DISCOVERY OF THE THIRD TRANSIENT X-RAY BINARY IN THE GALACTIC GLOBULAR CLUSTER TERZAN 5

We monitored Terzan 5 up to a few times per week for part of 2012 with the Swift/XRT, covering the 0.3–10 keV energy range (Burrows et al. 2005), as part of our monitoring campaigns of globular cluster X-ray transients (see Altamirano et al. 2012b; D. Altamirano et al. 2013, in preparation). This monitoring enabled us to observe the rising outburst of a new transient (and the 3rd known transient LMXB in this cluster) first detected on 2012 July 6 (Wijnands et al. 2012).

We used Swift/XRT's photon counting (PC) mode, which produces two-dimensional images, and windowed timing (WT) mode for which CCD data is collapsed into a one-dimensional image for fast readout. PC mode data should be checked for pile-up when the count rate exceeds 0.5 count s⁻¹. Pile-up is the recording of multiple photons as a single event, leading in the worst case to rejection of all events from the center of the point-spread function, or PSF. Our Swift/XRT observations include 22 observations during the outburst, with 8 observations in WT and the rest in PC mode (Table 1).

We used HEASOFT 6.12 and FTOOLS¹² (Blackburn 1995) to reduce and analyze the data. We reprocessed the data with the FTOOLS xrtpipeline and manually extracted data for spectral analysis. We investigated every observation for pile-up, following the Swift/XRT pile-up thread,¹³ and extracted data from an annulus around the source in PC mode observations that suffered from pile-up. We subtracted background from a circular region in the vicinity of the source in all PC observations. The extraction region for WT data was chosen to be a box around the event array (background subtraction was unnecessary for these count rates), as discussed in the Swift/XRT data reduction guide.¹⁴ We extracted spectra in the 0.5–10 keV bandpass using FTOOLS xselect, and created ancillary response function (ARF) files for each observations using FTOOLS xrtmkarf. We performed spectral analysis using XSPEC 12.7.1 (Arnaud 1996).

For heavily absorbed sources, WT data show low energy spectral residuals, which look like a "bump" in the spectrum, and cause spectral uncertainties in the ≲ 1.0 keV region.¹⁵ We compared Swift/XRT WT mode data to Chandra data (Section 2.2) taken within a few days, fitting them with the same model to find the energy range in which discrepancies appear. Based on this comparison, we ignored data below 1.4 keV in all WT observations during the outburst.

We observed Terzan 5 X-3 during outburst with Chandra/ACIS in full-frame and FAINT telemetry mode with no grating (Obs. ID: 13708, PI: Pooley). We also used Chandra archival data for our analysis of this source, details of which can be found in Table 1. All archival observations were taken in FAINT telemetry mode with the ACIS-S3 CCD at the focus. We analyze Chandra/ACIS data in the 0.5–10 keV energy range.

Data was reprocessed using CIAO 4.4 (Fruscione et al. 2006), with CALDB 4.4.8, following the standard CIAO science threads.¹⁶ We used observations during which all sources in the globular cluster were in quiescence. We reprocessed the data, corrected the relative astrometry and ran CIAO reproject_events, and then stacked the event files together using CIAO dmmerge.

Spectra were then extracted from both the archival and new data using CIAO task dmextract. Terzan 5 X-3 was heavily piled up in the new observation (in outburst), and so we extracted a spectrum from the readout streak. Finally, we combined all archival (quiescent data) using FTOOLS addspec.¹⁷ Combining the quiescent data resulted in 240 ks of exposure time.

The MAXI all sky X-ray monitor's (Matsuoka et al. 2009) GSC detector data covers the 2–20 keV energy range, and one-day averaged light curves are publicly provided in four bands: 2–4 keV, 4–10 keV, 10–20 keV and 2–20 keV (Mihara et al. 2011). We noticed two problems with MAXI/GSC light curves. Due to the low spatial resolution of MAXI/GSC and bright sources in the crowded field of Terzan 5, there is the possibility of background contamination from nearby sources like GX 3+1 and GX 5-1 (Figure 1). Since these two sources showed stable X-ray brightness with no obvious variations in the MAXI/GSC data during Terzan 5 X-3 outburst, the background contamination may lead to a constant enhanced background.

Zoom In Zoom Out Reset image size

We also noticed that MAXI/GSC light curves show periodic behavior with a period of ≈35 days for various well-known stable X-ray sources (e.g., the Crab nebula). This is caused by calibration issues regarding the 70-day precession of the International Space Station's orbit (MAXI team 2013, private communication). This problem principally affects the 2–4 keV data, with less effect on the 4–10 and 10–20 keV light curves. Thus we ignored the 2–4 keV light curves for this research. We also ignored MAXI/GSC 10–20 keV band light curves, due to the low statistical significance of Terzan 5 X-3's detection there. We decontaminated the MAXI/GSC 4–10 keV light curve assuming a constant background count rate of 0.023, calculated based on a weighted average of the count rates before the outburst for a period of ∼120 days, with the addition of a systematic error based on the rms variations in the light curve before the outburst. We used the corrected values of the statistical uncertainties in MAXI/GSC data, as the MAXI team announced an erratum in the reported statistical uncertainties on 2013 April 26 (MAXI team 2013, private communication) noting that the corrected uncertainties are a factor of two larger than previously reported.

The Swift/BAT telescope data covers 15–150 keV energy range (Barthelmy et al. 2005). We used daily light curves from the Swift/BAT transient monitor results provided by the Swift/BAT team (Krimm et al. 2013). Data points on these daily light curves are from Swift/BAT survey data and are represented in a single band (15–50 keV). These points are the weighted average of all observations performed each day. The Swift/BAT has better angular resolution (20 vs.1 degree FWHM for Swift/BAT versus MAXI/GSC). Such an improvement in the angular resolution limits the chances of contamination occurring. Although we cannot rule out some contamination, it seems reasonable that it does not pose a serious problem. As such, it is not surprising that we do not see evidence of contamination of the Swift/BAT data by nearby sources as is identifiable in the MAXI data.

We accurately and precisely located the position of Terzan 5 X-3, using the Chandra/ACIS data. We compared the Chandra/ACIS observation taken during the outburst and a stacked image of 7 Chandra/ACIS observations taken when all Terzan 5 sources were in quiescence (Section 2.2, Table 1). We corrected the astrometry in the outburst observation by comparing the positions (using the CIAO wavdetect tool) of three other sources in this observation with their astrometrically corrected positions as reported in Heinke et al. (2006b). Using a weighted average of the required shifts (+023 for R.A. and +004 for Decl.), we find the position of Terzan 5 X-3 to be R.A. = 17:48:0541 ± 002 and Decl. = −24: 46: 380 ± 02, in agreement (2σ) with the published position of the X-ray source CXOGLB J174805.4-244637 (Heinke et al. 2006b), at R.A. = 17:48:05413 ± 0001 and Decl. = −24:46:3767 ± 002 (Figure 2).

Zoom In Zoom Out Reset image size

We used MAXI/GSC and Swift/BAT hard X-ray transient monitor light curves to study the evolution of Terzan 5 X-3's outburst in the soft X-ray (4–10 keV, MAXI/GSC) and hard X-ray (15–50 keV, Swift/BAT) bands. We converted count rates into equivalent fluxes from the Crab Nebula to make these light curves suitable for comparison. For this purpose, we used conversion coefficients given for each instrument¹⁸: for the Swift/BAT hard X-ray transient monitor 1 Crab = 0.22 count cm⁻² s⁻¹ and for the MAXI/GSC 4–10 keV band 1 Crab = 1.24 count cm⁻² s⁻¹. We plot the light curves for Terzan 5 X-3's outburst as seen by both MAXI/GSC and Swift/BAT, and their ratio, in Figure 3.

Zoom In Zoom Out Reset image size

We distinguish four phases of Terzan 5 X-3's outburst. (a) Rise: the hard X-ray brightness of the source increases, the source eventually becoming significantly detected in the soft X-ray as well. (b) Hard state: hard X-ray brightness reaches its peak. (c) Soft state: soft X-ray brightness peaks while hard X-ray brightness drops. (d) Decline: the source briefly gets brighter in the hard X-ray again before turning off. Unfortunately there is insufficient data from the MAXI /GSC during the decline of the outburst. Therefore, we are unable to confirm that Terzan 5 X-3 returns to the hard state during its decline.

We used the soft X-ray (MAXI/GSC 4–10 keV, S) and hard X-ray (Swift/BAT 15–50 keV, H) light curves, to create a color–luminosity diagram (as an analogy to a hardness–intensity diagram, Fender et al. 2004) for the outburst (Figure 4). We defined our color to be (H − S)/(H + S), and defined luminosity as (H + S), converting count rates to luminosities in each band before their summation or subtraction. We converted count rates to luminosities with the assumption of a 5.9 kpc distance and power-law spectra, using power-law index values inferred from Swift/XRT data spectral fitting (Section 3.3.2). We extrapolated the MAXI/GSC 4–10 keV band flux to 0.1–12 keV and the Swift/BAT 15–50 keV band flux to 12–50 keV, and calculated luminosities in the 0.1–50 keV band. We cannot measure the spectral index above 10 keV, but since in the hard state the spectra are typically reasonably described by a power-law up to 50 keV, and in the soft state the flux above 12 keV is a minor contribution to the total, this is unlikely to have a large effect. In the 0.5–10 keV band, the source was bright for approximately 20 days, reaching a maximum luminosity of 7 × 10³⁷ erg s⁻¹ and an average luminosity of 3 × 10³⁷ erg s⁻¹ in this time interval. The evolution of the outburst and phases mentioned above can be clearly seen in this color–luminosity diagram (Figure 4). Spectral evolution during the outburst, including at fainter fluxes but with a more limited bandpass, can also be seen in Swift/XRT observations (Section 3.3.2).

Zoom In Zoom Out Reset image size

3.3.1. Hydrogen Column Density of Terzan 5

3.3.2. Outburst

The spatial resolution of Swift/XRT is such that we must account for possible contamination due to emission form other XRBs in the cluster. To assess these levels, we fit the spectra of pre-outburst observations (using the same extraction region as for the outburst spectra), finding a background level of L_X ∼ 10³⁴ erg s⁻¹. We used 6 Swift/XRT observations taken before the outburst to estimate the combined spectrum from the other cluster sources. Due to the low number of counts per observation, we combined their spectra. We fit the resulting pre-outburst spectrum with an absorbed power-law model (Table 2) with N_H fixed to 1.74 × 10²² cm⁻², based on the results in Section 3.3.1.

Table 2. Swift/XRT and Chandra/ACIS Observations Fit to an Absorbed Power-Law Model

Obs. ID	MJD	Photon Index	Flux	LX	$\chi _\nu ^2$/D.O.F.(nhp)
32148002-91445005	55966-56108	2.4$_{-0.4}^{+0.5}$	2.4$_{-0.6}^{+0.7}$	1.0$_{-0.2}^{+0.3}$	0.68/7(0.69)
91445006	56114.8	2.6 ± 0.7	10$_{-3}^{+5}$	4$_{-1}^{+2}$	0.83/6(0.54)
32148003	56115.8	2.5 ± 0.4	25$_{-5}^{+7}$	10$_{-2}^{+3}$	0.85/5(0.51)
32148004	56117.0	2.3 ± 0.3	37 ± 6	15 ± 2	1.86/9(0.052)
32148005	56118.1	1.9 ± 0.4	90 ± 12	37 ± 5	0.54/6(0.78)
32148006	56120.7	1.4 ± 0.1	658 ± 32	273 ± 13	1.58/19(0.05)
526511000	56121.7	1.4 ± 0.1	1658 ± 91	688 ± 38	1.20/14(0.27)
91445008	56124.3	1.46 ± 0.08	1920 ± 63	797 ± 26	0.69/25(0.87)
526892000	56124.9	1.39 ± 0.06	1936 ± 46	803 ± 19	1.56/34(0.03)
32148007	56125.9	1.40 ± 0.04	2125 ± 40	882 ± 17	1.05/37(0.38)
91445009a	56129.1	1.8 ± ?	10440 ± ?	4333 ± ?	3.94/56(1.9 × 10−21)
91445010	56134.1	1.62 ± 0.05	16070 ± 330	6669 ± 137	0.77/36(0.84)
13708a(CXO)	56138.4	1.68 ± 0.03	6680 ± 80	2772 ± 33	1.76/102(3.4 × 10−6)
91445011a	56140.1	1.8 ± ?	7330 ± ?	3042 ± ?	2.10/80(3.4 × 10−6)
91445012	56144.8	2.11 ± 0.03	3651 ± 70	1515 ± 29	1.12/38(0.27)
91445013	56149.4	2.07 ± 0.07	514 ± 20	213 ± 8	1.10/58(0.28)
32148008	56150.1	1.7 ± 0.1	295 ± 16	122 ± 7	0.81/45(0.81)
32148011	56152.2	1.8 ± 0.1	260 ± 16	108 ± 7	0.70/37(0.91)
530808000	56152.4	1.9 ± 0.2	266 ± 30	110 ± 12	1.12/10(0.34)
32148010a	56153.2	1.8 ± 0.1	219 ± 13	91 ± 5	1.79/44(9.5 × 10−4)
91445014	56154.3	1.9 ± 0.2	215 ± 18	89 ± 7	1.06/23(0.38)
32148012	56158.3	2.0 ± 0.1	149 ± 11	62 ± 5	1.06/33(0.37)
91445015	56159.1	1.8 ± 0.3	110 ± 13	46 ± 5	0.97/10(0.47)
91445016	56163.2	2.0 ± 0.2	45 ± 5	19 ± 2	1.49/16(0.09)
32148013	56169.3	2.4$_{-1.2}^{+1.5}$	2.7$_{-1.5}^{+4.8}$	1.1$_{-0.6}^{+1.9}$	⋅⋅⋅

Notes. Unabsorbed flux in units of 10⁻¹² erg s⁻¹ cm⁻² and luminosity in 10³⁴ erg s⁻¹, both in 0.5–10 keV. All the flux and luminosity values are background subtracted. "a" indicates spectra which are not well described by a power law (null hypothesis probability is <10⁻²); if χ² > 2, no errors are calculated (indicated by "?"). The first row shows a spectral fit to a merged spectrum of 6 Swift/XRT observations of Terzan 5 before the outburst of Terzan 5 X-3. For Obs.ID 32148013, due to low counts, we used cstat statistics in the fitting. Reported uncertainties are 90% intervals.

Download table as:  ASCIITypeset image

For our initial spectral analysis, we used an absorbed power-law model to fit Terzan 5 X-3's spectrum, including a second power-law component with values fixed to the pre-outburst results to model the background (Table 2, Figure 5). A simple power-law model provided a good fit to most of the spectra (Table 2), while physically motivated complex spectral models could not be well-constrained for the high-quality bright outburst spectra (see below), so we focus on the results from power-law fits. This spectral analysis shows a significant drop of photon index from 2.6 ± 0.7 to 1.4 ± 0.1 during the outburst rise (Figure 5), showing a clear hardening of the spectrum during the rise from quiescence. During the hard state, the photon index shows no significant variations and is ≈1.4. After the phase transition to the soft state, the photon index softens to ≈1.9, with significant variations and several poor fits.

Zoom In Zoom Out Reset image size

Several spectral models have been suggested in the literature to model the detailed spectra of transient NS LMXBs in outburst, see e.g., Lin et al. (2007). These models usually contain a soft component for the radiation from the disk and/or boundary layer(i.e., a multi-color black body) and a hard component for the radiation from the hot corona around the accreting object (i.e., Comptonized radiation). We attempted to perform analyses of Terzan 5 X-3's outburst spectra using a variety of complex models with multiple components (e.g., DISKBB + COMPTT, BBODY+COMPTT, BBODY+BKNPOW). We could not obtain strong constraints on the spectral parameters due to the limited energy band available and limited statistical quality of the XRT data. The Swift/XRT WT spectra suffered particularly from calibration issues below ∼1 keV (Section 2.1). The Swift/BAT survey mode hard (>15 keV) X-ray spectra had a low signal to noise ratio, which prohibited spectral analysis. We defer further detailed spectral fitting, e.g., to clearly distinguish thermal versus nonthermal components, to future work.

3.3.3. Thermonuclear Burst

During Swift/XRT ObsID 32148007, which started at 20:54:00 UT on 2012 July 17, we detected an eightfold count rate increase over ∼3 s, starting at ≈21:06:40, followed by a slower decline over ∼1 minute (Figure 6, bottom panel), suggestive of a thermonuclear burst (Altamirano et al. 2012c). Using FTOOLS xselect, we divided the data from this observation into time intervals of 4 s each. We extracted spectra from each time interval using FTOOLS xselect and analyzed the spectra.

Zoom In Zoom Out Reset image size

During this thermonuclear burst, the count rate reached ≈160 count s⁻¹. There is a possibility of pileup in Swift/XRT observations in windowed timing mode when the count rate exceeds 100 count s⁻¹ (Romano et al. 2006). Following the Swift/XRT pile-up thread for WT data, we extracted spectra while excluding increasingly large fractions of the central PSF. Doing this, we found that the fitted photon index did not change, and thus we found no significant signs of pileup in this observation during the burst.

We fit an absorbed power-law to the spectrum extracted from a pre-burst interval, and considered this fit as a fixed component of the spectral model for time intervals during the burst (cf., Worpel et al. 2013; our statistics are insufficient to determine whether this assumption is correct, and moderate changes should not affect our conclusions). We fit an absorbed blackbody model (BBODYRAD) to the burst emission, finding decent fits for all intervals, and show the spectral evolution in Figure 6. We find clear evidence of cooling (between 770 and 790 s in Figure 6), while the inferred radius remains essentially constant. This is a clear signature of a thermonuclear burst, and thus of a NS.

We estimated the timescale of this burst using two methods. We fit an exponential model (count rate $\propto e^{-t/\tau _1}$) to the light curve of the burst after the peak, estimating the timescale τ₁ ≈ 16  ±  1s. Following Galloway et al. (2008), we estimated an alternative timescale for thermonuclear bursts τ₂ = E_burst/F_peak, where E_burst is the total fluence during the burst and F_peak is flux at the peak, finding τ₂ ≈ 29s. Galloway et al. (2008) divide bursts into those with τ₂ longer than 10 s, and those with τ₂ shorter than 10 s. The longer bursts are generally powered by hydrogen burning (with the exception of "giant" bursts involving photospheric radius expansion, which was not seen here), and the short bursts involve only helium burning, since hydrogen burning proceeds more slowly than helium burning (Fujimoto et al. 1981; van Paradijs et al. 1988; Cornelisse et al. 2003). Our measured burst timescale indicates that hydrogen is being accreted, and thus that the donor star is hydrogen-rich, which requires an orbital period ≳1.5 hr (e.g., Nelson et al. 1986) and excludes a WD donor.

3.3.4. Quiescent Behavior

We used 7 Chandra/ACIS observations taken when all sources were quiescent (Table 1) to study the behavior of Terzan 5 X-3 in quiescence before its outburst. We extracted source and background spectra from each observation using CIAO dmextract. We used a combination of a power-law (PEGPWRLW) model and a hydrogen atmosphere for a NS (NSATMOS), with absorption (PHABS) set to our preferred cluster value, the NS radius to 10 km, mass to 1.4 M_☉, and distance set to 5.9 kpc. This model has been previously used to fit its spectrum in one quiescent observation (Heinke et al. 2006b). To study possible spectral variations, we simultaneously fit spectra from each observation in four different Trials, each with different parameters free (Table 3): (I) constraining the NSATMOS and PEGPWRLW components to have the same values between all observations; (II) letting only the power-law normalization vary between observations, while constraining the NSATMOS temperature to be the same; (III) letting only the NSATMOS temperature vary between observations, while constraining the power-law normalization to be the same; (IV) letting both the power-law normalization and the NSATMOS temperature vary between observations. We found no evidence for variation in the power-law photon index Γ between observations if we allowed it to vary. Therefore, we tied its value between observations in each Trial (Figure 7).

Zoom In Zoom Out Reset image size

Table 3. Spectral Fitting of 7 Chandra/ACIS Observations of Terzan 5 X-3 During Quiescence (See Table 1), Using PHABS(PEGPWRLW+NSATMOS) in XSPEC

Trial	ObsID	log T	FX, NS(0.5–10 keV)	Γ	FX, PL(0.5–10 keV)	LX, total(0.5–10 keV)	$\chi _\nu ^2$/D.O.F.(nhp)
		(K)	(10−13 erg cm−2 s−1)		(10−13 erg cm−2 s−1)	(1033 erg s−1)
I	all	6.143$_{-0.011}^{+0.015}$	2.4$_{-0.2}^{+0.3}$	1.9 ± 0.4	1.1$_{-0.1}^{+0.3}$	1.4$_{-0.1}^{+0.2}$	1.53/76(0.002)
II	03798	6.138$_{-0.018}^{+0.012}$	2.4$_{-0.5}^{+0.3}$	2.1 ± 0.4	1.9$_{-0.4}^{+0.6}$	1.8 ± 0.3	1.09/70(0.27)
	10059	t	t	t	1.3$_{-0.3}^{+0.5}$	1.5 ± 0.2	t
	13225	t	t	t	0.8$_{-0.3}^{+0.4}$	1.3 ± 0.2	t
	13252	t	t	t	1.2$_{-0.3}^{+0.5}$	1.5 ± 0.2	t
	13705	t	t	t	0.7 ± 0.6	1.3 ± 0.3	t
	14339	t	t	t	0.8$_{-0.3}^{+0.5}$	1.3 ± 0.2	t
	13706	t	t	t	1.4$_{-0.3}^{+0.4}$	1.6 ± 0.2	t
III	03798	6.172$_{-0.015}^{+0.011}$	3.3$_{-0.5}^{+0.4}$	1.9 ± 0.4	1.0$_{-0.2}^{+0.3}$	1.8 ± 0.2	1.22/70(0.099)
	10059	6.149$_{-0.019}^{+0.014}$	2.7$_{-0.5}^{+0.3}$	t	t	1.5 ± 0.2	t
	13225	6.120$_{-0.029}^{+0.019}$	1.9 ± 0.5	t	t	1.2 ± 0.2	t
	13252	6.144$_{-0.019}^{+0.014}$	2.4$_{-0.5}^{+0.3}$	t	t	1.4 ± 0.2	t
	13705	6.115$_{-0.042}^{+0.025}$	1.9$_{-0.7}^{+0.5}$	t	t	1.2 ± 0.3	t
	14339	6.131$_{-0.023}^{+0.017}$	2.2$_{-0.5}^{+0.2}$	t	t	1.3 ± 0.2	t
	13706	6.155$_{-0.017}^{+0.012}$	2.7$_{-0.5}^{+0.3}$	t	t	1.5 ± 0.2	t
IV	03798	6.161$_{-0.021}^{+0.015}$	3.0$_{-0.6}^{+0.3}$	1.7 ± 0.4	1.5 ± 0.4	1.9$_{-0.3}^{+0.2}$	1.10/64(0.27)
	10059	6.148$_{-0.021}^{+0.016}$	2.7$_{-0.5}^{+0.6}$	t	1.1$_{-0.3}^{+0.4}$	1.6$_{-0.2}^{+0.3}$	t
	13225	6.134$_{-0.024}^{+0.015}$	2.2$_{-0.5}^{+0.2}$	t	0.7$_{-0.2}^{+0.3}$	1.2 ± 0.2	t
	13252	6.146$_{-0.021}^{+0.015}$	2.7$_{-0.5}^{+0.3}$	t	1.0$_{-0.3}^{+0.4}$	1.5 ± 0.2	t
	13705	6.121$_{-0.043}^{+0.025}$	1.7 ± 1.0	t	1.3 ± 0.9	1.2 ± 0.5	t
	14339	6.142$_{-0.019}^{+0.014}$	2.4$_{-0.5}^{+0.3}$	t	0.7$_{-0.3}^{+0.4}$	1.3 ± 0.2	t
	13706	6.154$_{-0.021}^{+0.014}$	2.7$_{-0.5}^{+0.3}$	t	1.1$_{-0.2}^{+0.3}$	1.6 ± 0.2	t

Notes. In Trial I both components are constrained to have the same values between observations. In Trials II and III one of the components may vary between observations, while in Trial IV both components are free. We use a "t" whenever values of a parameter are tied between observations. kT is the NS surface temperature in the star's frame, Γ is the power-law photon index, F_{X, NS} is the unabsorbed flux from the NS atmosphere component, and F_{X, PL} is the unabsorbed flux from the power-law component. Uncertainties are 90% confidence intervals. nhp is the null hypothesis probability (otherwise known as the p-value).

Download table as:  ASCIITypeset image

Trial I gives a poor fit, with χ² = 116.3 for 76 D.O.F. An F-test confirms the improvement from allowing the power-law flux to vary (Trial I to II), giving an F-statistic of 6.0 and probability of 4 × 10⁻⁵ of obtaining such an improvement by chance. Alternatively, allowing the NS temperature to vary (Trial III) also gives an improvement compared to Trial I (F-statistic of 4.2, chance improvement probability of 1 × 10⁻³). Letting both components vary is a substantial improvement compared to allowing the NS temperature alone to vary (comparing III to IV, F-statistic=2.27, chance improvement probability of 4.7 × 10⁻²), while letting both components vary is not preferred over allowing the power-law component alone to vary (comparing II to IV, F-statistic=0.89, chance improvement probability 0.50). Thus, we identify clear variation in the nonthermal component, but no evidence for variation in the thermal component.

3.3.5. Rise of the Outburst

We fit the Swift/XRT spectra from the rise of the outburst with a two-component model including a thermal component (BBODYRAD in XSPEC) and a nonthermal component (PEGPWRLW in XSPEC). We found good fits permitting only the relative normalizations of the thermal and nonthermal components to vary, with a photon index tied between observations ($\Gamma =1.1^{+0.2}_{-0.4}$) and a blackbody radius tied between observations ($R=4.3^{+1.4}_{-1.2}$ km). When the power-law index is left free between observations, the values are consistent, though they are poorly constrained in several spectra. Comparing our two-component model fits (Table 4) to our power-law fits (Table 2), a clear improvement in the fit is seen. Simultaneous fits to the first five Swift/XRT spectra of the outburst (listed in Table 4) with an absorbed power-law (with the photon index free between observations) give a reduced χ² of 1.32 for 45 degrees of freedom, while fits with an absorbed power-law plus blackbody give a reduced χ² of 1.09 for 43 degrees of freedom. An F-test gives an F-statistic of 5.55 and chance improvement probability of 0.007, supporting the addition of the thermal component. Protassov et al. (2002) showed that the F-test is often inaccurate for testing the necessity of adding an additional spectral component. We therefore chose the spectrum with the clearest evidence of a thermal component (ObsID 32148004), which shows a Δχ² of 6.1 between the power-law and power-law plus thermal spectral fits (going from 12 degrees of freedom to 11). We simulated 1000 data sets using a best-fit absorbed power-law model, and fit them both with a power-law model and with a power-law plus thermal component model. None of our simulations showed a larger Δχ² than that produced by our model, allowing us to conclude that the probability of incorrectly concluding that a thermal component is required is less than 99.5%.

Table 4. Results of Spectral Analyses for the Rise of the Outburst with a Two-Component Model: Thermal (BBODYRAD in XSPEC) Plus Nonthermal (PEGPWRLW in XSPEC)

Obs. ID	MJD	kT	FX, BB (0.5–10 keV)	FX, PL (0.5–10 keV)	FX, PL/FX, total	LX, total(0.5–10 keV)	$\chi _\nu ^2$/D.O.F.
		(keV)	(10−12 erg s−1 cm−2)	(10−12 erg s−1 cm−2)		(1034 erg s−1)
91445006	56114.8	0.31 ± 0.03	5 ± 2	5 ± 2	50 ± 20%	4 ± 1	0.53/6
32148003	56115.8	0.36 ± 0.03	9 ± 3	13 ± 4	59$_{-16}^{+15} \%$	9 ± 2	0.68/5
32148004	56117	0.41 ± 0.02	15$_{-3}^{+4}$	17 ± 6	53$_{-16}^{+12} \%$	13 ± 3	1.19/9
32148005	56118.1	0.44$_{-0.07}^{+0.05}$	20 ± 10	70 ± 20	78$_{-15}^{+12} \%$	37 ± 9	0.55/6
32148006	56120.7	0.67 ± 0.06	110 ± 40	500$_{-70}^{+60}$	82$_{-8}^{+7} \%$	250 ± 30	1.39/19

Notes. F_{X, BB} and F_{X, PL} are the blackbody and power-law unabsorbed fluxes, respectively. The power-law photon index was tied between spectra and was found to be 1.1$_{-0.4}^{+0.2}$. The thermal component normalization (which is proportional to blackbody radius) is assumed constant and is tied between observations. Uncertainties are 90% confidence intervals. $\chi ^2_\nu$ and degrees of freedom in this table are found by fitting each data set individually based on values found in simultaneous fit.

Download table as:  ASCIITypeset image

This indicates that the hardening during the outburst rise is likely caused by the decreasing relative contribution of a thermal component. With increasing time, and thus with increasing L_x, the fractional contribution of the thermal component decreases, but its kT increases monotonically. In the next section, we suggest that the thermal component is due to low-level accretion onto the surface of the NS.

We observed clear evidence of hardening of the spectrum during the outburst rise from L_X ∼ 4 × 10³⁴ up to 10³⁶ erg s⁻¹. We have evidence that this hardening is due to the relative reduction in strength of a thermal component in the spectrum with increasing brightness. This is the first time that such hardening during the outburst rise has been detected, made possible by our program of Swift/XRT globular cluster monitoring allowing early detection of the outburst below L_X = 10³⁵ erg s⁻¹. The trend of inferred photon index (for a fit to a power-law model) versus L_X is clear from the data in the rise, and is consistent with the data in the decay (which are not well-sampled below L_X = 10³⁵ erg s⁻¹); see Figure 5.

Softening during outburst decays has been seen from other (likely) NS LMXBs, in the L_X range of 10³⁴ − 10^35.5 erg s⁻¹, especially when the soft (<2 keV) X-ray energy range is included (Jonker et al. 2003, 2004; Cackett et al. 2011; Fridriksson et al. 2011; Armas Padilla et al. 2011, 2013c). RXTE observations have shown marginal softening during the decay of Aql X-1 down to L_X = 5 × 10³⁴ erg s⁻¹ (Maitra & Bailyn 2004), only in the part of the spectrum below 6 keV. RXTE observations of SAX J1808.4-3658 showed almost no spectral changes from L_X ∼ 2 × 10³⁶ down to 2 × 10³⁴ erg s⁻¹ (Gilfanov et al. 1998). These apparently contrasting observations are consistent if the softening is due to the increasing importance of the thermal component at lower L_X. Evidence in favor of an increasing relative thermal component can also be seen in Swift/XRT spectra of SAX J1808.4-3658 declining from L_X ∼ 10³⁶ down to 10³³ erg s⁻¹ (Campana et al. 2008). Thus, we interpret the hardening we observe in Terzan 5 X-3's rise as due primarily to the decreasing importance of a thermal component, rather than to the same physics responsible for the softening of black hole LMXBs during their decay, which show a steepening power-law spectrum (e.g., Plotkin et al. 2013).

Comparing the spectra observed from Terzan 5 X-3 to those of other NS LMXBs, we find a common pattern, that below L_X ∼ 1–3 × 10³⁵ erg s⁻¹ a thermal component is often required. For instance, Armas Padilla et al. (2013b) find, using XMM-Newton spectra, that two LMXBs at L_X = 1–10 × 10³⁴ erg s⁻¹ require a strong thermal component, while this is not critical for another LMXB at L_X ∼ 10³⁵ erg s⁻¹. Wijnands et al. (2002b), using Chandra  find that SAX J1747.0-2853, at L_X ∼ 3 × 10³⁵ erg s⁻¹, does not need a thermal component. Armas Padilla et al. (2013c) measure a thermal component to comprise ∼20% of the 0.5–10 keV luminosity for a transient at L_X ∼ 9 × 10³⁴ erg s⁻¹ (using XMM), with no evidence (from poorer Swift/XRT spectra) for a thermal component above L_X = 2.6 × 10³⁵ erg s⁻¹. Jonker et al. (2003, 2004) study the return to quiescence of XTE J1709-267, finding a thermal component to comprise ∼40% of the flux at L_X ∼ 4 × 10³⁴ erg s⁻¹, increasing to >90% at 2 × 10³³ erg s⁻¹. All these results suggest that there is a physical transition operating around L_X ∼ 10³⁵ erg s⁻¹ which changes the energy spectra.

Such a transition can be provided by the declining optical depth of a hot Comptonizing atmosphere, as seen in numerical calculations of NSs accreting at low rates (Deufel et al. 2001; Popham & Sunyaev 2001). Deufel et al. 2001 show temperature profiles and emergent spectra for NSs illuminated by high-temperature protons, such as are produced by radiatively inefficient accretion flows. In their Figures 4 and 5 they show that the emergent spectrum is a featureless Comptonized spectrum extending to ∼100 keV above L_X ∼ 10³⁶ erg s⁻¹, which develops a clear 0.5 keV thermal component at ∼10³⁵ erg s⁻¹, and loses the Comptonized tail by 10³³ erg s⁻¹. This transition is a strikingly accurate match to our observations of Terzan 5 X-3's spectral variations, and to other NS transients discussed in the literature. However calculations of Deufel et al. 2001 underpredict the observed hard power-law components seen in many quiescent LMXBs at low L_X (<10³⁴ erg s⁻¹, including Terzan 5 X-3 in quiescence). This may arise from their not including the Comptonizing effects of the overlying accretion flow on the observed spectrum. Popham & Sunyaev (2001) compute solutions for a hot boundary layer, which becomes optically thin for accretion luminosities below ∼10³⁶ erg s⁻¹, suggesting that some additional Comptonization can be performed by the accretion flow.

The temperature of the thermal component increases monotonically with the total X-ray luminosity during the rise (Table 4), as expected if the thermal component during the rise is produced by accretion. A correlation of thermal component temperature with total luminosity has been suggested from comparisons of multiple sources (Armas Padilla et al. 2013c), but this measurement clearly confirms this correlation in a single source. Furthermore, heating of the NS crust by accretion during the outburst will give rise to a rapidly decaying surface temperature at the end of the outburst (e.g., Cackett et al. 2006b). This effect of an accretion-heated crust could be confused with changing thermal emission from low-level accretion onto the NS surface during the outburst decline, but is not an issue during the outburst rise.

The detection of a thermonuclear burst during this outburst showed that the accreting object is a NS. Furthermore, the timescale of this thermonuclear burst indicates that the accreted matter contains hydrogen, evidence that the donor is not a white dwarf. With this information, we are now capable of classifying 15 of the 18 known bright Galactic globular cluster LMXBs as either ultracompact (P_orb < 1 hr, accreting from a hydrogen-deficient and/or degenerate donor) or not ultracompact (accreting from a nondegenerate, H-rich star). Five are known to be ultracompact by direct detection of their orbital periods, and seven systems are known not to be ultracompact by direct measurement of their orbital periods. On the basis of X-ray burst behavior indicative of hydrogen burning (Galloway et al. 2008), another three systems can be identified as not ultracompact (Table 5). (4U 1722-30, in Terzan 2, has shown some evidence, by its persistent low-luminosity accretion and burst behavior, in favor of an ultracompact nature; in't Zand et al. 2007.) Thus, the fraction of observed bright globular cluster LMXBs that are ultracompact appears to be between 28 and 44% (for 5 or 8 of 18). This fraction is believed to be higher than in the rest of the Galaxy (Deutsch et al. 2000), but uncertainties in selection effects mean that we cannot confidently extrapolate the true underlying fraction of ultracompact systems and make clear comparisons to binary population synthesis models (e.g., Ivanova et al. 2008).

We have identified the quiescent counterpart to Terzan 5-X3 with the brightest previously suggested candidate quiescent LMXB in the cluster, CXOGLB J174805.4-244637 (Heinke et al. 2006b). Our spectral analysis reveals evidence for a variable power-law contribution to the quiescent spectrum over timescales of years, but exhibits no evidence for changes to the thermal component. It is fascinating to see clearly here that the quiescent spectral properties appear to lie on a continuum with the outburst properties, with increasing hardening from quiescence, through the early rise, up to the hard state at L_X > 10³⁶ erg s⁻¹ (Figure 5).

Chandra quiescent X-ray counterpart searches have now been performed for nine transient cluster LMXBs, of which the three with the faintest outbursts have been identified with very faint (L_X < 10³² erg s⁻¹) quiescent counterparts (M15 X-3, Heinke et al. 2009; NGC 6440 X-2, Heinke et al. 2010; IGR J17361-4441 in NGC 6388, Pooley et al. 2011b), two have spectrally hard counterparts with L_X ∼ 10³²–10³³ erg s⁻¹ (EXO 1745-248 in Terzan 5, Wijnands et al. 2005; IGR J18245-2452 in M28, Papitto et al. 2013; Linares et al. 2013), and four have spectrally soft counterparts with L_X ∼ 10³²–10³³ erg s⁻¹ (SAX J1748.9-2021 in NGC 6440, in't Zand et al. 2001; X1732-304 in Terzan 1, Cackett et al. 2006a; IGR J17480-2446 in Terzan 5, Degenaar & Wijnands 2011; and Terzan 5 X-3). These identifications support the idea that the faint soft X-ray sources identified as candidate quiescent LMXBs in globular clusters are indeed transient LMXBs, between (relatively bright) outbursts (Heinke et al. 2003c; Wijnands et al. 2013). The brightest of the faint soft X-ray sources should experience relatively high long-term average mass accretion rates, which will cause relatively large amounts of deep crustal heating and keep their cores warm. Such high mass accretion rates suggest frequent outbursts, and thus it is comforting that we have identified the brightest of the faint soft X-ray sources in the clusters NGC 6440, Terzan 5, and Terzan 1 with observed transients. The suggestion that roughly half the quiescent LMXBs in each cluster are easily identifiable in short Chandra observations by showing soft, primarily thermal X-ray spectra and X-ray luminosities between 10³² and 10³³ erg s⁻¹ (Heinke et al. 2005) continues to seem reasonable, though it remains unproven.

From the quiescent NS thermal bolometric luminosity (L_{NS, (0.01–10 keV)} = 1.5 × 10³³ erg s⁻¹, kT = 118 eV at the surface) of Terzan 5 X-3, and its outburst properties, we can make some general statements about its outburst history or neutrino cooling properties, assuming that the quiescent thermal flux is due to heat deposited in the core during multiple accretion episodes (Brown et al. 1998). We estimate the total mass transfer rate onto the NS during this outburst by converting the daily MAXI/GSC 4–10 keV flux estimates (in Crab units) into 0.1–12 keV fluxes (assuming a power-law with photon index set by the nearest Swift/XRT observations), converting the daily Swift/BAT flux estimates into 12–50 keV fluxes (using the same power-law photon index as for the MAXI data), adding these together, and assuming a 1.4 M_☉, 10 km NS. This gives us a total energy release over the outburst of 9 × 10⁴³ erg, and a total mass transfer of 2.4 × 10⁻¹⁰ M_☉.

If we assume "standard" modified Urca cooling (Yakovlev & Pethick 2004; Wijnands et al. 2013), then we can estimate (using the quiescent NS luminosity) a mass transfer rate onto the NS of $\dot{M}\sim 3\times 10^{-11}$ M_☉ yr⁻¹ (though this value might vary depending on the choice of crustal composition, e.g., whether a thick light-element layer is present; Page et al. 2004). Assuming this outburst is typical, we derive an average interval of ∼8 yr between outbursts. If the average interval between outbursts were much longer than 10 yr, then Terzan 5 X-3 would be brighter in quiescence than expected under even the slowest cooling processes. One could attribute its quiescent thermal luminosity to continued accretion, but our analysis of the quiescent observations identifies no evidence for variability in the thermal component, arguing against this explanation. All Terzan 5 X-ray outbursts since 1996 have been identified with their quiescent counterpart with arcsecond precision, except one in 2002 (Wijnands et al. 2002a). The 2002 outburst had an average luminosity of L_X ∼ 2 × 10³⁷ erg s⁻¹, peak L_X ∼ 4 × 10³⁷ erg s⁻¹, and lasted for ∼30 days (Degenaar & Wijnands 2012). The 2012 outburst of Terzan 5 X-3 had a similar average luminosity of L_X ∼ 2 × 10³⁷ erg s⁻¹, peak L_X = 7 × 10³⁷ erg s⁻¹, and lasted for 30 days above L_X ∼ 10³⁶ erg s⁻¹ (comparable to the RXTE/ASM detection limit for the 2002 outburst). We therefore suggest that the 2002 X-ray outburst is likely to have also been produced by Terzan 5 X-3. This would nicely fit the ∼8 yr recurrence time we inferred above.