TRANSIENT EXTREMELY SOFT X-RAY EMISSION FROM THE UNUSUALLY BRIGHT CATACLYSMIC VARIABLE IN THE GLOBULAR CLUSTER M3: A NEW CV X-RAY LUMINOSITY RECORD?

Supersoft X-ray sources (SSSs) were first detected by the Einstein Observatory and were later established as a prominent X-ray binary class on the basis of ROSAT observations (Greiner et al. 1991). SSSs are distinguished from other X-ray sources by their soft spectra (kT ∼ 15–80 eV) and high bolometric luminosities, generally ranging from 10³⁶ erg s⁻¹ to 10³⁸ erg s⁻¹. These attributes are most suitably explained by a binary system with a white dwarf (WD) primary in which hydrogen-rich matter is burned upon the surface of the WD, accreting at a rate of ∼10⁻⁷ M_☉ yr⁻¹ (Kahabka & van den Heuvel 1997).

Systems portraying characteristics of SSSs have been considered as possible progenitors of Type Ia supernovae (Rappaport et al. 1994). Type Ia supernovae are triggered by one of two pathways: the steady accumulation of matter onto an accreting WD, until it passes the Chandrasekhar limit (Whelan & Iben 1973), or the merger of two WDs (Webbink 1984; Iben & Tutukov 1984). During the evolution of single-degenerate binaries, the system undergoes a nuclear burning phase, appearing as an SSS (Rappaport et al. 1994). The second pathway has recently been shown to also give rise to SSS characteristics, as pre-double-degenerate systems undergo a period in which the initial WD accretes and burns matter from the companion's red giant wind (Di Stefano 2010b). However, for either case, observations show a lack of soft X-ray sources (Gilfanov & Bogdán 2010; Di Stefano 2010a). The majority of observed SSS systems are now understood to be novae in which previously accreted matter is burned on the WD surface (Pietsch et al. 2005), but WDs that burn H in nova eruptions are thought to lose, not gain, mass (Nomoto 1982). It is also clear that the majority of known Galactic non-nova SSSs do not contain red giants as donors, but rather main-sequence stars—with inferred donor masses smaller than expected (Cowley et al. 1998). Recently, massive (Be?) companions have been inferred for some SSSs in M31 (Orio et al. 2010). It remains possible that a large number of SSSs emit predominantly in the extreme UV, below Chandra detection limits (Southwell et al. 1996; Hachisu et al. 1996; Hachisu & Kato 2003). This possibility makes observations of apparently transient SSSs critical to the understanding of Type Ia progenitors.

Currently, five or six SSSs (besides novae) are known within our galaxy; however, only one has been seen within a globular cluster (Greiner 2000). Located within the M3 cluster, the transient SSS 1E1339.8+2837 (hereafter 1E1339) was first observed with the Einstein Observatory, possessing a 0.5–4.5 keV luminosity of 4 × 10³³ erg s ⁻¹ (Hertz & Grindlay 1983). The ROSAT all-sky survey later detected a very soft X-ray source within the M3 globular cluster, with a maximum bolometric luminosity of 2.0 × 10³⁵ erg s ⁻¹ (Verbunt et al. 1995). In 1992 January, the ROSAT High Resolution Imager (HRI) observed 1E1339 to have a soft spectrum (kT ∼ 20 eV) and a 0.1–2.4 keV luminosity of 2 × 10³⁵ erg s ⁻¹; however, within six months the source had faded below HRI detection limits (Hertz et al. 1993). During its peak, 1E1339 was roughly one to two orders of magnitude less luminous than an average SSS (see Kahabka & van den Heuvel 1997).

After the 1992 outburst, 1E1339 fell into a period of relative quiescence, adopting a hard X-ray spectrum more typical of cataclysmic variables (CVs; e.g., Baskill et al. 2005). A follow-up ASCA observation taken in 1997 confirmed the luminosity decrease, and found the spectrum of 1E1339 to be best approximated by a 5 keV bremsstrahlung model, with a 0.5–10 keV luminosity of 1.6  ×  10³³ erg s ⁻¹ (Dotani et al. 1999). The optical counterpart for 1E1339 was identified by Edmonds et al. (2004) near the subgiant branch in V–I color–magnitude diagrams, but to be extremely bright in the U band, with strong variability. Its X-ray to optical flux ratio (F_X/F_opt ∼ 1) is far too low for X-ray binaries containing neutron stars, high for normal CVs, but consistent with magnetic CVs (Verbunt et al. 1997). The extremely blue colors of 1E1339's optical counterpart (Edmonds et al. 2004) allow us to infer a high rate of accretion, which combined with the relatively low-luminosity X-ray emission rules out a neutron star or black hole accretor; we regard a WD accretor as certain. The small inferred radius of the SSS episode, and the high F_X/F_opt ratio, suggested a magnetic CV nature (Edmonds et al. 2004). In this paper, we report the results of three Chandra ACIS-S observations of 1E1339.

We observed the X-ray source 1E1339 on three separate occasions with the Chandra ACIS-S, using three active CCDs (S2, S3, and S4) with a 1/2 subarray (to reduce pileup in case 1E1339 were to become extremely bright) using the ACIS very faint mode. ObsID 4542 was taken on 2003 November 11 for 10.4 ks, ObsID 4543 on 2004 May 9 for 10.3 ks, and ObsID 4544 on 2005 January 10 for 9.8 ks. In addition, we obtained Hubble Space Telescope (HST) observations of 1E1339 in several bands, from F814W (I) to F220W (near-UV), coincident with the 2004 Chandra observation. An analysis of the HST observations will be published elsewhere.

The data were reduced using CIAO version 4.2,⁵ as well as FTOOLS.⁶ A bad pixel file was created with the CIAO acis_run_hotpix tool. The observations were then reprocessed (using acis_process_events) and further filtered⁷ to create new level 2 event files. In order to create a source map, the images upon the S3 chip were adjusted, then merged together. The merged image was then divided by the merged set of exposure maps, taken from each observation (see Figure 1). The position of 1E1339 was determined to be R.A. = 13^h42^m098, decl. = +28°22'47'' (J2000), in agreement with previous HST and ROSAT HRI observations (see Edmonds et al. 2004). We detected multiple other X-ray sources, which will be reported separately.

Zoom In Zoom Out Reset image size

From each individual observation, we selected a circle with a radius of 2'' around 1E1339 and chose an annulus of radius 10'' that excluded the source for the background. Using the task psextract, we generated RMF and ARF files for the source and background spectra. For each observation, an improved RMF file was created using the mkacisrmf task, over an energy range from 0.3 to 10 keV.

2.1. X-ray Variability

The three observations were found to have notable differences within their average count rates, ranging from a maximum of 6.6 × 10⁻² count s⁻¹ seen in the second observation (ObsID 4543) to a minimum of 3.0 × 10⁻² count s⁻¹ seen in the first observation (ObsID 4542, see Table 1). We further extracted light curves from each of the three observations of 1E1339.8+2837 with use of the CIAO dmextract tool, using a binning of 500 s (see Figure 2). We then tested the variability for each individual observation, making use of the dither_region tool and the glvary tool, which applies the Gregory–Loredo algorithm. The algorithm divides the observation in various time bins and looks for significant deviations, reporting a variability index (Gregory & Loredo 1992). We found the first and second observations to be variable (variability index of 7 and 6, corresponding to a variability probability of 0.998 and 0.989, respectively), while the third observation (ObsID 4544) showed no variability (variability index of 0, corresponding to a variability probability of 0.050).

Zoom In Zoom Out Reset image size

Table 1. Count Rates for Observations of 1E1339.8+2837

Date	ObsID	Exp. Time	Count Rate
			0.5–10 keV	<0.3 keV
2003 Nov 11	4542	10.4	2.7+0.2 − 0.2 × 10−2	8.6+5.0 − 3.4 × 10−4
2004 May 9	4543	10.3	6.1+0.3 − 0.3 × 10−2	1.7+0.7 − 0.5 × 10−3
2005 Jan 10	4544	9.8	2.3+0.2 − 0.2 × 10−2	1.0+0.1 − 0.1 × 10−2

Notes. A notable low-energy emission is seen in the 2005 observation. Errors are 90% confidence. Exposure times in ks. All count rates in counts s⁻¹.

Download table as:  ASCIITypeset image

We extracted X-ray spectra from each of the three observations of 1E1339 and used XSPEC version 12.5 to model them. The spectra from the first and third observation (ObsID 4542 and 4544, respectively) were binned using the GRPPHA tool so that each bin contained a minimum of 15 counts. The second observation (ObsID 4543) contained significantly more counts and was grouped to contain a minimum of 20 counts per bin. The first two observations displayed hard spectra typical of CVs; however, the third observation contained a large low energy tail (63 counts between 0.2–0.3 keV). Neither of the other observations showed this tail, containing minimal counts over the same range. The sensitivity of the ACIS-S detector, and the accuracy of the Chandra response files, fall off at low energies (below ∼0.5 keV). This introduces difficulty when trying to fit a spectrum over this range. For this reason, we ignored all counts below 0.5 keV for our primary fits.

We included a photoelectric absorption component within our X-ray analysis, using a hydrogen column density, N_H, frozen at 5.5 × 10¹⁹ cm⁻², inferred (Predehl & Schmitt 1995) from the measured color excess E(B − V) = 0.01 (Harris 2010).⁸ We used the convolution component cflux to calculate the flux over a range of 0.5–10 keV. Each spectrum was then fit to a hot X-ray plasma emission model (XSPEC model mekal). The metal abundance was fixed, using [Fe/H] = −1.57 (Harris 1996).

The three observations were fit with varying levels of success (see Table 2). From the hot diffuse gas emission model, the largest observed flux (0.5–10 keV) was (6.0 ± 0.8) × 10⁻¹³ erg cm⁻² s⁻¹, seen in the second observation, while the smallest flux was (1.8 ± 0.4) × 10⁻¹³ erg cm⁻² s⁻¹, seen in the third observation. Using a distance of 10.4 kpc to the M3 cluster (Harris 1996)⁸, these correspond to luminosities of 7.7 ± 1.0 × 10³³ erg s⁻¹ and 2.4^+0.6 _{− 0.5} × 10³³ erg s⁻¹. The individual fits can be seen in Figures 3–5. In addition, the observations were fit to an absorbed power-law spectrum. In the case of the first two observations this gave a worse fit; however, for the third observation this gave a slightly better fit and a marginally higher luminosity (L_X = (2.6 ± 0.5) × 10³³ erg s⁻¹ over the 0.5–10 keV range, again see Table 2). We also tested the models leaving the photoelectric absorption as a free parameter; however, the N_H values were consistent with the cluster value, and an F-test showed insignificant improvement. Finally, we tested the models with a pileup component. This served to worsen the fit in all three observations and slightly altered the luminosities (L_X = 3.9^+0.6 _{− 1.0} × 10³³ erg s⁻¹ for the first observation, L_X = 7.4^+2.2 _{− 1.0} × 10³³ erg s⁻¹ for the second observation, and L_X = (2.4 ± 0.6) × 10³³ erg s⁻¹ for the third observation).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We created light curves for each observation over the 0.5–10 keV range. Assuming a linear relation between the luminosity from the mekal models and the count rate, we used the maximum and minimum count rates from the modified light curves to estimate a maximum and minimum luminosity. From this, we found an estimated maximum 0.5–10 keV luminosity of (1.4 ± 0.3) × 10³⁴ erg s⁻¹, observed in the second observation, and a minimum luminosity of 8.5^+4.9 _{− 4.6} × 10³² erg s⁻¹, observed in the third observation, indicating that 1E1339's hard X-ray flux has varied by roughly an order of magnitude.

We can estimate the mass transfer rate implied using the formulae of Patterson & Raymond (1985b), converting to the 0.5–10 keV band using our power-law fit to get L_X(0.5–10 keV)$=1.6\times 10^{32}\dot{M}^{1.0}_{16}\,M^{1.8}_{0.7}$ erg s⁻¹, where M_0.7 is the WD mass in units of 0.7 M_☉ and $\dot{M}_{16}$ is the mass transfer rate in units of 10¹⁶ g s⁻¹. This gives a peak mass transfer rate of (2.5–8.8) × 10¹⁷ g s⁻¹, or (3.7–13.3) × 10⁻⁹ M_☉ yr⁻¹ (for WD masses of 0.7–1.4 M_☉). Such high-mass transfer rates are typically seen in nova-like CVs, or during the outbursts of dwarf nova CVs. However, the hard X-ray flux of such systems is typically much smaller (Patterson & Raymond 1985b; Wheatley et al. 1996). The CVs with the highest hard X-ray luminosities are typically magnetic systems (Verbunt et al. 1997) since matter flows onto the WD in a stream following the magnetic field (rather than a disk) and loses half of its gravitational potential energy in an optically thin post-shock region (Cropper et al. 1999) rather than an accretion disk which becomes optically thick (Patterson & Raymond 1985b). Thus, the peak hard X-ray luminosity of 1E1339 can be taken as evidence in favor of the magnetic nature of the accretion flow, as suggested by Edmonds et al. (2004).

3.1. Low-energy Emission

4.1. Comparison to Previous Observations

4.2. Comparison to GK Persei

4.3. Comparison to Very Faint X-ray Transients

From three observations of the X-ray source 1E1339, we found the source to be considerably variable, with a 0.5–10 keV luminosity ranging from (1.4 ± 0.3) × 10³⁴ erg s⁻¹ to 8.5^+4.9 _{− 4.6} × 10³² erg s⁻¹. The maximum observed X-ray luminosity gives 1E1339 the largest well-measured hard X-ray luminosity for any CV, providing a reliable estimate on the X-ray luminosity achievable by CV-type systems. The peak X-ray luminosity falls into the range of VFXTs, suggesting that the most X-ray luminous CV-type systems may account for some VFXTs.

Above 0.5 keV, we were able to fit each observation to a hot diffuse gas emission model with varying levels of success, showing a typical hard CV spectrum. Our observations over 0.5–10 keV are in agreement with previous observations which have shown 1E1339 to have fallen into a period of relative quiescence after its 1992 SSS period (see Dotani et al. 1999). However, a luminous low-energy component, below 0.3 keV, was seen in the 2005 observation. We were unable to successfully model this component of the observation due to the uncertain calibration of the ACIS-S detector below 0.3 keV. However, it is possible that the bolometric luminosity of the low-energy component may exceed that of the hard component.

1E1339's unusual behavior raises a number of questions. What is the origin and luminosity of the luminous low-energy component? Is this (largely hidden) low-energy flux due to nuclear burning on (part of?) the WD surface (Edmonds et al. 2004), or to a blackbody-like accretion-powered component, like those seen in magnetic CVs? What is the geometry of the accretion flow? (Can this be tested with longer, continuous X-ray observations?) Was the 1992 outburst an isolated nuclear flash, an increase in the accretion flow, or a change in the photospheric radius? Our near-simultaneous near-UV HST observations may help address these questions. There is also a clear need for XMM observations, as the pn camera has better-calibrated responses down to 0.2 keV to characterize the soft X-ray emission, and a long XMM observation could search for pulsations and a WD spin period, expected for a magnetic CV. Understanding the behavior of low-energy components such as that reported here may play a critical role in understanding the mismatch between observed and predicted numbers of SSSs reported by Gilfanov & Bogdán (2010) and Di Stefano (2010a).