THE FIRST DETECTION OF [O iv] FROM AN ULTRALUMINOUS X-RAY SOURCE WITH SPITZER. I. OBSERVATIONAL RESULTS FOR HOLMBERG II ULX

The data presented here are archival Spitzer data, taken on 2004 October 4, as part of the Legacy Program Spitzer Infrared Nearby Galaxies Survey (SINGS Program 193; see Kennicutt et al. 2003). The IRS spectral map is described by three parallel pointings and five perpendicular at half-slit steps, for both Short High (SH, in the range 9.9–19.6 μm) and Long High (LH, in the range 18.7–37.2 μm) modules. The total exposure time was 240 s for both SH and LH observations. The data were preprocessed by the IRS pipeline (ver. 11) at the Spitzer Science Center prior to download. Spectral maps were constructed using "BCD-level" processed data in conjunction with CUBISM (CUbe Builder for IRS Spectra Maps) version 1.5⁵ (Smith et al. 2007; Kennicutt et al. 2003). The absolute calibration error on fluxes obtained with CUBISM is ∼25% (Smith et al. 2004).

We adopt a distance of 3.05 Mpc to Holmberg II (KWZ; 1'' is 15 pc). The LH maps obtained with CUBISM have intrinsic sizes of 446 × 288 (670 pc × 430 pc), while the SH maps are 27'' × 136 (405 pc × 205 pc). However, in order to avoid aperture effects, the spectra presented here were extracted from like-sized apertures. Thus, the SH and LH spectral data result from an equally sized aperture of 23'' × 136 (340 pc × 205 pc), which is only slightly smaller than the full SH aperture. We also extract spectra from the smallest aperture permitted by CUBISM: an 89 × 89 aperture centered on the ULX (i.e., 2 × 2 LH pixels, or 130 pc × 130 pc). We will refer to these apertures as the standard aperture and the small aperture, respectively.

The high-resolution spectra (10–37 μm) for the two apertures are plotted in Figure 1. Since the spectra are extracted from matched apertures, the spectral lines in both the SH and LH modules are properly calibrated and the continuum is properly matched. In Figure 2, we show in detail the four detected lines, with the theoretical line centers. We used SMART (Spectroscopic Modeling Analysis and Reduction Tool; see Higdon et al. 2004) version 6.2.6 to measure line fluxes. Only four lines were detected, the fluxes for these and upper limits for nondetections are presented in Table 1. Here, we show the results for both the standard aperture and the small aperture. We note that the [O iv] line is detected at a signal-to-noise ratio (S/N) > 10. The spectral lines for the Holmberg II galaxy are also presented in Dale et al. (2009). The focus of that paper was not the ULX in Holmberg II, and as a result, the extraction region for the relevant spectral lines is larger than the apertures presented here. However, with this in mind, we find that our line fluxes agree reasonably well with those presented in Dale et al. (2009).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The SH and LH maps for the four detected lines are shown in Figure 3. A false-color Hubble Space Telescope (HST) image shows the Foot Nebula in Hα emission, together with the standard and small extraction apertures. Because of the poor spatial resolution of the Spitzer spectral maps, the emission for most lines is difficult to disentangle from the nearby star-forming regions in the Foot Nebula. This is especially true of the low-ionization lines ([Ne iii], [S iii], and [Si ii]). The morphologies of these maps indicate that the emission is likely coming from both the ULX and the surrounding star-forming regions. However, while the low-ionization lines [Si ii] and [S iii] are very extended, the [O iv] emission is visibly compact. The peak of the of the [O iv] emission is coincident with the Chandra position for the ULX, with the majority of the emission concentrated in the 2 × 2 pixel region, overlapping with the X-ray point source. There is a slight extension to the west and south that is visible on the spectral map and which may imply that the [O iv] emission is extended. These faint extensions correspond very well with two regions of young star formation visible in the HST image in Figure 3.

Zoom In Zoom Out Reset image size

KWZ found that the He ii line, which has an ionization potential just slightly lower than [O iv] (54.4 eV, compared to 54.9 eV), is also concentrated in a small region of 50 pc diameter around the identified optical counterpart of the ULX. Because the spatial resolution is limited, a detailed comparison of our maps with the optical maps presented by KWZ cannot be made. However, we find that the morphology of our [O iv] map appears similar to He ii, being brighter in the east and fainter and more extended in the west.

We use X-ray spectroscopy to model the X-ray emission from the ULX. We also use it to estimate the UV emission by extrapolating the model to lower energies. X-ray observations of the Holmberg II galaxy are available from ROSAT, ASCA, XMM-Newton, and Chandra (Zezas et al. 1999; Miyaji et al. 2001; Kerp et al. 2002; Kaaret et al. 2004; Goad et al. 2006). The X-ray luminosity of the ULX is in excess of 10⁴⁰ erg s⁻¹ in all observations, with the exception of a peculiar low and soft state in 2002 September, when the luminosity dropped to ∼5 × 10³⁹ erg s⁻¹ (Dewangan et al. 2004).

To model the X-ray spectrum, we used the best available data from the longest (∼100 ks) XMM-Newton EPIC PN observation in the archive (Obs. ID 0200470101, taken on 2003 December 5). The data were processed by Winter et al. (2006). We used XSPEC version 12.4 to model the spectrum between 0.3 and 8.0 keV. The X-ray spectrum is the most important component of the total SED since it is the only observed emission that directly emanates from the ULX, and is not contaminated by the surrounding star-forming region. For this reason, we have taken many precautions in modeling the X-ray spectrum.

The (far) UV emission from the ULX is the least known, but the most important for ionization. The only available method to estimate the shape of the UV is to extrapolate from X-ray models. Therefore, it is very important to find the right X-ray model to fit the data, especially in the soft X-ray band. Three physically different models were used. Each model contains different information about the accretion process. The first model is very basic and is widely used to model the accretion process for stellar-mass black holes, ULXs, and other accreting objects (Miller et al. 2004; McClintock & Remillard 2006; Winter et al. 2006; Feng & Kaaret 2006). The second is a modification of the first model that is more physically realistic when extrapolated to lower energies. Finally, the third model is more flexible than the other two models in the UV and allows for a tighter fit in the UV while staying within the X-ray error bars. The intrinsic (unabsorbed) X-ray luminosity is >10⁴⁰ erg s⁻¹ for all models. The results from the X-ray spectral fits are listed in Table 2, and a summary of the models is as follows:

• 1.  
  Power-law Plus Multicolor Disk (PLMCD) model. This is a two-component model with a multi-color accretion disk component (MCD; Mitsuda, et al. 1984) and a power law for the hard tail. Following KWZ, the absorption column density was fixed at the Galactic value of 3.42 × 10²⁰ cm⁻², and an additional local absorption (N_H) model was added. We assume a local metallicity of 0.1 Z_☉ (see Kaaret et al. 2004; Pilyugin et al. 2004). This is a very simplistic model commonly used to fit X-ray data for ULXs. However, the model falls apart when extrapolated to lower energies, where the power law dominates. The extrapolation is actually physically unrealistic, since it is impossible for the disk component to Comptonize lower energy photons. Done & Kubota (2006) pointed out that the power-law component in such cases needs to have a break at low energies, in agreement with the theoretical model where the photons from the disk are upscattered in the hot corona. Moreover, if the power law is produced by another mechanism (e.g., a jet), it would not extend down to the soft UV and optical spectrum because the fluxes here would be over predicted (see below, Section 4). Nevertheless, the model continues to be used extensively in X-ray spectral fitting of ULXs. In conclusion, this model does fit the X-ray data well (Δχ²/dof = 1.14), but it does not make physical sense when extrapolated to lower energies. It is shown here as a base, from which alterations can be made, resulting in more physically meaningful models. The following model is one such modification to the PLMCD model, and is more physically realistic when extrapolated into the UV.
• 2.  
  Modified PLMCD model. In order to isolate the intrinsic disk emission from the Comptonized disk emission, we fitted the MCD and power law separately for two different regions of the X-ray spectrum. We first fitted the hard tail with a power law. The spectrum departs from this power law around ∼1 keV. Then, we fitted an MCD model to the 0.3–1.0 keV band. The disk temperature (0.38 keV) is slightly higher than other published fits (KWZ obtained ∼0.2 keV), because our fitting methods are slightly different. Lower temperature accretion disks are generally more effective ionizing sources. Therefore, this higher temperature model may be viewed as a lower bound to the SED in the FUV and soft-X-ray bands. We note that the Modified PLMCD differs from the PLMCD model in that the UV band is predicted from the disk component rather than the extrapolated PL component. This difference yields a much more physically realistic interpretation for the emission.
• 3.  
  Broken power-law (BPL) model. Finally, we also fitted the X-ray spectrum to a simple BPL model. Kaaret et al. (2003) fit a similar model to the ULX in NGC 5408. Such a model could be explained if the emission from the companion star were Comptonized by a jet (Georganopoulos et al. 2002). This model is much more flexible in the UV than either the PLMCD or the Modified PLMCD models, and can predict the slope of the SED in the UV while simultaneously fitting the hard power-law component in the X-rays within the error bars of the X-ray measurements. Surprisingly, this simple model provides a very good fit to the X-ray data, better than the PLMCD model. The break energy is at 1.0 keV, the same energy at which we split the two components in the Modified PLMCD model (2 above). Finally, we note that, even if this model has a power law less steep than the PLMCD model at lower energies, it is still overpredicting the soft UV and optical emission according to our photometry results (Section 4). A similar overprediction at lower energies was found for this model by Done & Kubota (2006) for the Galactic binary XTE J1550−564 when at high accretion rates (in the very high state).

In order to constrain ULX multi-wavelength SED, we performed photometry in the IR and UV using Spitzer, XMM-Newton, and GALEX data. We also incorporated previously published optical and radio data from HST and the Very Large Array (VLA). A short description of the data and the measurements are shown in Table 3. All our photometric measurements were extracted from an aperture of 50 pc diameter (33). We used the optical counterpart identified by KWZ to center the extraction aperture (see Figure 3). Aspect correction for the IR and UV images was performed using bright sources in field, such as the foreground star seen in Figure 3. The measurements presented in Table 3 represent the integrated fluxes over this region, including the nebular emission and the contamination from other nearby sources. They should therefore be considered upper bounds to the intrinsic ULX contribution.

IR photometry. IRAC and MIPS images (see Figure 4) were taken as a part of SINGS Program 195. We extracted 3.6 μm and 8.0 μm fluxes using 50 pc diameter apertures centered on the ULX position. For the MIPS 24 μm photometry, we used a 100 pc extraction aperture and then divided the flux by four to obtain an approximate flux for an equivalent 50 pc aperture. We followed the standard extraction procedure and conversion factors provided in the IRAC Data Handbook.⁶ The pipeline mosaics for IRAC and MIPS are accurate to within 20%.

Zoom In Zoom Out Reset image size

UV photometry. XMM-Newton OM images of Holmberg II (Figure 4) are available in the UVW1 band, which is centered at 2910 Å (FWHM 620 Å). We use the mosaicked image created for the Optical Monitor Catalog (OMCat; Kuntz et al. 2008), which has an improved coordinate correction (better than 05) over the automated pipeline. To measure the UVW1 count rate, we used the XIMAGE software included in the HEAsoft package. We converted the count rate into flux using the conversion factors listed in the XMM-Newton Science Analysis System User's Guide.⁷ We obtained a UVW1 magnitude of 18.42 ± 0.06, which corresponds to a luminosity of 8.3 × 10³⁷ erg s⁻¹ and a flux density of 33.1 ± 0.1 μJy. The OM image in Figure 4 shows that the Foot Nebula has a strong UV emission which peaks on the OB association of stars seen in the HST image of Figure 3 west of the ULX. The total UVW1 flux for this UV bright region (including the ULX and its optical counterpart) is equivalent to 29 O5V stars. Figure 3 shows that these bright stars have apparently blown a cavity in the NW part of the nebula, thus producing the "Sole" and the "Toes" of the Foot Nebula. We note that these cavities are correlated with the faint extended [O iv] emission seen in the CUBISM spectral maps (Figure 3). The strongest emission of Hα is, however, in the Heel that hosts the ULX.

We also used archived GALEX observations, taken as a part of the GALEX Nearby Galaxies Survey (NGS; Bianchi et al. 2003). Photometric fluxes were extracted from the FUV (1350–2013 Å) and NUV (1750–2013 Å) bands with the XIMAGE tool. The GALEX images have a lower resolution compared to OM, but they show similar features.

Published radio data. Finally, we searched the literature for published photometric data for the Holmberg II ULX. Using radio VLA data, Miller et al. (2005) measured 1.174 ± 0.0085 mJy at 1.4 GHz and 0.677 ± 0.207 mJy at 4.86 GHz. Miller et al. (2005) found extended emission covering the Chandra position of the ULX: 37 × 27 at 1.4 GHz and 34 × 19 at 4.86 GHz. These are comparable to the 50 pc extraction region used in the IR and UV photometry.

Published optical photometry. In addition, KWZ measured a V-band magnitude of 22.04 ± 0.08 for the stellar optical counterpart from observations with the Advanced Camera for Surveys (ACS; F550M filter, see Figure 3) on the HST. The optical flux is likely less contaminated by the nebular emission or emission from other sources compared to our measurements, because of the much smaller aperture.

Extinction corrections. The Milky Way reddening in the direction of Holmberg II is E(B − V) = 0.032 (from the NASA Extragalactic Database (NED)). KWZ estimated an additional local reddening of E(B − V) = 0.07 ± 0.01 based on the X-ray absorption. Thus, the total extinction for R_V = 3.1 is A_V = 0.32. We show the projected extinction and the extinction-corrected values for the UV and optical data in the last two columns of Table 3. These were calculated using the Cardelli et al. (1989) extinction curves. For the infrared data, we estimate from Draine (1989) that the extinction is A_λ < 0.02, and therefore negligible. We note that the photometric fluxes presented here are likely contaminated by the surrounding nebular emission as is evidenced by the HST images in Figure 3. Moreover, KWZ caution that the local extinction could in fact occur very close to the compact object and therefore not affect our measurements at all. We will therefore take the (extinction-corrected) photometric measurements as upper bounds.

In Section 4 and Figure 6, we constructed the full SED for the Holmberg II ULX using multi-wavelength data and previously published data. The bolometric luminosity for this SED is 1.34 × 10⁴⁰ erg s⁻¹, when the far-UV spectrum is extrapolated from the X-rays, using the Modified PLMCD model fit. Table 2 illustrates the degeneracy resulting from the three methods for X-ray spectral fitting. All three show a fairly good fit to the X-ray data, making it very difficult to distinguish between the different physical interpretations that each model represents.

The optical and UV fluxes plotted in Figure 6 can, in principle, come from three different sources: the stellar companion, the accretion disk, and from a jet. However, the overall SED shape and in particular the fact that the optical flux is much lower than the UV fluxes, disfavor the jet scenario as discussed below. On the other hand, the emission from the disk in the Modified PLMCD model is too faint at optical and UV wavelengths if extrapolated from X-rays and requires an additional source. Previously published optical data and our UV measurements suggest that a ∼B2Ib star is the source of these fluxes as the companion to the accreting BH. We explore this further in Paper II.

The fact that the PLMCD model overpredicts the UV and optical fluxes if extrapolated to these wavelengths is expected, because the power law cannot extend too much below the peak of the accretion disk (e.g., Done & Kubota 2006). On the other hand, it is intriguing that the simple BPL model provides such a good fit to the high-quality X-ray data. Moreover, the location of the break (∼1 keV) in this case is in agreement with the jet model. Kaaret et al. (2003) fit a very similar model to another famous ULX in NGC 5408. It has been shown that jets can contribute significantly to the X-ray emission, both by optically thin synchrotron emission and Compton scattering of the photons from the disk (or the companion) by relativistic electrons in the jet (Georganopoulos et al. 2002). However, even if the power law is less steep compared to the PLMCD model, it is still overpredicting the UV and optical fluxes and is therefore not the preferred scenario.

Much better constraints are provided by photoionization modeling of the Spitzer spectra. The [O iv] line is only produced by high-energy photons (≳54 eV), and is clearly associated with the ULX (Figure 3). For instance, the "classical" PLMCD model predicts a power-law component with photon index of 2.4 dominating into the UV. This model has the strongest emission of the three models in the far-UV and soft X-rays, and should produce the strongest [O iv] emission. In this case, photoionization modeling can be used to predict the [O iv] flux predicted from such a model, and this predicted flux can be compared with our observed values. In a companion paper (Paper II), we present the results of such a photoionization analysis in an attempt to distinguish between the three X-ray models presented here.

The ionization potential of [O iv] is just above the He ii edge so both likely result from similar regions in the nebula and a quantitative comparison between previously published optical observations and the mid-IR observations can be made. KWZ found that the He ii recombination line is consistent with photoionization by the ULX. The flux of our [O iv] line is almost three times higher than the He ii flux measured by KWZ (2.7 × 10³⁶ erg s⁻¹).

We compared the mid-IR line ratios of the ULX with those in AGN. NGC 4395 is one of the few known galaxies that is likely powered by an IMBH (Filippenko & Ho 2003). It has a bolometric luminosity very close to our ULX, and high-excitation emission lines were detected. The narrow-line He ii measured by Kraemer et al. (1999) for NGC 4395 relative to Hβ is 0.12, a value typical for Seyfert1 galaxies (Dietrich et al. 2005). For the Holmberg II ULX, He ii/Hβ≈ 0.2 (KWZ), slightly higher than for NGC 4395. The [O iv] measured from NGC 4395 is almost 16 times larger than the He ii flux (Kraemer et al. 1999). However, the gas density is close to the critical density for this line (10⁴ cm⁻³)—much higher than the density of the gas surrounding the Holmberg II ULX (KWZ estimate 10 cm⁻³, based on the line surface brightness predicted with CLOUDY). The [O iv] is expected to increase as it approaches the critical density, while the He ii is not sensitive to the gas density. An even more important difference is the abundance of O, which is ∼5 times higher for NGC 4395 (Kraemer et al. 1999). The dependence of the [O iv] on the abundance is expected to be roughly linear. In spite of these two differences, we consider the [O iv] (and He ii) detected from the Holmberg II ULX to be consistent with photoionization by a high energy source similar to the AGN in NGC 4395.

The infrared [O iv] line (together with [Ne v]) is a signature of high excitation usually associated with actively accreting black holes (e.g., Lutz et al. 1998). This line is often used to disentangle the starburst emission in composite type galaxies, and is a good indicator of the AGN power (Genzel et al. 1998; Sturm et al. 2002; Satyapal et al. 2004, 2007; Smith et al. 2004; Meléndez et al. 2008). Genzel et al. (1998) and Sturm et al. (2002) have used the ratio of [O iv]/[Ne ii] to distinguish between starbursts and AGN-dominated galaxies in the mid-IR. In Figure 7(a), we compare the Holmberg II ULX with data published by Genzel et al. (1998). The comparison clearly indicates high excitation for the ULX, similar to AGN.

Zoom In Zoom Out Reset image size

Other infrared line diagnostics have been proposed by Dale et al. (2006, 2009) to disentangle AGN type ionizing sources from star formation, using the ratios [S iii] 33.48 μm/[Si ii] 34.82 μm and [Ne iii]/[Ne ii]. Figure 7(b) shows the four regions defined by Dale et al. (2006) in their Figure 5, and their published data for different sources in the SINGS sample. The Holmberg II ULX is located well within region I, together with most AGNs and LINERs. We note that Vázquez et al. (2007) found similar results in their Spitzer IRS study of the NGC 4485/90 ULXs. However, they did not find clear evidence for [O iv] in their sample of objects. In examining this data further, we find slightly discrepant ratios from Vázquez et al. (2007). From our analysis, only ULX-2 in their sample actually falls marginally inside region I, the other five ULXs being located in regions II and III. In communicating with the authors, we find that the discrepancy is likely do to differences in reduction pipeline used for processing or the apertures used in either analysis.

Both diagnostic diagrams suggest that the [O iv] emission in Holmberg II is dominated by the ULX and that the surrounding star-forming regions provide little contamination to this line. While this comparison provides good evidence that the line ratios for the ULX are very similar to AGN, we note that AGNs are different than ULXs, both in their intrinsic emission and their environment. For example, AGNs emit copiously in the UV and optical, and are surrounded by gas at high densities.

Dasyra et al. (2008) found a good correlation between the luminosity of the [O iv] line and the mass of the central BH in standard AGN. We estimate a BH mass of 3.5 × 10⁴ M_☉ if this correlation is extrapolated to the Holmberg II ULX. We can also estimate the BH mass using the inner disk temperature from the Modified PLMCD model (0.38 keV). We obtain a BH mass of 994 M_☉. However, the flux of the MCD disk component does not dominate the X-ray emission (is only 49% of the total flux). This scenario would imply that the ULX is not in a typical high state, and therefore the inner disk temperature cannot be used reliably to estimate the mass of the BH (Stobbart et al. 2006). Finally, using the bolometric luminosity calculated from our SED (1.34 × 10⁴⁰ erg s⁻¹), we estimate that the lower limit to the mass of the central BH, within the Eddington regime, is 85 M_☉.