Evidence for Relativistic Disk Reflection in the Seyfert 1h Galaxy/ULIRG IRAS 05189–2524 Observed by NuSTAR and XMM-Newton

The NuSTAR data were processed using v.1.6.0 of the NuSTARDAS pipeline with NuSTAR CALDB v20160731. For each observation, the source spectra were extracted at the position of IRAS 05189–2524 within the radius of 60''. Corresponding background spectra were extracted from source-free areas on the same chip using polygonal regions. As discussed in Teng et al. (2015), the NuSTAR data show minor count rate variability (∼20%) and no obvious change in spectral shape between the two epochs. Therefore, we combined the three NuSTAR observations using the ADDSPEC script in HEASoft v6.19 to maximize the signal-to-noise ratio (S/N) of the spectra. We binned the data taking into account the background level to provide a nearly constant S/N over the NuSTAR bandpass, with a minimum S/N of 3 and median S/Ns of 5.5 and 5.3 for the two focal plane modules (FPMA and FPMB, respectively).

The XMM-Newton observation was reduced with the XMM-Newton Science Analysis System v14.0.0 following standard procedures. The raw event files were filtered using EPCHAIN and EMCHAIN to produce cleaned event lists for each of the EPIC-pn (Strüder et al. 2001) and EPIC-MOS (Turner et al. 2001) detectors, respectively. Science products were then produced using XMMSELECT, considering only single and double events for EPIC-pn, and single to quadruple events for EPIC-MOS. We extracted the source from a circular region of radius $40^{\prime\prime}$, while background was estimated from a large region of blank sky on the same detector as IRAS 05189–2524. Instrumental response files were generated with RMFGEN and ARFGEN for each detector. The good exposure is ∼31 ks for EPIC-pn and ∼36 ks for each of the two EPIC-MOS detectors. After performing the reduction separately for MOS1 and MOS2, we combined these data into a single spectrum using ADDASCASPEC. The MOS and PN spectra were grouped to have 25 counts per bin.

We then test fitting the reflection features with a phenomenological model by adding a neutral reflection component pexrav (Magdziarz & Zdziarski 1995) and one Gaussian emission line with variable width using the model zgauss (model 2; see Table 1). We freeze the centroid of the Gaussian emission line at 6.4 keV, the disk inclination angle at the default value (cos i = 0.45), and the iron abundance at solar and include no high-energy cutoff (foldE = 0), as this simple model is not sensitive to these parameters. The fit is improved considerably with ${\rm{\Delta }}{\chi }^{2}/{\rm{\Delta }}\mathrm{dof}=-231/-\,2$ (the best-fit parameters are listed in Table 2). A broad Gaussian line with line width $\sigma ={0.71}_{-0.11}^{+0.13}$ keV and equivalent width (EW) $\simeq \,0.60\,\mathrm{keV}$ is required to provide a decent fit for the iron line emission feature in the spectrum. However, a clear excess still remains in the residuals between 3 and 6 keV, as would be expected from the red wing of a relativistically broadened iron line (Figure 1(c)). Model 2 also gives a very steep power law of ${\rm{\Gamma }}=2.47\pm 0.04$. We do not attempt to interpret the intrinsic continuum steepness at this point, as this is only a phenomenological model.

We note that for both the simple absorbed power-law and the phenomenological reflection model, at least two extra emission features are present in the residuals around 0.9 and 1.9 keV. The 0.9 keV excess might be associated with O Lyα or Fe–L shell emissions, and the feature around 1.9 keV could be caused by ionized Si emission lines. The excesses cannot be adequately fitted with simple Gaussian emission line models, as they could be a complicated combination of ionized absorption, overabundant metal lines, and X-ray contribution from the high-mass X-ray binary population in the star-forming regions. X-ray point-source emission correlates with galaxy star formation rate (SFR) in star-forming galaxies (Lehmer et al. 2010). At an SFR of $\sim 80\,{M}_{\odot }\,\,{\mathrm{yr}}^{-1}$ (Westmoquette et al. 2012), the 2–10 keV luminosity from the binary population is estimated to be $1.4\times {10}^{41}\,\mathrm{erg}\,{{\rm{s}}}^{-1}$, about 1% of the total luminosity, which is enough to dominate the soft emission of IRAS 05189–2524. Therefore, the soft part of the spectrum likely helps little to constrain the X-ray continuum from the AGN. In order to avoid the possible situation where the soft emission from the host galaxy is driving our modeling of the disk reflection features, we ignore the data below 2 keV from here on.

In Teng et al. (2015), the authors attempted to explain the reflection features observed in IRAS 05189–2524 by neutral reflection from the torus, which is commonly the case for Compton-thick AGNs. However, they found that neither the MYTorus (Murphy & Yaqoob 2009) nor the BNTorus model (Brightman & Nandra 2011) could provide an adequate fit for the broadband spectrum, which ruled out the possibility of distant reflection. In order to explore the relativistic reflection hypothesis, we replace the pexrav and zgauss components in the phenomenological model with the self-consistent relativistic reflection model relxill (Dauser et al. 2014; García et al. 2014) (model 3a; see Table 1). We use relxilllp in the relxill model family; instead of using an empirical emissivity law, the emissivity is calculated directly in the lamppost geometry (Dauser et al. 2013). Relxilllp calculates a self-consistent reflection fraction. This could help us constrain the BH spin by shrinking the parameter space, as results with unphysical solutions of low BH spins with high reflection fractions can be ruled out by the assumption of a lamppost geometry (for discussion, see Dauser et al. 2014). In order to measure the BH spin, we assume that the inner disk radius extends down to the ISCO. We fix the outer radius at the default value of 400 ${R}_{{\rm{g}}}$ (${R}_{{\rm{g}}}\,\equiv \,$GM/c² is the gravitational radius), as the model is not sensitive to this parameter, and the cutoff energy is allowed to vary in the fitting.

The relxilllp model yields a good fit to the 2–30 keV spectrum with ${\chi }^{2}/$ dof $\,=\,862.0/859$ and no obvious excess in the residuals (Figure 3(b)). The intrinsic X-ray continuum is steep, with the power-law photon index ${\rm{\Gamma }}={2.04}_{-0.03}^{+0.02}$. It is in the upper range of the photon index distribution observed in local Seyfert galaxies (e.g., Nandra & Pounds 1994; Winter et al. 2009; Rivers et al. 2013), but less extreme than the previously published values for IRAS 05189–2524: e.g., ${\rm{\Gamma }}={2.68}_{-0.13}^{+0.30}$ (Teng et al. 2009), ${\rm{\Gamma }}=2.51\pm 0.02$ (Teng et al. 2015). Albeit lacking high-quality data above 30 keV, the model is able to put a modest constraint on the high-energy cutoff ${E}_{\mathrm{cut}}={55}_{-7}^{+10}$ keV, as discussed in García et al. (2015a); such sensitivity could come partly from the low-energy part of the reflection spectrum (<3 keV). The location of the primary radiation source is found to be close to the disk $h={2.39}_{-0.32}^{+0.80}$ ${R}_{{\rm{g}}}$, with a rapid BH spin (dimensionless spin parameter $a={0.96}_{-0.03}^{+0.02}$), and the self-consistent reflection fraction derived by the model is high (${R}_{\mathrm{ref}}$ = 3.24). The best-fit results also find an intermediate inclination of $\theta =42^\circ \pm 2^\circ$ and an ionization parameter¹⁰ log$\,\xi =3.02\pm 0.02\,\,\mathrm{erg}\,\mathrm{cm}\,{{\rm{s}}}^{-1}$ for the accretion disk. We note that the iron abundance ${A}_{\mathrm{Fe}}$ is unusually high, with the best-fit value hitting the upper bound of the model of 10. But if we fix the iron abundance at solar in the relxilllp model, the fitting degrades considerably (${\rm{\Delta }}{\chi }^{2}\sim 130$), with a visible excess in the Fe K band.

Zoom In Zoom Out Reset image size

A supersolar metal abundance is not uncommon among ULIRGs (e.g., Rupke et al. 2008). In the case of IRAS 05189–2524, based on the diagnostics of [N ii]/Hα, [S ii]/Hα, and [O i]/Hα line ratios, an uncommonly high metallicity¹¹ ($\gt 4\,{Z}_{\odot }$) is estimated for the outer galactic disk, or else some peculiar ionization effects are needed to explain the line ratio measurements (Westmoquette et al. 2012). If the metallicity is indeed high on the galactic scale in IRAS 05189–2524, it is more physically consistent to use neutral absorption models with supersolar abundances in our spectral fitting, as neutral absorbers normally reside far from the central BH.

In an attempt to further constrain the iron abundance, we tie the iron abundances of the neutral absorber and the disk reflection component by replacing the zpcfabs model with an improved model for neutral absorption tbnew_feo ¹² convolved with the partial covering model partcov (model 3b; see Table 1). Both Fe and O abundances are variable parameters in the tbnew_feo model. We freeze the O abundance at solar, since it would not have a significant influence on our spectral fitting above 2 keV. As a result, the model finds a best-fit iron abundance ${A}_{\mathrm{Fe}}={5.5}_{-1.1}^{+2.5}$, which is consistent with the measurement from Westmoquette et al. (2012). This model provides a slightly better fit for the data (${\chi }^{2}/$dof $\,=\,860.6/859$), with a higher coronal height $h={4.70}_{-1.99}^{+4.04}$, a less steep power-law continuum (${\rm{\Gamma }}={1.79}_{-0.14}^{+0.17}$), and a cutoff at ${34}_{-10}^{+21}$ keV, even lower than in the previous model. As a result, the reflection fraction is reduced to ${R}_{\mathrm{ref}}=1.99$. We note here that the cutoff energy is uncommonly low for an AGN, even compared with the lowest NuSTAR measurements (e.g., Baloković et al. 2015; Ursini et al. 2015; Tortosa et al. 2016). This model generally has more uncertainty in the parameters; most notably, it places no constraint on the BH spin. Although the metallicities in the whole galaxy should be related, there is no solid justification to assume that the metallicities are the same in the BH vicinity and in the outer disk of the galaxy. However, we believe that it is a reasonable simplification given the data quality; untying the iron abundances brings no obvious improvement to the fitting.

Given the possible ionized absorption features in the Fe K band and motivated by the presence of large-scale outflows in the host galaxy (e.g., Westmoquette et al. 2012; Bellocchi et al. 2013; Rupke & Veilleux 2015), we add an ionized absorption component to account for any outflowing ionized absorbers. We modify the spectrum with an ionized absorption table model calculated by the XSTAR code (Kallman & Bautista 2001) and tie its iron abundance with that of the relxilllp component. It is a physically reasonable requirement, since if the wind has extreme velocity, it most likely arises from the accretion disk; thus, the same chemical abundances would naturally be expected. We first assume that the partial covering neutral absorber has a solar iron abundance (model 4a; see Table 1). The fitting results reveal an ionized absorber with a velocity ${v}_{\mathrm{out}}={0.13}_{-0.06}^{+0.02}\,c$ in the source rest frame, which is well above the typical outflowing velocities of warm absorbers, about the common value found for UFOs (e.g., Tombesi et al. 2010). With an ionization parameter log$\,\xi =3.06\pm 0.07\,\mathrm{erg}\,\mathrm{cm}\,{{\rm{s}}}^{-1}$, the absorption should be dominated by Fe xxv. However, the detection of a high-velocity ionized outflow is not significant, as it only brings marginal improvement to the fitting, ${\rm{\Delta }}{\chi }^{2}/{\rm{\Delta }}\mathrm{dof}$ $=-7.2/-3$, and the difference is not visually evident in the ratio plot. Other best-fit parameters are consistent with model 3a (without the ionized absorber). A rapidly spinning BH with $a\gt 0.89$ is required to provide an adequate fit for the data.

We then reintroduce the neutral absorption model with a variable iron abundance (model 4b; see Table 1) and force the iron abundances of the disk reflection, the outflowing wind, and the neutral absorber to be the same. The model brings no further improvement for the fitting. But again, it reduces the iron abundance ${A}_{\mathrm{Fe}}$ to ∼5, and since the iron abundance and the column density in the XSTAR model are degenerate parameters, the column density of the ionized outflow is estimated to be higher. Best-fit parameters of the disk reflection agree well with those given by model 3b, except for a slightly higher inclination of $\theta ={51}_{-5}^{+4}{\circ} .$ Only a low BH spin can be ruled out at the 90% confidence level by this model.

In our spectral analysis of IRAS 05189–2524, all acceptable fits require a supersolar iron abundance for the disk reflection component, but the value is not very well constrained. The primary effect of the iron abundance parameter is to change the relative strength of the iron line and the Compton hump. Because of the low net number counts at the high-energy end of the NuSTAR spectra, the shape of the Compton hump is not well constrained, which could make it difficult to obtain a tight constraint of the iron abundance. We note that a similar overabundance in iron has been reported in a number of AGNs (e.g., Risaliti et al. 2009; Patrick et al. 2011) and stellar-mass BHs (e.g., García et al. 2015b), with the examples of well-known narrow-line Seyfert 1 galaxies (NLSy1s) 1H 0707–495 (Dauser et al. 2012; Kara et al. 2015) and IRAS 13224–3809 (Fabian et al. 2013). In the case of IRAS 05189–2524, the galaxy is undergoing intense star formation, so it is expected to be iron enriched. Since supersolar metallicities were indicated via spatially resolved optical spectroscopy in the outer disk and the nuclear region of the galaxy (Westmoquette et al. 2012), it is physically reasonable for the iron abundance to be high in the accretion disk.

Absorption from material lying relatively distant from the BH has frequently been proposed as an alternative interpretation of relativistic disk reflection (e.g., Miller et al. 2008, 2009). To investigate whether the excesses in the Fe K band and above 10 keV observed in IRAS 05189–2524 can be fully accounted for by absorption-dominated models (without the reflection component), we test modeling the 2–30 keV spectrum with single and dual partial covering neutral absorption models with variable iron abundances. For the single-absorber model,¹³ we fix the covering faction at 99% to avoid large deviation from the data below 2 keV, the same way as in all the previous modelings. With ${\rm{\Delta }}{\chi }^{2}=215.6$ compared to model 3b, this scenario can be easily ruled out. The dual-absorber model¹⁴ with an iron abundance ${A}_{\mathrm{Fe}}={2.2}_{-0.4}^{+0.2}$ provides a better fit for the data, which requires one full covering absorber with the column density ${N}_{{\rm{H}},\mathrm{abs}1}={6.01}_{-0.93}^{+0.14}\times {10}^{22}$ cm⁻² and another of the coverage fraction ${f}_{\mathrm{abs}2}=64 \% \pm 5 \%$ with ${N}_{{\rm{H}},\mathrm{abs}2}={1.92}_{-0.18}^{+0.58}\times {10}^{23}$ cm⁻². However, it is still statistically worse than the reflection-dominated model with ${\rm{\Delta }}{\chi }^{2}=75.6$. The model leaves a broad excess in the Fe K band and also fails to produce the shape of the Compton hump above 10 keV (see Figure 5(b)), resembling the case in NGC 1365 (Risaliti et al. 2013).

The time-averaged 2–30 keV X-ray flux observed by NuSTAR is $7.1\times {10}^{-12}$ erg cm⁻² s⁻¹, corresponding to an intrinsic luminosity of $3.0\times {10}^{43}$ erg s⁻¹ during the 2013 observation. Using the central velocity dispersions measured from the Ca ii triplet line widths (Rothberg et al. 2013) and the ${M}_{\mathrm{BH}}$–σ relation (Tremaine et al. 2002), the BH mass of IRAS 05189–2524 is estimated to be ${M}_{\mathrm{BH}}=4.2\times {10}^{8}\,{M}_{\odot }$. We note that the BH mass derived from the Ca ii triplet here is more than 10 times larger than the value measured from CO bandheads in the near-infrared (Dasyra et al. 2006). As discussed in Dasyra et al. (2006), CO bandheads are often representative of young stellar populations, which could lead to systematically lower BH masses for ULIRGs. Combined with the bolometric luminosity inferred from the infrared (Teng et al. 2015), IRAS 05189–2524 is measured to be accreting at an Eddington rate ${\lambda }_{\mathrm{Edd}}={L}_{\mathrm{bol}}/{L}_{\mathrm{Edd}}$ of 0.12. Based on the correlation between the photon indices and the Eddington ratios of AGNs (Shemmer et al. 2008; Brightman et al. 2013, 2016), a photon index of ${\rm{\Gamma }}\sim 2.0$ would be expected. The photon index obtained by our disk reflection modeling with a low cutoff energy is ${\rm{\Gamma }}\simeq 1.7\mbox{--}2.0$, which is consistent with the expected value given the scatter of the relation.

In our spectral modeling, we did not detect significant disk winds. Fitting the possible absorption features in the spectrum with a simple Gaussian absorption line using the gabs model would reach similar ${\chi }^{2}$ to the physical XSTAR grid, with the best-fit line centroid at ∼7.33 keV. Although the effect of this ionized absorber on the continuum is subtle, we include it in the spectral fitting, as even mild absorption lying around the Fe K edge could influence the BH spin measurement, which is sensitive to the profile of the broad iron line. The power of an outflow scales with the amount of matter being ejected (King & Pounds 2015). With the rather small column density measured for the ionized absorption compared with that of massive disk winds found in other AGNs (e.g., IRAS 11119+3257; Tombesi et al. 2015), even if a UFO indeed exists in IRAS 05189–2524 at the epoch of the observation, it is probably not energetic enough to have a large impact on its galactic environment.