2. The X-ray - radio content of the sample
This study's X-ray data are from the second processing of the ROSAT All-Sky Survey (RASS-II). The list contains 80.000 X-ray sources with a detection likelihood and a positional accuracy such that 68% of the sources are found within 18" of their corresponding optical counterparts. The Survey has a limiting sensitivity of a few times erg cm-2 s-1 in the 0.1 - 2.4 keV energy band over the whole sky, although this depends slightly on the spectral slope, the amount of Galactic absorption and ecliptic latitude. For details of the RASS-II processing, see Voges et al. (1999).
The radio data are from the April 24, 1997 version of the FIRST VLA catalogue 3 (White et al. 1997). Data were taken with the VLA in its B-configuration and this version of the catalogue contains 268 000 sources in the north and south Galactic caps. It covers approximately 3000 square degrees and consists of the following areas: RA(2000) , RA(2000) , Dec (south).
The accuracy of a radio position depends on the brightness and size of the source and the noise in the map. Point sources at the detection limit of the catalogue have positions accurate to better than 1 arcsec at 90% confidence; 2 mJy point sources typically have positions good to 0.5 arcsec. The best possible positional uncertainty is limited to about 0.1 arcsec, although the systematic position errors are smaller than 0.05 arcsec.
All sources whose peak flux () is greater than 5 the RMS are included in the catalogue, corresponding to a flux limit of 1 mJy (or slightly better) over 99% of the survey region. The peak and integrated flux densities are derived by fitting elliptical Gaussian models to the source. Complex radio sources therefore appear in the catalogue as multiple entries. The uncertainty in is given by the RMS noise at the source position while uncertainty in can be considerably greater depending on the source size and morphology. For bright sources the accuracies of and are limited to about 5% by systematic effects. For sources that are not well-described by an elliptical Gaussian model, is not an accurate measure of the integrated flux density. In particular, sources whose extent is larger than 30 arcsec have likely been over-resolved and the reported measures of the integrated flux should be used with caution. Additional information on the FIRST catalogue data products and reduction procedures can be found in White et al. (1997).
We correlated the FIRST and RASS catalogues, keeping all sources whose radio and the X-ray positions differed by less than 60". This selection criterion matched 2588 FIRST sources with 1649 RASS sources out of a possible total of 5520 RASS sources residing in the FIRST survey area. A considerable number of X-ray objects had more than one possible radio counterpart. The cumulative distribution of the separations between the X-ray and radio positions is shown in Fig. 2. The top line gives the distribution of angular distances for all matches between the X-ray and radio positions. The shaded area shows the one-to-one matches between radio and X-ray positions, i.e., it excludes X-ray sources with multiple matches in FIRST. The dashed line shows the expected spurious match rate for all (i.e., multiple and single) radio sources.. We examined in greater detail those fields with multiple FIRST sources and found that instead of the usual two radio counterparts expected if occasional background sources fell purely by chance in the 1 arcmin field, most "multiply matched" ROSAT sources had three counterparts. Closer examination of the radio maps showed these sources often had a complex morphology, where identification of a radio core component is highly problematic. A physical picture for these associations might be a distant X-ray emitting cluster of galaxies where the radio sources are individual galaxies in the cluster. A first support for this scenario is coming from ROSAT HRI observations of the source RX J1234.6+2350 (Gliozzi et al. 1999). Because of the considerable uncertainty in identifying an optical counterpart, we have chosen to exclude these objects from much of the detailed analysis which follows, although we give the relevant radio and X-ray data in Table 2. McMahon et al. (2000) discuss in some detail the distribution of optical counterparts for double (and triple) FIRST sources.
The distribution of the angular separation of the one-to-one or "single" FIRST/RASS matches is well represented by a Gaussian with " for angular separations up to 30". From simple geometric arguments, using the catalogue sizes and areas covered in both wave bands, we estimate the number of chance coincidences to be about 35, i.e., 2% up to angular separations of 30". The increasing slope in the "one-to-one" matches beyond 30-40" is a direct indicator for the increasing number of chance coincidences at larger extraction radii (Fig. 2).
The number of X-ray - radio correlations found in a certain radio flux interval is subject to a strong selection effect caused by the limiting sensitivity of the X-ray survey. In Fig. 3 we show the total number of FIRST sources as function of the measured peak flux as an open histogram; the corresponding number of radio-X-ray matches is given as a shaded histogram. The detection probability rises from about 0.25% at the lowest radio flux levels to a few percent of the sources with a flux of 500 mJy. These numbers are considerably smaller than found for previous correlations where less sensitive radio surveys were used and thus reflect the influence of the RASS flux limit on the source detection.
For all singly matched FIRST sources, optical counterparts were determined from Automatic Plate Measuring (APM) scans of the O and E POSS plates (McMahon 1991). Following White et al. (2000, W00), we have used the FIRST survey positions to correct the APM positions on a plate-by-plate basis. The increased accuracy of the optical positions yields a far more successful match rate (McMahon et al. 2000) since it excludes many spurious matches and eliminates few true matches. Past experience has shown that the agreement between radio and optical positions depends on the optical morphology of the counterpart where the offsets for galaxies tend to be larger than for stellar counterparts. Therefore, we associate the radio source with an optical counterpart if the angular distance between the two objects is 1.5" for a starlike counterpart and 2" for a resolved optical counterpart. In addition, we have relaxed these radio-optical position offset criteria for heavily resolved optical objects (optical sizes ). The criterion we use for these objects is that the radio/optical position difference be less that 25% of the total optical extent. This ensures that objects such as NGC 6173 which are significantly optically extended, are not considered unidentified.
A rigorous analysis of the reliability and completeness of the current sample is desirable; however, an analysis relying on the positional coincidence of the objects in the different wavelength bands only, for example with a Likelihood Ratio analysis as done for flux-limited radio - optical surveys (Windhorst et al. 1984), seems to be insufficient for a quantitative assessment as we are dealing with a `pre-selected' sample (via the radio - X-ray correlations). The X-ray selection is spatially and spectrally inhomogeneous and depends in a not well understood way on the classes of the sources. To assess the reliability of the radio and optical matching, we estimated the number of chance coincidences by shifting the radio positions by + 6" in declination and counted the number of cases where the nearest optical candidate was found inside a given angular distance.
In Fig. 4 we plot the distribution of the angular distances between the radio sources and the proposed optical counterparts (open histogram). This distribution peaks at separations 1". The probability of finding an unrelated point-like optical object within a given radius is indicated by the hatched histogram which shows the match rate when the radio positions were shifted by 6". This distribution peaks at the offset position of 6", but we can use the wings of this distribution to estimate the false match rate. Specifically, at a matching radius of 2" only 1% of the optical - radio matches (7 sources) are likely chance coincidences.
Finally, there is another direct argument for the correctness of the association of the X-ray, radio and optical sources for the majority of the objects of the sample: while the rate of optical - radio source coincidences is rather low (approximately 15% of the FIRST point sources have an optical counterpart at angular distances less than 2"; McMahon et al. 2000), 706 of the 843 singly matched RASS/FIRST sources (84%) have a counterpart on the POSS plates.
Raw APM photometric magnitudes are accurate to mag, but zero point uncertainties, saturation effects, and systematic errors for bright extended objects (galaxies) which vary from plate to plate make the effective uncertainty higher. However, the O-E colors can be used for a rough characterization of the objects (McMahon, 1991) and are reliable in the range ; outside this range non linearity effects start to dominate.
The APM magnitudes overestimate the brightness of optical sources by mag (e.g, W00). To improve the photometric accuracy and uniformity of the sample, the APM magnitudes are being recalibrated plate-by-plate (McMahon et al. 2000) using magnitudes from the Minnesota Automated Plate Scanner POSS-I catalogue (APS 4, Pennington et al. 1993). In addition, although this is a high Galactic latitude sample, some sources lie in areas of high reddening. We therefore used these re-calibrated magnitudes, and list an approximate A(E) extinction correction (which has not been applied to the tabulated magnitudes) for each candidate object. Although the zero-point corrections can be quite large (0.5 mag), the extinction corrections are usually quite small (the median values are and , see also W00) but even these corrections can become significant at the high and low RA edges of the survey reaching up to 0.3 mag in the E bandpass (see Fig. 3 in W00). The recalibration of the APM magnitudes is currently available only for the northern RASS/FIRST sample (see Sect. 2). Using the NASA/IPAC Extragalactic Data Base (NED), recent FIRST-related optical identification programs, and our own followup spectroscopic observations (see below), we have determined spectroscopic classifications for a total of 454 objects. The majority of unclassified objects are found at the lower end of the radio flux scale (see next section), where large scale surveys with high sensitivity have recently become available, further indicating that previous attempts for an optical identification of radio sources have been biased towards the brightest radio sources. We also see changes in the typical source population: at low radio fluxes the fraction of galaxies and AGN compared to the number of quasars is higher than at high radio fluxes. X-ray detection biases and selection effects in the optical identification of the sources also influence the sample composition.
2.1. New spectroscopic observations
We obtained low dispersion optical spectra of 108 objects in the RASS-FIRST catalogue over the course of 13 observing runs from 1996 through April 1998 using the Lick 120" Shane reflector plus Kast double spectrograph and the Kitt Peak National Observatory's 2.1-m telescope with the GoldCam spectrograph and F3KC Ford CCD. While most of our observations were of previously unknown bright (18.5 mag) objects, some previously known and fainter unclassified objects were also observed as time and observing conditions allowed. Spectra were taken through a 2" slit, resulting in a resolution of 6 Å for the Lick spectra and 4 Å for the Kitt Peak spectra. Wavelength coverage generally extends from 3700Å to 7500Å. At Lick, the slit was always rotated to the parallactic angle. This was not done at Kitt Peak since it adds significant overhead to the observing, but an effort was made to observe objects as they crossed the meridian.
The goal of these observations was to be able to spectroscopically classify these radio- and X-ray-emitting objects as belonging to one of several broadly defined classes - Galactic stars, galaxies, starbursts, BL Lac objects, and broad or narrow-lined AGN. We therefore required spectra with sufficiently high S/N ratios to unambiguously detect features characteristic of these different classes. Simulated spectra consisting of various line widths and strengths plus a Poisson noise contribution show that a broad emission line of 5 Å can be detected at a 4 confidence level in a 10 Å resolution spectrum when the S/N30.
Reduction proceeded in the standard manner using IRAF V2.11. Wavelength calibration was carried out using comparison lamps generally taken at the beginning of the night. Comparison of our measured redshifts with published values for previously observed sources shows our values are accurate to for z1.0 and for z1.0. The latter limit is more uncertain because the high-redshift objects generally exhibit only broad emission lines. Fig. 6 shows example spectra corresponding to each of the major spectroscopic classifications (see below). Source name and redshift are given at the top of each panel.
In Table 1 we present the relevant data for all 843 unique RASS-FIRST matches. We present only a sample page of the table here; a full copy of the table is available from the CDS via anonymous ftp to cdsarc.u-strasbg.fr.
Table 1. ROSAT sources with single FIRST matches
Column 1 gives the ROSAT All-Sky Survey identification of the X-ray source, followed by the J2000 position of the FIRST source, which is assumed to be the radio counterpart. Column 3 lists the distance in arcsec between the radio and the X-ray source followed by a common name obtained from NED, truncated to the first 14 characters.
The next two entries give the type of the object and the reference for its classification: "QSO" is any object with broad ( 1000 km s-1) lines and ; "A" are narrow line objects ( 1000 km s-1) with [OIII]/H 2.0 and/or [NII]/H 0.6. "BA" are objects with broad (1000 km s-1) lines and ; "H" are starbursts with [OIII]/H 2.0 and/or [NII]/H0.6; "G" denote galaxies, objects with no emission lines and Ca II break contrasts 30% (but see the defintion of BL Lacs given below) and "B" are BL Lacs as defined in Laurent-Muehleisen et al. (1998). "S" are stars and "Cl" are possible clusters (based on the proximetry of a known cluster). It should be noted that the galaxy / cluster designation is based on either visual inspection of the optical environment of the source or NED classification of a cluster. For these sources, the multiwavelength fluxes should be treated with caution as the radio and optical emission might be from a galaxy in a cluster whereas the X-ray flux may be dominated by emission from extended cluster gas. Objects denoted by `rad' are spectroscopically unclassified radio sources, which were part of the RGB survey (B95). For previously known objects, the data found in NED were used to avoid confusion. For this reason, some objects are designated as "Sy", "Irs", "Vis", "UvE", etc. In the following analyses, objects with these classifications were either grouped with one of the standard classifications (e.g., quasar) or were excluded if it proved impossible to determine a standard classification. A small number of objects are most likely associated with bright Galactic stars (from Simbad: `Smb') although an optical identification is not always unambiguous so care should be taken with these tentative associations. Objects classified with spectra taken as part of this program are designated "New" in the tables, objects referenced as "id" have been spectroscopically classified and will be discussed in Becker et al. (2000). There also exist several cases where the given identification is ambiguous and/or where there are more than one candidate for the X-ray object at small distances. References for classifications include NED, Simbad, this paper, and the FIRST spectroscopic followup papers W00, Becker et al. (2000), and Gregg et al. (1996).
Following the redshift in Column 5, we give the FIRST 1.4 GHz peak flux (in mJy) in Column 6. The X-ray flux in the 0.1 - 2.4 keV energy range with its statistical error in units of erg cm-2 s-1 (for details see below) is given in Column 7, followed by the power-law photon index of the X-ray source deduced under the assumption of Galactic absorption. If no values are given the quality of the data does not allow the determination of a meaningful spectral index. In the following four columns, the E- and O-magnitudes of the optical counterpart are given as well as a value for the classification of the optical object: 1 for `non-stellar', -1 for `star-like', 0 for `noise-like', 2 for `possible blend'. We give the extinction correction in Column 13 (see Sect. 2) and the the angular distance between radio and optical position (in seconds of arc) in the last column.
Table 2 contains similar information (sans data about optical counterparts; see Sect. 2) for the 518 RASS fields with multiple FIRST counterparts. The first line contains the X-ray flux, classification/redshift (if known) and the following lines contain the X-ray-radio angular separation, position (J2000) and peak radio flux (mJy).
Table 2. ROSAT sources with multiple FIRST matches
Table 1 shows that about 63% of the 1400 classified sources are `star-like', 25% are `non-stellar', and 12% are blends on the O plates. Very few objects are `noise-like'. These ratios are slightly different on the E-plates with more objects classified as `non-stellar' and `blend' instead of `star-like'. Both classifications are, therefore, given in the table.
We obtained the X-ray fluxes from the measured count rates by assuming an average photon index of for the underlying X-ray spectrum and Galactic absorption (Dickey & Lockman 1990, Stark et al. 1992; for details see Brinkmann et al. 1994). The stated errors reflect the errors in the counting statistics of the survey sources and do not incorporate deviations from the assumed power-law slope, additional absorption, or systematic errors depending on the form of the local X-ray background or on details of the detection algorithm. A reasonable estimate of the total error of the X-ray flux is therefore of the order of .
The quoted photon indices were estimated using the two hardness ratios given by the RASS II processing (Voges et al. 1999) and applying the method described in Brinkmann et al. (1994), for fixed Galactic absorption. The errors of the power-law indices were estimated from the errors of the hardness ratios (Schartel 1995).
© European Southern Observatory (ESO) 2000
Online publication: April 10, 2000