EVIDENCE FOR BLACK HOLE GROWTH IN LOCAL ANALOGS TO LYMAN BREAK GALAXIES

To select our sample for X-ray observations, we begin with the 31 LBAs with HST imaging (Overzier et al. 2009) and Spitzer (IRAC+MIPS) photometry (Overzier et al. 2011). In Figure 1 we show the standard "BPT" diagram (Baldwin et al. 1981) used to classify the optical emission-line spectra of galaxies, where the emission-line ratios are from Table 2 in Overzier et al. (2009). We plot all the galaxies in the GALEX-SDSS cross-matched sample and then highlight the 31 LBAs in our sample (large symbols). We note that seven⁵ of the LBAs lie in the composite region in the "BPT" diagram, which is known to harbor "composite objects," which are typically (but not necessarily) objects having both star formation and a Type 2 AGN (see Kauffmann et al. 2003; Kewley et al. 2006). Hereafter, we refer to this sample as "LBA composites." In this paper, a flat Λ cold dark matter (ΛCDM) cosmology with h = 0.7, Ω_Λ = 0.7 is assumed.

Zoom In Zoom Out Reset image size

XMM-Newton observations were performed for 6 of the 7 LBA composites. SDSS J0054+1554 was given a "C" priority and not observed (marked as green asterisk in Figure 1). A summary of observations is shown in Table 1.

The reduction of the XMM-Newton data was done by XAssist (ver. 0.97)⁶, which is a software for automatic analysis of X-ray data. XAssist runs the XMM-Newton Science Analysis System (SAS) packages automatically to filter the data, generate the light curves, and extract the spectra of sources in the field, as well as compute the associated redistribution matrix files (RMFs) and ancillary region files (ARFs). The LBA composites are weak in X-ray emission, and their optical positions were used to define the source regions in X-ray data reduction. XAssist determines the source extraction sizes for source by fitting an elliptical Gaussian model to the source image, which in the case of on-axis XMMobservations typically results in regions of size 20''–25''. Background regions were chosen as annuli centered on the sources to extract the background spectra. Table 4 lists the PN and MOS counts for each source.

The Optical Monitor (OM) aboard on XMM-Newton allows us to make simultaneous observations in optical/UV and X-ray bands. We processed the OM data with the omichain package in SAS to obtain UV count rates and the corresponding magnitudes of each object, as well as their associated uncertainties. We obtained exposures in the U (344 nm), UVW1 (291 nm), UVM2 (231 nm), and UVW2 (212 nm) band filters, but only the UVW1 was used for all six LBAs. We list the UV detections with each filter and the luminosity measured in UVW1 in Table 2. The conversions from UV count rates to specific fluxes (in units of erg s⁻¹ cm⁻² Å⁻¹) were calculated based on the white dwarf standard stars⁷.

Table 3 lists the observed and extinction-corrected [O iii]λ5007 line luminosity (L_{[O iii],obs} and L_{[O iii],corr}, respectively) for each LBA composite. We obtain L_{[O iii],obs} from the SDSS Data Release 7 MPA/JHU catalog. We derive L_{[O iii],corr} from L_{[O iii],obs} using the observed Hα/Hβ ratio, an intrinsic (Hα/Hβ)₀ ratio of 2.9, and the standard extinction curve for galactic dust (Osterbrock & Ferland 2006). In Table 3, we also list the mid-infrared (MIR) and far-infrared (FIR) luminosity, L_24 μm and L_FIR, respectively (Overzier et al. 2011).

Table 3. [O iii] and IR Luminosities

ID	$L_{\rm {[O\,\mathsc {iii}],obs}}$	$L_{\rm {[O\,\mathsc {iii}],corr}}$	L24 μm	LFIR
	(1041)	(1041)	(1044)	(1044)
SDSS J0808+3948	0.43 ± 0.01	0.66	0.76	1.1 ± 0.2
SDSS J2103−0728	5.00 ± 0.06	12.03	4.8	8.8 ± 1.3
SDSS J1434+0207	1.32 ± 0.03	2.84	0.48	2.0 ± 0.6
SDSS J0214+1259	0.57 ± 0.04	1.22	1.9	9.1 ± 1.7
SDSS J0922+4509	2.05 ± 0.06	4.75	2.4	14.3 ± 2.2
SDSS J0802+3915	1.54 ± 0.05	4.47	1.2	6.5 ± 2.0

Notes. [O iii] luminosities are obtained from MPA/JHU catalog of SDSS DR7. The mid- and far-IR luminosities are based on Spitzer photometry (Overzier et al. 2011). All luminosities are in units of erg s⁻¹.

Download table as:  ASCIITypeset image

X-ray spectral fitting was performed with XSPEC (ver. 12). X-ray photons are collected by three detectors on XMM-Newton, i.e., PN, MOS1, and MOS2. The data of three detectors can be fitted simultaneously in XSPEC by tying all parameters together except a multiplicative constant factor. We froze this factor at 1 for PN, and allowed it to vary between 0.8 and 1.2 for MOS1 and MOS2. Since our targets have limited counting statistics (see Table 4), we first fit their spectra using the C-statistic (Cash 1979), which is often used in this low counts regime. The spectra were grouped to 1 count bin⁻¹, to improve the performance of XSPEC (Teng et al. 2005). The simplest spectral model often fit to AGNs is an absorbed power-law model. The X-ray photons suffer from two sources of absorption. The first is due to the neutral hydrogen in our own galaxy and its column density (N_H,G) is listed in Table 1. The other is the intrinsic obscuration from the gas at the redshift of the source (N_H,in). N_H,in is a free parameter in the spectral fitting as well as the photon index Γ of the power law. However, due to the low signal-to-noise ratio (the number of photons detected by the PN and MOS detectors are listed in Table 4), we were not able to constrain both N_H,in and Γ simultaneously in several cases. Therefore for sources with low S/N, we alternately froze N_H,in and Γ and let the other parameter vary. The same methodology was applied when adding an additional thermal component into the power-law continuum.

Table 4. X-ray Spectral Fits Using the C-statistic

ID	Counts	Γ	C-stat/dof	NH,in (Γ = 1.9)	C-stat/dof	$L_{\rm {0.5\hbox{--}2 keV}}$a	$L_{\rm {2\hbox{--}10 keV}}$a	$L_{\rm {2\hbox{--}10 keV}}/L_{\rm {[O\,\mathsc {iii}],corr}}$
	PN/M1/M2			(1022 cm−2)		(1041 erg s−1)	(1041 erg s−1)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
SDSS J0808+3948	75/34/44	1.66 (1.41–1.92)	116.11/138	<0.14	115.82/138	2.11 (1.55–2.80)	4.01 (2.94–5.33)	6.08
SDSS J2103−0728	76/26/17	2.48 (2.08–2.94)	112.78/100	<0.04	118.60/100	2.93 (2.14–3.95)	4.83 (3.53-6.51)	0.40
SDSS J1434+0207	86/41/43	1.46 (0.88–2.04)	134.41/156	<0.27	135.80/156	2.65 (1.53–4.28)	4.41 (2.55–7.13)	1.55
SDSS J0214+1259	83/25/28	2.72 (1.91–3.68)	133.53/122	<0.09	136.31/122	2.39 (1.48–3.68)	4.30 (2.66–6.62)	3.51
SDSS J0922+4509	54/70/62	1.89 (1.50–2.30)	172.57/165	<0.23	171.02/165	12.35 (6.64–19.94)	22.70 (12.21–36.66)	4.78
SDSS J0802+3915	15/8/19	1.71 (−0.58–4.17)	28.56/38	<1.37	28.62/38	9.29 (0.81–31.96)	16.80 (1.47–57.81)	3.76

Notes. The errors and upper limit correspond to the 90% confidence level. Column 1: LBA names; Column 2: numbers of counts detected by PN, MOS1, and MOS2; Column 3: derived photon indices (Γ), no intrinsic absorption assumed; Columns 4 and 6: reduced C-statistic; column (5): derived intrinsic column densities at the redshifts of the sources, by freezing Γ at 1.9; Column 7: luminosities in soft X-ray band (0.5–2.0 keV); Column 8: luminosities in hard X-ray band (2.0–10.0 keV); Column 9: ratio of hard X-ray and extinction-corrected [O iii] luminosities. ^aThe errors of luminosities correspond to the errors of normalizations of power-law components in the spectral fit.

Download table as:  ASCIITypeset image

3.1.1. Absorbed Power-law Model

As mentioned above, we fitted the spectrum in two ways to derive the power-law photon index and the intrinsic absorption. First, we freed the power-law photon index and assumed no absorption at the redshift of the source (N_H,in = 0 cm⁻²) and that the continuum power-law spectra is only absorbed due to the Galactic column density (listed in Table 1). Again in many cases there was not sufficient signal for the absorption and slope to both be free parameters. We list the photon indices derived from this simple absorbed power-law model in Column 3 in Table 4, as well as the corresponding C-statistic in Column 4. Second, the spectra were fitted with the power-law model this time with the intrinsic absorption as a free parameter while the photon index was fixed as Γ = 1.9, which is a mean value for X-ray detected LBGs (Laird et al. 2006) and is also a typical spectrum for an unobscured AGN. The values of N_H,in are shown in Column 5 in Table 4. The errors of photon indices in Column 3 and the upper limits of intrinsic absorptions in Column 5 were calculated at a confidence level of 90% (i.e., single interesting parameter ΔC = 2.7). We also list in Table 4 their derived luminosities in soft X-ray (0.5–2.0 keV) and hard X-ray (2–10 keV) bands (Columns 7 and 8, respectively), and the ratio of hard X-ray and [O iii] luminosities (Column 9). The associated uncertainties for luminosity correspond to the 90% confidence range of normalization of the continuum power-law component in spectral fit. See Figure 2 for the spectral fits of each source with a fixed photon index Γ = 1.9.

Zoom In Zoom Out Reset image size

We also fitted the X-ray spectra of our targets using the χ²-statistic to see how different the fits are compared with those by C-statistic. The fits were performed in the same way as those using the C-statistic above. The spectra were grouped in 10 counts bin⁻¹ for both PN and MOS detectors, and the spectra of MOS1 and MOS2 were combined in the fits. Table 5 lists the photon index and intrinsic column density of each source derived while the other one is frozen in the spectral fitting. SDSS J0802+3915 is excluded due to insufficient bins for fitting after grouping into 10 counts bin⁻¹. Also, spectral plots are shown in Figure 3, which are fits with photon indices fixed at 1.9. Fitting the spectra using χ² does not make any improvements or significant difference in estimating the absorption and photon index compared to the C-statistic method. Thus, we use the X-ray luminosity derived from the C-statistic fit in the following discussion.

Zoom In Zoom Out Reset image size

Table 5. X-ray Spectral Fits Using the χ²-statistic

ID	Γ	χ2ν	NH,in (for Γ = 1.9)	χ2ν
			(1022cm−2)
(1)	(2)	(3)	(4)	(5)
SDSS J0808+3948	1.71 (1.33–2.13)	1.01	<0.24	0.553
SDSS J2103−0728	3.19 (2.40–4.28)	0.19	<0.08	2.15
SDSS J1434+0207	1.24 (0.40–2.06)	1.09	0.16 (0–2.44)	1.17
SDSS J0214+1259	>2.05	0.79	<0.68	1.22
SDSS J0922+4509	2.83 (2.01–4.18)	0.72	<0.18	0.99

Notes. The errors and upper limit correspond to the 90% confidence level. Column 1: LBA name; Column 2: derived photon indices (Γ), no intrinsic absorption assumed; Columns 3 and 5: reduced χ²; Column 4: derived intrinsic column densities at the redshifts of the sources, by freezing Γ at 1.9.

Download table as:  ASCIITypeset image

Additionally, as SDSS J0808+3948, SDSS J1434+0207, and SDSS J0922+4509 have relatively more photons detected, we also fitted their spectra by thawing both the parameters of photon index and intrinsic absorption (using both the C and χ² statistics). Table 6 shows the values of these parameters and their associated errors or upper limits. Since we have two interesting parameters in the fits, the errors at the 90% confidence level are calculated according to Δχ² or ΔC = 4.6.

Table 6. X-ray Spectral Fits With Both Photon Indices and Column Densities as Free Parameters

ID	Statistic	Γ	NH,in	χ2ν or
	Method		(1022cm−2)	C-stat/dof
SDSS J0808+3948	χ2	1.85 (1.22–3.05)	<0.53	0.69
	C	1.79 (1.35–2.50)	<0.22	0.86
SDSS J1434+0207	χ2	1.23 (0.02–3.28)	<21.8	1.18
	C	1.41 (0.713–2.47)	<0.35	0.87
SDSS J0922+4509	χ2	4.56 (1.87–10.0)	0.59 (0–2.91)	0.71
	C	3.35 (1.94–6.56)	0.42 (0.05–1.41)	1.01

Notes. The errors and upper limit correspond to the 90% confidence level. The errors of photon indices and column densities correspond to Δχ² or ΔC = 4.6, since there are two interesting parameters.

Download table as:  ASCIITypeset image

3.1.2. Power-law Plus Thermal Model

The LBAs are the sites of strong starbursts, so we would expect to see thermal X-ray emission from the hot gas that is a characteristic features of such systems (e.g., Grimes et al. 2005, 2006, 2007). We therefore fit models in which we added a thermal component to the absorbed power-law model aforementioned. We used the thermal plasma model apec in XSPEC instead of the older model, vmekal, used in Grimes et al. (2007). The apec model has three free parameters: the plasma temperature, the metal abundance (in solar units), and the normalization. Due to the low number of photon counts, we fixed the abundance at the solar value. The parameter of temperature and normalization were free in fitting. However, due to insufficient photon counts, we fixed the temperature at 0.5 keV for SDSS J0808+3948, SDSS J0922+4509, and SDSS J0802+3915 which is a typical value for starburst galaxies (e.g., Grimes et al. 2005, 2006, 2007). Here, we kept the photon index fixed at 1.9 and allowed the intrinsic absorption at the redshift of the LBA to vary. The results are shown in Table 7, which lists the temperature of plasma, the column density, and the derived luminosities of both the thermal and power-law component, as well as the C-stat for comparison with the single power-law fitting. The errors for parameters were calculated according to ΔC = 4.6, and the uncertainty of luminosity was proportional to the corresponding error of normalization of each component. For the three LBAs where the temperature was permitted to vary the temperature ranged between 0.3 and 0.6 keV. The thermal component luminosities were on the order of 10⁴¹ erg s⁻¹, with 10⁴⁰ erg s⁻¹ for SDSS J0808+3948 and about 10⁴² erg s⁻¹ for SDSS J0802+3915. The power-law component luminosity in the hard X-ray range (2–10 keV) is within a factor of two of those derived from the single power-law model spectral fit mentioned above. This power-law plus thermal model fit makes no statistically significant difference in the derived column density compared to that of the single power-law model.

Table 7. Apec Plus Power-law Model

ID	Temperature	NH	Lthermala	${L_{\rm {0.5\hbox{--}2\, keV,PL}}}$b	${L_{\rm {2\hbox{--}10\, keV}}}$c	C-stat/dof
	(eV)	(1022cm−2)	(1041 erg s−1)	(1041 erg s−1)	(1041 erg s−1)
SDSS J0808+3948	500 (fixed)	0.10 (0.01–0.41)	0.29 (0.15–12.69)	1.78 (1.09–2.74)	3.76 (2.31–5.79)	114.8/133
SDSS J2103−0728	422 (276–755)	0.17 (0–1.03)	1.74 (0.16–3.50)	1.45 (0.56–2.63)	3.85 (1.49–6.75)	101.9/96
SDSS J1434+0207	339 (217–689)	1.07 (0.18–2.55)	1.48 (0.39–2.98)	0.86 (0.19–1.75)	6.61 (1.47–13.44)	129.3/152
SDSS J0214+1259	570 (326–844)	266.5 (13.9–555.5)	2.23 (1.05–3.84)	1.95 (1.0–4.73)	6.20 (3.10–15.03)	125.7/118
SDSS J0922+4509	500 (fixed)	0.16 (0–0.65)	2.98 (1.50–11.10)	6.04 (1.13–12.51)	21.84 (4.10–45.23)	169.4/162
SDSS J0802+3915	500 (fixed)	1.96 (0.81–50.35)	12.64 (1.97–31.67)	14.23 (0.23–166.63)	40.58 (0.66–475.17)	24.42/35

Notes. ^aThermal emission luminosity. ^bPower-law luminosity in 0.5–2 keV band. ^cTotal luminosity in 2–10 keV.

Download table as:  ASCIITypeset image

Since the signal-to-noise of the XMM-Newton spectra is low, we attempted to constrain any thermal component by simultaneously fitting the spectra of all LBAs together assuming that they have approximately the same hot gas temperature, i.e., we tied the temperature parameters together between the fits but let the apec normalization vary for each LBA. We also fixed the power-law index at 1.9, and tied the normalization of power-law component in our fit (although note that overall normalization for each LBA was allowed to be free). The temperature given by this joint fit is 0.56 keV and its 90% confidence level is 0.43–0.69 keV. We list the intrinsic column density, the thermal luminosity, and the soft and hard X-ray luminosities contributed by the power-law component in Table 8. We see that the column densities and thermal luminosities given by this joint fit are consistent with those in the individual fits, but with improved constraints.

Table 8. Apec Plus Power-law Model in a Joint Fit

ID	$N_{\rm {H}}$	${L_{\rm {thermal}}}$a	${L_{\rm {0.5\hbox{--}2keV,PL}}}$b	${L_{\rm {2\hbox{--}10keV}}}$c
	(1022 cm−2)	(1041 erg s−1)	(1041 erg s−1)	(1041 erg s−1)
SDSS J0808+3948	0.14 (0.07–0.28)	0.39 (0.13–0.75)	2.18 (1.33–3.36)	4.87 (2.98–7.50)
SDSS J2103−0728	0.35 (0.05–1.17)	1.83 (0.60–3.48)	0.97 (0.37–1.76)	3.31 (1.28–6.00)
SDSS J1434+0207	0.59 (0.06–1.86)	1.12 (0.37–2.12)	1.57 (0.35–3.19)	6.91 (1.53–14.06)
SDSS J0214+1259	769 (45–1240)	1.94 (0.64–3.69)	⋅⋅⋅	9.11 (4.67–22.1)
SDSS J0922+4509	0.19 (0.05–0.52)	1.27 (0.42–2.42)	8.29 (1.55–17.17)	19.43 (3.64–40.24)
SDSS J0802+3915	2.94 (0.50–9.30)	6.34 (2.08–12.05)	2.31 (0.04–27.05)	63.49 (1.03–743.45)

Notes. ^aThermal emission luminosity. ^bPower-law luminosity in 0.5–2 keV band. ^cTotal luminosity in 2–10 keV.

Download table as:  ASCIITypeset image

In Figure 4, we compared the two compositions of soft X-ray luminosity, i.e., the thermal and power-law luminosities in 0.5–2.0 keV, where the dashed line indicates where both values equate. This comparison was done separately for both the cases in which the spectra were fit individually and those that were fit simultaneously. These are shown in the left and right panels of Figure 4, respectively. The results derived from the individual spectra fits imply that these two mechanisms play a nearly equal role in soft X-ray emission, except in J0808+3948 where the power-law soft X-ray luminosity is about 1 dex higher than that of thermal emission. For the joint spectral fit, the luminosities from the two emission mechanisms are comparable, within the range of 1 dex. The obscuration of SDSS J0214+1259 is Compton thick in the joint fit, and the power-law luminosity at 0.5–2 keV is negligible (not shown).

Zoom In Zoom Out Reset image size

We list the UV luminosity for each LBA measured from the OM observation in Table 2, and their UV images are shown in Figure 5. We derive the SFR from our UV luminosities using the formula from Rosa-González et al. (2002), which is based on evolutionary synthesis models of starbursts by assuming a Salpeter initial mass function (IMF). For consistency with Overzier et al. (2009), we divide the Rosa-González et al. relation by a factor of 1.5 to reflect the use of a Kroupa rather than a Salpeter IMF (as described in Calzetti et al. 2009). The resulting formula is as follows:

$$\begin{equation} {\rm SFR(UV)}=9.3\times 10^{-29}{\it L}_{\nu } \ ({\rm erg\ s}^{-1} {\rm Hz}^{-1}) \, M_{\odot }\, {\rm yr^{{-1}}}. \end{equation} \tag{ 1 }$$

Zoom In Zoom Out Reset image size

In Table 9 and Figure 6, we compare our SFR derived from the UV luminosity with that derived from a combination of the 24 μm and Hα luminosities (see Overzier et al. 2009, for details). Due to the uncertain infrared SEDs, no K-correction was applied to the 24 μm luminosity in Overzier et al. (2009), and the actual SFRs might be somewhat higher than their derived values. The UV-derived SFR is lower than that derived from the IR continuum plus optical emission lines. This is consistent with the fact that the UV photons suffer significant dust attenuation and are re-emitted in the IR (see also Overzier et al. 2011).

We note that any AGN contribution to the UV continuum is negligible in these objects. This is demonstrated most directly by HST COS FUV spectra of two of our targets (Heckman et al. 2011).

The far-infrared (FIR) luminosity is considered to be a reliable indicator of the global SFR of star-forming galaxies. In particular, Ranalli et al. (2003) found that the 2–10 keV X-ray luminosity has a tight linear correlation with the FIR luminosity by investigating a sample of local star-forming galaxies. We show our LBAs in the $L_{\rm {2\hbox{--}10 keV}}\hbox{--}L_{\rm {FIR}}$ diagram (see Figure 7) with the Ranalli relation, which is indicated by the solid line with two dashed lines indicating a factor of two above and below. The hard X-ray luminosities of the LBAs are taken from Table 4, which are derived from the absorbed power-law model fit with the photon index fixed at 1.9. For comparison, we also show the sample of Luminous Infrared Galaxies (LIRGs) from Iwasawa et al. (2009) and Lehmer et al. (2010), where Iwasawa et al's sample was divided into two groups of AGNs and "Hard X-ray Quiet" galaxies (HXQs) indicated by diamond and asterisk symbols, respectively, and Lehmer et al's sample was indicated by triangle symbols⁸. We can see that our LBAs are about 0.5 to 1 dex above the Ranalli et al. relation, and even more displaced from the non-AGN LIRGs. This implies that a normal star-forming process would only account for a minority of the hard X-ray emission of these LBAs.

We also fit a $\log (L_{\rm {2\hbox{--}10 keV}})\hbox{--}\log (L_{\rm {FIR}})$ correlation from our LBA sample by applying the survival analysis (ASURV Rev 1.2; Isobe & Feigelson 1990), which gives

$$\begin{equation} \log (L_{\rm {2-10 keV}})=(0.41\pm 0.35)\times \log (L_{\rm {FIR}})+(23.42\pm 11.28). \end{equation} \tag{ 2 }$$

The dash-dotted line shows the best fit for the correlation.

We find similar results in the soft X-ray band. We fit the correlation between the soft X-ray (0.5–2 keV) and FIR luminosity of our LBA sample, and compared it with the result of Ranalli et al. (2003). The survival analysis gave a free slope fit

$$\begin{equation} \log (L_{\rm {0.5-2 keV}})=(0.42\pm 0.24)\times \log (L_{\rm {FIR}})+(22.89\pm 10.76). \end{equation} \tag{ 3 }$$

Besides the linear slope fit, Ranalli et al. (2003) gave a free-slope fit of $L_{\rm {0.5\hbox{--}2 keV}}\propto L^{0.87\pm 0.08}_{\rm {FIR}}$ for star-forming galaxies. In Figure 8, the dash-dotted line indicates the best fit of LBAs, and the Ranalli curve is shown as the dashed line below it. As in the case of the hard X-ray emission, only part of the soft X-ray flux is likely to be contributed by normal processes associated with star formation.

Given that the LBAs suffer from less dust extinction than typical local starbursts with similar SFRs (Overzier et al. 2011), one might worry that the high ratio of X-ray to far-IR luminosity seen in Figures 7 and 8 is due to the incomplete conversion of far-UV into far-IR light. This is not the case. As discussed in Overzier et al. (2011), the SFRs derived from the observed far-UV are much smaller than those derived from the far-IR data, implying that the far-IR is a good proxy for the SFR. Our results in Section 3.2 and Figure 6 further support this.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The results from Section 4.1 are consistent with interpreting the optical spectra as indicating the presence of an AGN. The best tracer of both Type 1 and Type 2 AGNs in the optical spectra is the [O iii]λ5007 emission line. In this case, we would expect that the amount of hard X-ray emission from the LBAs would be consistent with the values found for AGN with similar [O iii] luminosities.

In Figure 9, we plot a histogram of the ratio of the hard X-ray (2–10 keV) luminosity obtained from the power-law model spectral fit of our LBAs to its observed [O iii] luminosity. We also show the empirical distribution of the luminosity ratios of Type 1 (dashed blue line) and Type 2 AGNs (dot-dashed red line) in this plot (Heckman et al. 2005a; LaMassa et al. 2011). The ratios of hard X-ray to [O iii] luminosities in the LBAs are too low (by about an order of magnitude) to be consistent with the values seen in Type 1 AGNs. However, they do lie within the relatively broad range of values seen in Type 2 AGNs. The relatively weak hard X-ray luminosity of Type 2 AGNs is due to obscuration of the hard X-rays by the high column densities of gas associated with the AGN torus. In contrast, the [O iii] line is formed in the 100 to 1000 pc scale narrow line region and is not affected by the torus surrounding the black hole.

Zoom In Zoom Out Reset image size

At the first sight, the presence of significant obscuration of the hard X-ray emission in the LBAs would not be consistent with the relatively low absorbing column densities listed in Table 4. Here, we see that the column density derived from the spectral fitting of an absorbed power law gives the 90% confidence level upper limit at the order of 10²¹ cm⁻² (10²² cm⁻² for SDSS J0802+3915). These columns are much too low to attenuate the hard X-ray emission by ∼ an order of magnitude. Such a seeming paradox is often seen in Type 2 AGNs and is most likely due to fitting an overly idealized X-ray spectral model. More realistic models (e.g., partial covering models and/or models with reprocessing; Turner et al. 1997; LaMassa et al. 2009, 2011) can reconcile the observed X-ray spectrum with heavy obscuration.

Alternatively, the actual absorbing column densities derived from the fits to the X-ray spectra may be correct. In this case, the low ratio of hard X-ray to [O iii] luminosities relative to unobscured AGN would require that the [O iii] line is predominantly produced by the starburst rather than the AGN. As discussed by Overzier et al. (2009), this possibility cannot be excluded.

Assuming for the moment that the relatively high ratios of X-ray to far-IR luminosity indicate that an AGN is present in these LBAs, our results still do not establish whether this AGN is more luminous and heavily obscured in hard X-rays or less luminous and relatively unobscured. An additional diagnostic of an obscured AGN is a strong iron Kα emission line with a large equivalent width (EW) of about 1 keV (e.g., LaMassa et al. 2009, 2010). In Figure 2, we show the comparison between the data and the spectral fit from an absorbed power law. We find that there is a large data-to-model ratio at the Fe Kα energy in 1434+0207, 0214+1259, and 0802+3915. We used a power-law plus Gaussian model to fit the PN spectra of 1434+0207 and 0214+1259 (insufficient photon counts for 0802+3915). We removed the 5–8 keV photons first to determine the power-law index, then froze this parameter, and added the 5–8 keV photons with a Gaussian component. The line energy is 6.6 keV for 1434+0207 and 6.2 keV for 0214+1259, and the spectral plots are shown in Figure 10. The plots indicate the possibility of an iron emission line in these LBAs. However, the photon counts are too few to give good estimates of the EW of iron lines.

Zoom In Zoom Out Reset image size

In order to further investigate the presence of a Fe line in our sample, we loaded the spectra of all six LBAs in XSPEC to construct an average spectrum due to the small number of photons of each LBA (effectively stacking the spectra in model space). This was done in the same way by which we fitted the thermal component simultaneously in Section 3.1.2. This is not a fit to a stacked spectrum, but rather a joint fit to the set of observed spectra without shifting them to their rest-frame. The averaged spectrum was modeled as a continuum power law in the range of 3–8 keV plus a Gaussian emission line. The intrinsic line width (σ) in the Gaussian component was fixed at 0.01 keV, and the photon indices of the continuum power law were set to 1.9. By tying the Gaussian normalization of all data sets, we obtain the emission-line energy at 6.40 keV, with a 90% confidence region (ΔC-stat=4.6 for two model components) of 6.09–6.68 keV. The EW is 1.0 keV from this joint fit, and the associated 90% confidence upper limit of EW is about 4.4 keV, where the upper limit is obtained by scaling the confidence region of the line intensity by the best-fit value of the EW, which is likely to be overestimated (Yaqoob 1998). However, the data has low signal-to-noise ratio, and the emission-line profile is not significant. When we removed the Gaussian component in the spectral fitting, the C-stat only increased by 5.5 compared to the value of 109.6/102 (dof) for the model with Gaussian.

We then performed the Markov Chain Monte Carlo (MCMC) simulation to test the significance of the iron emission line from the joint fit. We generated six chains, and the length of each chain is 5 × 10⁵. The 90% confidence level for the line energy from the MCMC is 6.10–6.93 keV, and the estimated upper limit of EW is 2.5 keV. The histogram distribution of the line energy and normalization parameters is shown in Figure 11. The significance of the Fe Kα line is calculated as the fraction of non-zero line normalization in the chains, which is found to be 95.3%, i.e., at the ∼2 σ level.

Zoom In Zoom Out Reset image size

One other test for the presence of a Type 2 AGN in the LBAs is to compare the amount of mid-IR emission expected from such an AGN with the observed [O iii]λ5007 luminosity. LaMassa et al. (2010) have shown that both of these are good proxies for the bolometric luminosity of an obscured AGN (unlike the 2–10 keV X-ray luminosity).

In Figure 12 we plot histograms of the ratio of mid-IR/[O iii] luminosities for the sample of Type 2 AGN in LaMassa et al. (2011) and for the LBAs in this paper. The distribution for the Type 2 AGN has a mean of 1.89 dex and a dispersion of 0.57 dex compared to 2.71 and 0.34, respectively for the LBAs. All six of the LBAs lie above the mean value for the Type 2 AGN.

Zoom In Zoom Out Reset image size

This suggests that if the LBAs are indeed starburst/AGN composite systems, the starburst dominates over the AGN in the mid-IR in most cases. In a future paper, we will examine this issue in more detail spectroscopically using Spitzer intensified Reticon spectrograph (IRS) data. Note that AGN emission from the torus peaks in the mid-IR, while the emission from dusty starbursts peaks in the far-IR (as is also the case for the LBAs). We therefore conclude that the bolometric luminosity of the LBAs is primarily due to the starburst.

As noted above, it is possible that the [O iii] emission in these LBAs is primarily produced by the starburst and that the hard X-ray source suffers little attenuation. In this case, the AGN contribution to the bolometric luminosity will be even smaller.

As discussed in the introduction, these LBA composites may offer us an unparalleled opportunity to study the processes involved in the formation of SMBHs in the early universe (e.g., Elmegreen et al. 2008). In this section, we will infer the properties of the SMBHs in our sample if galaxies using standard methods.

We will consider two possibilities. The first is the minimal case in which the hard X-ray source is not obscured. In this case, the bolometric correction to the hard X-ray luminosity in Marconi et al (2004) leads to a range of AGN bolometric luminosities from 1 to 6 × 10⁹ L_☉. The implied black hole masses then range from ∼3 × 10⁴ to ∼2 × 10⁵(L/L_Edd)⁻¹ M_☉. If these black holes are indeed radiating near the Eddington limit, the implied masses are then in the range of the "intermediate-mass" black holes found in some nearby low-mass galaxies (e.g., Peterson et al. 2005). In the context of the high-redshift universe, similar objects could represent the seeds for the subsequent growth of more massive systems.

The second possibility is that the hard X-rays are significantly obscured in these LBAs and that the [O iii]λ5007 emission line is powered by a Type 2 AGN. Adopting the bolometric correction to the extinction-corrected [O iii] luminosity from Kauffmann & Heckman (2009) then yields AGN bolometric luminosities of ∼10¹⁰ to ∼2 × 10¹¹ L_☉ and black hole masses of ∼3 × 10⁵ to ∼6 × 10⁶(L/L_Edd)⁻¹ M_☉. In this case, an Eddington-limited black hole would have a mass of order 10⁻³ of the DCO stellar mass (Overzier et al. 2009). This is similar to the mass ratio seen in present-day galactic bulges (Haring & Rix 2004; Marconi & Hunt 2003). However, if the [O iii] line is contaminated or dominated by the starburst, the AGN bolometric luminosity and implied black hole mass would be overestimated.