THE UNUSUAL X-RAY BINARIES OF THE GLOBULAR CLUSTER NGC 6652

The 462 events extracted from source A's readout streak were grouped into nine bins, with 50 counts bin⁻¹ to maximize spectral resolution while minimizing the error in the normalized counts in each bin (Figure 2). We ignore events above 8 keV, as the ACIS response is poorly understood at these energies. The spectrum is well fit by an absorbed power-law model with a photon index of Γ = 1.7^+0.3_{− 0.2} (Table 2). The N_H value of 2.7^+0.1_{− 0.1} × 10²¹ cm⁻² exceeds the accepted cluster value of 5× 10²⁰ cm⁻² (Harris 1996). We derive an unabsorbed luminosity for source A of L_X(0.5–6.0) = 4.4^+0.6_{− 0.5} × 10³⁵ erg s⁻¹. Mukai & Smale (2000) observed XB 1832−330 with ASCA, finding that a partial covering model was required, instead of a single absorption column, to fit the spectrum. A similar model was required by BeppoSAX (Parmar et al. 2001) and XMM (Sidoli et al. 2008) observations. Our spectrum is of insufficient quality to distinguish between a partial covering model versus a single absorption column.

Zoom In Zoom Out Reset image size

We detect 9205 counts for source A in the 2000 HRC observation of NGC 6652. We convert the average source count rate into an unabsorbed bolometric X-ray luminosity with PIMMS using our best ACIS spectral fit, finding L_X(0.5–6.0) = (1.61 ± 0.02) × 10³⁶ erg s⁻¹. A's luminosity appears to have decreased by almost a factor of four from 2000 to 2008. Sidoli et al. (2008) report two XMM observations in 2006 September and October. Converting their X-ray luminosities to the 0.5–6 keV range, they find L_X(0.5–6 keV) = 1.2 × 10³⁶ and 1.1 × 10³⁶ erg s⁻¹. Parmar et al. (2001) report a BeppoSAX measurement in 2001 March, which we convert to L_X(0.5–6 keV) = 1.5 × 10³⁶ erg s⁻¹. Tarana et al. (2007) report 2003–2005 INTEGRAL measurements (restricted to >20 keV, and thus not directly comparable to our fluxes) that are consistent with the BeppoSAX flux. RXTE proportional counter array (PCA) Galactic Bulge Scan monitoring of X1832−330¹⁰ indicates substantial variability on ∼6 month timescales, but also a general declining trend from 1999 to 2010, by about 30%.

3.2.1. Timing Analysis

We produced a light curve by binning the 0.3–7 keV data into 103 bins each with 50 s bin⁻¹ (Figure 3), which shows clear variability, by factors >10 on timescales <100 s. (A Kolmogorov–Smirnov (K-S) test indicates variability at >99.9% confidence.) Such variability is unusual in low-mass X-ray binaries. To test whether the variation might be caused by changes in the obscuring column (perhaps, if the system is edge-on, caused by material at the rim of an accretion disk), we group the light curve by count rate, with boundaries when the count rate crosses 0.15 and 0.3 counts s⁻¹ (as the count statistics are low for 50 s bins). Within each larger bin, we compute the hardness ratio of 0.3–1 keV photons (the most likely to be absorbed) over 0.3–7 keV photons (Figure 3, bottom), with binomial 1σ error bars derived from Gehrels (1986). If the dips were due to obscuration, we might expect the lowest-count-rate bins to have the lowest hardness ratios. This is suggested by the first 1000 s, but is less clear from the data set as a whole. We quantify this by comparing the ratio of 0.3–1 keV counts to total counts in the low-count-rate portions (<0.15 counts s⁻¹), versus medium- and high-count-rate portions (below and above 0.3 counts s⁻¹), finding the fraction to be 0.15 ± 0.03 for the low-count-rate portions versus 0.25 ± 0.04 and 0.23 ± 0.03 for the higher-count-rate portions. This is a significant effect indicating that obscuration may play a role in the dipping.

Zoom In Zoom Out Reset image size

The peaks in the light curve reach count rates that suggest substantial pileup. We estimate the amount of pileup and intrinsic luminosities using the PIMMS tool, for the best-fit absorbed power-law spectrum (see below; the choice of spectrum does not appreciably affect results). At the peak observed count rate of 0.46 counts s⁻¹, the estimated fraction of recorded events that are actually multiple is 27%, and we infer the intrinsic L_X(0.3–7) ∼ 9 × 10³⁴ erg s⁻¹, a factor of two higher than would be extrapolated without pileup. However, the pileup model in PIMMS is systematically uncertain, so the true peak L_X is uncertain, presumably by less than a factor of two. The inferred minimum L_X is <2 × 10³³ erg s⁻¹, for 0.02 counts s⁻¹ (see Figure 3).

Power spectra (produced using XRONOS¹¹) show no evidence of periodicity during the 5.6 ks ACIS observation. We re-examined the 2000 HRC observation to check for long-term variability. Using PIMMS and our best ACIS spectral fit (below), we estimate the unabsorbed luminosity to be L_X(0.5–6.0) = (1.1 ± 0.1) × 10³⁴ erg s⁻¹ versus our average ACIS unabsorbed luminosity of L_X(0.5–6.0) = 1.1^+0.1_{− 0.1} × 10³⁴ erg s⁻¹ (Section 3.2.2). Heinke et al. (2001) found significant variability during the HRC observation. The short, low-sensitivity light curve suggests variation by at least a factor of two. As the HRC sensitivity and exposure length are each three times smaller than those for the ACIS observation, its usefulness for studying variability is limited.

An archival ROSAT HRI observation (1994 March 27, for 817 s exposure time) shows marginal evidence for source B. The image (Figure 4) shows one bright source, which we attribute to A, setting our astrometry. We measure 6 photons within 2'' of source B's position, while only 1.4^+2.5_{− 1} photons are expected at this position, using the average background rate between 65 and 10'' from source A, giving 4.6⁺¹_{− 2.5} (at 1σ; 4.6^+1.3_{− 4.6} at 2σ; Gehrels 1986) photons from B. This ∼2σ detection suggests (correcting for the HRI point-spread function enclosed energy of 38% within 2'') L_X(0.5–6 keV) = 1.5^+0.3_{− 0.8} × 10³⁴ erg s⁻¹. Thus, we have no evidence for B's variability on long timescales.

Zoom In Zoom Out Reset image size

3.2.2. Spectral Analysis

We created both an overall time-averaged spectrum of B (680 counts, binned by 30 counts bin⁻¹) and spectra for the three count rate ranges listed above. We include the Chandra CCD pileup model (Davis 2001), with the grade morphing parameter α fixed to 0.5, and photoelectric absorption (XSPEC phabs). B's time-averaged spectrum can be fit well with an absorbed power-law model with photon index Γ = 1.24^+0.10_{− 0.23} (Table 2). We also try adding an NS hydrogen atmosphere model (NSATMOS; Heinke et al. 2006a) to the absorbed power-law model. We fix the mass and radius to canonical values (1.4 M_☉, 10 km), so kT is the only free parameter. The NSATMOS component does not improve the quality of the spectral fit, and the upper limit on its kT implies an upper limit on L_{X, NS} (0.5–10.0 keV) <7  × 10³² erg s⁻¹. This limit may not be applicable if the system is edge-on, as the NS could suffer higher N_H than other X-ray emitting regions.

We next fitted the three count-rate-selected spectra, with 240, 186, and 253 counts (low to high), binned by 15 counts each, to investigate what parameters may be changing. An absorbed, piled-up power-law spectrum is best fit with (at least) two parameters varying; the power-law normalization, and either the N_H or power-law spectral index. We give parameters for both such fits in Table 3. The alternative of allowing only the power-law normalization to vary gives significant residuals in the spectra (Figure 5), and a higher chi-squared: 42.93 for 38 degrees of freedom (dof) versus 30.3 or 30.6 (for 36 dof) from the other two models (respectively). An F-test gives probabilities of 0.2% of attaining such an improvement in the chi-squared by chance, indicating that another quantity besides the normalization is also varying.

Zoom In Zoom Out Reset image size

Table 3. Count-rate-resolved Spectral Fits to B

Source	Model	NH	Γ	χ2υ/dof	LX
		(1020cm−2)			(erg s−1)
Allowing PL norm and index to vary
B (High)	PHABS(POW)	(5)	1.5+0.2− 0.2	0.84/36	5.1+1.1− 1.0 × 1034
B (Med)	...	...	1.3+0.3− 0.3	...	2.7+0.7− 0.7 × 1034
B (Low)	...	...	1.0+0.2− 0.2	...	1.2+0.2− 0.2 × 1034
Allowing PL norm and NH to vary
B (High)	PHABS(POW)	5.0+3.3− 0	1.4+0.7− 0.4	0.83/37	5.5+1.1− 1.0 × 1034
B (Med)	...	5.9+7.6− 0.9	...	...	2.4+0.5− 0.5 × 1034
B (Low)	...	22+12− 9	...	...	1.1+0.2− 0.2 × 1034

Notes. Spectral fits to NGC 6652 B, split by count rate (see the text). Errors are 90% confidence for a single parameter. Luminosities are unabsorbed in erg s⁻¹ for 0.5–10 keV.

Download table as:  ASCIITypeset image

An unbinned 0.3–7 keV light curve gives a 0.01 K-S probability of source constancy, indicating that C is variable. This is confirmed by its (0.3–7 keV, 500 s binning) light curve, which shows two clear peaks (Figure 6). A power spectrum shows no evidence of periodicity. Using the best-fit double MEKAL spectral model (below), we find a peak L_X(0.5–10.0) = 3.4^+1.1_{− 1.1} × 10³³ erg s⁻¹, and a minimum L_X < 3.2  × 10³² erg s⁻¹, for an average L_X = 1.1 × 10³³. Using the power-law spectral model (a poor fit) to the ACIS data to infer the HRC spectrum, we find L_X = 2 × 10³² erg s⁻¹, half the ACIS estimate with this spectrum, but the spectral uncertainties render this conclusion uncertain.

Zoom In Zoom Out Reset image size

We group C's spectrum by 10 counts bin⁻¹ for χ² statistics (Figure 7), and also fit the unbinned spectrum with C-statistics (finding similar results). No single model we tried can fit this very soft spectrum, with reduced χ² values well above 2 (Table 2). The spectrum can be fit by a model containing two MEKAL (thermal plasma; Liedahl et al. 1995) components with the cluster absorption (to simplify fitting we fixed N_H), with temperatures of <0.096 and >2.3 keV. Although a double MEKAL model is often used to describe the spectra of CVs (e.g., Baskill et al. 2005), such a strong low-temperature component is rarely seen. We are aware of such components only in nova remnants (Balman 2005). Alternatively, an absorbed blackbody and MEKAL represents a simplified form of a model for polar CVs (e.g., Ramsay et al. 2004). C's spectrum is reasonably fit by a blackbody plus MEKAL model, with a blackbody temperature of 67⁺¹⁴_{− 14} eV and inferred radius of 46⁺⁵⁸_{− 23} km, and a MEKAL temperature of >3.2 keV.

Zoom In Zoom Out Reset image size

A K-S test on the unbinned light curve gives a probability of constancy of 0.46. The estimated luminosity from the HRC data using the ACIS power-law fit below is L_X(0.5–10.0) = 1.1× 10³³ erg s⁻¹. This is comparable to the luminosity from the ACIS data, L_X (0.5–10.0) = 8⁺⁵_{− 3} × 10³² erg s⁻¹, so no variability is apparent.

We extract a 69 count spectrum, binned with 10 counts bin⁻¹. Source D is relatively soft, with all six bins below 2.0 keV (Figure 8), motivating us to freeze N_H to the cluster value. An NSATMOS model is a poor fit, but an absorbed power law (photon index of 2.3^+0.6_{− 0.6}) is an adequate fit to the binned spectrum; a single-temperature MEKAL is a somewhat worse fit (Table 2; null hypothesis probabilities of 25% and 11%, respectively). The power-law fit, however, shows residuals suggesting strong line emission around 1 keV (Figure 8). Adding a Gaussian line of zero width to the power-law fit gives a line energy of 1.03^+0.12_{− 0.04} keV, but does not improve the fit by a statistically significant amount (an F-test indicates a probability of 43% that such an improvement could happen by chance). Fitting the unbinned data (using the C-statistic) with a power law gives a slightly larger photon index (2.7^+0.4_{− 0.4}), and the fraction of simulated spectra with lower C-statistic values is 97%, suggesting a relatively poor fit. (The single-temp MEKAL fit to the unbinned spectrum gives 96% of simulations with lower C-statistic values, as well.)

Zoom In Zoom Out Reset image size

We detect only 15, 7, and 15 events for source E, F, and G, respectively, too few for detailed spectral and timing analyses. We extract background-subtracted count rates in energy bands of 0.5–6.0 keV, 0.5–1.5 keV, and 1.5–6 keV, using the WAVDETECT regions and PIMMS to estimate unabsorbed source luminosities. While sources E and F have roughly equal numbers of soft and hard counts, all but one of source G's counts lie above 1.5 keV, suggesting strong absorption.

Using PIMMs, we estimate the unabsorbed luminosities of the faint sources in both the 0.5–1.5 keV and 1.5–6.0 keV energy bands, using an absorbed power law of photon index 1.4. We determine the total unabsorbed luminosity from 0.5 to 6.0 keV for each source by adding the soft and hard luminosities (Table 1). We extract the counts at the ACIS WAVDETECT coordinates for each source from the HRC data set, finding 4, 1, and 0 counts from E, F, and G, respectively. These upper limits are consistent with their measured ACIS luminosities.

We create an X-ray color–magnitude diagram (Figure 9) containing the sources observed in the 2008 ACIS data, by plotting the unabsorbed 0.5–6.0 keV luminosities versus the X-ray color, defined as 2.5 log [(0.5–1.5 kev counts)/(1.5–6.0 keV counts)] (Grindlay et al. 2001). We plot an X-ray color–luminosity relation for the NSATMOS model, and X-ray color predictions for power-law and MEKAL spectral models. The spectra of qLMXBs are frequently modeled with a neutron star atmosphere (e.g., NSATMOS; Heinke et al. 2006b) component, with a harder power-law component with a photon index of 1–2 (Rutledge et al. 2002b), making up anywhere from <10% to >90% of the 0.5–10 keV flux (Jonker et al. 2004a). We plot a representative NSATMOS plus power-law model in which the power-law component contributes 50% of the 0.5–6.0 keV luminosity. Note that B's position is shifted to the left by pileup (considering its rapid variability, it is hard to correct for this).

Zoom In Zoom Out Reset image size

No sources lie near the NS atmosphere cooling track, but sources C and D have colors and luminosities in the range of qLMXBs in other clusters (e.g., Pooley & Hut 2006; Heinke et al. 2006b). C's rapid variability and unusual spectrum likely rule out a qLMXB nature for it, but D could be a qLMXB with a dominant power-law spectral component.

Surveys of X-ray sources in globular clusters indicate that most nonmagnetic CVs display hard spectra, typically consistent with MEKAL temperatures >6 keV or colors <0.5 (Grindlay et al. 2001; Pooley et al. 2002). Sources E, F, and G are consistent with the X-ray luminosities and spectra of CVs observed in many other globular clusters (Pooley & Hut 2006; Heinke et al. 2005), though quiescent LMXBs, active binaries, and millisecond pulsars cannot be ruled out. G's extremely hard spectrum, though based on few counts, indicates a high intrinsic absorption (>10²² cm⁻²), and thus an edge-on CV (e.g., W8, W15, W33, and AKO9 in 47 Tuc; Heinke et al. 2005) or a background AGN.

Source B, one of the brightest low-L_X cluster sources, can be classified as a very faint X-ray transient (VFXT). VFXTs are X-ray transients that have peak luminosities of 10³⁴–10³⁶ erg s⁻¹ and quiescent luminosities at least one order of magnitude lower (Wijnands et al. 2006; Muno et al. 2005a), generally containing an accreting neutron star or black hole. Some VFXTs have "normal" outbursts as well as very faint outbursts, but it is not clear if all do, or if VFXT behavior has a variety of causes (Degenaar & Wijnands 2010).

Figure 3 illustrates that source B reaches peak X-ray luminosities up to L_X(0.5–10.0 keV) = 9 × 10³⁴ erg s⁻¹ and minimum L_X < 2 × 10³³ erg s⁻¹ on timescales of minutes. B's high peak luminosity is strong evidence that the system must contain a neutron star or a black hole, but the variability is unusual for LMXBs. Perhaps the simplest explanation for this variability is a high inclination angle, so that we observe B's accretion disk edge-on, and our view of the central X-ray source is interrupted by structures at the accretion disk rim (White & Holt 1982; Xiang et al. 2009). Obscuration by an accretion disk should lead to changes in the N_H value, and our spectral analysis gives evidence in favor of this (Section 3.2.2). Our spectral fitting requires intrinsic L_X changes along with N_H changes, which seems to argue against this explanation. However, our spectral fitting is constrained by a lack of data to only three count rate ranges, which contain substantial variability, and we think it likely that variability also occurs on timescales shorter than we can accurately probe. These two effects could prevent us from accurately measuring how N_H changes with L_X in B's light curve, and thus we cannot yet rule out that obscuration is fully responsible for the variability.

If we only see scattered light from a central source (in this picture the variability is due to obscuration of an accretion disk corona), the true isotropic X-ray luminosity could be higher than the observed luminosity, by a factor of up to 100 (e.g., Muno et al. 2005b). Thus, B could be a "normal" LMXB, which would remove the difficulty of explaining its unusually low accretion luminosity. We have been awarded a deeper Chandra observation of NGC 6652 (50 ks, to be taken mid-2011), which should clarify the spectral variability of this source. We have also been awarded a Gemini time-series imaging observation of NGC 6652, designed to search for evidence of short periods in sources A and B suggested by previous Hubble Space Telescope imaging (Deutsch et al. 2000; Heinke et al. 2001).

The very soft spectral component of C is unlike anything seen in quiescent LMXBs, but is consistent with the spectra of some CVs. The (low-T) blackbody + (high-T) MEKAL spectral fit suggests a polar CV nature, where the blackbody describes soft X-rays emitted from the white dwarf's polar cap, and the MEKAL describes hard X-rays from the accretion column shock front (Ramsay & Cropper 2004). The inferred polar cap radius of 25⁺⁴⁰_{− 14} km (from the blackbody spectral fit) compares well with the typical radius of an accreting pole of a WD of ∼75 km (for a 0.6 M_☉WD; Ishida et al. 1997). The spectrum and luminosity are similar to the likely polar CV X10 in the globular cluster 47 Tuc (Heinke et al. 2005), which has L_X (0.5–6 keV) = 2.6 × 10³² erg s⁻¹ and was modeled with a kT = 53 eV blackbody and two MEKAL plasmas of temperatures 0.39 and >14 keV. Thus, we suggest source C may be the second polar CV identified in a globular cluster. Knowing the frequency of magnetic accretion channeling in cluster CVs (by identifying magnetic CVs) is important for understanding their unusually rare outbursts (Grindlay 1999; Dobrotka et al. 2006; Ivanova et al. 2006).

Alternatively, a very low-T thermal plasma has been seen in historical novae, from the nova shell (e.g., Balman 2005). However, the nova eruption would probably have been seen if recent, or if old, the shell should have expanded to a resolvable size (e.g., the Nova Per 1901 shell would be 25 in radius if located in NGC 6652). Our upcoming Chandra observation should provide sufficiently high-quality spectra to distinguish between the spectral models discussed, and search for spectral variations over time.

Source D's luminosity and spectrum are consistent with both CVs and quiescent LMXBs. The feature of particular interest is the hint of an emission line around 1 keV. If this feature is real (which will be tested by our upcoming Chandra observation), it would make D a rather odd X-ray source. There is another X-ray bright CV with a similarly strong low-energy line, the luminous CV X9 in 47 Tuc (Heinke et al. 2005).

The pronounced spectral hardness of G indicates it is an edge-on CV, or a background AGN. E and F cannot be clearly classified, but as the majority of globular cluster X-ray sources with their X-ray colors and L_X are CVs (Pooley & Hut 2006), we may suggest that CVs are the most likely possibility. Deeper Chandra observations should detect additional sources, enabling a comparison of the amazingly rich X-ray population of this globular cluster with other globular clusters.