THE VLA SURVEY OF THE CHANDRA DEEP FIELD–SOUTH. III. X-RAY SPECTRAL PROPERTIES OF RADIO SOURCES

Among the most fundamental issues in astrophysics are when and how galaxies formed and how they evolved with cosmic time. In particular, it is crucial to understand the relation between the star formation processes and the mass accretion history onto the central supermassive black holes in elliptical galaxies and the bulges of spirals, as traced by the tight relation between the mass (or the velocity dispersion) of the bulges and the mass of the supermassive black holes associated with the active galactic nucleus (AGN) phase (Kormendy & Richstone 1995; Magorrian et al. 1998). Since these processes have different signatures throughout the electromagnetic spectrum, multiband observations are needed to unravel the complex history.

In particular, deep multiwavelength surveys help us to reconstruct the cosmic evolution of AGNs and star formation processes. In this respect, X-ray and radio emission are good tracers of both processes. The radio properties of the X-ray population found in deep surveys have been studied in a few papers based on VLA data in the Chandra Deep Field–North (CDFN; Richards et al. 1998; Richards 2000; Bauer et al. 2002; Barger et al. 2007), combined MERLIN and VLA data in the CDFN region (Muxlow et al. 2005), and Australia Telescope Compact Array (ATCA) data in the Chandra Deep Field–South (CDFS; Afonso et al. 2006; Rovilos et al. 2007). Deep radio surveys are also realized in shallower but wider X-ray fields like COSMOS (see Schinnerer et al. 2007; Smolcic et al. 2008a; 2008b). In this paper, we use the deep radio data obtained with the VLA in the CDFS and Extended Chandra Deep Field South (E-CDFS) fields. The comparison of the properties of the radio sources (whose catalog is presented in Kellermann et al. 2008, hereafter Paper I) and of the X-ray sources (see Giacconi et al. 2002; Alexander et al. 2003; Lehmer et al. 2005) allows us to characterize both processes over a wide range of redshifts.

In this paper, we present a systematic study of the X-ray properties of the radio sources in the CDFS radio catalog. The radio catalog includes 266 sources (see Paper I) and constitutes one of the largest and most complete samples of μJy sources in terms of redshift information. We have redshifts for 186 (∼70%) of the sources, 108 spectroscopic and 78 photometric. We have reliable optical/near-IR identifications for 94% of the radio sources, and optical morphological classifications for ∼61% of the sample. Optical and near-IR properties of the radio sources are discussed by Mainieri et al. (2008, hereafter Paper II), while a multiwavelength approach to studying the source population is presented by Padovani et al. (2009, Paper IV).

The present paper is organized as follows. In Section 2, we briefly describe the radio, X-ray, and optical data sets. In Section 3, we describe the procedure used to identify the X-ray counterparts of the radio sources. In Section 4, we describe the X-ray properties of the radio sources with X-ray counterparts in the catalog of Giacconi et al. (2002) for the CDFS and of Lehmer et al. (2005) for the E-CDFS. In Section 5, we show the average X-ray properties of radio sources without individual X-ray counterparts, obtained by stacking techniques. Our conclusions are summarized in Section 6. Luminosities are quoted for a flat cosmology with Ω_Λ = 0.7 and H₀ = 70 km s⁻¹ Mpc⁻¹ (see Spergel et al. 2007).

2.1. The Radio data

2.2. The X-ray data

2.3. The optical data

The radio catalog is presented in Paper I. To investigate the X-ray properties of the radio sources, first we match the radio sources with the X-ray catalogs of Giacconi et al. (2002, after applying the positional shift correction as in Alexander et al. 2003), whenever their exposure time in the X-ray image is larger than 25% of the maximum exposure of the CDFS-1 Ms field. For all the remaining radio sources, we match them with the catalog of Lehmer et al. (2005) in the E-CDFS field. To identify X-ray counterpart candidates, we initially selected all the pairs of radio and X-ray sources with separation less than 3σ_d, where σ²_d = σ²_r + σ²_X and σ_R and σ_X are the rms error of the radio and X-ray positions, respectively. Typically, σ_x ranges from 02 to 15 depending on the off-axis angle, as computed by Giacconi et al. (2002), while σ_r ranges from 05 to 2''.¹⁶ If more than one source satisfy this criterion, the preferred counterpart is the one with the smallest offset. However, this criterion for radio/X-ray source matching was refined as described below.

First, we examined the 126 radio sources in the CDFS-1 Ms observation. Following the matching procedure discussed above, we searched for X-ray counterparts and find 55 matching candidates. Then, we use the optical identifications both of the radio and the X-ray sources (see Paper II) to refine the positions and check for possible false matches. We then produced thumbnails of the radio, X-ray, and optical images for all the candidates. Optical images are chosen from the available bands, using a priority-based on depth and spatial resolution: R band with FORS at VLT (see Giacconi et al. 2002), z-band GEMS (Rix et al. 2004), i-band ACS for GOODS (Giavalisco et al. 2004), wide field images (WFI) R deep (Hildebrandt et al. 2006). A close visual inspection, with the help of the optical images which have the best resolution, allowed us to discard three likely false matches, whose X-ray counterpart candidates are associated with optical sources different from the optical counterparts of the radio sources (see the contour maps of the extended sources overlaid over the WFI images with the position of the X-ray sources shown in Paper I). We also visually investigated the radio sources without X-ray match candidates to look for missed matches, but found none. The X-ray counterparts from the Giacconi and Lehmer catalogs are shown in Table 1 of Paper I

Down to the CDFS-1 Ms flux limits (5.5 × 10⁻¹⁷ erg cm⁻² s⁻¹ in the soft 0.5–2 keV band and 4.5 × 10⁻¹⁶ erg cm⁻² s⁻¹ in the hard 2–10 keV band) we have 52 radio sources with X-ray counterparts (corresponding to 40% of the whole radio sample) and 74 radio sources without X-ray counterparts.

For the remaining radio sources, we searched for X-ray counterparts in the E–CDFS data with the same procedure. We find 37 (corresponding to 26%) X-ray matches (after removing one false match). In total, we thus have 89 radio sources with X-ray counterparts (52 from the CDFS-1 Ms data, and 37 from the E-CDFS data). This number is exactly the same found by Rovilos et al. (2007), where the ATCA data by Afonso et al. (2006) are used. However, we compared our X-ray-detected radio sources with those of Rovilos et al. (2007), and we found significant differences: we have 31 X-ray-detected radio sources not included in Rovilos et al. (2007), while Rovilos et al. (2007) include 31 sources which we do not include. Among them, 23 are radio sources not present in our catalog, one has been discarded as a false match, and seven are not included because of our matching criterion. The main difference in the two samples is accounted by the differences of the two radio surveys, both in sensitivity (Kellerman et al. 2008 is more sensitive in the center, while Afonso et al. 2006 in the outer regions) and in the accuracy of the radio positions.

The majority of radio sources (177) do not have X-ray counterparts in the X-ray catalogs, but were studied with stacking techniques. In summary, about 1/3 of the radio sources cataloged in Paper I have an X-ray counterpart in the present analysis.

For all of the radio sources with an X-ray counterpart and redshift information, we analyze the X-ray spectrum as described in Tozzi et al. (2006). We have 89 radio sources with a cataloged X-ray counterpart, 76 of them with known redshifts. Among them, 31 have spectroscopic redshifts and optical classification based on the detection of optical emission lines, while the remaining 45 sources have only photometric redshifts or uncertain optical classification.

The normalized distribution of X-ray fluxes of the radio sources with X-ray counterparts found in the 1 Ms field is shown in Figure 1 for the soft (left panel) and hard (right panel) bands. The distributions are compared with those of the whole X-ray sample in the 1 Ms exposure of the CDFS. We find that the radio selection at the current flux density limit marginally tends to consist of the brightest X-ray sources. From a Kolmogorov–Smirnov (K–S) test, we find that the probability of the two distributions being extracted from different parent populations are ∼87% and ∼95% for the soft and hard bands, respectively. None of these probabilities are significant (i.e., >95%); therefore, we conclude that the additional brightness limit introduced by the radio selection is not affecting much the X-ray flux distribution. On the other hand, the normalized redshift distribution of the radio sources with X-ray counterparts is significantly shifted toward lower redshifts with respect to the distribution of the whole X-ray sample, as shown in Figure 2. The average redshift of the radio sources is 〈z〉 = 1.01 (median 0.73), while that of the X-ray sources is 〈z〉 = 1.28 (median 1.03). The two redshift distributions are inconsistent at more than the 99% confidence level. This is mostly due to the larger fraction of star-forming galaxies among the X-ray sources with radio counterpart, since the radio emission is often associated with star formation. Since star-forming galaxies are intrinsically fainter in the X-ray band, they are typically found at lower redshift with respect to AGNs among the CDFS X-ray sources. This is the main reason of the shift toward lower redshift among radio sources with X-ray counterparts. Note that the peak at z ∼ 0.7 is due to the large-scale structure noted in Gilli et al. (2003).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Most of the sources with spectroscopic redshifts have also an unambiguous optical classification based on the detection of high-ionization emission lines, as described in Section 2.2. However, only 31 sources in the Szokoly et al. (2004) sample have spectroscopic redshifts and optical classification based on the detection of optical emission lines. We note that among the radio sources with X-ray counterparts, a wide range of optical types are present. We find 12 radio sources distributed among BLAGN (five) and HEX (seven), which corresponds roughly to Type I and Type II AGNs, respectively (see Szokoly et al. 2004; Tozzi et al. 2006). The most common optical species corresponds to LEX, which includes 14 radio sources. Among these sources we expect a larger number of star-forming galaxies or narrow line emission Galaxies, but also hidden AGNs. Only five sources are in the ABS spectroscopic class.

The distribution of radio luminosities is different for the four optical types, as shown in Figure 3. The radio-luminosity density is computed using the measured radio-spectral slope when available (see Paper I), while we assume the average value α_R = 0.7 for sources detected only at 20 cm. The rest-frame radio-luminosity density was calculated as $L_{1.4\,{\rm GHz}} = 4\pi d_{\rm L}^2 S_{1.4\,{\rm GHz}} 10^{-33} (1+z)^{\alpha _{\rm R}-1}$ W Hz⁻¹, where d_L is the luminosity distance (cm), and S_1.4 GHz is the flux density (mJy). The radio luminosity of BLAGN ranges from 10²³ to a few ×10²⁶ W Hz⁻¹, while for HEX sources it reaches only 10²⁵ W Hz⁻¹. The LEX radio sources are mostly in the range from 10²¹ to 10²⁴ W Hz⁻¹, typical of star-forming galaxies. Sources in the ABS optical class have radio luminosities consistent with Faranoff–Riley Type I galaxies and star-forming galaxies for the lower luminosity sources, except one bright radio galaxy.

Zoom In Zoom Out Reset image size

All of the information for the 76 radio sources with an X-ray counterpart and redshift are shown in Tables 1 and 2 for the 1 Ms and the E-CDFS fields, respectively. X-ray luminosities are obtained from the X-ray spectral analysis, and refer to the intrinsic (de-absorbed) emitted power. For 13 radio sources with X-ray counterpart in the CDFS or in the E-CDFS fields, we do not have redshift information. These sources are shown in Table 3 and are excluded from the X-ray spectral analysis.

In Figure 4, we plot the 20 cm radio and hard-band X-ray luminosities for the 76 radio sources with X-ray spectral analysis and redshift information. We recall here that the X-ray luminosities represent the intrinsic power emitted in the corresponding rest-frame X-ray band (after removing intrinsic absorption), and the X-ray K-correction is already accounted for via the detailed spectral analysis. We find a clear trend in the luminosity range typical of star-forming galaxies (10⁴⁰ ⩽ L_X ⩽ 10⁴² erg s⁻¹) and a larger scatter for higher X-ray luminosities. This result also reinforces the expectation that the majority of sources in our sample with L_X ⩽ 10⁴² erg s⁻¹ are powered by star-forming activity both in the radio and in the X-ray bands. Indeed, by looking at the sources with optical classification, the star-forming galaxies sector may be conservatively defined by the conditions log(L_x) ≃ 41.5 and log(L_1.4) ≃ 23. However, this result is based only on 40% of the sources and must be taken with caution. Among 14 sources in the SF-galaxy luminosity range, the dominant optical types are LEX (10 sources) and ABS (2 sources), with only two HEX sources.

Zoom In Zoom Out Reset image size

We also note that the correlation expected for star-forming galaxies, as found by Ranalli et al. (2003) for the X-ray luminosity range 10³⁸ < L_X < 10^41.5 in the hard band, appears to be a factor ∼2 higher. On the other hand, the radio-X-ray luminosity relation by Persic & Rephaeli (2007) for the integrated X-ray hard luminosity at high redshift (see their Equations (10) and (11)) is in better agreement with the X-ray-detected radio sources. The same holds for the L_R–L_X relation found for 122 late type galaxies in the CDFN studied by Bauer et al. (2002). A censored statistical analysis including the X-ray upper limits, confirms the linear slope and normalization of Ranalli et al. (2003), as shown in Paper IV. The presence of a robust L_X–L_R relation agrees also with the recent analysis by Lehmer et al. (2008) on a sample of late-type galaxies in the CDFN and E-CDFS, suggesting that X-ray emission can be used as a robust indicator of star formation activity out to z ∼ 1.4. All the mentioned studies, including our own, are at variance with the argument of Barger et al. (2007) that the L_X–L_R relation is spurious.

On the other hand, the wide scatter at higher luminosities reflects the wide range of radio-to-X-ray-luminosity ratio found in the AGN. Within the 17 sources with L_X > 10⁴² erg s⁻¹, the dominant optical types are distributed among BLAGN (five sources), HEX (five sources), LEX (four sources), and ABS (three sources). While X-ray emission in this luminosity range is mostly associated with the AGN, the radio emission may still be associated with star formation activity in the majority of sources, as found by Rovilos et al. (2007) using the Spitzer 24 μm luminosity.

Variability of the X-ray and radio luminosity may increase the scatter for the AGN sources. In order to check this possible bias, we considered the X-ray and radio luminosities separately for the sources with and without detected variability (Paolillo et al. 2004). Only the 46 sources for which variability has been measured (in the 1 Ms exposure field) are included. We do not find any statistical evidence for a different behavior (i.e., a larger scatter) among the two subsamples, which include 14 and 32 variable and nonvariable sources, respectively. Therefore, it seems unlikely that X-ray variability can account for a significant part of the large observed scatter in the L_X–L_R relation for the AGN. However, we cannot completely exclude some effect, since, as shown in Paolillo et al. (2004), probably the large majority (>90%) of the CDFS sources are X-ray variable, their variability being undetected due to the low S/N.

We also compute the radio loudness with respect to the hard-band X-ray flux as defined in Terashima & Wilson (2003). Here we use the 20 cm luminosity rather than the 6 cm luminosity used by Terashima & Wilson: R_X ≡ νL_R(5 GHz)/L_2–10. On average, radio loudness defined by the 20 cm luminosity is shifted by −0.16 with respect to R_X defined at 6 cm. Therefore, we take log(R_X) = −2.9 as the boundary between radio-loud and radio-quiet AGNs found by Panessa et al. (2007) with a sample of local Seyfert galaxies and low–luminosity radio galaxies. Since this criterion applies only to sources with nuclear activity, we consider only sources with L_X > 10⁴² erg s⁻¹. As shown in Figure 5, the radio loudness distribution of the whole sample of sources with X-ray spectral analysis shows some bimodality. Using the same radio-loud/radio-quiet boundary, among the ∼50 sources with L_2–10 > 10⁴² erg s⁻¹, ∼1/3 are radio loud and ∼2/3 radio quiet.

Zoom In Zoom Out Reset image size

In Figure 6, left panel, we plot the fractional distribution of intrinsic absorbing columns of equivalent N_H for the 76 radio sources with X-ray spectral analysis and redshift information. The shape of the distribution for N_H > 10²¹ cm⁻² is similar to that of the entire X-ray sample (dashed line), but the two distributions are inconsistent at more than 3σ, due to the significantly larger number of radio sources with low-intrinsic absorption. This is reflected also in the larger fraction of sources with L_X < 10⁴² erg s⁻¹ in the radio sample (35%) compared with the whole X-ray sample (20%). These X-ray sources are mostly powered by starbursts, and therefore the X-ray emission does not show the intrinsic absorption found in sources with nuclear emission.

Zoom In Zoom Out Reset image size

We divided sources into two subsamples according to the X-ray luminosity, and we find that radio sources with L_2–10 > 10⁴² erg s⁻¹ have a distribution of N_H consistent with that of X-ray-selected sources in the same luminosity range (see Figure 6, right panel). Again, this shows that the subsample of radio sources with X-ray luminosity L_2–10 > 10⁴² erg s⁻¹ is representative of the X-ray-selected AGN population. This does not show any significant difference in the intrinsic absorption properties of X-ray sources with and without radio counterparts. This allows one to discard a simple model in which the radio emission is associated with starbursts which in turn would absorb the X-ray emission from the AGN, as discussed by Rovilos et al. (2007). As a general remark, we recall that the gap in the distribution at N_H < 10²¹ cm⁻² is not due to difficulties in measuring low values of N_H, especially at high redshifts. Indeed, as shown in Figure 9 of Tozzi et al. (2006), the resampling of N_H values according to the statistical error, decreases only by 20% the number of sources with N_H < 10²¹ cm⁻². Here, for simplicity, we do not correct for this effect, nor for incompleteness (see, again, Tozzi et al. 2006), since we mostly focus on the comparison between radio sources with X-ray counterpart and the parent X-ray population.

In Figure 7, we plot the intrinsic absorption versus redshift. The apparent increase of N_H with redshift is due to the difficulty of measuring N_H at high redshift as discussed in Tozzi et al. (2006). In this Figure, we see clearly that the large number of N_H upper limits, causing the difference in Figure 6 (left panel), are mostly at low redshifts. The two effects are clearly the same, and are due to the X-ray flux limit, which introduces a sharp cutoff around z ∼ 1 for sources with L_X ⩽ 10⁴² erg s⁻¹, where all the star-forming galaxies, showing no intrinsic absorption, are found.

Zoom In Zoom Out Reset image size

To summarize the properties of X-ray-detected radio sources, we show in Figure 8 a simple but efficient classification based on X-ray properties only. We assume L_2–10 = 10⁴² erg s⁻¹ as the threshold luminosity separating star-forming galaxies and AGN, and N_H = 10²² cm⁻² as the conventional threshold intrinsic absorption for unabsorbed and absorbed AGNs. We find 23 star-forming galaxies, 15 unabsorbed AGNs and 29 absorbed AGNs. In the star-forming regime, the presence of absorption, not necessarily with high N_H, is the signature of nuclear emission. Therefore, if we adopt N_H = 10²¹ cm⁻² as a conservative threshold, we can also tentatively identify nine low-luminosity AGNs at L_2–10 < 10⁴² erg s⁻¹. The optical classification for 40% of the sources shows that several AGNs are missed by optical spectroscopy, while only two HEX are included in the star-forming galaxies sector.

Zoom In Zoom Out Reset image size

Broadly speaking, we find that X-ray emission of 1/3 of the X-ray-detected radio sources is consistent with being associated with star formation in the host galaxy, while the remaining 2/3 are AGNs. We also find a weak correlation between the radio loudness and the intrinsic absorption among the sources with L_2–10 > 10⁴² erg s⁻¹. As shown in Figure 9, there is a large scatter, but the radio loudness is significantly higher at lower N_H. The Spearman rank correlation coefficient is −0.2, with a significance weaker than 2σ. This is a hint that radio emission is decreasing with increasing absorption among X-ray-detected AGNs.

Zoom In Zoom Out Reset image size

Finally, we find a correlation between the radio-spectral index α_R, computed between 20 and 6 cm, and the intrinsic absorption measured in X-ray. The correlation is significant at the 3σ level in a Spearman rank correlation test for the 59 sources which have both measured 6 cm flux densities and X-ray spectral analysis. The relation is shown in Figure 10 for only 59 sources which have both measured 6 cm fluxes and X-ray spectral analysis. On the other hand, we do not find a correlation between the radio-spectral index and hard X-ray luminosity, nor between radio-spectral index and radio-X-ray loudness. The apparent trend of having flatter radio spectra at lower intrinsic absorption may be due to a component of thermal radio emission dominating at low N_H due to star formation processes, while AGNs, with significant intrinsic absorption, show steeper radio spectra typical of nonthermal transparent synchrotron emission (see Richards 2000). Indeed, the correlation we found in our sample is partially due to the presence of star-forming galaxies, since it becomes significant only at the 90% level when only sources with L_2–10 > 10⁴² erg s⁻¹ are included. Finally, we note that the average X-ray spectral slope of the X-ray-detected radio sources is Γ = 1.8 ± 0.1, in agreement with that of the X-ray sample.

Zoom In Zoom Out Reset image size

We have 174 radio sources in the field of view of the CDFS 1 Ms exposure or in the complementary area of the E-CDFS without an X-ray counterpart (we note that the sources with RID = 1, 21, and 266 are outside the E-CDFS field). To retrieve X-ray information about these sources, we performed aperture photometry on the X-ray images at the radio positions. We use X-ray images obtained by masking the cataloged X-ray sources, replacing the removed regions with a Poissonian background based on the measured value of the local background. In this way, we avoid including the emission from any detected X-ray source. We performed the photometry separately in the soft (0.5–2 keV) and hard (2–7 keV) X-ray bands. The results are given in Table 4. The net counts and the S/N are computed in the extraction radius which is dependent on the off-axis angle of the X-ray images, as described in Giacconi et al. (2001).

As we see from Table 4, there are 26 sources whose S/N from aperture photometry in one of the two bands is higher than the signal-to-noise limit for X-ray-detected sources. However, these sources should not be considered X-ray-detected since, on the basis of the X-ray detection algorithm, they have a very low probability of being real sources, and therefore were not included in the Giacconi et al. (2002) or the Lehmer et al. (2005) catalogs. Eventually, we have included their contribution in the stacking analysis of all the X-ray-undetected radio sources. For all the other sources, we quote the 3σ upper limits both in counts and fluxes.

In general, the histogram distributions of the net counts, shown separately in the 1 Ms exposure (Figure 11) and in the complementary area of the E-CDFS (Figure 12) show a clear excess with respect to a distribution of photometry based on random positions (dashed line) in the soft band (left panel), and a marginal excess in the hard band (right panel). The probability that the measured and random net counts distributions are different is more than 5σ in the soft band, while it is marginal in the hard band for sources in the 1 Ms CDFS (80%) and in the E-CDFS (60%). In any case, there are no sources which dominate the X-ray photometry of the radio sources with no cataloged X-ray sources, allowing us to perform a meaningful stacked analysis to obtain their average properties.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

A visual impression of the total X-ray emission from the radio sources without counterparts in the X-ray catalog can be obtained simply by stacking the X-ray image at the position of the radio sources (see, e.g., Georgakakis et al. 2003). The stacked images in the soft and hard bands are shown in Figure 13. Overall, 74 radio sources in the 1 Ms field are detected with 460 ± 75 and 300 ± 90 net counts in the soft and hard bands, respectively. We performed a Monte Carlo simulation to assess the significance of the detection of the stacked image. The detection in the soft band is at more than 99.9% confidence level, while in the hard band it is at 99%. Energy fluxes are computed as in Rosati et al. (2002), using conversion factors from the measured net count rate to the energy flux, assuming an average spectral slope of Γ = 1.4, equal to 5.07 × 10⁻¹² and 2.97 × 10⁻¹¹ erg s⁻¹ cm⁻² (counts s⁻¹)⁻¹ in the soft and hard bands, respectively. After correcting for the effective average X-ray exposure time, the photometry of the stacked images corresponds to a flux of (4.4 ± 0.7) × 10⁻¹⁷ erg s⁻¹ cm⁻² and (1.7 ± 0.3) × 10⁻¹⁶ erg s⁻¹ cm⁻² per source in the soft and hard bands, respectively.

Zoom In Zoom Out Reset image size

With the same analysis, the 100 radio sources in the E-CDFS field yield 260 ± 30 and 90 ± 60 net counts in the soft and hard bands, respectively. The stacked images in the soft and hard bands are shown in Figure 14. The detection in the soft band is highly significant, while in the hard band it is only marginal. After correcting for the effective average exposure, and adopting the appropriate conversion factors for the E-CDFS fields, the average flux per source is (5.4 ± 0.6) × 10⁻¹⁷ erg s⁻¹ cm⁻² and (1.7 ± 1.1) × 10⁻¹⁶ erg s⁻¹ cm⁻² in the soft and hard bands, respectively. These values are consistent with those found in the CDFS, supporting the accuracy of the stacking procedure.

Zoom In Zoom Out Reset image size

To investigate further the nature of the weak X-ray emission, we evaluate the average hardness ratio defined as HR = (H − S)/(H + S), where H and S are the net counts in the hard (2–7 keV) and soft (0.5–2 keV) bands, respectively, corrected for vignetting. If we stack the net counts of sources in four bins of radio flux density, we find a roughly constant value of HR ∼ −0.5 ± 0.1, indicating that the statistics are not able to indicate a significant change in the average X-ray spectral properties of the radio sources as a function of their radio flux density.

We have redshifts for 64% (110) of our radio sources with no X-ray detection. The redshift distributions of the sources with and without X-ray counterparts are consistent with each other, as shown in Figure 15. If we split the sample in four redshift bins with about 27 sources each, we can measure the average X-ray luminosities for radio sources, using the average fluxes, and assuming a power-law spectrum with Γ = 1.8. The four redshift bins are: 0.0 < z < 0.4, 0.4 < z < 0.7, 0.7 < z < 1, 1 < z < 2.3. The results, listed in Table 5, show that the X-ray luminosity for these sources lie in the range of star-forming galaxies. Only in the higher redshift bin (〈z〉 ∼ 1.4) is the average hard X-ray luminosity at the high end of the typical starburst galaxies. However, to evaluate the contribution from a population of low-luminosity AGNs, a multiwavelength approach, as discussed in Paper IV, is needed. If we plot these sources in the L_R–L_X plane, we find that, on average, they are consistent with the relation expected for star-forming galaxies (see Figure 16). This is consistent with the censored analysis presented in Paper IV, confirming the results by Lehmer et al. (2008) supporting the L_X–L_R correlation holding at high redshift; this is at variance with the claim of Barger et al. (2007).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We present detailed X-ray spectral properties of 76 VLA sources with X-ray counterparts in the CDFS (Giacconi et al. 2002) and the E-CDFS (Lehmer et al. 2005). We also present the average X-ray properties of the radio sources without X-ray counterparts in the 1 Ms CDFS exposure and in the E-CDFS field. Our main results are summarized as follows.

• 1.  
  One third of the radio sources are detected in the X-ray bands. Among them, ∼1/3 of the radio sources are consistent with being star-forming galaxies, while the remaining 2/3 are AGNs, by assuming L_X = 10⁴² erg s⁻¹ as the threshold between star-forming galaxies and AGNs.
• 2.  
  In the AGN luminosity range, L_2–10 > 10⁴² erg s⁻¹, ∼1/3 of the sources are radio loud and ∼2/3 radio quiet, where radio loud is defined as log(R_X)> − 2.9 (with R_X ≡ νL_R(5 GHz)/L_2–10).
• 3.  
  The intrinsic absorption in the X-ray band of the radio sources is shifted to lower values with respect to the X-ray-selected sample, showing that radio selection tends to find a larger number of star-forming galaxies; when selecting source with L_2–10 > 10⁴² erg s⁻¹ the distribution is similar to that of the X-ray sample.
• 4.  
  We find a weak anticorrelation of radio loudness as a function of intrinsic absorption, adding support to the finding that radio emission is not efficient in selecting more absorbed AGNs.
• 5.  
  The stacked X-ray images of 174 radio sources without cataloged X-ray counterparts show a clear detection in the soft band and a marginal detection in the hard band.
• 6.  
  The average X-ray luminosities of radio sources without cataloged X-ray counterpart are consistent with being powered by star formation.

Deeper X-ray and radio data in the CDFS will allow us to extend this analysis toward lower levels, and to obtain additional constraints on the role of star-forming galaxies as opposed to AGNs in the sub-mJy radio population.

P. Tozzi acknowledges support under the ESO visitor program in Garching during the completion of this work. We thank Massimo Persic and Piero Ranalli for discussion on the X-ray/Radio luminosity correlation, and Isabella Prandoni for valuable comments. We acknowledge financial contribution from contract ASI–INAF I/023/05/0 and from the PD51 INFN grant. The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc.