NUSTAR AND XMM-NEWTON OBSERVATIONS OF THE EXTREME ULTRALUMINOUS X-RAY SOURCE NGC 5907 ULX1: A VANISHING ACT

Ultraluminous X-ray sources (ULXs) are off-nuclear point sources with X-ray luminosities that exceed the Eddington limit for the ∼10 M_☉ "stellar-mass" black holes observed in Galactic black hole binaries (BHBs; e.g., Orosz 2003), i.e., L_X > 10³⁹ erg s⁻¹. Multi-wavelength observations have largely excluded highly anisotropic emission that could artificially increase the estimated luminosity (e.g., Berghea et al. 2010; Moon et al. 2011). The observed luminosities therefore require either the presence of larger black holes than those observed in our own galaxy (e.g., Miller et al. 2004; Strohmayer & Mushotzky 2009; Zampieri & Roberts 2009), or super-Eddington modes of accretion (e.g., Poutanen et al. 2007; Finke & Böttcher 2007). For recent reviews focusing on ULXs, see Roberts (2007) and Feng & Soria (2011).

Although the majority of ULXs only have luminosities marginally in excess of 10³⁹ erg s⁻¹ (Walton et al. 2011b; Swartz et al. 2011), and therefore likely represent a high luminosity extension of the stellar mass BHB population (Middleton et al. 2013; Liu et al. 2013; Motch et al. 2014), a smaller population of extreme sources have observed X-ray luminosities of L_X > 10⁴⁰ erg s⁻¹ (e.g., Farrell et al. 2009; Walton et al. 2011a; Jonker et al. 2012). The extreme luminosities displayed by these sources are of substantial interest, and mean they remain among the best candidates to host black holes more massive than those observed in Galactic BHBs.

NGC 5907 ULX1 is a luminous member of this population of extreme ULXs. The source was initially reported in the ULX catalog presented in Walton et al. (2011b) with a peak X-ray luminosity of ∼5 × 10⁴⁰ erg s⁻¹ (see also Sutton et al. 2012). Although X-ray data on the edge-on spiral galaxy NGC 5907 is relatively sparse, since the discovery of ULX1 a number of follow-up observations have been undertaken with XMM-Newton, Chandra, and Swift, revealing the source to be variable by a factor of a few, confirming that a single, accretion powered source dominates the observed X-ray flux (Sutton et al. 2013a). Based on data with a bandpass limited to E < 10 keV, the observed characteristics of NGC 5907 ULX1 below 10 keV appear to be broadly consistent with a BHB in the sub-Eddington low/hard state (Sutton et al. 2012; see Remillard & McClintock 2006 for details on the standard BHB accretion states), implying the possible presence of a very massive black hole. However, the highest quality soft X-ray data available tentatively suggest the presence of a spectral break above ∼5 keV (Sutton et al. 2013a), which, if confirmed, would be inconsistent with this accretion regime, and potentially identify NGC 5907 ULX1 as a high-Eddington source (Gladstone et al. 2009).

The Nuclear Spectroscopy Telescope Array (NuSTAR; Harrison et al. 2013), in conjunction with XMM-Newton, Suzaku, and Chandra, has been providing the first ever high quality broadband X-ray spectra for a sample of known ULXs (see Bachetti et al. 2013, 2014; Rana et al. 2014; Walton et al. 2013a, 2014). Here we present results from the recent NuSTAR and XMM-Newton observations of NGC 5907 ULX1. The paper is structured as follows: Section 2 describes our observations and data reduction procedure, and Sections 3, 4, and 5 describe the analysis performed. Finally, we discuss our results and conclusions in Section 6.

During 2013 the NuSTAR and XMM-Newton X-ray observatories performed two coordinated observations of NGC 5907 ULX1, with some portion of the NuSTAR observation simultaneous with XMM-Newton in both cases (see Table 1 for details). Here, we outline our general data reduction procedure for these observations; details specific to each epoch are given in Sections 3 and 4.

2.1. NuSTAR

2.2. XMM-Newton

Based on the previously published observations, the X-ray emission from NGC 5907 ULX1 appears to be fairly persistent (Sutton et al. 2013a). However, during the first epoch reported here, no single point source obviously dominates the emission at the position of the ULX (Figure 1, left panel). Instead, the soft X-ray emission observed by XMM-Newton from the immediate vicinity of the ULX appears to be comprised of a series of faint sources (Figure 1, right panel), which even in combination do not result in a detection with NuSTAR. One of these sources is close to the known position of the ULX, and we therefore extract the XMM-Newton spectrum of this source from a circular region of radius 12'' to avoid contamination from the other sources. Fewer than 50 net source counts are detected in total (EPIC-pn+EPIC-MOS) in the 0.3–10.0 keV bandpass, so we rebin the data from this epoch to have a minimum of five counts per bin (before background subtraction) to maintain spectral coverage, and minimize the Cash statistic (Cash 1979) when analyzing these data. Here, the background contributes ∼45% of the total counts from the source region.

Zoom In Zoom Out Reset image size

With so few counts it is not possible to undertake detailed spectral modeling, but to estimate a source flux we model the spectrum with a simple absorbed powerlaw, assuming Γ = 1.7 based on previous XMM-Newton observations (Sutton et al. 2012). We find an observed 0.3–10.0 keV flux of $F_{\rm {Epoch1}}=11^{+5}_{-3}\times 10^{-15}$ erg cm⁻² s⁻¹ (throughout this work fluxes are estimated with CFLUX), substantially lower than any flux observed from NGC 5907 ULX1 to date (Sutton et al. 2013a). For a distance to NGC 5907 of 13.4 Mpc (Tully et al. 2009), the corresponding 0.3–10.0 keV luminosity is $2.4^{+1.1}_{-0.7}\times 10^{38}$ erg s⁻¹.

We find that NGC 5907 ULX1 is very clearly detected by both XMM-Newton and NuSTAR in the second epoch (see Figure 1), suggesting a remarkable flux transition in a fairly short space of time (the end of the first NuSTAR observation and the start of the second are separated by ∼4 days). We extracted spectra using a circular region of radius 21'' for XMM-Newton, chosen to simultaneously maximize the signal-to-noise for the ULX and minimize the contamination from the fainter sources nearby, and of radius 50'' for NuSTAR, given its larger point-spread function (PSF). In order to improve the statistics at the highest energies, we combine the spectra obtained by FPMA and FPMB with ADDASCASPEC (after confirming their consistency), and rebin all the spectra from this epoch to a minimum of 25 counts per bin, minimizing χ² in our analysis of these data. We obtain a detection in NuSTAR up to just over 20 keV.

Figure 2 shows the broadband 0.3–25.0 keV spectrum obtained from epoch 2. The NuSTAR data confirm the presence of a spectral break in the ∼3–10 keV bandpass, similar to other ULXs observed by NuSTAR to date (Bachetti et al. 2013; Rana et al. 2014; Walton et al. 2013a, 2014). As the data for NGC 5907 ULX1 from this epoch are of much lower quality compared to the broadband observations of those other ULX targets, we limit our spectral analysis in this work to simple continuum modeling of the time averaged spectrum; the count rates obtained unfortunately do not permit detailed variability studies of these data (0.12 and 0.04 ct s⁻¹ in 0.3–10.0 keV from EPIC-pn and each EPIC-MOS detector, and 0.008 ct s⁻¹ in 3–25 keV from each NuSTAR FPM). Both Galactic and intrinsic neutral absorption are included in all models. In addition, given the larger extraction region used for the NuSTAR data, we also investigate whether the results presented below are influenced by undetected contamination from the other sources seen in the epoch 1 XMM-Newton data (Figure 1), and we repeat our analysis using the epoch 1 NuSTAR spectrum extracted from the epoch 2 source region as an alternative background spectrum. We obtain consistent results with both background estimations, and we therefore only present those obtained with the epoch 2 background for simplicity.

Zoom In Zoom Out Reset image size

We first fit the broadband spectrum with an absorbed powerlaw model, but find that this provides a poor fit ($\chi ^{2}_{\nu }$ =χ²/DoF = 403/277; see Table 2), and results in systematic residuals across the whole bandpass (see Figure 2, right panel). In particular, the model overpredicts the highest energies, confirming the existence of spectral curvature above ∼3 keV. We also find that a model consisting of a geometrically thin, optically thick accretion disk (DISKBB; Mitsuda et al. 1984; see also Shakura & Sunyaev 1973) provides a poor fit ($\chi ^{2}_{\nu }$ =393/277), clearly underestimating the high energy flux. However, a slightly more complex accretion disk model in which the radial temperature index (p) is free to vary (DISKPBB; Mineshige et al. 1994) does provides a good fit across the full bandpass ($\chi ^{2}_{\nu }$ =286/276), and there is no obvious requirement for any additional spectral components. The radial temperature index of p = 0.55 ± 0.02 obtained (see Table 2) is shallower than expected for a thin accretion disk (p_thin = 0.75), consistent with an accretion disk in which strong advection of radiation occurs, as may be expected at very high accretion rates where radiation pressure should dominate and modify the disk structure (e.g., Abramowicz et al. 1988).

Based on the DISKPBB model we find an observed 0.3–20.0 keV X-ray flux for NGC 5907 ULX1 during the second epoch of F_Epoch2 = (7.2 ± 0.3) × 10⁻¹³ erg cm⁻² s⁻¹, ∼87% of which falls in the 0.3–10.0 keV bandpass. The observed fluxes of the various data sets considered in this work are summarized in Table 3. This equates to a broadband X-ray luminosity of (1.55 ± 0.06) × 10⁴⁰ erg s⁻¹ even before any absorption correction. While extreme for the ULX population, this is still a factor of ∼2–3 lower than the peak luminosity of ∼5 × 10⁴⁰ erg s⁻¹ (0.3–10.0 keV) that has been observed from this source (Sutton et al. 2013a).

Table 3. Observed Fluxes from the Data Sets Analyzed in This Work

Data Set	0.3–10.0 keV Flux	0.3–20.0 keV Flux
	(10−15 erg cm−2 s−1)
Epoch 1a	$11^{+5}_{-3}$	⋅⋅⋅
Epoch 2	$630^{+30}_{-20}$	720 ± 30
2013 Swift	$30^{+20}_{-10}$	⋅⋅⋅
Source S (Chandra)	7 ± 3	⋅⋅⋅

Notes. The Chandra flux for Source S is also listed for comparison. ^aLikely dominated by emission from Source S, see Section 5.1.

Download table as:  ASCIITypeset image

5.1. Chandra Imaging and Astrometry

5.2. 2013 Swift Snapshots

We have presented an analysis of two epochs of broadband X-ray observations of the extreme ULX NGC 5907 ULX1 undertaken with NuSTAR and XMM-Newton. These observations reveal an astonishing level of X-ray flux variability between the two epochs. Although the source is not detected by NuSTAR in the first epoch, a clear detection is obtained in the second, providing the first constraints on the hard X-ray (E > 10 keV) emission. Broadly similar to the results obtained for other ULXs observed by NuSTAR to date (Bachetti et al. 2013; Rana et al. 2014; Walton et al. 2013a, 2014; E. S. Mukherjee et al., in preparation), we find the hard X-ray emission from NGC 5907 ULX1 to be very weak relative to that at lower energies (see Figure 2). During this epoch the broadband spectrum is not consistent with a ∼10^3–4 M_☉ intermediate mass black hole accreting in the low/hard state displayed by Galactic BHBs. Instead, the spectrum is well modeled with an advection dominated accretion disk (e.g., Abramowicz et al. 1988), confirming the spectral cutoff tentatively suggested by the archival XMM-Newton data (Sutton et al. 2013a), and implying that NGC 5907 ULX1 may be accreting at a very high, possibly super-Eddington rate.

The most intriguing aspect of these observations is the extreme level of flux variability. We show in Figure 3 a long-term X-ray light curve for NGC 5907 ULX1, adapted from Sutton et al. (2013a) to include our XMM-Newton and NuSTAR observations (and also the additional 2013 Swift data). Even assuming the faint source detected by XMM-Newton in epoch 1 is NGC 5907 ULX1, its observed flux varied by ∼2 orders of magnitude in ∼4 days (the flux appears stable throughout epoch 2, see the inset in Figure 3). However, given the flux observed by Chandra and the astrometric offset between the epoch 1 and epoch 2 XMM-Newton observations, the emission detected in epoch 1 is likely dominated by Source S (see Section 5.1), implying an even more extreme variation from ULX1. While there are transient ULXs that show orders of magnitude of variation (similar to low mass X-ray binaries, e.g., Middleton et al. 2013), even at peak luminosity these tend to be the fainter members of the ULX population. The brighter ULXs tend to be variable only by a factor of ∼a few, broadly similar to high mass X-ray binaries, (e.g., Kaaret et al. 2009; Kaaret & Feng 2009; Kong et al. 2010; Walton et al. 2013a). Until this work, the behavior observed from NGC 5907 ULX1 was similar to this latter population, consistent with its extreme luminosity.

Zoom In Zoom Out Reset image size

Given the relatively persistent behavior observed previously one tantalizing possibility for the extreme low state observed in epoch 1 is that it may represent an eclipse of the X-ray source, perhaps by a stellar companion or by a warped outer accretion disk. If this were the case, the lack of a hard X-ray detection with NuSTAR suggests the medium eclipsing NGC 5907 ULX1 is extremely thick (N_H ∼ 10²⁴ cm⁻² or more). However, the duration of the NuSTAR observation is ∼2 days (the low-earth orbit results in a ∼50% observing efficiency; Harrison et al. 2013), and we do not know how long this low flux persisted before the NuSTAR observation, thus if this were an eclipse by the companion star the orbital period would have to be ≫2 days. For comparison, the orbital period of the eclipsing black hole binary IC 10 X-1, the most massive dynamically constrained stellar remnant measured to date, is ∼34 hr (Prestwich et al. 2007; Silverman & Filippenko 2008), and the X-ray eclipses last ∼30% of this period, significantly shorter than the NuSTAR observation. The other eclipsing BHB system known, M 33 X-7, has an orbital period of ∼3.5 days, with the X-ray eclipses lasting ∼15% of this (Pietsch et al. 2006; Orosz et al. 2007). Nevertheless, GRS 1915+105, one of the Galactic sources widely considered most analogous to ULXs, has a much longer orbital period (∼34 days; Steeghs et al. 2013), and several works have suggested that ULXs might plausibly have very long orbital periods (up to ∼100 days or more; Podsiadlowski et al. 2003; Pooley & Rappaport 2005; Madhusudhan et al. 2008) if they accrete from evolved stellar companions via Roche-lobe overflow, as suggested by the lack of iron emission (Walton et al. 2012, 2013b). If the eclipse is by some other material, e.g., the outer regions of a warped accretion disk rather than the stellar companion, this may precess on long, super-orbital timescales, and would ease the requirement for a long orbit. A scenario along these lines was proposed as a potential explanation for the rare (and generally less extreme) dips observed from the well studied, soft-spectrum ULX NGC 5408 X-1 (lasting up to a few days; Grisé et al. 2013). This behavior is perhaps analogous to the "dipping" low-mass X-ray binaries (e.g., Díaz Trigo et al. 2009).

The Swift observations earlier in 2013 represent an additional period of low flux that appears to last ∼6 weeks (although we note the infrequent observing cadence during this period). However, the flux during epoch 1 is fainter than that from the combination of these Swift observations though, and likely substantially so given that the epoch 1 emission is dominated by Source S. It may therefore be possible that we are observing the effect of an eclipse imprinted on top of a strong level of intrinsic variability. While it may not be possible to conclusively rule out an explanation along these lines with the current data, we do not consider this scenario to be particularly likely. Nevertheless, it is interesting to note that if this were the case, it would imply that we are viewing NGC 5907 ULX1 at a high inclination. This is at odds with the suggested framework of the wind dominated "ultraluminous state" for super-Eddington accretion in which hard spectrum ULXs similar to NGC 5907 ULX1 are viewed close to face on, such that the hot inner regions of the accretion flow are visible. For edge-on sources, the inner regions would instead be obscured by cooler material in a large scale-height wind launched from the disk, resulting in a soft X-ray spectrum (Sutton et al. 2013b; Middleton et al. 2014), inconsistent with that observed. If the variability does result from a high inclination, this would thus favor the "patchy disk" scenario suggested by Miller et al. (2014).

If the extreme rise in flux is not related to the end of an obscuration event, it must instead be caused by a rapid increase in the mass accretion rate onto NGC 5907 ULX1. Such rapid increases in flux are occasionally seen in Galactic X-ray binaries, and appear in some instances to be related to increased accretion, rather than obscuration (e.g., XTE J1701-407: Degenaar et al. 2011; Pawar et al. 2013, 4U 1700-377: Smith et al. 2012). In the most extreme cases, super-fast X-ray transients can flare by orders of magnitude in an extremely short period of time (hours or days, e.g., Sidoli et al. 2009), although these are very short-lived flare events at much lower luminosities which differ markedly from the behavior of NGC 5907 ULX1.

One scenario that can result in rapid rises in accretion rate is for the system to have a highly eccentric orbit, with large accretion bursts triggered close to periastron, broadly similar to Be/X-ray binaries (e.g., Reig 2011; Casares et al. 2014). This scenario is currently the leading interpretation for the almost periodic outbursts exhibited by the most extreme ULX observed to date, ESO 243-49 HLX1 (L_{X, peak} ∼ 10⁴² erg s⁻¹), which displays a repeated "fast-rise-exponential-decay" (FRED) outburst profile with similarly large and rapid flux increases to those seen in our observations of NGC 5907 ULX1 (e.g., Lasota et al. 2011; Webb et al. 2014, although see King & Lasota 2014). Given the poor sampling to date of the long-term behavior of NGC 5907 ULX1, it is possible that something similar could be occurring in this case, albeit at a lower absolute luminosity, and with the source rising into a potentially super-Eddington state (in contrast, HLX-1 seems to exhibit the standard evolution shown by sub-Eddington Galactic binaries during its outbursts; Servillat et al. 2011; Godet et al. 2012). Thus it may be possible for the extreme variability observed to be related to the orbit of the system, without being associated with an eclipse event. In light of these observations, extended monitoring of this remarkable source is strongly recommended in order to test this exciting possibility.

The authors would like to thank the reviewer for positive and useful feedback, which helped improve the manuscript, as well as Diego Altamarino for useful discussions. M.B. and D.B. acknowledge financial support from the French Space Agency (CNES). This research has made use of data obtained with the NuSTAR mission, a project led by the California Institute of Technology (Caltech), managed by the Jet Propulsion Laboratory (JPL) and funded by NASA, and has utilized the NuSTAR Data Analysis Software (NUSTARDAS) jointly developed by the ASI Science Data Center (ASDC, Italy) and Caltech (USA). This research has also made use of data obtained with XMM-Newton, an ESA science mission with instruments and contributions directly funded by ESA Member States and NASA, and NASA's Chandra and Swift satellites.

Facilites: NuSTAR - The NuSTAR (Nuclear Spectroscopic Telescope Array) mission, XMM - Newton X-Ray Multimirror Mission satellite, Chandra - , Swift - Swift Gamma-Ray Burst Mission