X-RAY PROPERTIES OF THE z ∼ 4.5 Lyα EMITTERS IN THE CHANDRA DEEP FIELD SOUTH REGION

In this paper, we focus on the X-ray data with an off-axis angle <8', because the spatial resolution of ACIS-I data degrades rapidly for off-axis angles >8'. This excludes 22 LAEs from our sample, leaving a total of 91 LAEs at z ∼ 4.5 covered by Chandra images with off-axis angle θ < 8'. Of these, 22 are covered by the 2 Ms CDF-S exposure and 86 by the shallower ECDF-S exposures, with 17 sources covered by both (see Figure 1). We choose a radius of 3'' to match X-ray counterparts to our LAE sample, as our narrowband data have a seeing of 09, and the radius of 50% point-spread function (PSF) regions of Chandra ACIS-I reaches 28 at the edge of our selection area. Only one LAE (J033127.2-274247) has an individually detected X-ray counterpart (ECDF-S-J033127.2-274247), with a spatial offset ⩽04 between the NB673 and X-ray coordinates. This object has previously been spectroscopically identified as an unobscured z = 4.48 quasar (Treister et al. 2009), with full-band luminosity of L_{0.5–10 keV} = 4.2 × 10⁴⁴ erg s⁻¹ (assuming Γ = 1.4; Lehmer et al. 2005). We measured the f_Lyα/f_{0.5–10 keV} ≈ 0.065, consistent with expectations from a quasar template (f_Lyα/f_{0.5–10 keV} ∼ 0.05; Sazonov et al. 2004).

Zoom In Zoom Out Reset image size

There are two LAEs (J033204.9-280414 and J033154.1-274159) located in 95% PSF circles of two ECDF-S sources with offsets between X-ray and optical of 45 and 41, respectively. These offsets are too large to reliably associate the Lyα and X-ray sources, thus we do not classify them as X-ray detections, and we exclude these sources from our X-ray stack (Section 3.2).

We plot the X-ray signal-to-noise ratio (S/N) distribution of the remaining X-ray flux measurements in Figure 2. This comprises 106 exposures on 88 distinct LAEs (22 are covered by the 2 Ms CDF-S exposure and 84 by the shallower ECDF-S exposures, with 17 sources covered by both and one source covered by two ECDF-S pointings, see Figure 1). The S/Ns were calculated as $\mbox{S/N} = S/(\sqrt{T+0.75}+1)$ (Gehrels 1986), where S and T are the net counts and total counts extracted from their 50% PSF circles⁷ at 0.5–2, 2–7, and 0.5–7 keV bands, respectively. When converting from PSF-corrected count rate to flux, the full and hard bands were extrapolated to the standard upper limit of 10 keV. All X-ray fluxes have been corrected for Galactic absorption (Dickey & Lockman 1990). To convert from X-ray counts to fluxes, we have assumed power-law spectra with photon index of Γ = 2 (except where explicitly stated otherwise), which generally represents the X-ray spectra of both starburst galaxies and type 1 AGNs. For the LAEs without individual X-ray detections, we derived 3σ upper limits to their X-ray fluxes (see Figure 4).

Zoom In Zoom Out Reset image size

To determine the mean X-ray properties of the high-redshift LAEs that are too weak to be directly detected, we employed a stacking technique similar to that described in Wang et al. (2004) and Laird et al. (2006). The only difference was in the count extraction, we chose the 50% PSF regions here for the non-detections rather than 80% PSFs (as in Wang et al.) or fixed radius of 15 (as in Laird et al.). Small apertures give better upper limits on non-detections, and constant size is difficult for flux estimation here. After masking out the detected X-ray sources, the net and background counts were measured in the CDF-S and ECDF-S separately (see Figure 1 and Table 1). We computed two stacks: (1) the CDF-S data alone and (2) all available data (CDF-S and ECDF-S). For objects in the overlap between the CDF-S and ECDF-S coverage, only their CDF-S data were included in stack (1), while data from both images were included in stack (2).

A marginal signal was found from the stacking of the CDF-S data in the soft band, while no signals were found in the other band of CDF-S or from the ECDF-S stacking. When cumulating the 22 LAEs in the central CDF-S in the soft X-ray band, we measure net and background counts of 26.6 and 74.4, which yields an S/N = 2.4, with an effective exposure time of ∼36 Ms. The net counts and exposure time can be converted to an average flux of F_{0.5–2.0 keV} = (8.8 ± 3.7) ×10⁻¹⁸  erg cm⁻² s⁻¹. Including the ECDF-S data, the net exposure rises to 52 Ms for 106 exposure of 88 objects, while the net and background counts only increased 0.2 and 26 at soft band, respectively, corresponding to a slightly decreased S/N of 2.2. In principle, the increase in effective exposure of a factor of 1.44 (52 Ms/36 Ms) should imply an expected value of S/N = 2.4$\times \sqrt{1.44}$ = 2.9. However, given the small numbers of X-ray photons involved, the count rates in the CDF-S and ECDF-S are in fact consistent at <2σ. The effective 52 Ms exposure decreases our soft band signal to 〈f_0.5–2〉 = (6.4 ± 2.9) × 10⁻¹⁸  erg cm⁻² s⁻¹, corresponding to an average luminosity of 〈L_{0.5–2 keV}〉 = 1.3 × 10⁴² erg s⁻¹ for LAEs at z ∼ 4.5. We also stacked the images of our LAEs together. Since the stacked samples in ECDF-S were within their background fluctuations, we only stacked the 22 LAEs located in CDF-S. The resulting stacked image shows a signal consistent with the analysis above, which becomes apparent to visual inspection when smoothed with a Gaussian matched to the ACIS PSF size (see Figure 3).

Zoom In Zoom Out Reset image size

We performed Monte Carlo simulations to check the significance of the stacked signal. By randomly choosing 22 positions on the source-masked CDF-S image, then cumulating source and background counts, we obtained distributions of both the net counts and the soft band S/N distribution (Figure 2). The net counts in the simulations agreed very well with a Poisson distribution having a mean of 74 (the expected total background counts in 22 apertures). Both the Monte Carlo simulation and the Poisson distribution gave a probability of P(S/N < 2.4) = 99.83% for obtaining a signal as strong as the observed one by chance. We also use a jackknife test on our stacking result (see the insets in Figure 2). This test is to validate the sample by using subsets of the data from which one or two sources have been excluded. The jackknife test shows that there are two sources that contribute about half of the stacked signal. We regard these as suspected X-ray sources.

We have recently obtained optical spectroscopy of ∼75% of our LAE sample (Z. Y. Zheng et al. 2010, in preparation), using the IMACS spectrograph on the Magellan 6.5 m telescope, including the two LAE candidates (NB673-27 and NB673-62) which have S/N_soft > 1 in CDF-S and contributed about half of the X-ray signal. One (NB673-27) is confirmed as a low-redshift emission-line galaxy based on strong continuum flux blueward of the emission line. The other object (NB673-62) was confirmed as an LBG at z = 4.4 with little-to-no Lyα flux present. As our aim is to analyze the X-ray properties of the LAEs at z = 4.5, we excluded these two objects in the following stacking analysis. The remaining 20 LAEs in the central CDF-S had net and background counts of 14 and 61, which yields an S/N = 1.4, with an effective exposure time of 32.7 Ms. Our Monte Carlo simulations give a probability of P(S/N < 1.4) = 96.03% for obtaining a signal as the observed one by chance. The stacked X-ray image of the 20 LAEs is also plotted in Figure 3, which is less distinguishable as being above the noise. We thus give a 3σ upper limit on the average flux as 〈f_0.5–2〉 < 1.6 × 10⁻¹⁷  erg cm⁻² s⁻¹ for the 20 LAEs in CDF-S. Including the ECDF-S data, the effective exposure increases to 48.9 Ms, implying a decreased 3σ (1σ) upper limit of average flux as 〈f_0.5–2〉⩽ 1.2 × 10⁻¹⁷  erg cm⁻² s⁻¹ (6.3 × 10⁻¹⁸  erg cm⁻² s⁻¹), corresponding to a luminosity of 〈L_{0.5–2 keV}〉⩽ 2.4 × 10⁴² erg s⁻¹ (1.2 × 10⁴² erg s⁻¹).

One LAE (J033127.2-274247) was detected in X-ray in ECDF-S, which was spectroscopically identified as a z = 4.48 unobscured AGN with Lyα luminosity of L_Lyα = 2.4 × 10⁴³ erg s⁻¹. This yields a direct high-z Lyα quasar density of 2.7^+6.2_−2.2 × 10⁻⁶ Mpc⁻³ (1σ Poisson error; Gehrels 1986), which is consistent with X-ray luminosity function of AGNs at high redshift (the comoving space density for all spectral type AGNs with 43 < log L_x < 45 at redshift 4 < z< 5 is 2.3 × 10⁻⁶ Mpc⁻³; Yencho et al. 2009). Since Chandra ACIS does not have uniform sensitivity across the field of view, it is hard to directly get the fraction of galaxies hosting a quasar with X-ray luminosity above some value of L_X (e.g., see Figure 4, there are three LAEs with X-ray upper limit fluxes higher than the detected one in ECDF-S). If we only consider the LAEs in CDF-S, then the type 1 quasar fraction should be ⩽5% with L_{0.5–2 keV} >2 × 10⁴³ erg s⁻¹.

Zoom In Zoom Out Reset image size

Following Wang et al. (2004) and Malhotra et al. (2003), we compare the X-ray to Lyα flux ratios of LAEs with three known high-redshift type 2 quasars (see Figure 4), CDF-S 202 (z = 3.7; Norman et al. 2002), CXO 52 (z = 3.288; Stern et al. 2002), and HDFX 28 (z = 2.011; Dawson et al. 2003), and with a type 1 quasar template derived from Sazonov et al. (2004).⁸ The Lyα-selected AGNs at z = 4.5 (a type 1 AGN, this work), at z = 3.7 (Ouchi et al. 2008), at z = 3.1 (a type 1 AGN from Gronwall et al. 2007 and one from Ouchi et al. 2008), and at z = 2.25 (nine AGNs;⁹ Nilsson et al. 2009) are also plotted in Figure 4. Since there are many values from different redshifts, Figure 4 is plotted in luminosity, and the soft X-ray luminosities are converted by assuming a photon index of Γ = 2. We only consider the soft band observations because they are more sensitive than the total band 0.5–10 keV. Also, high-redshift AGNs have effective power-law indexes are often different, as Γ = 1.8 (type 1 AGNs) or Γ < 1 (type 2 AGNs). This introduces at least 50% difference in X-ray photometric flux normalization in the 0.5–10 keV band, but less than 10% in the 0.5–2 keV band (see Figure 2 of Wang et al. 2007). So in Wang et al. 2004, who choose Γ = 2 to get the 1σ upper limit of 0.5–10 keV band flux to Lyα ratio of the z = 4.5 LAEs at Large Area Lyman Alpha field, their type 2 AGN fraction of <4.8% should be two times larger, as <9.6% compared with type 2 AGN like CXO 52.

After scaling the X-ray luminosities with Lyα line luminosities, most of the Lyα-selected AGNs are located within the region where type 1 and type 2 quasars are located. All the 20 LAEs in CDF-S are fainter in X-rays than HDFX 28, our type 1 quasar and Sazonov's template, greater than 50% and 70% of them are fainter than CDF-S 202 and CXO 52. This indicates that about half of our LAEs at z ∼ 4.5 can be type 2 quasars like CDF-S 202. By comparing with LAEs in ECDF-S region, we can find that only CDF-S allows us to resolve almost all of the type 1 AGN as well as some kind of type 2 AGN in our LAE sample. However, the average X-ray (1σ upper limit) to Lyα ratio is 16 and 20 times below those of type 2 quasars such as CDF-S 202 and CXO 52, and 31, 40, and 78 times below our LAE–QSO, the type 1 quasar template and LAE–QSO at z = 3.1. This implies that <6.3% of our LAEs can be type 2 AGNs like CXO 52 and CDF-S 202 and <3.2% of our LAEs can be type 1 quasar like our LAE–QSO.

The average flux (3σ upper limit) of our stacking analysis in the soft band corresponds to an average X-ray luminosity of 〈L_{0.5–2 keV}〉 = 2.4 × 10⁴² erg s⁻¹. If we assume that this is due to HMXBs, using the empirical relation between the 0.5–2 keV luminosity and SFR of the nearby star-forming galaxies (Ranalli et al. 2003), we derive the upper limit of SFR as SFR_X⩽ 530 M_☉ yr⁻¹. This SFR is much higher than previously measured for LAEs. For comparison, Gronwall et al.'s X-ray-undetected LAEs at z = 3.1 can be translated to a 3σ upper limit of SFR of <70 M_☉ yr⁻¹, only ∼13% of our upper limit SFR_X at z = 4.5. If we adopt the more recent X-ray to SFR calibrations from Rosa-Gonzalez et al. (2009) and Mas-Hesse et al. (2008), based on the XMM-Newton observation and the synthetic model of starbursts, respectively, the SFR upper limit estimated above will decrease by a factor of 1/3 ∼ 2/3. Even then it is more than 30 times larger than the SFR from the Lyα emission for our z ∼ 4.5 LAEs, which has a median of SFR_Lyα = 5.2 M_☉ yr⁻¹, compared to ⩽10 times at z = 3.1 (Gronwall et al. 2007) and z = 2.1 (Guaita et al. 2010). Our larger upper limits stem from a combination of factors—the larger luminosity distance at z = 4.5, a somewhat smaller sample, and the presence of a nearly significant signal in our stack, which may indicate the presence of weak AGN among our sample.

Hayes et al. (2010) reported that the average escape fraction of Lyα photons f_esc,Lyα from star-forming galaxies at redshift z = 2.2 is f_esc,Lyα = (5.3 ± 3.8)% by performing a blind narrowband survey in Lyα and Hα. Since the X-ray emission from star-forming activities is essentially unaffected by intergalactic medium (IGM) and intrinsic dust in the galaxy at high redshift,¹⁰ we can choose SFR_X as the upper limit of the unobscured intrinsic SFR. (SFR_X could overestimate the intrinsic star formation in the case where AGNs provide part of the X-ray flux.) Then, we have SFR_X⩾ SFR_intr and SFR_Lyα = SFR_intr× f_IGM× f_esc,Lyα. Songaila (2004) measured the transmission of the Lyα forest produced by IGM up to redshift 6.3. The transmitted fraction f_IGM is ∼0.3 at redshift z = 4.5 and ∼0.7 at z = 3.1. So we can get the lower limit of escaping fraction of Lyα photons as f_esc,Lyα∼ 3.2% for z = 4.5 LAEs and ∼ 9% for z = 3.1 LAEs¹¹ with the SFR_X relation from Ranalli et al. (2003). The lower limit of f_esc,Lyα could rise by an additional factor of 2–3 based on the recent SFR_X calibrations from Rosa-Gonzalez et al. (2009) and Mas-Hesse et al. (2008), which show that more X-ray photons are produced through star-forming activities.

Any weak AGNs at high redshift should be captured in the narrowband surveys, provided their Lyα emission is strong enough. However, the AGN fractions among LAE samples reported in the literature refer to quasar fractions (L_x > 4 × 10⁴³ erg s⁻¹) for all samples at redshifts z > 3. This is mainly due to the inadequate depths of X-ray exposures, apart from the two CDFs. At z ∼ 2.1, Guaita et al. did not report the X-ray luminosities, which can be as low as 10⁴² erg s⁻¹ in CDF-S. In the local universe, a large fraction of weak AGNs were reported based on multiple methods including X-rays (Finkelstein et al. 2009c). Although the LAEs of Gronwall et al. are located in the CDF-S where the X-ray luminosity is complete above 8 × 10⁴² erg s⁻¹ at z ∼ 3.1, they only found one X-ray-detected LAE in ECDF-S, with L_X = 2.8 × 10⁴⁴ erg s⁻¹. They used a stacking analysis to derive a 3σ upper limit of 3.8 × 10⁴¹ erg s⁻¹ on the mean 0.5–2 keV luminosity of their LAEs. Their stacking is based on the old 1 Ms CDF-S, but when we repeated this stack using the 2 Ms CDF-S data, we also found no signal (S/N < 1). The new data decreased the 3σ upper limit luminosity to 3.1 × 10⁴¹ erg s⁻¹ at z ∼ 3.1. In contrast, our z ≈ 4.5 stacking analysis gives a 3σ upper limit luminosity to luminosity of 〈L_{0.5–2 keV}〉 = 2.4 × 10⁴² erg s⁻¹, where weak AGNs with luminosity of ∼10⁴² erg s⁻¹ might be hidden.

As mentioned in Section 1, AGNs could be the cause of high EWs of LAEs. The rest-frame EWs of two Lyα-selected AGNs in CDF-S (see Figure 5) are ∼30 Å (our work) and ∼100 Å (Gronwall et al. 2007). In Subaru/XMM Deep Field survey, the two Lyα-selected AGNs at z = 3.7 and 3.1 show rest-frame EWs of 60–70 Å. At z = 2.25, all the nine Lyα-selected AGNs in the COSMOS field show a rest-frame EW range of 25 Å < EW_rest< 160 Å (Nilsson et al. 2009). Although the intrinsic Lyα EW for AGN is uncertain, the rest-frame Lyα EWs of bright AGNs are typically in the range 50–150 Å (Charlot & Fall 1993, and references therein). Charlot & Fall (1993) show that AGNs which are completely surrounded by neutral hydrogen gas have rest-frame Lyα EWs of 827α⁻¹(3/4)^α Å (ignoring absorption by dust), where α is the spectral index blueward of the Lyα line. According to the template of Sazonov et al. (2004), α = 1.7, which yields EW_rest ∼ 300 Å. Considering the scattering in the IGM, Dijkstra & Wyithe (2006) show that the intrinsic distribution of EW should be centered on EW_rest = 100 Å with σ_EW = 30 Å. We also check the three type 2 quasars in Figure 5. Only type 2 quasars like CXO 52 would be selected as an LAE candidate with a large EW; the other two are either too faint or have an insufficient narrowband-to-broadband contrast to be selected as LAEs. Prior to examination of the optical spectra of our LAEs, only from the view of X-ray and optical images, we found that LAEs in CDF-S with EW_rest(Lyα) < 400 Å dominate the signal as shown in Figure 3—indeed, this subsample has a soft band S/N as high as 2.7 (see Table 1). This is mainly due to the two LAEs which show S/N_soft > 1 (Figure 2) and contribute about half of the net counts. As mentioned in Section 3.2, spectroscopic results show that the two LAE candidates are not Lyα galaxies at z ≈ 4.5. Excluding these two objects from the stack, the subsample with EW_rest(Lyα) < 400 Å decreased to an S/N of 1.7. This level of signal could be simply a Poisson fluctuation in the photon statistics. Alternatively, it may be due to some low-luminosity AGN in the sample (as seen in the low-redshift Lyα-selected AGNs at z ∼ 0.3) or to star formation in the modest number of foreground and LBGs that enter the sample. Low-luminosity AGN entering our sample could be either type 1 or type 2. The type 1 AGNs are most likely confined to the EW_rest(Lyα) < 400 Å subsample, while the type 2 AGNs show a larger dispersion in both EW_rest (Figure 5) and other properties (Figure 4). This can be explained by the distinct mechanisms for the extinction of Lyα photons and the X-ray absorption for type 2 AGN, e.g., extinction of Lyα photons by narrow line region and absorption of X-ray photons by dust torus. Then, most Lyα-selected AGNs are likely to be hidden in the low EW_rest region.

Zoom In Zoom Out Reset image size