THE GALACTIC BULGE SURVEY: OUTLINE AND X-RAY OBSERVATIONS

X-ray observations are excellent probes of coronally active and accreting sources. Whereas studies of X-ray sources in our Galaxy have to a large extent focussed on bright systems, investigations of fainter source classes have typically been done in the Galactic center (e.g., Muno et al. 2003). There, however, crowding and extinction make it more difficult to identify the correct optical and/or infrared counterparts for large fractions of the sources (e.g., Mauerhan et al. 2009).

It is clear from previous surveys that multi-wavelength observations are vital for classifying faint X-ray sources since X-ray spectral information alone is rarely sufficient. Classification is important for various science goals as we outline below.

The Galactic Bulge Survey (GBS) we present in this paper was designed to allow multi-wavelength observations of the detected X-ray sources. The GBS consists of Chandra  and optical imaging of two strips of 6° × 1°, one centered 15 above the Galactic plane and the other 15 below the plane. We have chosen this area as the source density is still high, but by excluding |b| < 1° we avoid the regions that are most heavily affected by extinction and source confusion (see Figure 1). In Table 1 we provide a reference table for the acronyms used in this paper.

Zoom In Zoom Out Reset image size

The discovery of a large sample of X-ray sources will allow us to identify rare X-ray binaries such as quiescent eclipsing black hole (BH) and neutron star (NS) low-mass X-ray binaries (LMXBs) and ultra-compact X-ray binaries (UCXBs; LMXBs that have orbital periods <1 hr). For dynamical mass measurements in LMXBs one needs to measure three parameters: the radial velocity amplitude of the companion star (K₂), the ratio between the mass of the companion star and the NS or BH mass (q), and the inclination (i). A measurement of the rotational broadening of the stellar absorption lines (vsin i) combined with K₂ gives this determination of q. The reason is that in Roche lobe overflow systems, like LMXBs, the companion star is tidally forced to co-rotate with the binary orbit (Tassoul 1988).

The system inclination can be determined through modeling of ellipsoidal variations caused by distortion of the companion star using multi-color optical light curves (e.g., Orosz & Hauschildt 2000, Cantrell et al. 2010). In systems with favorable viewing angles, the (X-ray) eclipse duration can be used to accurately determine the inclination (Horne 1985). Since the inclination is constrained by the geometry, mass measurements in eclipsing systems are independent of the modeling that lies behind inclinations derived from ellipsoidal variations. Quiescent eclipsing systems are prime targets for optical mass measurements. Such mass measurements provide constraints on the NS equation of state (EoS) and on the dividing line between NS and BH systems (e.g., Özel et al. 2010). Constraining the NS EoS remains one of the ultimate goals for NS studies (Lattimer & Prakash 2004 and references therein) and mass measurements of BH systems would improve our estimates of the stellar mass BH mass distribution with implications for supernova and Gamma-ray burst modeling. Furthermore, the BH sample selected in quiescence, as in the GBS, is not susceptible to potential X-ray outburst duty cycle based selection effects that occur when selecting BHs after they have become X-ray bright and returned to quiescence, which is now common practice. Finding eclipsing LMXBs with optical or near-infrared counterparts bright enough that optical or near-infrared spectroscopy can provide accurate mass measurements is the principal goal of the GBS.

The second main goal of the GBS is to study the origin and evolution of the population of X-ray binaries. By comparing the observed number of sources per source class (e.g., cataclysmic variables, CVs and LMXBs) with those that have been predicted on the basis of population synthesis models, one can constrain binary evolution models, in particular the nature of the common envelope phase. Most compact binary sources responsible for high energy phenomena went through one or two phases of common-envelope evolution, such as CVs, AM Canum Venaticorum stars (AM CVns), and UCXBs (Taam & Sandquist 2000). However that phase is not yet well understood. Compact binaries have much lower orbital energy and angular momentum than the progenitor binary that contained giants (Paczynski 1976). The binary semi-major axis is thought to shrink mainly during a phase of unstable mass transfer and ejection, i.e., the spiral-in. If the outcome of this process is derived by assuming that the change in orbital energy is enough to eject the giant's mantle, the predicted properties do not match the observations of double white dwarf binaries. These properties can be matched with the assumption that the giant's mantle is ejected, carrying a fixed fraction of the total angular momentum (Nelemans et al. 2000), but this begs the question how the required energy is provided. A more complete theoretical description is required that takes into account both energy and angular momentum.

From the properties of single radio pulsars it has been inferred that they receive a kick at birth, due to asymmetries in the supernova explosion (Lyne & Lorimer 1994). This has important consequences for the evolution and properties of X-ray binaries (e.g., Brandt & Podsiadlowski 1995; Kalogera 1996) so in principle X-ray binaries can be used to constrain the properties of the kick. In particular, the question is whether BHs also receive a kick at formation and, if so, what its properties are (e.g., White & van Paradijs 1996, Jonker & Nelemans 2004, Willems et al. 2005, Miller-Jones et al. 2009).

Four main steps must be followed in binary population synthesis modeling (Postnov & Yungelson 2006). First, a set of initial conditions must be chosen. Key parameters which must be set include the initial mass function, the initial distribution of mass ratios in binaries, the initial binary fraction, and the initial distribution of orbital periods. Second, any evolution with time of these initial conditions, such as that due to changes in star formation rate, should be taken into account. Next, a recipe for (binary) stellar evolution must be specified. The major uncertainties in this recipe include the aforementioned uncertainties in common envelope evolution, uncertainties in the NS and BH natal kick distributions, and uncertainties in mass and angular momentum loss. Finally, a recipe must be developed for converting accretion rates, typically calculated on timescales of centuries or longer, into instantaneous observables such as the X-ray luminosity. Key uncertainties in this step largely involve understanding the spectra of accreting sources at different accretion rates and understanding how disk instabilities will affect the distribution of luminosities of those sources which undergo disk accretion.

To make progress on these issues we envisage a two-pronged approach that includes detailed studies of individual systems on the one hand, and of the population on the other hand. Any viable evolutionary scheme must be able to reproduce the specific properties (such as component masses, orbital period, age, system velocity) of each observed individual system. Any viable evolution scheme must also reproduce the distributions of and correlations between these properties in the population of X-ray binaries. Observationally this requires the discovery of large homogeneous samples of CVs and (ultra-compact) X-ray binaries, and the detailed follow-up of a number of individual systems.

The large number of sources will also allow us to investigate the spatial distribution of LMXBs and this provides input to the LMXB formation scenarios. Jonker & Nelemans (2004) found that there is a significant excess of LMXBs at −10° < l < 0° with respect to 0° < l < 10°. Weidenspointner et al. (2008) found an asymmetric distribution of the 511 keV line emission using INTEGRAL data which they suggested was due to an asymmetric LMXB distribution (although see Bandyopadhyay et al. 2009). However, any spatial non-uniformity in the Galactic birth distribution of LMXBs is thought to be washed out by the natal kick imparted on the NS in the LMXB formation models presented by van den Heuvel (1983) and Kalogera (1998). On the other hand, the LMXB evolutionary model, involving triple star evolution, discussed by Eggleton & Verbunt (1986) provides a channel for the formation of LMXBs without a large kick velocity. In general, except for formation scenarios involving either direct collapse to a BH or an accretion induced collapse, a Blaauw kick should be imparted regardless of the type of compact object formed during the supernova event (Blaauw 1961). Evidence for velocity kicks is found in the z-distribution of LMXBs (van Paradijs & White 1995; Jonker & Nelemans 2004) and in the velocity distribution of radio pulsars (Lyne & Lorimer 1994; Hansen & Phinney 1997). An improved spatial distribution of LMXBs in the Bulge will help to test if such kicks happen commonly or rarely.

Previous X-ray surveys of the Galactic bulge provide some general expectations for our survey. The Advanced Satellite for Cosmology and Astrophysics (ASCA) observed the central region of our Galaxy with |l| < 45° and |b| < 04 in the 0.7–10 keV band, albeit with a resolution of 3'  which led to a large number of unclassified sources (Sugizaki et al. 2001). The ASCA survey discovered 163 sources down to a flux of ≈3 × 10⁻¹² erg cm⁻² s⁻¹  which as a group have properties in common with CVs, high-mass X-ray binaries (HMXBs), quiescent LMXBs, and Crab-like pulsars. A recent paper by Anderson et al. (2011) finds a number of massive stars within the ASCA Galactic plane sources. The XMM-Newton Galactic plane survey covered 3 deg² along the Galactic plane down to F_X (2–10 keV) 2 × 10⁻¹⁴ erg cm⁻² s⁻¹, finding roughly one-third to be soft sources suggestive of nearby coronally active stars (Hands et al. 2004), while two-thirds were hard, absorbed sources. The latter group is dominated by active galactic nuclei (AGNs) but it also has a substantial Galactic component, especially between 10⁻¹² erg cm^-2 s⁻¹ > F_X > 10⁻¹³ erg cm⁻² s⁻¹. Optical identifications of 30 sources with F_X > 7 × 10⁻¹⁴ erg cm⁻² s⁻¹  led to the classification of 16–18 coronally active stars, three CVs, and two LMXB candidates (Motch et al. 2010). Finally, spectroscopic follow-up of medium-duration Chandra observations of numerous Galactic plane fields (the ChaMPlane survey, Grindlay et al. 2005; Rogel et al. 2006) finds roughly one-third to be soft sources, largely stars, and two-thirds to be hard sources, including many AGNs and a few CVs.

In this paper, we describe the modeling that provides rough estimates for the number of sources that we expect in the GBS (Section 2). The Chandra X-ray data, the X-ray source list, and analysis of the X-ray source properties are described in Section 3. A brief outline of the multi-wavelength imaging data is given in Section 4, followed by a comparison of ROSAT sources in the GBS area with the Chandra-discovered sources in Section 5. We end with a short summary (Section 6) and outlook (Section 7).

In the tradeoff between a homogeneous survey depth and total survey exposure time we used 7'  as an effective radius of the Chandra  field of view in designing the survey. The size and shape of the point-spread function for off-axis angles larger than 7' have degraded such that many optical and/or near-infrared stars will fall inside the X-ray error circle, even for sources detected with more than 10 counts, making multi-wavelength follow-up more difficult (cf. Hong et al. 2005). As noted above, this yields the possibility of multiple detection of sources discovered in the overlap regions. Indeed, 105 sources are detected more than once, where we have taken source positions within 3'' of each other as multiple detections of the same source. Out of these 105 sources, 95 sources are detected two times, 9 sources are detected three times, and 1 source is detected four times. The properties that we list in Table 3 for these sources are those of the detection that gave rise to the largest number of X-ray counts. In Table 3 we also list the number of times that sources are detected. In addition, we found that although Chandra sources 33 and 230 are 48 apart, they are conceivably two detections of the same source. Both detections are rather far off-axis (5$.\hspace{-2.22214pt}^\prime$2 and 8$.\hspace{-2.22214pt}^\prime$8, respectively) explaining the relatively large uncertainty in both measurements of the position.

We investigate the relationship between the number of sources detected as a function of off-axis angle and the number of source counts. In this way we can determine the approximate number of source counts where the survey sensitivity for off-axis angles less than 7' is constant. We find (Figure 4) that when normalized to the surface area, the number of detected sources drops as a function of off-axis angle for angles larger than ≈2'. The probable reason for the slightly lower number of sources detected per unit surface area inside the 2'  radius is that the chip gaps between the four ACIS–I CCDs subtend a larger fraction of the total solid angle within 2'  than they do in the outer regions. That area has a lower effective exposure time as can be seen in Figure 5. For source counts larger than 10 the normalized number of detected sources is approximate constant for off-axis angles less than 7' (blue dot-dashed line in Figure 4). Thus, the GBS coverage is approximately homogeneous down to source fluxes of ≈7.7 × 10⁻¹⁴ erg cm⁻² s⁻¹, whereas the sensitivity is decreasing with off-axis angle for fainter sources.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We extract source counts using circular source extraction regions of 10''. Background extraction regions are annuli with inner and outer radii of 15'' and 30'', respectively. We plot the 89 sources for which we detected more than 20 counts in a hardness–intensity diagram (Figure 6). To mitigate the effects that small differences in exposure time across our survey can have, we use count rates as a measure of intensity. We define the hardness ratio as the ratio between the count rate in the 2.5–8 keV minus that in the 0.3–2.5 keV band to the count rate in the full 0.3–8 keV energy band (after Kim et al. 2004). We derived the hardness using XSPEC version 12.6 (Arnaud 1996) by determining the count rates in the soft and hard bands taking the response and ancillary response file for each of the sources. The three brightest sources suffer from photon pileup, where more than one photon is registered during the CCD integration time (see, e.g., Davis 2001). Photon pileup will artificially harden those three sources in a hardness–intensity diagram, therefore, we have not plotted these three sources in Figure 6 nor in Table 3. For the other 86 detected sources photon pileup is unimportant (less than a few percent). Naively, one would expect most hard sources to be more distant and more reddened than the soft sources, as the intrinsic spectral shape of the most numerous classes of sources we expect to find does not differ much.

Zoom In Zoom Out Reset image size

The most interesting aspect from Figure 6 is perhaps the paucity of bright hard sources. Conversely, most bright sources are probably nearby with low extinction. As foreseen, the spectral information is insufficient for source classification for the majority of the total number of detected sources, therefore, classification will have to come from multi-wavelength observations. Finally, there seems to be a dichotomy in the hardness with one peak centered on a hardness of 0.4 and another centered on −0.5 with a clear paucity of sources with hardness 0. A similar dichotomy was reported in Warwick et al. (2010).

A conceivable explanation for the nature of this dichotomy is that the soft sources are dominated by nearby sources, such as RS CVns. The harder sources are further away and are either intrinsically hard spectrum sources or they appear hard due to the effects of extinction. The latter increases strongly toward the Galactic center for distances between ≈3 and 5 kpc (Marshall et al. 2006). The strong increase in extinction over a small distance interval would, in this scenario, be responsible for the lack of sources near a hardness of 0.

The brightest Chandra  source, source 1 is a transient (see below). The four ROSAT Bright Source Catalog (BSC) sources, 1RXS J173728.0−290759, 1RXS J174043.1−281806, 1RXS J173933.4−291001, 1RXS J173826.7−290140, probably correspond to Chandra  source number 2, 3, 4, and 7 in Table 3, as the nominal offset between the Chandra  and ROSAT position is 576, 426, 297, and 116, respectively. The ROSAT positional (1σ) uncertainty of these four sources is 9'', 8'', 13'', and 16'', respectively. Source 2 of the Chandra  GBS survey (1RXS J173728.0−290759) is a well-known Seyfert 1 galaxy. Source 3 (1RXS J174043.1−281806) is the persistent LMXB and UCXB-candidate SLX 1737−282. Source 4 is associated with HD 316072; the angular distance between the Chandra  position and HD 316072 is 046. HD 316072 is a bright, V = 9.92, G9III star. This source is conceivably a bright active star or a long-period LMXB. Taking the observed and absolute V-band magnitude and assuming A_V = 0 yields a maximum source distance of 675 pc. Modeling the Chandra X-ray spectrum consisting of 238 photons, we find that the source spectrum is soft. It can be described by an absorbed black body spectrum with kT = 0.22 keV and N_H = 0.4 × 10²² cm^-2. This provides a good fit at an absorbed/unabsorbed 0.5–10 keV flux of 5.6 × 10⁻¹³/1.2 × 10⁻¹² erg cm⁻² s⁻¹, respectively. This converts to a 0.5–10 keV luminosity upper limit of 6.5 ×  10³¹ erg s⁻¹. Such a soft spectrum would be consistent both with a quiescent NS LMXB as well as a white dwarf accretor in a CV. Optical time series spectroscopy should allow us to distinguish between an active (binary) star or an X-ray binary scenario. Source 7 is 034 away from a bright optical source identified as a pre-main-sequence star by Torres et al. (2006).

Similarly, sources 5, 6, 8–10 are bright enough that they should have been detected by the RASS BSC. Source 5 is 278 away from AX J1740.1−2847, which is identified as a low-luminosity high-mass X-ray binary pulsar (Kaur et al. 2010). Source 6 is a known Be X-ray binary which is listed in the RASS faint source catalog as 1RXS J174444.7−271326, indicating some moderate X-ray variability. Source 8 is nominally 298 away from AX J1735.1−2930, which is currently unclassified. Source 9 is 033 away from HD315997, which is an A5 star in a known eclipsing binary in a 2.8723 d orbital period (Nesterov et al. 1995; Otero et al. 2006). The V-band magnitude varies between 11.21 and 11.40 with a secondary eclipse minimum of 11.24 (Otero et al. 2006). The small size of a NS and BH implies that an accretion disk should be present if one wants to explain the amplitude and duration of the secondary eclipse in an X-ray binary context. Taking the observed and absolute V-band magnitude and assuming A_V = 0 yields a maximum source distance of 695 pc for the A5 star. This implies a maximum source 0.5–10 keV luminosity of ≈6 × 10³¹ erg s⁻¹. It is also possible that the A5 star is in orbit with a late type star that is not seen in the optical spectrum but which is responsible for the X-ray emission. Again, orbital phase resolved spectroscopic observations are necessary to reveal the nature of this object.

Source 10 is 047 away from HD 315992 (F8, V = 9.98) and nominally 2638 away from 1RXS J173628.8−291055 which is listed in the RASS faint source catalog. In the ASCA Galactic center survey the source is found to have a 0.7–10 keV flux of 1 × 10⁻¹² erg cm⁻² s⁻¹ (Sakano et al. 2002). As for several of the other sources discussed above, multiple optical spectra of this source will allow us to investigate if they are in a binary and, if so, what the nature of the second star is.

To summarize, the Chandra  point sources 2–10 correspond to: the Seyfert 1 galaxy 1RXS J173728.0−290759, SLX 1737−282, HD 316072, AX J1740.1−2847, 1RXS J174444.7−271326, 1RXS J173826.7−290140, AX J1735.1− 2930, HD 315997, and HD 315992.

5.1.1. Chandra  Light Curves

We inspect the Chandra  light curves of sources 1–10. Sources 1, 3, and 5 show evidence for flare-like variability. Fitting the light curve with a constant gives a χ² value of 359, 57, and 43 for 20, 17, and 20 degrees of freedom, respectively.

There is marginal evidence in the light curve of source 4 for the presence of an eclipse-like feature (see Figure 7), although, fitting the light curve with a constant gives a χ² value of 27.5 for 20 degrees of freedom.

Zoom In Zoom Out Reset image size

The light curves of sources 2, 6, 7, 8, 9, and 10 are consistent with being constant with χ² values of 24.3, 27.2, 22.7, 27.4, 12.4, and 20.5 for 20 degrees of freedom, respectively (except source 9 for which there are 19 degrees of freedom).

The only other two BSC RASS sources in the GBS area, 1RXS J175113.2−293842 and 1RXS J174220.8−273736, are not detected in our GBS survey. Their nominal ROSAT positional errors are 26'' and 12'', respectively. This means that, especially for 1RXS J175113.2−293842, it is not inconceivable that the source has a Chandra  counterpart further away than 30''. However, even in a search radius of 1'  there is no Chandra  source possibly related to 1RXS J175113.2−293842. Whereas the band passes of ROSAT (0.1–2.5 keV) and Chandra (0.3–8 keV) are different, Chandra is much more sensitive and both sources should have been exceptionally soft for it not to be detected in Chandra; this is not the case. The ROSAT hardness ratio 1 (HR1) for 1RXS J175113.2−293842 is 1.0 ± 0.3 and hardness ratio 2 (HR2) is 0.54 ± 0.27 (where HR1 = (B−A)/(B+A) and HR2 = (D−C)/(D+C), with A = 0.11–0.41 keV, B = 0.52–2.0 keV, C = 0.5–0.9 keV, and D = 0.9–2.0 keV count rate). Assuming a power-law spectrum with index 2, an interstellar extinction of 1 × 10²² cm⁻² (consistent with the observed hardness ratios HR1 and HR2) and using the RASS-measured 0.1–2.4 keV count rate of 0.07 counts s⁻¹, we calculate the absorbed 0.1–2.4 keV flux to be approximately 1 × 10⁻¹² erg cm⁻² s⁻¹  at the time of the ROSAT observation. Assuming Chandra  detected zero counts and following Gehrels (1986), we take an upper limit of 3 counts for a 95% upper limit over the 2 ks exposure. This makes the approximate Chandra  0.1–2.4 keV upper limit for such a source spectrum 5 × 10⁻¹⁵ erg cm⁻² s⁻¹, implying a decay in flux of at least a factor 200.

For 1RXS J174220.8−273736 the detected RASS count rate is 0.16 counts s⁻¹, the HR1 is 1.00 ± 0.03 and the HR2 is 1.00 ± 0.08. For the same assumed source spectrum as above the absorbed 0.1–2.4 keV ROSAT flux was 3 × 10⁻¹² erg cm⁻² s⁻¹, implying that the source decayed in flux more than a factor of 500.

Conversely, the brightest source in our survey, SAX J1750.8−2900 (Natalucci et al. 1999), was not detected in the RASS BSC. This source is a recurrent transient LMXB that happened to be in outburst when our Chandra  observation was obtained but that was apparently in quiescence during the RASS observations.

We list ROSAT sources discovered in pointed observations when they fall inside the covered GBS area. Sources found using both the position sensitive proportional counter (PSPC) detector as well as those found using the high-resolution imager (HRI) observations are included (see Table 4). We exclude pointed observations of sources discussed above (Chandra  source numbers 1–10).

Sometimes multiple ROSAT observations of the same source region provide slightly different source coordinates, and thus different ROSAT names, whereas using Chandra  we find one source. Although it is possible that there was indeed more than one source and these have faded below the Chandra  detection level, it is more likely that these multiple ROSAT detections are in fact of one and the same source. As an example, the six ROSAT sources 2RXP J173826.2−290147, 2RXP J173827.4−290138, 2RXP J173827.5−290152, 2RXP J173826.7−290205, 2RXP J173827.2−290144, 2RXP J173824.0−290146 are probably all related to Chandra  source 7 (at α = 17:38:26.18 and δ = −29:01:49.4). The nominal angular distance between the ROSAT positions and the single Chandra  position is 245, 184, 168, 167, 149, and 283, respectively. The last ROSAT position was derived from an observation that had the source far off-axis (50$.\hspace{-2.22214pt}^\prime$8). Similarly, the smallest positional offset was obtained for the observation where the source was only 14$.\hspace{-2.22214pt}^\prime$8 off-axis.