EVIDENCE FOR ELEVATED X-RAY EMISSION IN LOCAL LYMAN BREAK GALAXY ANALOGS

The launch of GALEXenabled rest-frame UV-selection of relatively nearby (z < 0.3) galaxies in order to discover potential local LBG analogs. Heckman et al. (2005) introduced such a sample of UV luminous galaxies (UVLGs) with FUV (λ ∼ 1550 Å) luminosities, L_FUV > 2 × 10¹⁰ L_☉. This selection was further refined by adding an FUV surface brightness criterion (I_FUV > 10⁹ L_☉ kpc⁻²; Hoopes et al. 2007) in order to select "super-compact UVLGs" with similar global properties as LBGs. Since this time, several studies have established these galaxies, henceforth called LBAs, as excellent analogs to LBGs (Basu-Zych et al. 2007, 2009; Overzier et al. 2008, 2009, 2010, 2011; Gonçalves et al. 2010; Heckman et al. 2011). LBAs are also notably different from other local star-forming galaxies, the former having lower dust attenuations per SFR (Basu-Zych et al. 2007; Overzier et al. 2011) and metallicities per stellar mass (Overzier et al. 2009) than the latter. LBAs are physically compact (half-light diameters ≈1–2 kpc; see contours and inset image in Figure 1), often disturbed systems whose Sloan Digital Sky Survey (SDSS) optical spectra indicate that they are starbursts (see Figure 4).

Zoom In Zoom Out Reset image size

We selected the sample of LBAs to be studied by Chandra by cross-matching the SDSS DR7 spectroscopic and photometric data with the GALEX GR3 data to find galaxies with L_FUV > 2 × 10¹⁰ L_☉ and I_FUV > 10⁹ L_☉ kpc⁻². In this process, we used the SDSS u-band half-light radius (r_{u, 50}), based on the SDSS DR7 expRad_u catalog value, to approximate the FUV surface brightness ($I_{\rm FUV}\equiv L_{\rm FUV}/\pi r_{u, 50}^2$; see the Appendix in Overzier et al. 2008 for further discussion). To determine L_FUV, we k-corrected the observed GALEX magnitudes using version 4 of the k-correct (Blanton et al. 2003) IDL code with the seven-band GALEX+SDSS (FUV, NUV, u', g', r', i', z') magnitudes. These steps for selecting LBAs are identical to the ones taken by Hoopes et al. (2007) in the original selection of LBAs, with the sole exception that we started with updated catalogs, resulting in 301 LBAs. For this pilot program, we targeted the four closest (z < 0.1) and most UV-luminous LBAs with Chandra, discussed in the following Section 2.1.1. In addition, we include in our LBA sample two other Chandra-observed galaxies that are considered good analogs of LBGs: Haro 11 and VV 114. Haro 11 satisfies the aforementioned LBA selection criteria; VV 114 has an FUV luminosity of 2.5 × 10¹⁰ L_☉ and FUV surface brightness of ∼7.8 × 10⁸ L_☉ kpc⁻², comparable to LBAs and LBGs.

2.1.1. X-Ray Measurements of LBAs

Chandra observed four SDSS+GALEX-selected LBAs in Cycle 12, using the Advanced CCD Imaging Spectrometer (ACIS)-S camera in the very faint observing mode. We used the Chandra Interactive Analysis of Observations (CIAO) package version 4.4.1 with Chandra Calibration Database version 4.5.3 for our analysis. Table 1 summarizes the Chandra observations for the LBAs. The exposure times (see Column 3) ranged from ∼9 to 20 ks. We produced exposure maps following the procedure described in Hornschemeier et al. (2001, Section 3.2). Briefly, these maps were normalized to the effective exposures of sources located at the aim points and include the effects from vignetting, gaps between the CCDs, bad column and pixel filtering, and the spatially dependent degradation of the ACIS optical blocking filter. A photon index of Γ = 1.4 was assumed in creating the exposure maps.

Table 1. Lyman Break Analogs: Summary of Chandra X-Ray Observations

ID	D	ObsID	Exposure	Aperture	Net Counts		Flux (2–10 keV)a	ϕ2–8/ϕ0.5–2
	(Mpc)		(ks)	('')	0.5–2 keV	2–10 keV	(10−15 erg s−1 cm−2)
J002101.0+005248.1	452	13014	19.4	5	27.0$^{+6.4}_{-5.3}$	6.4$^{+4.0}_{-2.8}$	7.0$^{+4.4}_{-3.1}$	0.2$^{+0.2}_{-0.1}$
J080619.5+194927.3	315	13015	20.0	5	24.0$^{+6.1}_{-5.0}$	9.3$^{+4.4}_{-3.3}$	9.9$^{+4.7}_{-3.5}$	0.4$^{+0.3}_{-0.2}$
J082355.0+280621.8	210	13012	9.0	10 × 20b	42.9$^{+7.8}_{-6.8}$	13.0$^{+5.4}_{-4.3}$	31.1$^{+1.3}_{-1.0}$	0.3$^{+0.2}_{-0.1}$
J225140.3+132713.4	279	13013	19.8	5	42.8$^{+7.7}_{-6.6}$	9.0$^{+4.4}_{-3.3}$	9.7$^{+4.8}_{-3.5}$	0.2$^{+0.2}_{-0.1}$

Notes. ^a2–10 keV flux assuming simple power law spectrum with N_H = 3 × 10²⁰ cm⁻² (average Galactic extinction for the sample, with variations in this value between different galaxies contributing to ∼3% uncertainty) and Γ = 1.9. ^bWe used a slightly tilted (position angle of −76) elliptical aperture, based on the optical structure, to account for any extended emission in this galaxy.

Download table as:  ASCIITypeset image

For our range of off-axis angles, the 90% encircled energy fraction radius is ∼1.2''. However, we typically extracted counts within a 5'' aperture to sufficiently include the extent of the optical emission, except in the case of J082355.0+280621.8 where we used a 10 × 20'' (radius) elliptical aperture to account for the larger extent of emission in this galaxy (see the third panel in Figure 1). The insets in Figure 1 show the extraction aperture compared to the optical extent of the galaxy. To determine the background counts, we measured the counts within a 15'' radial aperture at 50 random locations between ∼30''–60'' away from our target to properly sample the background statistics such that the error in the background contributes negligibly to the calculation of the net counts. In order to ensure that this random background region did not coincide with another source, we repeated this process (i.e., selected a new background region randomly) if any pixels within the background region exceeded 1.5 times the mean of the background distribution. The total counts from this background region were scaled by the ratios of the source to background aperture areas to give the net counts. We determined fluxes using PIMMS, assuming a power law with Γ = 1.9, since it was the power law index that best fit our combined LBA spectrum (see Section 3.2.2), and Galactic extinction, N_H = 3 × 10²⁰ cm⁻², which is the average value for these four LBAs, using the COLDEN Galactic neutral hydrogen density calculator. The N_H values range from 1–5× 10²⁰ cm⁻², corresponding to negligible (∼3%) error on the flux calculations. Based on the SDSS spectroscopically determined redshift, we apply a (1+z)^{Γ − 2.0} = (1 + z)^−0.3 k-correction to the 2–10 keV fluxes to calculate 2–10 keV X-ray luminosities (provided in last column of Table 2).

Table 2. Lyman Break Analogs: Derived Quantities

ID	R.A.	Decl.		ru, 50	log M⋆b	SFRc			LXe
	(deg)	(deg)	za	(kpc)	(M☉)	(M☉ yr−1)	IRX	12+log (O/H)d	(1040 erg s−1)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
J002101.0+005248.1	5.254	0.880	0.098	0.84	9.75	23.70	0.65	8.19	17$^{+10}_{-7}$
J080619.5+194927.3	121.581	19.824	0.070	0.66	9.52	15.16	0.96	8.15	12$^{+6}_{-4}$
J082355.0+280621.8	125.979	28.106	0.047	0.44	9.46	17.11	0.88	8.23	16$^{+7}_{-5}$
J225140.3+132713.4	342.918	13.454	0.062	1.54	9.39	8.65	0.53	8.15	9$^{+4}_{-3}$
VV114	16.946	−17.507	0.020	2.3f	10.65	37.82	1.0	8.45	24f
Haro11	9.219	−33.555	0.020	1.1f	9.84	10.88	0.52	8.33	12f

Notes. ^aSpectroscopic redshifts taken from the SDSS DR7 spectroscopic catalog, with Δz/(1 + z) < 0.008. ^bStellar mass uncertainties can be as large as 0.4 dex (Zibetti et al. 2009). ^cUV+IR SFR. ^dWe used gas-phase metallicities from the SDSS DR7 MPA/JHU VAC (the SDSS DR7 MPA/JHU value-added catalogs are available from http://www.mpa-garching.mpg.de/SDSS/), which were determined using the technique outlined by Tremonti et al. (2004). ^e2–10 keV X-ray luminosity. ^fOther LBAs studied by Grimes et al. (2006, 2007).

Download table as:  ASCIITypeset image

Grimes et al. (2006, 2007) have conducted detailed X-ray studies on VV 114 and Haro 11 and we refer to these papers for the information presented in Columns 2–4 and 9 of Table 2. We also compare our sample with a sample of z ∼ 0.2 LBAs, whose X-ray emission was studied by Jia et al. (2011, henceforth J11) using XMM.

For comparison with the LBAs, we include other samples of "normal" (not containing AGNs) local star-forming galaxies, studied using Chandra. These local star-forming galaxies include 24 nearby spiral and irregular galaxies studied by Colbert et al. (2004) and 22 additional star-forming galaxies from Mineo et al. (2012a, hereafter referred to as M12; 7 of the original 29 galaxies are also in the Colbert et al. 2004 sample). In both studies, the galaxies are sufficiently nearby to spatially resolve the XRB population; these authors extracted the X-ray luminosities from XRBs (specifically, HMXBs in the M12 sample) separate from the rest of the X-ray emission in these galaxies.

In addition, our comparison samples include luminous and ultraluminous infrared galaxies (LIRGs and ULIRGs), galaxies which have infrared (IR) luminosities exceeding 10¹¹ L_☉ and 10¹² L_☉, respectively. These IR-selected galaxies are dusty, high SFR galaxies, whose X-ray properties have been explored in a number of X-ray studies (Lehmer et al. 2010; Iwasawa et al. 2011; Symeonidis et al. 2011). We include 13 LIRGs/ULIRGs from Lehmer et al. (2010, hereafter L10), after eliminating 4 likely AGNs from their sample of 17, and 29 LIRGs/ULIRGs from the Great Observatories All-Sky LIRG Survey (GOALS; Iwasawa et al. 2011), excluding 15 classified AGNs from their total sample of 44. While many of the LIRGs/ULIRGs are spatially resolved with Chandra, the X-ray luminosities refer to galaxy-wide 2–10 keV emission.

We obtain 2–10 keV X-ray luminosities (L_X) from the references above, converting luminosities from other bands to the 2–10 keV band in some cases (e.g., 0.3–8 keV, Colbert et al. 2004; 0.5–8 keV, M12). In cases where L_X was derived using a power law with Γ ≠ 1.9 or using other X-ray bands (e.g., Γ = 1.8, Colbert et al. 2004; Γ = 2.0, M12), we converted to Γ = 1.9 for the sake of consistency. We note that L_X values from both LIRG/ULIRG samples (L10; Iwasawa et al. 2011) are calculated from fits to the individual galaxy spectra and include a hot-gas component (kT ≲ 0.8 keV) plus power law (Γ = 1–3). Since these are likely to be more accurate than using an assumed power law, we use their quoted values of L_X.

Therefore, the complete sample of comparison galaxies includes 88 galaxies.

While UV radiation traces recent star formation, it is also easily absorbed by dust and reradiated as IR radiation. Therefore a proper accounting of the total SFR includes both UV+IR, as described in the following relation from Bell et al. (2005):

$$\begin{equation} {\rm SFR} (M_\odot \ {\rm yr}^{-1}) = 9.8 \times 10^{-11} ({L}_{{\rm IR}}+3.3 {L}_{{\rm UV}}), \end{equation} \tag{ 1 }$$

where L_IR and L_UV correspond to the total IR (8–1000 μm) and UV (νL_ν at 2800 Å) luminosities, respectively, in units of solar luminosity. The factor of 3.3 is the typical correction factor that brings νL_ν (at 2800 Å) to the integrated L_UV. All of the LBAs have been observed by the Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010) and detected in the 22 μm band. We estimated L_IR by converting the monochromatic 22 μm luminosity to the total 8–1000 μm luminosity by using the Chary & Elbaz (2001) IR spectral energy distribution template. SFRs have been calculated using the L_UV determined from GALEX data,⁸ the L_IR determined from WISE data,⁹ and Equation (1). The amount of dust attenuation in the UV can be approximated by the IR-excess (IRX), defined as

$$\begin{equation} \rm IRX = \log ({\textit {L}}_{IR}/{\textit {L}}_{UV}). \end{equation} \tag{ 2 }$$

SFRs and IRX values for the LBAs are presented in Columns 7 and 8 in Table 2. We estimate SFRs and IRX in a consistent way for the comparison samples using the available values of L_IR and L_UV from the related papers (see Section 2.2; given in Howell et al. 2010 for the GOALS sample of LIRG/ULIRGs and obtained from S. Mineo 2013, private communication for the M12 sample).

We estimated stellar masses using Two Micron All Sky Survey (2MASS) K_s magnitudes and following the relations given by Table 7 of Bell et al. (2003), which are calculated for different choices of optical/NIR colors. The colors used in the stellar mass measurements are SDSS u' − z', B−K and B−V (where B and V photometric data come from the Third Reference Catalog of Bright Galaxies (RC3); de Vaucouleurs et al. 1991) colors for the LBAs, Colbert et al. (2004) star-forming and L10 LIRG samples, respectively. Howell et al. (2010) use 2MASS K_s magnitudes and IRAC 3.6 μm to estimate M_⋆ in the GOALS (U)LIRGs, which agrees well (slightly lower by ∼4%, see L10) with our method. M12 estimate M_⋆ from 2MASS K_s magnitudes, applying the relation given by Equation (2) of Gilfanov (2004), which is based on models from Bell & de Jong (2001), and we adjust these values to our adopted IMF (Kroupa). We note that Zibetti et al. (2009) show that the stellar masses estimated using different methods can vary by ∼0.4 dex, which does not affect any of the final results that we present in this paper.

We measure the metallicities using the method outlined in Pettini & Pagel (2004), using the O iii λ5007 and N ii λ6584 emission-line ratios (referred to, hereafter, as the PP04 O3N2 method). Kewley & Ellison (2008) compare different metallicity measurement techniques and find that different estimated gas-phase metallicities can differ systematically by 0.7 dex depending on the method. While calibration from various methods to an absolute metallicity value remains uncertain, relative uncertainties are minimized when using one consistent method and the PP04 method proves to be one of the most robust methods (Kewley & Ellison 2008). Emission-line measurements are available for all 6 of the LBAs and 38 of the 88 comparison galaxies, which we henceforth refer to as the "partial comparison sample." Using published emission-line values (28 comparison star-forming galaxies, 10 LIRGs, Haro 11, and VV 114; Kim et al. 1995; Wu et al. 1998; Moustakas & Kennicutt 2006; Bergvall & Östlin 2002) and SDSS spectroscopic data (4 Chandra-observed LBAs and 1 LIRG) we determine gas-phase metallicities.

The stellar masses and gas-phase metallicities for the LBAs are provided in Columns 6 and 9 (respectively) in Table 2.

We show the relation between the 2–10 keV X-ray luminosity and SFR (L_X versus SFR) for the LBAs compared to the full star-forming sample of 88 galaxies in Figure 2 (see legend for symbol descriptions). We also compare with results from stacked z = 1.5–4 LBGs (Basu-Zych et al. 2013, open squares), where individually detected X-ray sources (potential AGNs) were eliminated in order to study the X-ray emission related to star formation. On average, the UV-selected samples (LBAs and the stacked z = 1.5–4 LBGs) appear to have L_X per SFR elevated by a factor of ∼1.5σ compared to the comparison sample, which is described by the dashed line and gray region (showing 1σ) based on the fit to L_X/SFR from M12 (converted to the 2–10 keV band, using Γ = 1.9) and given by

$$\begin{equation} \log L_{\rm X} {(\rm 2\hbox{--}10\ keV)}=39.4(\pm 0.4) \log {\rm SFR}. \end{equation} \tag{ 3 }$$

The error bar shown in the lower right corner of Figure 2 describes the uncertainty attributed to our choice of Γ = 1.9 ± 0.2, which accounts for the different assumed values for the power law index in the different samples (e.g., Γ = 2.0, M12; 1.8, Colbert et al. 2004). We note that the uncertainty is ∼0.1 dex, which does not change the result that the LBAs have significantly elevated L_X per SFR compared to the other samples. We also test whether biases are introduced from comparing samples that used different techniques for measuring X-ray luminosities; comparing luminosities for galaxies in multiple samples (i.e., there are seven galaxies that are in M12 and Colbert et al. 2004), we find the data agree within 10% with no systematic differences. The likelihood that all six LBAs would randomly scatter this far from the local relation is ∼10⁻⁵ (based on χ²), almost certainly implying there is something fundamentally different about LBAs compared to typical normal galaxies. We note that there are some other galaxies which appear to have elevated L_X/SFR, which we discuss in Sections 4.1 and 4.2.

Zoom In Zoom Out Reset image size

The J11 sample of z ∼ 0.2 LBAs, having optical emission-line signatures between starbursts and obscured AGNs (see further discussion in Section 3.2.1), are shown in Figure 2 as crosses. Two of the J11 sources have L_X/SFR that are extreme (factors ∼10–20 above the relation), likely suggesting that AGNs power the X-ray emission in these sources; two other sources appear similarly elevated (by ∼1.5σ, or factor of ∼4) as the LBAs in our sample.

Several studies have shown that 2–10 keV emission in local galaxies correlates well with SFR (e.g., Ranalli et al. 2003; Persic & Rephaeli 2007; Lehmer et al. 2010; Mineo et al. 2012a). The 2–10 keV emission mainly originates from accreting XRBs (including ultraluminous X-ray sources (ULXs)). Although minimized by studying emission in the 2–10 keV band, we note that a number of other sources can also contribute to X-ray emission in normal galaxies: supernovae and their remnants, and hot gas from starburst-driven winds and outflows (see, e.g., review by Fabbiano 1989; Strickland et al. 2000; Grimes et al. 2005; Li & Wang 2013; Mineo et al. 2012b). Low-luminosity AGN activity may also contribute to the 2–10 keV X-ray luminosity in normal galaxies. In the next section, we assess the likelihood that the presence of AGNs cause the elevated L_X/SFR observed in LBAs.

Since AGNs also produce significant X-ray emission, we explore the possibility that hidden AGNs may cause L_X values per SFR that are ∼1.5σ higher in the LBAs. Benefiting from the high angular resolution capability of Chandra, three of the six LBAs in our sample, VV 114, Haro 11, and J082355.0+280621.8, appear to have extended X-ray emission (see Figure 3), and we can distinguish multiple X-ray point sources, all of which are luminous enough to qualify as ULXs (L_X >10⁴⁰ erg s⁻¹) based on the limiting luminosity of the available data (e.g., ULXs have L_X ≳ 10⁴¹ erg s⁻¹ in the case of J082355.0+280621.8). Using the images for these three LBAs, we measure the flux in the nuclear regions, potentially associated with AGNs, compared to the entire galaxy and find that the nuclear regions contribute only <15% to the 2–10 keV emission in each case, immediately indicating that the elevated L_X/SFR for those LBAs cannot be attributed to AGN activity.

Zoom In Zoom Out Reset image size

While VV114 may harbor an obscured AGN (in the eastern component, VV114E), Grimes et al. (2007) conclude that star formation dominates the 2–10 keV power in this galaxy. Based on Spitzer/IRS spectroscopic analysis of the [O iv] λ25.9 μm and [Ne v] λ14.3 μm emission features, Cormier et al. (2012) conclude that Haro 11 does not contain an AGN. To further test for the presence of AGNs in the four newly observed SDSS+GALEX-selected LBAs, we use optical spectral line diagnostics and X-ray spectral constraints (see below).

3.2.1. Optical Spectral Line Diagnostics

The Baldwin–Phillips–Terlevich (BPT; Baldwin et al. 1981) diagram separates galaxies whose optical emission is dominated by AGNs versus star formation using the ratios of forbidden lines (i.e., [N ii] λ6584, [O iii] λ5007), which are excited mainly by AGNs, to nearby Balmer lines. In the left panel of Figure 4, we show the distribution of >900, 000 sources from the SDSS DR7 catalog (shaded background). The solid gray line is based on a theoretical modeling of the upper limit that star-forming galaxies can occupy in this diagram by Kewley et al. (2001). Kauffmann et al. (2003) revise the classification criterion to further separate star-forming from composite galaxies. The LBAs in this study (black points) reside in the star-forming region of this diagram.

Zoom In Zoom Out Reset image size

Using XMM, J11 studied a sub-sample of six z ∼ 0.2 LBAs (shown as crosses in Figures 2, 4, and 5) that lie within or near the composite region of the BPT diagram. They also found that these galaxies are systematically offset with respect to the X-ray/SFR relation, having L_X =4–20 × 10⁴¹ erg s⁻¹ (see Figure 2). Based on comparing the X-ray luminosities with the far-IR and [O iii] λ5007 luminosities, J11 conclude that type 2 AGNs are likely present within their sample. One of the composite LBAs most offset in L_X was also shown to host a compact radio source using very long baseline interferometry (VLBI). The compact source was interpreted as an AGN having a small fraction of the total radio flux of the LBA (Alexandroff et al. 2012). Several other LBA composites were not detected at VLBI resolution, showing that the overall (i.e., large-scale) radio emission of LBAs is dominated by star formation.

Zoom In Zoom Out Reset image size

Motivated by the study by Hornschemeier et al. (2005) which showed that using the BPT diagram alone may still include AGNs (e.g., narrow-line Seyfert 1 galaxies), we confirmed that our sample of LBAs had Balmer line widths that were consistent with those of the forbidden lines (within 3σ compared to the distribution of the entire SDSS DR7 population).

Another emission-line test has been proposed by Trouille et al. (2011), referred hereafter by the TBT test (shown on right panel of Figure 4), which uses rest-frame g−z color and the ratio of [Ne iii] λ3869 to [O ii] λλ3726, 3729 to separate AGNs (upper region) from star-forming galaxies (below dashed line). Trouille et al. (2011) compare their method with the BPT diagnostic; the solid red, dashed blue, and cross-hatched gray regions in Figure 4 (right) correspond to the BPT-determined AGN, star-forming, and composite classifications, respectively. According to this classification, the LBAs reside in the star-forming region of the TBT diagram.

The TBT method is good for selecting sources that might be X-ray detected. In their analysis, Trouille et al. (2011) find that X-ray sources with L_X >10⁴² erg s⁻¹ appear in the top region, while few luminous X-ray sources fall below the dashed line. Moreover, they show that the TBT method is able to classify luminous, less luminous, and soft X-ray detected non-broad line AGN sources accurately, while ∼20% of X-ray detected AGNs are classified as star-forming galaxies. We note that the J11 "composite LBAs" (crosses), based on the BPT classification, were detected at L_X ∼10⁴² erg s⁻¹, yet fall in the star-forming region of the TBT classification.

Our sample of LBAs lies conservatively within the star-forming region of the TBT diagram, providing further support against AGNs. While using optical emission lines to detect AGNs is expected to be sensitive to even low-levels of AGN activity, this method may suffer from aperture dilution in some of the higher redshift sources or may miss heavily obscured AGNs.

3.2.2. X-Ray Spectral Properties of LBAs

In Figure 5, we show the relationship between L_X/SFR and specific SFR (sSFR =SFR/M_⋆) for the LBAs (black points), stacked z = 1.5–4 LBGs (blue squares) and comparison sample (gray symbols). LBAs, LBGs, and LIRGs/ULIRGs (gray triangles) have similarly high sSFRs, but L_X/SFR is ∼1.5σ higher for the LBAs , when compared to the fit and scatter of L_X/SFR for HMXB-dominated galaxies (M12, gray dashed line/shaded region; see Equation (3)), suggesting that the XRB population in LBAs may be unique, even compared to other galaxies that have high sSFRs (see Section 4.2).

Within the XRB population, HMXBs are short-lived, tracing recent star formation activity (on timescales ∼10^6–7 yr), while low-mass X-ray binaries (LMXBs) trace older stellar populations (for timescales >10^8–9 yr) and the 2–10 keV X-ray luminosity scales with the stellar mass (M_⋆) of the galaxy. Assuming the following analytic parameterization for local (within a distance of 60 Mpc) normal galaxies

$$\begin{eqnarray} L_{\rm X} & = & L_{\rm X}({\rm LMXB}) + L_{\rm X}({\rm HMXB}) \end{eqnarray} \tag{ 4 }$$

$$\begin{eqnarray} &=& \alpha M_\star + \beta {\rm SFR}, \end{eqnarray} \tag{ 5 }$$

L10 measure constants of α = (9.05 ± 0.37) × 10²⁸ erg s⁻¹ $M_\odot ^{-1}$ and β = (1.62 ± 0.22) × 10³⁹ erg s⁻¹ (M_☉ yr⁻¹)⁻¹ (see also Colbert et al. 2004). Rearranging Equation (5) to give L_X/SFR in terms of sSFR (SFR/M_⋆) L10 show that log  L_X/SFR is proportional to log (1/sSFR + constant), causing two regimes: at low sSFRs (⩽10⁻¹⁰ yr⁻¹), where both LMXBs and HMXBs contribute significantly to the X-ray luminosity, the relation is inversely linear; and at high sSFRs (⩾10⁻¹⁰ yr⁻¹), where the contribution of LMXBs is negligible and the X-ray luminosity scales with SFR, the relation flattens (see black curve in Figure 5).

M12 correct the observed 0.5–2 keV luminosities for contributions from LMXBs and hot gas in their sample of star-forming galaxies (gray points), and fit for the HMXB-driven L_X/SFR relation. After converting their relation to 2–10 keV, using Γ = 1.9, we show the HMXB-driven X-ray/SFR relation by the gray dashed line (shaded region marks the 1σ scatter) in Figures 2, 5, and 7. Hereafter, we refer to the relation as the local L_X/SFR relation and restrict our analysis to high sSFR, HMXB-dominated galaxies, with sSFRs >10⁻¹⁰ yr⁻¹ (right of the dotted line in Figure 5), to simplify the discussion by focusing on the relationship between galaxy properties and the HMXB population.

We note that most of the galaxies that appear to have elevated (>1σ) L_X/SFR in the low SFR end of Figure 2 have low sSFRs (LMXB-dominated; see Figure 5), where L_X/SFR is higher than for HMXB-dominated galaxies. In addition to the LBAs and stacked LBGs, Figure 5 reveals other HMXB-dominated galaxies with elevated L_X/SFR. These are discussed in more detail in the following section.

We return to the question whether metallicity differences between the LBA sample and other HMXB-dominated galaxies might drive the elevated L_X/SFR. In Figure 6 we show L_X/SFR versus gas-phase metallicity for the LBAs (black points) and the partial comparison sample (symbols are described in legend of Figure 2). The data are correlated at the 99.2% significance level, based on a Spearman's rank correlation test. Optical spectral data are available for the full LBA sample, but only 40% of the HMXB-dominated (with sSFRs >10⁻¹⁰ yr⁻¹) comparison sample.

Zoom In Zoom Out Reset image size

Based on XRB population synthesis models, the formation of HMXB populations is strongly metallicity-dependent (e.g., Dray 2006; Linden et al. 2010; Fragos et al. 2013a, 2013b). One explanation is that stellar winds are weaker in massive stars with lower metallicities, allowing them to retain most of their mass and end their evolution as black hole companions. Since black hole HMXBs can be more luminous than neutron star HMXBs, the luminosity of an HMXB per SFR is expected to be higher in low-metallicity environments (Mapelli et al. 2009; Zampieri & Roberts 2009; Fragos et al. 2013b). In addition, a reduced stellar wind would affect the secular evolution of the binary and would result in less orbital expansion due to angular momentum loss from the stellar wind, and in overall more systems that will encounter Roche-lobe overflow mass transfer instead of Bondi–Hoyle type wind accretion. Roche-lobe overflow systems can drive much higher accretion rates onto the black hole and hence be more luminous X-ray sources (Fragos et al. 2013a). Indeed, initial results have shown evidence for an increased number of HMXBs, or off-nucleus ULXs, per SFR in lower-metallicity environments (Swartz et al. 2008; Mapelli et al. 2009, 2010, 2011; Kaaret et al. 2011; Prestwich et al. 2013).

While the lower metallicities in the LBA sample are not as extreme as in dwarf galaxies, Overzier et al. (2010) find that z ∼ 0.2 LBAs have lower metallicities per stellar mass compared to the local mass–metallicity relation (e.g., Tremonti et al. 2004), but consistent to the mass–metallicity relation of LBGs at z ∼ 2 (Erb et al. 2006). Additionally, Neill et al. (2011) link LBAs with the host galaxies of luminous supernovae, concluding that their low metallicities and high sSFRs may be related to extremely massive progenitors (>100 M_☉) or wind-driven mass loss that may result in luminous supernovae.

We estimate metallicities for the z = 1.9 and 2.5 LBG samples from the X-ray stacking analysis in Basu-Zych et al. (2013) using the z ∼ 2 mass–metallicity relation (Erb et al. 2006; converting the metallicities to PP04 O3N2 units, using the conversions given in Kewley & Ellison 2008). The range of L_X/SFR versus metallicity values for z ∼ 2 LBGs is shown as a blue hatched region in Figure 6. Based on the theoretical predictions that metallicity affects the number of luminous HMXBs, we postulate that the higher X-ray luminosity per SFR in the LBAs may be attributed to a higher number of more luminous HMXBs.

In addition to the LBAs and stacked LBGs, we note that several other HMXB-dominated galaxies appear to have elevated L_X/SFR >1σ from the local L_X/SFR relation (see Figure 5). We have metallicity estimates for two of these cases (NGC 3310 and NGC 2139), which have relatively low metallicities and appear above the gray solid line in Figure 6. In the other cases, low stellar masses (<10¹⁰ M_☉) imply that the expected metallicities are also low based on the mass–metallicity relation; some of these galaxies are well-known hosts to multiple ULXs, potentially attributed to their low metallicities (e.g., Cartwheel, Mapelli et al. 2009; NGC 1313, Zampieri & Roberts 2009).

The gray line in Figure 6 shows the theoretical prediction, based on XRB population synthesis models (Fragos et al. 2013b). It is notable that these models predict that the X-ray luminosity from HMXBs per unit of SFR increases by approximately an order of magnitude going from solar metallicity to less than 10% solar. While theoretical models express metallicity in absolute Z units, we convert these to units of oxygen abundance (12 + log  [O/H]) by scaling solar metallicity (Z_☉) to 12 + log  [O/H] ∼ 8.69. Our analysis shows reasonable agreement with the theoretical prediction (gray curve), albeit with significant scatter, which is likely attributed to the issues discussed below.

We mention a couple of caveats to this analysis. First, we expect that the metallicities, measured using observed emission lines, are characteristic of star-forming regions where these lines are produced and HMXBs are presumed to be present. However, we are measuring the galaxy-averaged metallicity rather than the metallicities for the regions directly associated with HMXBs. The second caveat is that there is significant scatter between different techniques that measure metallicity and the absolute scaling is not well constrained (see, e.g., Kewley & Ellison 2008 and the discussion in Section 2.4), which complicates comparisons with theoretical models. Nevertheless, since the former issue adds scatter via stochastic variations between the galaxies and the latter introduces systematic errors, the observed correlation shows promising evidence that metallicity may be responsible for driving much of the scatter in the L_X/SFR relation, specifically offering a plausible explanation for the elevated L_X/SFR in the lower-metallicity LBAs.

In Figure 3, the images of extended X-ray emission in three LBAs (VV 114, Haro 11, and J082355.0+280621.8) display multiple off-nuclear point sources. Given the X-ray sensitivity limit of the data, detected sources (ULXs) have L_X >10⁴⁰ erg s⁻¹ for VV 114 and Haro 11 and >10⁴¹ erg s⁻¹ for J082355.0+280621.8. Based on their SFRs and the analysis of HMXBs in star-forming galaxies by M12, at these limiting luminosities, we would have expected one or fewer detected ULXs in each galaxy compared to what we observe (five, three, and three off-nuclear sources in VV 114, Haro 11, and J082355.0+280621.8, respectively). Accounting for possible source confusion, where multiple sub-ULX sources appear as a single, unresolved L_X >10⁴⁰ erg s⁻¹ ULX, implies an improbably high number density of HMXBs per SFR. While further analysis is required to confirm this result, the numbers of ULXs suggest that LBAs host an excess of extremely luminous ULXs, potentially because they have uniquely high SFRs and low metallicities.

Typically, local galaxies appear to have higher levels of dust attenuation with increasing SFR (Wang & Heckman 1996; Martin et al. 2005; Johnson et al. 2007), but the LBA sample appears to be a unique population that is offset toward lower IRX values per SFR (Basu-Zych et al. 2007; Overzier et al. 2011). We explore the possibility that LBAs appear to have elevated X-ray luminosities per SFR caused by dust effects. Since dust attenuation is uniquely low in LBAs, we might expect that typical galaxies suffer higher levels of dust obscuration, which might affect the observed L_X/SFR, i.e., if X-ray emission were significantly absorbed in other star-forming galaxies, such as IR-selected samples with similar SFRs. In Figure 7, we show our results for testing the hypothesis that L_X/SFR depends on the amount of dust and gas, measured using two different methods: IRX (see Equation (2)) to measure the dust attenuation (shown in left panel), and the SFR surface density, Σ_SFR, which is related to the gas surface density (Kennicutt 1998; henceforth referred to as the Schmidt–Kennicutt law, shown in the right panel).

Zoom In Zoom Out Reset image size

In the left panel of Figure 7, we find that there is a weak correlation between L_X/SFR and IRX, at an 86.4% confidence level, if we restrict our analysis to the HMXB-dominated (with sSFR >10⁻¹⁰ yr⁻¹) galaxies with measured metallicities (the partial comparison sample, 38 galaxies, and 6 LBAs). The correlation is much stronger (99.0%) if we include all of the HMXB-dominated galaxies (i.e., including galaxies without measured metallicities, 76 galaxies from the comparison sample and 6 LBAs). The gray error bars show the average values (in bins of IRX) from M12, which appear consistent with our fit (black line). Since dust and metallicity are also correlated in our data (at ∼95% confidence level; see also Skibba et al. 2011), we consider the expected impact that dust extinction alone, using IRX as a proxy, might have on L_X/SFR. We estimate the neutral hydrogen column density from IRX using the following steps: (1) we convert IRX into the FUV attenuation (A_FUV) following the relation given by Equation (15) in Hao et al. (2011): A_FUV = 2.5log [1 + 0.46 × 10^IRX]; (2) we estimate the attenuation in the V-band from A_FUV as A_V ≈ 0.4A_FUV, using the extinction laws from Gil de Paz et al. (2007): A_FUV = 7.9E(B − V) and Cardelli et al. (1989): A_V = R_VE(B − V), where the V-band reddening is R_V = 3.1; and (3) according to Güver & Özel (2009), the neutral hydrogen column density (N_H) relates to the V-band attenuation as N_H = 2.21 × 10²¹A_V cm⁻². Based on these steps, we derive a relation between N_H and IRX as

$$\begin{equation} N_{\rm H} \approx 2.21 \times 10^{21} [0.98 \log (1+0.46 \times 10^{\rm IRX})]{\rm cm}^{-2}. \end{equation} \tag{ 6 }$$

The top axis in Figure 7 (left panel) displays the corresponding neutral hydrogen column density and the gray dotted curve marks the nearly negligible effect (<0.05 dex, see right axis) of this X-ray absorption on L_X/SFR. Since the X-ray absorption is a product of the metallicity and hydrogen column density (our dotted curve assumes solar metallicity), the same values of N_H have a stronger effect on the level of X-ray absorption with increasing metallicity. The dark gray region shows the range of X-ray absorption expected given the observed range of metallicities in our samples, 12 + log [O/H] =8.1 (top edge)–8.85 (bottom edge).

Since IRX provides a global measure of the dust attenuation averaged over the full extent of the galaxy, by comparing the IR emission to the UV emission, the IRX value may not accurately capture the true extinction present along the lines-of-sight to specific star-forming regions where HMXBs reside. Therefore, we try another independent measure to quantify the extinction in these galaxies.

The SFR surface density (SFR/kpc², Σ_SFR = SFR/πab, where a and b are the semi-major and semi-minor axes), is related to neutral gas surface density (Σ_H) as Σ_SFR = 2.5 × 10⁻⁴[Σ_H/M_☉ pc⁻²]^1.4 M_☉ yr⁻¹ kpc⁻² (Kennicutt 1998). The neutral hydrogen gas column density is Σ_H/m_H, where m_H is the mass of a hydrogen atom. Therefore, we estimate N_H from the UV+IR-derived SFR surface density by

$$\begin{equation} N_{\rm H} = 4.6\times 10^{22} \Sigma _{\rm SFR}^{0.7} {\rm cm}^{-2}. \end{equation} \tag{ 7 }$$

We see no correlation of L_X/SFR with Σ_SFR (see right panel of Figure 7), as expected based on the small effect of N_H (derived from Σ_SFR, as given in Equation (7); gray dotted curve) on L_X/SFR. However, the Schmidt–Kennicutt law may not apply to galaxies with high SFR efficiencies, i.e., galaxies with high SFR surface density per gas density (e.g., LBAs, Basu-Zych 2009; and possibly LIRGS, García-Burillo et al. 2012), which explains why the estimated N_H values from the two techniques differ. However, the two techniques do provide comparable values of N_H for the other star-forming galaxies (gray squares and points in Figure 7; correlated at 85% confidence level according to a Spearman's rank correlation test).

Neither measure of extinction implies sufficient N_H to significantly absorb 2–10 keV X-ray emission. However, since both IRX and Σ_SFR are calculated using global properties of the galaxies, we are measuring the average absorption and are unable to account for heavily obscured sub-galactic regions. For example, in ULIRGs, Genzel et al. (1998) find incredibly high optical absorption values, A_V ≈ 50–1000 mag for some ULIRGs, where the optical dust column densities are comparable to the dense molecular cloud (CO) millimeter observations. Therefore, HMXBs found in such heavily obscured regions would be affected by significant X-ray absorption and in such specific cases the observed X-ray luminosities may be underestimated. While this scenario is plausible in some ULIRGs and LIRGs, X-ray absorption is not likely to fully explain the apparent correlation between L_X/SFR and IRX across the entire population of star-forming galaxies. However, the correlation between L_X/SFR and IRX is mainly influenced by the ULIRGs; i.e., the data are not correlated (<50% significance level) when ULIRGs are eliminated.