DISCOVERY OF A SECOND TRANSIENT LOW-MASS X-RAY BINARY IN THE GLOBULAR CLUSTER NGC 6440

NGC 6440 was observed with the Chandra ACIS-S detector for 49.1 ks, from 2009 July 28, 15:16 (TT) to 2009 July 29, 05:15, using a 1/2 subarray. We searched for periods of elevated background, but found none. We used the level-2 event files provided by the CXC and CIAO 4.1⁹ for our analysis. Images were produced in the 0.3–7 keV and 0.5–2.5 keV bands, both of which are dominated by the scattered halo (a combination of dust grain scattering and the intrinsic point-spread function of the Chandra mirrors) and readout streak from a bright transient LMXB, heavily affected by pileup.¹⁰ Several faint point sources are clearly visible, which can be confidently identified with the cluster X-ray sources identified by Pooley et al. (2002). In Figure 1, we show Chandra images of NGC 6440 during the 2009 outburst, shortly after the 2009 outburst, during quiescent observations in 2000–2003 (Pooley et al. 2002; Cackett et al. 2005) and during the 2001 outburst (in't Zand et al. 2001). Clearly this is a new transient.

Zoom In Zoom Out Reset image size

The CIAO detection algorithm wavdetect was run on a 0.3–7 keV image of the cluster core to identify the positions of known cluster sources, which we shift (ΔR.A. = +0.008s, Δdecl. = +031) to align with the (ICRS) astrometry of Pooley et al. (2002). We estimated (by eye) the center of the symmetric "hole" in the counts by matching circles to the doughnut-shaped locus of maximum count rate in the LMXB halo, in both wavebands. Our result is (J2000) R.A. = 17:48:52.76(2), decl. = −20:21:24.0(1) (1σ values, after our astrometric correction), giving it the IAU name CXOGlb J174852.7−202124 and (for shorthand) NGC 6440 X-2. Detailed analysis of the remaining cluster sources will be presented elsewhere.

To measure the spectrum and luminosity of the transient, we extracted a spectrum from the readout streak, excluding a 20'' radius circle around the piled-up transient. We extracted background from rectangular regions above and below the readout streak, computed response functions for the position of the transient, and corrected the exposure time of the spectrum. We binned the spectrum to 60 counts per bin to improve its statistics, and excluded data over 8 keV and below 0.5 keV. An absorbed power law fits the data (Table 1, Figure 2), with photon index 1.7 ± 0.1. We found no evidence for a 6.4 or 6.7 keV iron line, with a 90% upper limit of 0.4 keV on its equivalent width.

Zoom In Zoom Out Reset image size

The light curve from the readout streak events shows a clear decline during the Chandra observation by a factor of ∼40%. Power spectra from the readout streak events show a clear periodicity at 1000.0 s and its harmonics, identical to one of the spacecraft dither frequencies and thus a likely artifact (it is not seen in Rossi X-ray Timing Explorer (RXTE) data). No other periodicities (such as the 57-minute orbital period, Altamirano et al. 2010a) were identified, suggesting that the transient is not seen at high inclination.

The RXTE has conducted regular scans of the Galactic bulge region since 2000 (generally covering NGC 6440 twice/week), in part to search for faint transient sources below the sensitivity of RXTE's all-sky monitor (Swank & Markwardt 2001), with results promptly made available.¹¹ The PCA scans are significantly more sensitive in the Galactic bulge than the RXTE all-sky monitor data. (Although dwell-by-dwell all-sky monitor data suggest a few points at L_X ∼ 10³⁸ ergs s⁻¹ from NGC 6440, suggesting X-ray bursts, their 6σ maximum significance is matched by similar points from black hole candidates in the Galactic center. Thus, we cannot be assured of their reality.) Scans in late 2009 May showed evidence for an increased count rate from NGC 6440 (see the third red line in Figure 3). An outburst was confirmed by Swift observations in early June (see below). No further activity was observed until a scan showed a 5σ detection on 2009 July 28, near the beginning of the Chandra observation. A pointed RXTE observation on July 30 showed a significant decline from the Chandra and bulge scan fluxes (details below). Later bulge scans (on August 1 and 2) were consistent with a decay below the PCA's (background-limited, due to its non-imaging nature) sensitivity (Figure 3).

Zoom In Zoom Out Reset image size

A third bulge scan peak on 2009 August 29 triggered a pointed observation on August 30, which discovered millisecond pulsations from NGC 6440 X-2 with a frequency of 205 Hz (Altamirano et al. 2010a). Further RXTE observations starting on September 1 found fluxes returning to quiescence. Upon identifying a fourth outburst with Swift/XRT on 2009 October 1, a pointed RXTE observation on October 2 detected NGC 6440 X-2 near the detection limit (∼1.5 mCrab). Altamirano et al. (2010a) report the detection of pulsations at a 3.4σ level, further confirming the identification of this transient with NGC 6440 X-2. RXTE observations on October 3 found it below the PCA's sensitivity limit. A fifth outburst was caught by RXTE/PCA on October 28, returning to quiescence by November 1. Below, we describe spectral analysis of these data; all RXTE/PCA timing analysis is described in Altamirano et al. (2010a).

We extracted spectra from the PCA (Jahoda et al. 2006) detectors (PCU2 was consistently on, sometimes joined by other PCUs), excluding times when elevation angle ELV < 10°, source offset >002, or the time since the last South Atlantic Anomaly passage was <30 minutes. The background was modeled using the latest faint background model pca_bkgd_cmfaintl7_eMv20051128.mdl and the recorded times of South Atlantic Anomaly passage.¹² Response files were created with the PCARMF (v. 11.1) tool. For spectral fitting, we exclude data below 3 keV and above where the source can no longer be detected above our modeled background; this cut ranged between 10 and 30 keV, depending on the observation. HEXTE (cluster B, sensitive to the 20–200 keV energy range) did not detect the source even in the brightest (August 30) pointed RXTE observation. N_H was fixed to the cluster value due to RXTE's relative insensitivity to N_H. All our reported L_X values are for 0.5–10 keV (for consistency with other instruments; this extrapolation may incur additional systematic errors) assuming a distance of 8.5 kpc.

The July 30 pointed observation could be fit with an absorbed power law with photon index 2.2 ± 0.1 and a luminosity of (5 ± 1) × 10³⁵ erg s⁻¹, 3.5 times lower than the July 28 Chandra observation (Table 1; but see below). The August 30 observation showed a photon index of 1.84 ± 0.02 and L_X = 2.9 × 10³⁶ erg s⁻¹. The series of observations on September 1 and 2 showed much lower fluxes, averaging L_X = 4 × 10³⁵ erg s⁻¹; the longest one also shows clear evidence of an iron line at 6.7 ± 0.2 keV. However, simultaneous Swift observations (see below) find much lower fluxes from the cluster of 6 × 10³⁴ erg s⁻¹, consistent with the quiescent cluster emission and inconsistent with the transient position. Therefore, we attribute the RXTE/PCA flux observed on September 1 to the Galactic Ridge emission at this location. Subtracting this flux from the July 30 RXTE/PCA measurement gives L_X = (1 ± 1) × 10³⁵ erg s⁻¹, a very marginal detection, and makes its spectral parameters unreliable.

We refine our spectral fitting of the August 30 observation by extracting spectra from only the top layer of PCU2 and modeling the inferred Galactic Ridge emission using our fits to the August 1 data. Fitting with an absorbed power law gives a good reduced chi-squared, but a residual near 6.5 keV suggests the addition of an iron line (this is in addition to the Galactic Ridge iron line seen in the September data). Freezing N_H to the cluster value, we obtain Γ = 1.78 ± 0.03, L_X(0.5–10 keV) =(2.4 ± 0.1) × 10³⁶ erg s⁻¹, with an iron line at 6.6^+0.6_−0.4 keV, of equivalent width 54 ± 39 eV (90% confidence). An F-test suggests that this is reasonable, providing 3.6% probability of such a Δχ² by chance. Other AMXPs have displayed evidence of Fe K lines (e.g., SAX J1808.4−3658, Cackett et al. 2009; Papitto et al. 2009; HETE J1900.1−2455, Cackett et al. 2010; Swift J1756.9−2508, Patruno et al. 2010a), often showing relativistic broadening which our observations are unable to resolve.

Two RXTE/PCA observations were obtained during the early October outburst, both during the source's decline. We identified pulsations in the first (October 1) observation, confirming that pulsations are common from this object (Altamirano et al. 2010a). The second (October 3) occurred after NGC 6440 X-2 had dipped below the PCA's sensitivity limit. We model the Galactic Ridge contribution to the October 1 observation, using the October 3 observation, finding Γ = 1.78 ± 0.15, L_X(0.5–10) =8.4^+0.8_−0.5 × 10³⁵ erg s⁻¹ (see Table 1).

Three RXTE/PCA observations were able to detect the October/November outburst. The first (on October 28) identified pulsations again (Altamirano et al. 2010a), at L_X(0.5–10 keV) =8.6^+0.9_−0.8 × 10³⁵ ergs s⁻¹ (see Table 1), after subtraction of Galactic Ridge emission. RXTE/PCA observations detected emission over background on October 29 and (marginally) October 30, and observations on October 31 and after were consistent with Galactic Ridge emission.

The full RXTE/PCA Galactic bulge scan data reveals seven times from 2000 to 2009 when NGC 6440's count rates are 4σ above zero. The two brightest outbursts (L_X>10³⁷ erg s⁻¹) have been identified with the other transient in NGC 6440, SAX J1748.9−2021 = CX1 (Altamirano et al. 2008; in't Zand et al. 2001). The other five potential outbursts are much fainter and briefer (only a single bulge scan point each, so lasting less than a week), of which three were discussed above. The other two are 2007 April 15 and 2009 March 20. A period of 30.7 days (±0.3 days, from the uncertainties in the peak of the late October outburst), with reference date MJD 55040.5, reasonably predicts the peaks of the four well-studied outbursts. The model predicts missed outbursts on April 27 (weak activity is suggested by bulge scans on April 18–20), May 28 (a bulge scan detection occurred on May 25, followed by a faint Swift detection on June 4), and June 27 (no bulge scans conducted ±10 days around this date). The bulge scan point in March is off the prediction by 7 ± 1 days, indicating either that the outburst period has slightly decreased over time (as the April and May evidence also suggest) or that the March point is not a real outburst. Sensitive monitoring since November 2 (up to 2010 February 11) has not been possible due to solar constraints and an outburst of SAX J1748.9−2021 (Suzuki et al. 2010; Patruno et al. 2010b). We show the bulge scan data in the relevant date range (and detections and upper limits from Swift and Chandra) in Figure 3; see also Altamirano et al. (2010a). We use an absorbed power law of photon index 2 to convert RXTE/PCA bulge scan fluxes to 0.5–10 keV unabsorbed fluxes with PIMMS¹³ and assume a distance of 8.5 kpc to estimate luminosities.

Eighteen Swift/XRT observations were performed, tracking five outbursts from NGC 6440 X-2 (see Table 1). We extracted all Swift XRT spectra from 20 pixel radii (except for October 1 and 28, see below) and background from a surrounding annulus. We downloaded the response matrix swxpc0to12_20010101v009.rmf from the Swift Web site¹⁴ and created effective area files using the XRTMKARF tool. Spectra with more than 50 counts were binned with 15 counts/bin for χ² statistics, those with fewer used C-statistics (either binned with 5 counts/bin or unbinned), while we produced only luminosity limits for less than 10 detected counts.

A Swift X-ray Telescope (XRT) observation on 2009 June 4 found enhanced X-ray emission from NGC 6440. Using the FTOOL XRTCENTROID on the June 4 XRT source, we identified a position of R.A. = 17:48:52.73, decl. = −20:21:24.1 with an error radius of 5''. This position is consistent with NGC 6440 X-2 (see Figures 1 and 4), but not with other known X-ray sources in NGC 6440, so we conclude it is the same source. A second Swift observation on June 11 found much weaker emission, at position R.A. = 17:48:52.64, decl. = −20:21:29.9, with error radius 73, consistent with either NGC 6440 X-2 or with the cluster center. Spectral fitting of the few detected photons with an absorbed power law derives a photon index of 4.2^+1.2_−1.1 and L_X = 1.0^+1.0_−0.5 × 10³⁴ erg s⁻¹. This is consistent with emission from the rest of the cluster, containing a mixture of soft quiescent LMXBs and sources (likely cataclysmic variables) with harder spectra (Pooley et al. 2002; Heinke et al. 2003).

Zoom In Zoom Out Reset image size

We obtained four Swift observations soon after the initial Chandra discovery (Table 1). On July 31, Swift/XRT found enhanced emission from NGC 6440 X-2's position. Observations on August 4, 6, and 10 found fluxes and positions consistent with the cluster center, and soft spectra (Table 1, Figures 4 and 5). The centroid of the emission in the deepest of these observations (August 6, 1.9 ks) is R.A. = 17:48:529, decl. = −20:21:35.1, with an error radius of 61, which is consistent with the cluster center but not with NGC 6440 X-2.

Zoom In Zoom Out Reset image size

After the August 29 bulge scan detection, Swift observed NGC 6440 on September 1, identifying emission from the location (R.A. = 17:48:52.6, decl. = −20:21:24.9, error radius 48) of NGC 6440 X-2 at L_X = (6 ± 2) × 10³⁴ erg s⁻¹ (Figures 4 and 5). Further Swift observations on September 2 and 4 found faint emission, consistent in position and flux with the cluster.

Swift monitoring on October 1 found NGC 6440 X-2 back in outburst. The count rate was high enough to produce pileup (in photon counting mode); we fit the radial profile with a King model and identified a 6'' core, so we extracted a spectrum from an annulus from 6'' to 45'' (20 pixels). On October 2, Swift identified a declining flux from NGC 6440 X-2, returning to quiescence by October 4 (Figures 4 and 5).

A Swift observation on October 29 confirmed that NGC 6440 X-2 was back in outburst, and relatively bright; we dealt with pileup as described above. Observations on October 30 and 31 observed X-2's decay, which by November 1 was below Swift/XRT's detection limit (Figures 4 and 5).

We combined two Chandra ACIS-S observations of NGC 6440 (total exposure 48 ks) when no transients were in outburst (ObsIDs 947 and 3799; Pooley et al. 2002; Cackett et al. 2005) to look for evidence of NGC 6440 X-2's X-ray emission in quiescence. The observations were reprocessed and aligned, and filtered to produce a 0.3–7 keV band image (Figure 1). No photons lie within a 1'' circle around NGC 6440 X-2. Using 2.3 photons as our 90% confidence upper limit (Gehrels 1986), we compute limits on the unabsorbed quiescent luminosity. For a power law of index 2.2, L_X(0.5–10 keV)<7.4 × 10³⁰ erg s⁻¹, while for a hydrogen-atmosphere neutron star model (Heinke et al. 2006), L_X(0.5–10 keV)<1.6 × 10³¹ erg s⁻¹ and L_NS (0.01–10 keV)<6.0 × 10³¹ erg s⁻¹ (we use L_NS as the total emission from the NS surface).

We obtained a 5 ks ACIS-S follow-up Chandra Director's Discretionary Time observation on 2009 August 10, to see if NGC 6440 X-2 had returned to full quiescence, or was continuing to accrete at L_X(0.5–10 keV) ∼ 5 × 10³² erg s⁻¹, similarly to SAX J1808.4−3658 at the end of its 2008 outburst (Campana et al. 2008). Two photons were detected at the position of NGC 6440 X-2 (see Figure 1). As the background is quite low (<0.01 photons expected), both photons are probably from NGC 6440 X-2. We estimate (for an assumed spectrum similar to that in outburst) L_X = 6⁺⁹₋₄ × 10³¹ erg s⁻¹, with uncertainties from Gehrels (1986) at 90% confidence. This confirms that NGC 6440 X-2 returned to quiescence, although it appears brighter than the limits from the 2000 and 2003 quiescent observations (above).

The location of NGC 6440 X-2 has been previously observed by the European Southern Observatory's New Technology Telescope (NTT) and by the Hubble Space Telescope (HST) (Piotto et al. 2002). The NTT imaging in R and B was described by Verbunt et al. (2000), and the HST imaging in V and B was described by Piotto et al. (2002). We identified UCAC2 standards to place the NTT R astrometric frame onto the ICRS frame, and then by identifying common stars in the WFPC2 and NTT frames, to place the HST astrometric frame onto the ICRS frame, with an uncertainty of 02 (1σ). No star can be identified within 4σ of the transient position on the NTT and WFPC2 B frames, but one star in the WFPC2 V frame is located 04 from the transient position (Figure 6). Calibrating our photometry with that in Piotto et al. (2002), we find a magnitude of V = 21.0 ± 0.2 for this object. Based on objects in the HST image that are just barely detected in B, we estimate B>22.0. These images were probably taken during quiescence, and NGC 6440 X-2's 57 minute orbital period indicates that it will be very faint. As the ultracompact LMXB XTE J0929-314 in quiescence may have been detected with M_V = 13.2 (D'Avanzo et al. 2009; thus it would have V = 31 in NGC 6440), we think it unlikely that this star is the true counterpart.

Zoom In Zoom Out Reset image size

Li et al. (2009) reported that unfiltered images were taken of NGC 6440 on 2009 July 30 and 31 with the 0.76 m Katzman Automatic Imaging Telescope (KAIT). Their image subtraction against previous KAIT observations (2007, 2008) revealed no evidence of an optical transient, with limiting magnitudes of 19.5–20.0.

We obtained images with the CTIO 4-meter telescope using the ISPI infrared imager on 2009 August 2–4 in the J and K bands (Figure 7). The total exposure times were 12 and 8 minutes per night, giving theoretical magnitude limits of 19.5 and 18 in J and K, respectively (with a signal-to-noise ratio of 10). The airmass was around 1.1 and the seeing was 1'' the first night, increasing to 15 the last night. The images have been taken using a dithered pattern and an offset blank field has been observed to estimate the sky contribution to the emission. We used IRAF common packages and the PANIC package (Martini et al. 2004) to reduce the data and then SExtractor/Scamp (Bertin 2006) to calibrate the astrometry and photometry of the image, using the 2MASS catalog as a reference (Skrutskie et al. 2006). Based on the detection of sources around NGC 6440 X-2, a source of magnitude K = 17 and J = 18.5 would have been detected.

Zoom In Zoom Out Reset image size

We obtained three 500 s frames in each of g' (note that the g' bandpass is between V and B) and r' using Gemini-South GMOS-S on 2009 September 1, during the outburst decay (program GS-2009B-DD-2). Unfortunately the r' frames were saturated at the location of the transient, but the g' frames provided our best optical outburst limit (Figure 6). No star was seen within 08 (4σ) of the transient position. An approximate calibration of the g' images was obtained using a star-by-star comparison with the B, V photometry of Martins et al. (1980). We determined the detection limit for stars within an annulus about the cluster center containing the location of the transient. Based on the faintest 5% of the stars in the annulus, the estimated detection limit is g' ∼ 22.0. Thus, we adopt g'>22 as an approximate limit for NGC 6440 X-2, at a time when Swift found its X-ray luminosity to be 6 × 10³⁴ erg s⁻¹.