X-ray spectroscopy of the candidate AGN in Henize 2-10 and NGC 4178: Likely supernova remnants

TL;DR

This study re-analyzed X-ray data of candidate AGN in dwarf galaxies and found that the spectra are better explained by supernova remnants than active galactic nuclei, highlighting the importance of spectral analysis in source identification.

Contribution

The paper demonstrates that X-ray spectral analysis can distinguish supernova remnants from AGN in low-luminosity sources, challenging previous classifications.

Findings

Supernova remnant models fit the spectra better than AGN models.

X-ray lines indicate supernova remnants in candidate AGN sources.

Spectral analysis is crucial for accurate source identification in low-count X-ray data.

Abstract

Black holes in dwarf/bulgeless galaxies play a crucial role in studying the co-evolution of galaxies and their central black holes. Identifying massive black holes in dwarf galaxies suggests that the growth of black holes could precede that of galaxies. However, some of the most intriguing candidate active galactic nuclei (AGN) in small galaxies have such low luminosities that the sample is vulnerable to contamination by other sources, such as supernova remnants. We re-analysed Chandra X-ray Observatory observations of candidate AGN in Henize 2-10 and NGC 4178, considering the potential signals of emission lines in the minimally-binned X-ray spectra. We find that hot plasma models, which are typical of supernova remnants, explain the observed spectra much better than simple power-law models, which are appropriate for AGN. We identify clear signals of X-ray lines in the faint X-ray source identified with the radio source in Henize 2-10 by Reines et al. 2016. Combining our work with the MUSE measurement of the ionization parameter in this region by Cresci et al. 2017 indicates that this radio and X-ray source is more likely a supernova remnant than an AGN. A similar analysis of the low-count X-ray spectrum of a candidate AGN in NGC 4178 shows that a hot plasma model is about seventeen times more probable than a simple power-law model. Our results indicate that investigation of X-ray spectra, even in a low-count regime, can be a crucial tool to identify thermally-dominated supernova remnants among AGN candidates.

Tables

Table 1. Table 1: Summary of spectral analysis results for the candidate AGN in Henize 2-10.

Model	Parameters	Parameter Values	$𝚫$cstatpl/$𝚫$d.o.fpl	AICc value	log(CvM)
					(goodness)
tbabs*pegpwrlw	$N_H$	${2.3}_{-1.4}^{+1.7}\times {10}^{21}$ cm-2	$0.0/0$	$137.73$	$-6.32$ $\left(99\%\right)$
	$\Gamma$	${2.71}_{-0.46}^{+0.52}$
	Fluxunabsorb	${1.61}_{-0.48}^{+1.10}\times {10}^{-14}$ ergs cm-2 s-1
tbabs*(pegpwrlw +gauss+gauss +gauss+gauss)	$N_H$	${1.3}_{-0.4}^{+2.8}\times {10}^{21}$ cm${}^{-2\ast }$	$-45.26/-10$	$116.70$	$-9.37$ $\left(10\%\right)$
	$\Gamma$	${1.5}_{-0.9}^{+0.8}$
	Fluxunabsorb	${4.8}_{-1.6}^{+3.7}\times {10}^{-14}$ ergs cm-2 s-1
	$E_1$	${0.78}_{-0.22}^{+0.06}$ keV
	${\sigma }_1$	${0.18}_{-0.06}^{0.09}$ keV
	$E_2$	${1.31}_{-0.04}^{+0.02}$ keV
	$E_3$	${1.85}_{-0.05}^{+0.06}$ keV
	$E_4$	$4.56\pm 0.11$ keV
	${\sigma }_2,{\sigma }_3,{\sigma }_4$	${0.04}_{-0.04}^{+0.06}$ keV
tbabs*vapec	$N_H$	${9.1}_{-0.0}^{+15.0}\times {10}^{20}$ cm${}^{-2\ast }$	$-6.47/-2$	$135.67$	$-6.40$ $\left(99\%\right)$
	$kT$	${2.45}_{-0.83}^{+0.91}$ keV
	O	${960}_{-930}^{+40\ast \ast }$
	Ne	${970}_{-940}^{+30\ast \ast }$
	E.M.a	${1.39}_{-0.31}^{+25.4}\times {10}^{59}$ cm-3
tbabs*(vapec+vapec)	$N_H$	${2.7}_{-1.8}^{+2.8}\times {10}^{21}$ cm${}^{-2\ast }$	$-31.82/-3$	$112.59$	$-8.12$ $\left(59\%\right)$
	$k{T}_{\mathrm{hot}}$	$>8.36$ keV
	$k{T}_{\mathrm{cool}}$	${0.52}_{-0.11}^{+0.24}$ keV
	O, Ne, Mg, Si	${11}_{-7}^{+24}$
	E.M.${}_{}{}^a{}_{\mathrm{hot}}{}^{}$	${3.3}_{-2.2}^{+1.83}\times {10}^{60}$ cm-3
	E.M.${}_{}{}^a{}_{\mathrm{cool}}{}^{}$	${1.72}_{-0.96}^{+5.66}\times {10}^{60}$ cm-3
tbabs*vnei	$N_H$	${5.9}_{-2.2}^{+1.8}\times {10}^{21}$ cm-2	$-29.31/-2$	$112.83$	$-8.15$ $\left(38\%\right)$
	$kT$	$>22$ keV
	O, Ne, Mg, Si	${1.7}_{-0.7}^{+1.2}$
	$\tau$	${1.6}_{-0.7}^{+1.1}\times {10}^{10}$ s cm-3
	E.Ma	${3.9}_{-1.4}^{+2.0}\times {10}^{60}$ cm-3

Equations

$$\mathrm{AICc}=2k+{\mathrm{cstat}}_{min}+\frac{2k^2+2k}{n-k-1},$$

Full text

††pubyear: 2018††pagerange: X-ray spectroscopy of the candidate AGN in Henize 2-10 and NGC 4178: Likely supernova remnants–4

1 Introduction

The correlations between key properties of a galaxy(the mass of the bulge, and the velocity dispersion) and the mass of the supermassive black hole (SMBH) hosted by the galaxy (e.g. Magorrian et al., 1998; Gebhardt et al., 2000; Marconi & Hunt, 2003; Jahnke et al., 2009; Merloni et al., 2010) are of high importance in astrophysics. These observational results have motivated theories on how galactic bulges and SMBHs evolve together (see Fabian, 2012; Kormendy & Ho, 2013; Heckman & Best, 2014, for reviews). These theories suggest that the properties of galaxies are regulated, in part, through the feedback from active galactic nuclei (AGN) in the form of radiation and outflows. Irradiation from AGN can evaporate cool gas, while jets and winds heat the warmer gas. Both methods serve to terminate star-formation, as well as the fuel source for accretion onto the AGN.

While massive galaxies and bright cluster galaxies show clear evidence of AGN feedback, it has been harder to study AGN activity in low-mass galaxies that lack bulges, or have only pseudobulges (which are not classical bulges; e.g., Dashyan et al., 2018; Booth & Schaye, 2013). And some theories suggest that the observed connection between black hole and galaxy bulge mass might only be a result of repeated mergers ( i.e., they need not be physically coupled; Jahnke & Macciò, 2011). Volonteri & Natarajan (2009) points out that the relation of BH masses to galaxy and bulge masses on the low-mass end will provide key evidence as to the origin of massive BHs, as either direct collapse of $\sim 10^{4}$ M*⊙*  clouds versus initial seeds of $\sim 10^{3}$ M*⊙*  BHs from the collapse of Population III (first-generation) stars. Recent works have been divided as to whether the standard BH-bulge mass relation continues straightforwardly to lower masses (Barth et al., 2005; Xiao et al., 2011; Baldassare et al., 2015) or shows substantial changes (Greene et al., 2008; Jiang et al., 2011). Thus the detections of AGN powered by lower-mass black holes ($10^{3}$–$10^{6}$ M*⊙*, hereafter intermediate-mass black holes or IMBHs, as opposed to super-massive black holes) in low-mass bulgeless or pseudobulge galaxies are crucial for understanding how SMBHs and galaxies grow.

Increasing efforts have been devoted towards finding IMBHs in dwarf and/or bulgeless galaxies over the past 30 years. For instance, the small, bulgeless spiral galaxy NGC 4395 hosts a faint Seyfert I AGN, verified to be powered by a relatively small BH, of mass $(3.6\pm 1.1)\times 10^{5}$ M*⊙*  (Filippenko & Sargent, 1989; Peterson et al., 2005). The dwarf elliptical galaxy Pox 52 also hosts a weak AGN, with a central BH engine of mass ($3.2\pm 1.0)\times 10^{5}$ M*⊙* (Kunth et al., 1987; Barth et al., 2004; Thornton et al., 2008). A number of optical spectroscopic searches for “dwarf" Seyfert nuclei have been conducted (e.g. Ho et al., 1995; Greene & Ho, 2004, 2007; Reines et al., 2013; Lemons et al., 2015; Chilingarian et al., 2018; Liu et al., 2018), identifying a large number of candidate IMBHs.

A number of methodologies have been used to test whether emission line regions at the centres of galaxies are produced by AGN, including: the ratios of emission lines indicating the ionization state (and thus suggesting an origin either in star formation or from AGN activity Baldwin et al. 1981); the breadth of H lines (Greene & Ho, 2005; Reines et al., 2013); and the identification of X-rays from the candidate AGN. The Wide-Field Infrared Survey Explorer (WISE) provides mid-infrared colour information, which has been exploited to identify AGN regardless of obscuration (Stern et al., 2012; Satyapal et al., 2014). Abel & Satyapal (2008) proposed using the luminosity of a [Ne V] line in the mid-infrared, which also allows the study of obscured AGN.

However, young supernova remnants evolving in a high-density medium can also produce emission lines in a relatively high ionization state, and broad H lines (e.g. Filippenko, 1989; Terlevich et al., 1992; Baldassare et al., 2016). Hainline et al. (2016) showed that selection using only mid-infrared colours from WISE is vulnerable to confusion from star-forming galaxies. The formation of Wolf-Rayet and massive O stars in young starbursts can produce extreme ultraviolet flux that could lead to high ionization of Ne (Schaerer & Stasińska, 1999; Kewley et al., 2001; Lutz et al., 1998). Accurate measurement of the BH mass requires time-consuming reverberation mapping (Peterson, 1993; Peterson et al., 2005; Peterson, 2014, etc.) or dynamical spectroscopic measurements (Du et al., 2017; Songsheng & Wang, 2018), which have been done for relatively few candidate IMBH AGN in small galaxies. Thus, identification of X-rays from candidate AGN has become a crucial element of many campaigns to identify IMBH AGN, sometimes being regarded as the crucial piece of proof necessary to verify a candidate AGN (e.g. Greene & Ho, 2007; Secrest et al., 2012; Chilingarian et al., 2018).

The key element of the present work is the realization that moderate-resolution X-ray spectroscopy, even with relatively small numbers of counts, is capable of distinguishing between some hot plasma models, typical of young thermal supernova remnants (SNRs)111Not all SNR X-ray spectra are dominated by thermal ejecta; some, like the Crab and SN 1006, are dominated by non-thermal synchrotron emission from pulsar wind nebulae or shocks., and the power-law spectra typical of AGN. This opportunity is provided by the very strong emission lines in the metal-enriched thermal ejecta dominating typical SNR spectra, especially the strong Si lines around 1.8 keV. The ability to identify X-ray spectra produced by thermal SNR emission allows us to constrain whether X-ray emission arising from a candidate AGN is more consistent with an AGN or a SNR. In this paper, we study two test cases in detail.

1.1 Potential AGN in Henize 2-10 and NGC 4178

Henize 2-10 is a dwarf starburst galaxy without a central bulge, located at a distance of 9 Mpc (Vacca & Conti, 1992), with a stellar mass of $3.7\times 10^{9}$ M*⊙* (Reines et al., 2011). Reines et al. (2011) identified an X-ray source located in Henize 2-10 with a steep-spectrum radio source. Comparing the X-ray and radio luminosities of this source, Reines et al. (2011) argued that the radio and X-ray luminosities and their ratio excluded origins other than an accreting black hole. The Fundamental Plane of black hole activity relates black hole mass, X-ray luminosity, and radio luminosity, for black holes accreting in a radiatively inefficient state (e.g. Merloni et al., 2003). Using this plane, Reines et al. (2011) argued that the observed radio brightness requires a BH with a mass $\sim 2\times 10^{6}$ M*⊙*, as smaller black holes do not generate this radio brightness at the observed $L_{X}$. The identification of a BH with such a large mass in a dwarf starburst galaxy without a bulge was quite surprising, as it suggested that large BHs could appear first, before galactic bulges grow. Such a discovery would have strong implications for the evolution of galaxies and BHs.

Deeper Chandra observations of Henize 2-10 (Reines et al., 2016) resolved the X-ray-source identified as the candidate AGN into two components. One source, coincident with the radio point source, was suggested by Reines et al. (2016) to be the AGN. Another X-ray source, brighter (in some epochs) and spectrally hard, without a radio counterpart, was suggested to be a high-mass X-ray binary. The weaker X-ray source coincident with the radio source had an X-ray luminosity $L_{0.3-10\mathrm{keV}}\sim 10^{38}$ erg s*-1*. Reines et al. (2016) fit the X-ray spectrum of the nuclear source with a power-law, and a thermal plasma model (apec), but their spectral fitting did not discriminate between the models. We argue below that their choice of spectral binning lost information, thus preventing clear discrimination between these models.

Recently, Cresci et al. (2017) performed MUSE(Multi Unit Spectroscopic Explorer) integral field spectroscopy on this portion of Henize 2-10. They found that the optical emission line ratios of gas in the central region of Henize 2-10 were consistent with starburst models, with no indication of AGN ionization. They also show that the revised X-ray luminosity, and X-ray/radio flux ratio, of the candidate AGN identified by Reines et al. (2016) are consistent with those of young SNRs.

NGC 4178 is a SB(rs)dm galaxy with a HII nucleus (Ho et al., 1997) and H$\alpha$ emission similar to a star-forming galaxy (Koopmann & Kenney, 2004). This dwarf galaxy ($M_{*}=1.3\times 10^{10}$ M*⊙; Ho et al., 1997) is highly inclined ($i\sim 70\deg$) and located at a distance of $16.8$ Mpc (Tully & Shaya, 1984) in the Virgo cluster. Satyapal et al. (2009) argued for the existence of an AGN in NGC 4178, based on the high mid-IR luminosity observed in a [Ne V] line ($L_{14.32\mu\mathrm{m}}=8.23\times 10^{37}$ erg s-1*), using the arguments of Abel & Satyapal (2008). Secrest et al. (2012) used a Chandra X-ray observation of NGC 4178 to identify an X-ray source corresponding to the IR source. However, the observed luminosity of the X-ray source is about five orders of magnitude less than that expected based on the [Ne V] luminosity. They postulate that this could be due to strong absorption in the nucleus. Secrest et al. (2013) did not detect any optical signature of an AGN in Gemini observations. Thus, the X-ray emission and strong [Ne V] emission are the only evidence for an AGN in this galaxy.

In this paper we scrutinise the X-ray spectra of the candidate AGNs in Henize 2-10 and NGC 4178 to test whether these candidate AGN may be better explained as SNRs. In Section 2, we describe how we obtained X-ray spectra of the candidate AGNs in Henize 2-10 and NGC 4178. Section 3 compares various models fit to the observed X-ray spectra. We demonstrate that minimally-binned spectral analysis can robustly identify strong X-ray spectral lines, which can be helpful in discriminating between AGN and SNRs.

2 Observations and Data reduction

We used the high-resolution Chandra X-ray observations of Henize 2-10 in February 2015 (ObsID 16068, PI: Reines; Reines et al., 2016)) and of NGC 4178 in February 2011 (ObsID 12748, PI: Satyapal; Secrest et al., 2012)) to study these X-ray sources. Both observations were taken using the ACIS-S instrument, in the VFAINT mode and FAINT mode, respectively.

We used CIAO version 4.9 (Fruscione et al., 2006) for the data reduction and the construction of the X-ray spectra. We reprocessed the initial data according to CALDB 4.7.6 standards using the chandra_repro command. We set the check_vf_pha parameter to “no” to prevent the removal of any good events. The default option of the pix_adj parameter now uses the EDSER algorithm (Li et al., 2004), which results in more-precise sub-arcsecond resolution. We constructed X-ray spectra using the specextract command, which produces effective area files corrected for the fraction of the point-spread function extracted in the spectrum. We used photons with energies between $0.3-10.0$ keV for our analysis.

Because of the low photon counts, grouping these spectra with 15 photons in each bin would provide an inadequate number of bins to constrain different models. Such spectra would also not resolve any narrow features that might exist in the spectrum. Thus, we regrouped the spectra using the dmgroup command such that each group includes at least 1 event. The spectra generated were analysed using XSPEC version 12.9.1m, using the C-statistic (Cash, 1979) modified to deal with backgrounds, which can handle few to no counts per bin (note that binning to keep 1 count/bin is recommended222https://heasarc.gsfc.nasa.gov/xanadu/xspec/manual/node304.html).

2.1 Henize 2-10

The central region of Henize 2-10 is extremely compact with multiple X-ray sources within a $2\arcsec$ region, plus three other clear X-ray sources within a $50\arcsec$ region. The X-ray image of the central region of Henize 2-10 is dominated by diffuse emission and a highly variable off-nuclear hard X-ray source suggested by Reines et al. (2016) to be a high-mass X-ray binary. However, the likely HMXB is not visible in the 160 ks 2015 observation of Henize 2-10 used here. Thus, there is relatively little contamination of the weak X-ray source and candidate AGN, compared to the other X-ray observations of Henize 2-10 (The HMXB, which is $0\aas@@fstack{\prime\prime}7$ from the candidate AGN, is $\sim 10$ times brighter than our source in the remaining observations. Thus we cannot use the ObsIDs 2075 (20 ks) and 16069 (40 ks), taken in 2001 March and on 2015 February 16, respectively, for this analysis.). To clearly resolve the candidate AGN from the nearby diffuse emission and select the source region, we re-binned the image to $1/8^{\rm th}$ of the ACIS-S pixel size and convolved this image with an FWHM$=0\aas@@fstack{\prime\prime}25$  ($\sigma\approx 0.44$ pixels) Gaussian kernel using the aconvolve command (see Fig. 1). We extracted X-ray photons from a $0\aas@@fstack{\prime\prime}5$ circular region centred on this X-ray source ($\alpha=8^{\mathrm{h}}36^{\mathrm{m}}15\aas@@fstack{s}13$; $\delta=-26^{\circ}24\arcmin 34\aas@@fstack{\prime\prime}08$). This radius was chosen to encompass the maximum source photons while minimising photons from the diffuse emission around the nuclear source (see Fig. 1, left). We selected background photons from an annulus of radius $2-3\arcsec$, excluding the bright diffuse emission directly east of the AGN candidate.

2.2 NGC 4178

The sub-arcsecond angular resolution of the Chandra X-ray Observatory clearly resolves the nuclear X-ray source from nearby bright X-ray sources (see Fig. 1, right). We selected a circular region of $1\arcsec$ around the source ($\alpha=12^{\mathrm{h}}12^{\mathrm{m}}46\aas@@fstack{s}32$; $\delta=+10^{\circ}51\arcmin 54\aas@@fstack{\prime\prime}61$) to collect the source photons. We used a $10\arcsec$ region nearby, lacking bright X-ray sources, for the background.

3 Results and Discussion

In each fit, we calculated the best-fitting parameters using the C-statistic, which is the maximum-likelihood-based statistic for Poisson data (Cash, 1979), as the fit statistic (e.g. “statistic cstat" in XSPEC), with correction for background subtraction, aka the “W-statistic"333https://heasarc.gsfc.nasa.gov/xanadu/xspec/manual/node304.html. To evaluate the relative quality of different models, we used the Akaike Information Criterion values (Akaike, 1974), with correction for small sample sizes (AICc; see Cavanaugh, 1997):

[TABLE]

where $k$ is the number of parameters and ${\rm cstat}_{\rm min}$ is the negative of twice the logarithm of the likelihood of the model, with best-fitting parameters assuming Poisson statistics. A smaller AICc value indicates a better likelihood. The quantity $\exp$((AICci - AICcj)/2) gives the relative likelihood of model $j$ with respect to model $i$. A small value (e.g. 1/20) of this quantity indicates that model $i$ is preferred, by the inverse of this factor (e.g. by a factor of 20) over model $j$.

We also use XSPEC’s “goodness" simulations to calculate the quality of individual fits. These simulations use the Cramer-von Mises (CvM) statistic (Cramer, 1928; Von Mises, 1928) (e.g. we set “statistic test cvm" in XSPEC). They measure the fraction of simulations of the model with a CvM statistic smaller (better) than the CvM statistic of the data; a fraction close to 1 (e.g. 99%) indicates the model can be rejected with this level (e.g. 99%) of confidence.

For both sources, we chose a simple absorbed power law spectrum for an initial fit, and when that proved inadequate to explain the emission features (based on the quality of the fit), moved on to use hot thermal plasma models . For modelling photoelectric absorption, we used the tbabs model with typical interstellar (wilm) abundances (Wilms et al., 2000). For both sources, we find that the hot plasma models explain the observed spectra better than a simple power law. The results of our spectral fitting are summarized in Tables 3.1 & LABEL:table:spec_ngc4178. For easy visualization, the plotted spectra of Henize 2-10 and NGC 4178 in this paper were re-binned such that each bin had four and three counts, respectively. We also tested the effects of modelling the background for the Henize 2-10 candidate AGN, rather than subtracting it. As the background makes only a small ($\sim 10$%) contribution to the source flux, we found negligible differences in the fits that we tested, with all fitted parameters lying within the errors found with background subtraction.

3.1 Henize 2-10

From Fig. 1, we see that there is another bright X-ray source to the east of our candidate AGN, even in ObsID 16068. However, the convolved image shows that the X-ray emission from the two sources are distinct and clearly resolved by Chandra. We analyse the X-ray flux from this eastern source in later sections. We binned the $\sim 200$ photons observed from the candidate AGN into $\approx 96$ bins, each with one or more count per PHA (energy) bin. The minimum $N_{H}$ was fixed to $9.1\times 10^{20}$ cm*-2* to account for Galactic absorption in the direction of Henize 2-10 (Kalberla et al., 2005). Fitting an absorbed power law (tbabspegpwrlw) to the spectrum gives $\Gamma=2.7_{-0.5}^{+0.5}$ and $N_{H}=2.3_{-1.4}^{+1.7}\times 10^{21}$ cm-2*, consistent with the spectral fit found by Reines et al. (2016). This fit is shown in Fig. 2, achieving a C-statistic of 131.47. We created $10^{5}$ realizations of the power-law spectrum using the goodness command, and found that $\sim 99\%$ of them have a lower CvM statistic than the data. This indicates that a simple power-law is a poor description of the observed spectrum.

In Fig. 2, we can see that the power law model predicts the overall trend in counts versus energy, but fails to explain several narrow features clearly visible by eye in the observed spectrum. On studying the residuals, we see that the higher counts at $\approx 1.3$ keV and $\approx 1.9$ keV align with the K$\alpha$ emission lines of Mg and Si. The overall shape of the spectrum, with evidence of strong lines near 1.0, 1.3, 1.9, and (perhaps) 2.5 keV is reminiscent of the spectral shapes of supernova remnants, with lines around those energies due to Ne and/or Fe, Mg, Si, and S, respectively (e.g. Cas A SNR: Holt et al. 1994; Hughes et al. 2000; Tycho SNR: Badenes et al. 2006; SN 1987A: Michael et al. 2002).There also appears to be a feature at $\sim 4.5$ keV, which we discuss below.

As a first step, we model the spectrum with an absorbed power-law plus Gaussians at the locations of strong residuals (Fig. 3). We find that four lines are strongly required. Narrow ($\sigma<0.1$) lines are required at $1.85_{-0.05}^{+0.06}$ keV (which we associate with Si XIII K$\alpha$, lab energy 1.83 keV444http://www.atomdb.org/Webguide/webguide.php), $1.31_{-0.04}^{+0.02}$ keV (which we associate with Mg XI K$\alpha$, 1.33 keV), and $4.56\pm 0.11$ keV, which does not have an immediately obvious nature. A broad ($\sigma=0.18_{-0.06}^{+0.09}$ keV) emission feature is also required at 0.78${}^{+0.06}_{-0.22}$ keV, which may be explained by a combination of O VIII (0.65 keV), Ne IX K$\alpha$ (0.91 keV), and/or Fe L-shell (covering roughly 0.8–1.1 keV). When we add these four Gaussians to the power-law model, the C-statistic drops by $\sim 45$ with 10 fewer degrees of freedom. Comparing the AICc statistics (Table 3.1) for the power-law fit, vs. the fit including four Gaussians, indicates that the 4-Gaussian fit is preferred by a factor of $\sim$37,000.

These strong emission lines suggest thermal plasma at low temperatures and/or high abundances, consistent with thermally emitting ejecta from a supernova remnant. We therefore attempt to fit the spectrum of the Henize 2-10 AGN candidate with self-consistent physically motivated thermal plasma models, beginning with a simple collisionally ionized model.

The apec model has a single parameter for all metal abundances. Since we do not see a strong Fe K emission line in our spectrum (e.g. at 6.7 keV), the single apec model fails to fit the observed emission lines of Mg and Si. Thus we employed the vapec model that allows fits to individual relative abundances of metals. We allowed the abundances of O, Ne, Mg, Si and S to vary relative to solar values, since these are the abundant elements with strong lines in the observable spectral range. Fitting an absorbed collisionally ionized plasma emission model (tbabsvapec) gives $kT=2.5_{-0.8}^{+0.9}$ keV and prefers high abundances of O ($176_{-169}^{+824}$) and Ne ($190_{-175}^{+810}$), with Si ($5_{-5}^{+995}$), S ($7_{-7}^{+228}$), and Mg ($0^{+249}_{-0}$) abundances basically unconstrained. Since the errors on the relative abundances of Mg, Si and S are large, and consistent with solar abundances, we fix the abundance of these elements to solar (1). Fitting the X-ray spectrum with this constraint gives $N_{H}=9_{-0}^{+15}\times 10^{20}$ cm-2*, $kT=2.5_{-0.9}^{+0.9}$ keV, and super-solar abundances of O and Ne ($\sim 1000$). 555The lower limit on $N_{H}$ reaches our hard limit, while the abundances of O & Ne reach the hard upper limit. The spectrum matches the observed features with K$\alpha$ emission lines of O and Ne. Despite having two additional parameters in comparison to an absorbed power law model, the C-statistic decreases only marginally (by $6.47$). Thus, though this fit is better than a simple absorbed power-law, it is not a statistically significant improvement over the power-law model. We also performed “goodness" simulations using $10^{5}$ realizations of the model to test the quality of the fit. Due to the large error-bars in the abundance parameters, we used the “nosim” option in the goodness command in XSPEC for our simulations. Similar to the absorbed power-law spectra, we find that $\sim 99\%$ of the realizations have a lower CvM statistic confirming that a single temperature hot plasma model is a poor fit to the observed spectrum. Also, the high relative abundance of O and Ne with respect to Mg and Si seems unphysical.

The low abundances of Mg & Si in this fit seem unlikely, since we see evidence for apparent emission lines near the K$\alpha$ energies of Mg XI and Si XIII. Adding Gaussian emission lines, with $\sigma=0.01$ keV, at energies 1.31 & 1.85 keV decreases the C-statistic by $\sim 14$, while decreasing the degrees of freedom by 4. The change in the AICc values indicates that adding these Gaussian lines increases the likelihood by a factor of $\sim 20$. The data thus suggest that these lines are real, but that the single-temperature plasma model does not get their energies correct.

Another way of showing the statistical significance of these lines is by looking at the probability of obtaining the observed number of counts in these energy ranges, given the spectral fit to the remainder of the spectrum. While the single temperature hot plasma model predicts a total of 26.4 photons in the energy ranges 1.2-1.4 and 1.75 - 1.95 keV, the data show 46 photons in these bins. Assuming a Poissonian distribution, this has a single-trial probability of 0.2%.

The single-temperature model predicts the Mg K$\alpha$ line to be at slightly higher energies ($\sim 1.47$ keV) than observed (1.31 keV). At the required temperature to explain the high-energy photons as bremsstrahlung, the Mg will be principally in Mg XII, which has a K$\alpha$ line at 1.47 keV, rather than the observed 1.31 keV better fit by Mg XI. Thus, a lower temperature plasma could match the observed Mg emission line, but would not explain the observed continuum emission at higher energies. We consider three models to address this problem.

First, we consider a two-temperature plasma to model the observed spectrum (tbabs*(vapec+vapec)). Similar models have often been used to model SNR emission, and have been suggested to indicate cooler emission from the reverse shock and hotter emission from the blast wave (e.g. Jansen et al., 1988; Willingale et al., 2002). To minimise the number of free parameters, we only allowed the abundances of the (measurable) elements O, Ne, Mg & Si to vary. When fitting this model, we see that the error bars in the abundances of these four elements overlap. Numerical simulations of supernovae with progenitor masses from 9 to 120 M*⊙* in Sukhbold et al. (2016) show that the relative abundance ratios of these elements with respect to Si are $\sim O(1)$. Since the errors on the abundance values are large in our case, we linked the abundances of O, Ne, Mg and Si in both plasmas. This model gives $N_{H}=2.7_{-1.8}^{+2.8}\times 10^{21}$cm*-2*, $kT=0.52_{-0.11}^{+0.24}$ for the cold plasma and $kT\geq 8.36$ keV for the hot plasma (see Fig. 5). We find the relative abundances of the linked elements O, Ne, Mg, Si and S to be $11_{-7}^{+24}$ times solar. This model is shown in Fig. 5. The Fe abundance prefers to be smaller than the abundances of O, Ne, Mg, Si & S. Fixing the abundances of O, Ne, Mg, & Si to 11.0, we constrain the Fe abundance to be $1.0_{-0.85}^{+1.08}$. Compared to the absorbed power-law model, the C-statistic decreases by 31.82 while the degrees of freedom only decreases by 5. Thus, the AICc value decreases by 24.84, implying that this model is $\sim 2.5\times 10^{5}$ times more likely. The “goodness” simulations show that $\sim 59\%$ of the realizations have a lower CvM statistic that the observed data, showing that this model is a good fit to the observed data.

The lower ionization state of Mg & Si could also be explained by a plasma that is not yet in ionization equilibrium, as is often the case in young SNRs. A vnei model with a lower ionization time-scale (The ionization time-scale to attain equilibrium at $10^{7}$ K is $\sim 10^{11}-10^{12}$ s cm*-3* for Mg and Si; Smith & Hughes, 2010) can explain the continuum emission at higher energies, while explaining the low-ionization emission lines of Mg and Si. The relative abundances of O, Ne, Mg & Si were linked together as in the previous models. When fitting this model, we find $N_{H}=5.9_{-2.2}^{+1.8}\times 10^{21}$ cm*-2*, $kT>22$ keV and ionization time-scale $\tau=1.6_{-0.7}^{+1.1}\times 10^{10}$ s cm*-3*. The ionization time-scale is similar to that of the young supernova remnant Cassiopeia A (Willingale et al., 2002) . Though we find the abundance of O, Ne, Mg and Si to be consistent with solar values within their error limits ($1.7_{-0.7}^{+1.2}$), the low ionization time-scale of this model argues for the X-rays being generated by a SNR, rather than an AGN. This model has similar AICc and CvM statistic values as the two-temperature plasma model, suggesting that both models are equally likely.

The emission at higher energies could also be due to synchrotron emission, e.g. from a pulsar wind nebula or synchrotron-dominated shocks, instead of a hotter plasma component. Synchrotron spectra can typically be modelled using a simple power-law model, and many SNRs show such a strong power-law component. Thus we also try fitting a tbabs*(pegpw + vapec) model to the observed spectrum. Though the best-fitting model does indicate super-solar abundances, the power-law component is very hard($\Gamma=0.4_{-1.3}^{+1.1}$). Using $\Gamma=2.7$ (typical of forward shocks in SNRs; eg. Torii et al., 2006; Tamagawa et al., 2009), gives a poorer fit (the “goodness" exceeds 95%, indicating a poor fit). Thus, synchrotron emission from shocks in a SNR does not appear to explain the flux at higher energies well.

The observed X-ray spectrum also shows an increase in the photon count rate at $2.6$ keV, which is very close to the K$\alpha$ emission line of S XVI (2.62 keV). This can be incorporated into the tbabs*(vapec+vapec) model by allowing for an increased abundance of S in the hot plasma. Linking the abundance of S in both plasmas brings its abundance down, since we do not see any emission corresponding to the expected S XV line that should be produced by the plasma at the lower temperatures. Similarly, a higher ionization time-scale of S in the vnei model($\sim 10^{13}$ s cm*-3*) could also account for the relative over-abundance of S XVI as compared to S XV. Understanding why S behaves differently from the other elements requires further observations.