of the 1389 RRLs we have space velocities for, as we can clean halo interlopers from this sample. In the bottom panel, we can increase the statistics by a factor of 20 by using the full OGLE-IV bulge RRL sample (although ∼10% of the stars in this sample are halo interlopers). The results are the same—the bar/bulge RRLs trace an inclined prominent bar, similar to the distribution of red clump giants in the bulge (see also, e.g., Stanek et al. 1997; Cao et al. 2013). In contrast, the central/classical bulge RRLs do not. That there are two populations of bulge RRLs, one that has a bar-shaped spatial distribution and one that does not, could resolve the disagreement for the axisymmetric RRL geometry seen using near-infrared VVV observations (Dékány et al. 2013) and the view provided by optical OGLE photometry, in which the RRLs do appear to follow the elongated spatial distribution of the bar (Pietrukowicz et al. 2015).

Zoom In Zoom Out Reset image size

In Figure 8, the rotation curve of the central/classical bulge RRLs and the bar/bulge RRLs is shown. The bar/bulge RRLs rotate in a similar fashion to the giants (Kunder et al. 2012; Ness et al. 2013; Zoccali et al. 2014; Ness et al. 2016), whereas the central/classical bulge RRLs deviate slightly from this curve. The central/classical bulge rotates more similarly to the halo stars passing through the inner Galaxy, although they have a considerably lower velocity dispersion. The velocity dispersion of the halo found in the inner Galaxy agrees with models that suggest an exponentially increasing velocity dispersion with decreasing distance from the center of the Galaxy (Battaglia et al. 2005).

Zoom In Zoom Out Reset image size

Both bulge RRL populations have similar velocity dispersions of ∼100 km s⁻¹. This is important when considering if kinematic fractionation (Debattista et al. 2017) is the explanation for the two different RRL populations. In this scenario, all stars originate from the disk but, due to having different ages (difference of ∼1.5–2 Gyr) and initial velocity distributions, are manifest differently into the box/peanut bulge. For example, kinematic fractionation could give rise to a weakly barred spatial distribution for a hotter and older initial population and a barred spatial distribution for a cooler and younger population. We do not observe that the central RRLs have a larger velocity dispersion compared to the bar/bulge RRLs, although our velocity dispersions are subject to the exact selection criteria of the two populations, and this is not perfect.

It is also not believed that the age spread of the inner Galaxy RRL population is ∼2 Gyr. This is because the horizontal branch morphology changes as a function of metallicity, and this correspondingly changes the frequency of stars that will fall on the instability strip. Age differences of a few billion years produce dramatic changes in the morphology of the horizontal branch (HB; see, e.g., Lee 1992, Figure 6), and because the metallicity distribution of the bulge RRL is strongly peaked at ∼−1.0 dex with a dispersion of ∼0.3 dex (Walker & Terndrup 1991; Kunder & Chaboyer 2008; Pietrukowicz et al. 2015), this can be reproduced with these RRLs having a small age spread of ${\rm{\Delta }}t\,\sim$ 1 Gyr or less (Lee 1992; Lee & Jang 2016). An age spread larger than this would cause a wider spread of $[\mathrm{Fe}/{\rm{H}}]$ in the bulge RRLs than what is currently observed. It may be that kinematic fractionation does occur with a smaller age range than ∼1.5–2 Gyr (V. P. Debattista 2020, private communication) which would then alleviate this tension. Further and more detailed simulations of kinematic fractionation may prove successful in reproducing the RRL results presented here.

The high velocity dispersion of the inner Galaxy RRLs first reported in data release 1 (DR1) of BRAVA-RR ;(Kunder et al. 2016) arose from halo interlopers passing through the inner Galaxy, that now, with Gaia, we are able to isolate. It is likely that similarly high velocity dispersions seen from samples of very metal-poor stars ($-3\lt [\mathrm{Fe}/{\rm{H}}]\lt -$ 1.5 dex) toward the bulge by, e.g., Arentsen et al. (2020) are also caused by a large fraction of halo interlopers that were not able to be cleaned from the sample due to uncertainties in finding distances to these stars. The Gaia DR2 distances from parallaxes (e.g., Bailer-Jones et al. 2018) have typical uncertainties that are ∼40% or larger for stars at distances of the bulge. Further, bona fide bulge stars have been shown to have parallax distances peaking at a distance of ∼5 kpc instead of at ∼8 kpc (Kunder et al. 2019; Arentsen et al. 2020).

One method commonly used to categorize bulges into classical and pseudobulges is to use the ${V}_{\max }/\sigma$ parameter (Kormendy & Kennicutt 2004). This can be plotted against , the apparent flattening of the bulge, to be used to classify the nature of bulge systems (see Figure 9). Pseudobulges, i.e., bulges formed via secular processes, have in general ${V}_{\max }/\sigma \gt 0.6$, whereas those that are formed via mergers, like classical bulges or ellipticals, have ${V}_{\max }/\sigma \lt 0.4$ (Kormendy & Kennicutt 2004).

Zoom In Zoom Out Reset image size

The BRAVA survey showed that the bar/bulge that dominates the mass of the inner Galaxy has a ${V}_{\max }/\sigma =0.64\pm 0.50$ (Howard et al. 2008). Given the agreement of both the BRAVA radial velocities and radial velocity dispersions with the ARGOS, GIBS, and APOGEE surveys, the uncertainty in this ${V}_{\max }/\sigma$ value is likely smaller than this. Adopting the Weiland et al. (1994) minor-to-major axis ratio of ∼0.6 for , our Milky Way pseudobulge is similar to that of NGC 4565, a well-known edge-on spiral with a peanut-shaped bulge. This value is below the most rapidly rotating bars and pseudobulges, due to its minor-to-major axis ratio (see Figure 9). We do not attempt to find a new ${V}_{\max }/\sigma$ parameter using the bar/bulge RRLs presented here for the following reasons: (1) our BRAVA-RR sample is more than an order of magnitude smaller than the BRAVA sample so we do not have the statistics of other radial velocity surveys, (2) our selection between bar/bulge RRLs and central/classical bulge RRLs is almost certainly not exact, so our bar/bulge RRL sample is likely contaminated, and (3) the ARGOS, GIBS, and APOGEE surveys find almost identical radial velocities and radial velocity dispersions, so the ${V}_{\max }/\sigma$ from all large radial velocity surveys agree with that found from BRAVA.

To probe the properties of the central/classical bulge using the ${V}_{\max }/\sigma$ parameter, we adopt σ = 100–120 km s⁻¹ for the velocity dispersion and 20–40 km s⁻¹ for the rotation speed of the central bulge. Therefore, the central bulge component has ${V}_{\max }/\sigma =0.2-0.4\pm 0.1$. The minor-to-major axis ratio of the central bulge component has not been measured, but as it is thought to be more spherical than the bar/bulge, is likely less than ∼0.4. We have checked the Auriga simulations (e.g., Fragkoudi et al. 2020; Blázquez-Calero et al. 2020) for an indication of the minor-to-major axis ratio for a classical/accreted population of stars in the boxy/peanut bulge. These simulations do indicate that an accreted population of stars should also exist in boxy/peanut bulges, and that the fraction of accreted stars in the boxy/peanut region can be between 13% and 80% for stars with metallicities of $[\mathrm{Fe}/{\rm{H}}]\lt -$ 1 dex (similar to the RRLs), and between 0.2% and 10% for stars with all metallicities. Unfortunately, a specific value of the minor-to-major axis ratio for the accreted part of the bulge is not evident from these simulations, so the uncertainty on is substantial. Therefore, the ${V}_{\max }/\sigma$ of the central/classical bulge is in good agreement with ellipticals and a prolate stellar distribution. A quantitative measurement of for the spheroidal part of the bulge would be useful.

It has been shown that there are two sequences of RRLs manifest on the period–amplitude diagram (Pietrukowicz et al. 2015). A Kolmogorov–Smirnov (K–S) goodness-of-fit test was carried out to establish whether one can reject the null hypothesis that the periods of the inner (${R}_{\mathrm{GC}}\,\lt$ 0.9 kpc) and outer (3.5 kpc $\lt \,{R}_{\mathrm{GC}}\,\gt$ 1.5 kpc) RRLs come from the same distribution. The K−S test on the full OGLE-IV sample of RRL returns a probability of $P\,=$ 0.0004, below the default threshold ${P\,}_{\mathrm{th}}=$ 0.05, below which one rejects the null hypothesis. Changing the distance of the outer RRLs to ${R}_{{\rm{GC}}}\,\gt$ 0.9 kpc (instead of 1.5 kpc) results in a K−S test probability of $P\,=$ 0.02. Therefore, the pulsation properties of the central/classical RRLs appear to be different from the bar/bulge RRLs. In particular, a visual inspection shows that the bar/bulge RRL population exhibits more of a continuum of periods, whereas the central/classical RRLs exhibit a stronger double sequence in the period–amplitude diagram.

The double sequence in the period–amplitude diagram was recently interpreted to indicate that the bulge RRLs are an accreted population that formed from primordial building blocks in a hierarchical merging paradigm (Pietrukowicz et al. 2015; Lee & Jang 2016). This is because such a double sequence was linked to the same process of multiple population phenomena seen in globular clusters (Romano et al. 2010; Tenorio-Tagle et al. 2015). So, one period–amplitude sequence would be formed from the pristine gas in these building blocks that was globally enriched by supernovae. The second sequence would be formed as the stars were enriched by the helium-rich winds of the first generation of massive stars (Romano et al. 2010; Tenorio-Tagle et al. 2015). One test of this controversial interpretation would be to find globular cluster origin signatures, such as Na–O anticorrelations, in the bulge RRLs accreted in the early merger.

There are few RRLs in bulge globular clusters (GCs), as bulge GCs are more metal rich than GCs in the halo or disk, and RRLs preferentially form in metal-poor systems. The bulge GCs with the most RRLs are NGC 6441 and NGC 6388; these GCs are unusual in the sense that they not only have a large number of RRLs for their high metallicity (each of these GCs has ∼30 RRL; Clement et al. 2001), but because the periods of these RRLs are longer than seen elsewhere in the Galaxy (Pritzl et al. 2001, 2003). Very few field bulge RRLs have periods and amplitudes similar to those seen in NGC 6441 and NGC 6388 (e.g., Pietrukowicz et al. 2015; Kunder et al. 2018a), and it is unclear how NGC 6441 and NGC 6388 fit into the formation scenario of the Milky Way (e.g., Bellini et al. 2013).

There are ∼35 RRLs in the OGLE-IV data set in the bulge GCs NGC 6401, NGC 6553, NGC 6642, NGC 6304, NGC 6522, NGC 6453, and NGC 6569. With such small numbers, it is not easy to come to conclusions on the similarity of the bulge GC RRLs to the field bulge RRLs based on periods and amplitudes alone.

Distinct period–amplitude sequences, such as those seen in the bulge field RRL and the GC RRLs, have not been reported within the field RRLs in the Milky Way (e.g., Fabrizio et al. 2019). This is likely because the metallicity distribution of RRLs in the Milky Way field spans a larger range in $[\mathrm{Fe}/{\rm{H}}]$ than that of both the Milky Way bulge RRL population and the Milky Way GC RRL population (e.g., Fabrizio et al. 2019).

High-amplitude short-period (HASP) RRLs are signatures of massive systems with rapid metal enrichment (e.g., Fiorentino et al. 2015). The fraction of HASP RRLs, f_HASP, for the OGLE-IV RRLs with ${R}_{\mathrm{GC}}\,\lt$ 0.9 kpc and for the RRLs with 3.5 kpc $\lt {R}_{\mathrm{GC}}\gt$ 0.9 kpc is f_HASP = 15%. This is larger than the f_HASP = 11% found for the halo RRLs at a distance of 5 kpc from the Galactic center (Belokurov et al. 2018). Limiting ourselves to only those RRLs in the BRAVA-RR sample, which has the advantage that halo RRLs are cleaned from the sample, the fraction of HASP in the central/classical RRL is f_HASP = 19% compared to f_HASP = 15% for the bar/bulge RRL. These high fractions suggests that both the inner and the outer bulge RRLs had early chemical enrichment histories and were formed in a dissimilar manner to the surviving Galactic dwarf spheroidals (which have ${f}_{\mathrm{HASP}}\,\sim$ 6% or smaller). Both the bar/bulge and any potential centrally located classical bulge must have been made primarily from progenitor galaxies larger than those that survived to become today's dwarf spheroidals.