2. Cross-correlation of the ROSAT all-sky survey source list with SIMBAD OB stars
2.1. The ROSAT source list
The source list considered here arises from the first ROSAT all-sky survey data reduction as completed by 1991 October by the Scientific Analysis System Software (SASS; Voges et al. 1992). This software provides for each source the position, count rates and hardness ratios HR1 and HR2 defined as
where (A-B) is the raw count rate in the energy range A to B in keV. Because of spacecraft problems no data were available for about 5% of the sky located between ecliptic longitudes 41 and 49 and ecliptic longitudes 221 and 229 . The accumulation by 2 wide strips along the great scan circles yields variations of sensitivity perpendicular to the strip and overlapping at high ecliptic latitudes may produce multiple detection of the same source in adjacent strips. The lists of sources derived from each strip were then merged into a single master list totaling about 15,000 sources at 20 . The merging process assumed that a source detected in two or more strips was the same when the difference in position was less than a minimum value of one survey sky pixel () or less than the combined positional errors. The strip oriented analysis does not allow an easy estimate of the sensitivity of detection in a given part of the sky but allows quick detection. The errors on ROSAT X-ray positions have two different origins. First, the statistical uncertainty with which the centroid of the X-ray image is positioned on the pixel grid by the Maximum Likelihood source detection algorithm. Second, the systematic error in the knowledge of the satellite attitude for each photon collected in scan mode. The analysis of the subsample of the 13 known high mass X-ray binaries (HMXBs) detected in the RGPS by the SASS (see Sect. 2.2 and Table 10), led to the conclusion that the systematic attitude error to apply to survey positions was and that the 95% confidence radius could be expressed as
where is the Maximum Likelihood error.
2.2. The SIMBAD database
The master list used for the cross-correlation was extracted from the SIMBAD database in 1990 and not later updated. SIMBAD early type stars are mostly recognized from HD, CD, BD, HR and SAO spectral information and associated literature. We list in Table 1 the distribution in spectral types of the SIMBAD stars located within 20 from the galactic plane, subdwarfs excluded. Stars having general spectral types 'OB' mostly originate from the Luminous Star (LS) catalogues compiled by the Hamburger Sternwarte and Warner and Swasey Observatories for both hemispheres (Hardorp et al. 1959, Stephenson & Sanduleak 1971). Another important group of 'OB' stars without precise spectral classification arises from the Vatican Emission line Star (VES) catalogue (McConnell & Coyne 1983). Most of the stars classified as 'OB' are in fact earlier than B3 (Slettebak & Stock 1957). Because of the heterogeneousness of the various OB catalogues, it is difficult to quantify with accuracy the completeness in magnitude of our input sample. The Luminous Star catalogue which gathers most of the faintest early type stars is apparently complete down to B 12 over the whole galactic plane.
Table 1. Distribution in spectral types of the OB SIMBAD stars ( 20 )
We first extracted all SIMBAD entries which had an associated error circle overlapping a wide box centered on each ROSAT source. This rather large search area allows the correlation of candidate OB/X-ray sources with objects having inaccurate coordinates such as some variable stars, or with supernova remnants. In a second step we selected from the main correlation list all sources having a match within with a star of spectral type O or B, retaining for each selected OB/X-ray association the entire list of possible candidates extracted in the wide box. The restriction on angular distance allows to eliminate many spurious OB/X-ray matches since the 90% confidence ROSAT radius is usually less than (Voges et al. 1992, corresponding to a 95% confidence radius in the range of ) and optical positions of OB stars are known to better than . Thirteen sources which were thought to have a likelier identification than the proposed OB star were removed by hand from the correlation list (see Table 2). This happened for instance when a known active corona (RS CVn binary, pre-main sequence star) or a supernova remnant was also present in the ROSAT error circle. Pre-main sequence stars and RS CVn binaries are known to be sometimes bright soft X-ray sources with luminosities up to 1031 erg s-1 (e.g. Montmerle et al. 1983, Walter & Bowyer 1981). Keeping in mind that our aim was to select only promising accreting candidates leaving doubtful identifications for a later study we decided to ignore these cases for the moment. On similar grounds, compact OB associations (i.e. groups of OB stars located within few arcminutes and appearing blended at the X-ray spatial resolution of the RASS) were not considered in this analysis because of the difficulty in assessing an individual X-ray to bolometric luminosity ratio for such objects. The B2V star HD 63177 (V = 8.31) tentatively associated with RX J0744.9-5257 was also taken off from the final correlation list since optical follow-up observations have shown the presence of a cataclysmic variable away from the candidate B star (Motch et al. 1996a).
Table 2. List of OB/X-ray associations removed from the main correlation list on the basis of the existence of a possible alternative identification. Count rates are taken from the SASS analysis of the survey and the optical information listed is entirely extracted from SIMBAD as in 1990
Finally, for the sake of comparison, we keep aside the group of 13 known massive X-ray binaries detected during the ROSAT all-sky survey at (see Table 10). Of these, 9 were listed in the OB SIMBAD catalogues and retrieved correctly during the above selection process while 4 were added using data extracted from the literature (He 3 -640, Cen X-3, 1H1909+096, EXO 2030+375). The final list of OB/ROSAT source associations contained a total of 237 sources entries split as shown in Table 3.
Table 3. The final OB/RGPS correlation list
2.3. The / diagnosis for OB stars
All stars of spectral type earlier than B5 emit soft X-rays (0.2-4.0 keV) with a roughly constant X-ray to bolometric luminosity ratio independent of the luminosity class and age (e.g. Long & White 1980; Pallavicini et al. 1981). Sciortino et al. (1990) show that the mean value of / for O stars is close to 10-6.46 with about 10% of these early type stars having / in the range of 10-6 -10-5.5. From ROSAT survey data Meurs et al. (1992) find a mean log ( / ) of -6.8 0.5 for 43 O3 to B2.5 stars with no strong difference between OB and OBe stars. Using ROSAT PSPC pointed observations, Cassinelli et al. (1994) find that the / ratio decreases sharply with spectral types later than B1 and could reach 10-9 at B3V.
The physical origin of the X-ray emission is still a matter of debate. Several authors assumed a picture stretched from the solar case in which a hot corona located close to the stellar surface lies below the cooler, high speed velocity wind (see e.g. Cassinelli et al. 1981, Waldron 1984). Alternatively, Lucy & White (1980) proposed that blobs of high density formed in the expanding wind may produce shocks and consequently X-rays.
2.4. Computation of the X-ray to bolometric luminosity ratio
Interstellar extinction may alter quite significantly the ratio of X-ray to optical flux measured from these stars. Depending on the softness of the assumed intrinsic X-ray spectrum the decrease of the 0.1-2.4 keV PSPC count rate with increasing interstellar column density may be quicker or slower than that of the optical flux. For instance, the overall PSPC count rate produced by a T = 106 K thin thermal spectrum (Raymond & Smith 1977) will be steeply decreasing with column density and at = 1021 cm-2 it will be 100 times more dimmed than any optical flux crossing the same interstellar medium. On the opposite, the PSPC count rate resulting from a power law energy distribution with a photon index of 1, typical for Be/X-ray systems, will decrease less rapidly with than optical radiation.
In order to correct for interstellar extinction we computed the colour excess E(B-V) from the spectral types and magnitudes listed in SIMBAD. Intrinsic B-V were taken from Johnson (1966) and absolute magnitude and bolometric corrections are from Deutschman et al. (1976) and Humphreys & McElroy (1984). The corresponding column density was then used to compute the PSPC count rate to un-absorbed flux conversion factor assuming a 107 K thin thermal spectrum (e.g. Pallavicini et al. 1981). Finally, the X-ray luminosity was estimated using the distance derived from the colour excess and spectral type together with the catalogue V or B magnitudes.
Obviously, there exist several possible temperatures for normal OB X-ray emission. Chlebowski, Harnden & Sciortino (1989) find temperatures in the range of 3 to 9 106 K from Einstein data and Cassinelli et al. (1994) reach similar conclusions from ROSAT observations. By intentionally assuming a temperature on the hot side of the distribution we avoid a systematic overestimation of the X-ray luminosity when correcting for interstellar absorption. Only in the rare case of a soft component (e.g. 2 106 K) seen at very low do we expect overestimation of the un-absorbed X-ray emission. If the intrinsic spectrum is actually described by a power law distribution of photon index 1-2 as expected for young accreting neutron stars (e.g. White et al. 1983) the effect on the un-absorbed luminosities is not large since we overestimate them by at most a factor of 2 at low and underestimate them by the same factor at = 1022 cm-2. Finally we note that with a 107 K thin thermal energy distribution, the absorbed X-ray count rate to optical flux ratio remains constant within a factor of 2 for a large range of and that therefore, the estimated X-ray to bolometric luminosity ratio is relatively unaffected by errors on the intervening column density.
Another problem arises from the unavoidable incompleteness of the data listed in SIMBAD. Several stars lack precise spectral types and/or magnitudes. Keeping in mind our concern to select the most obvious accreting candidates, in order not to overestimate the X-ray to bolometric luminosity ratio we used as default value B0 for all stars without subtypes and luminosity class III for all stars lacking appropriate information. When the B-V colour was not available we arbitrarily assumed B-V = 0.0. These default assumptions imply that we may have missed a fraction of the low / candidates. In the few cases when a range of spectral types and luminosity classes was available, we used the average value. For early type stars, the difference in intrinsic colours and bolometric corrections between two consecutive spectral types or luminosity classes is always small compared to other uncertainties.
We show in Fig. 1 the distribution of all X-ray/OB star identifications having spectral types earlier than or equal to B5 and located within r95, the 95% confidence radius of the ROSAT position, in the / versus HR2 diagram. The hardness ratio HR2 is more sensitive to the intrinsic shape of the spectrum while the softer HR1 is rather an indicator of the photoelectric absorption. Most OB stars cluster at / between 10-7 and 10-6 as expected from previous studies carried out with the Einstein satellite (Pallavicini et al. 1981, Sciortino et al. 1990). Simulations show that indeed the slight variation in energy range from Einstein (0.2-4.0 keV) to ROSAT (0.1-2.4 keV) does not significantly change the X-ray luminosities derived from the two satellites. Similar values of / were also derived from a subsample of ROSAT survey data (Meurs et al. 1992). We did not investigate possible differences between OB and OBe stars in our sample. Most normal OB stars have HR2 values comprised between -0.7 and +0.3 which correspond to thin thermal temperatures in the range of 3 106 K to 3 107 K in agreement with those usually reported for normal OB stars. In contrast, the known massive X-ray binaries detected in the galactic plane exhibit larger luminosity ratios and a much harder HR2 which probably reflects the intrinsically harder power law-like energy distribution of accreting neutron stars and also maybe to a lower extent the often large interstellar and intrinsic column densities. We note, however, that luminous soft X-ray components were detected in some massive X-ray binaries accreting from the wind of the primary (e.g. 4U 1700-37; Haberl et al. 1994) or through Roche lobe overflow (e.g. LMC X-4; Dennerl 1991). This last source is located in a direction of low interstellar absorption and exhibits HR1 = 0.36 0.01; HR2 = -0.20 0.01. Because of the sometimes large orbital phase dependent circumstellar absorption, the presence of a soft X-ray excess, although clearly detectable from the relative strength of the hard (2-10 keV) and soft (0.1-2 keV) un-absorbed components, does not necessarily imply a very soft value of HR1 or HR2 in the ROSAT band (e.g. 4U 1700-37; Haberl et al. 1994). On the other hand, Be/X-ray systems seem to lack a soft component (Haberl 1994).
By contrast, it can be seen on Fig. 2 that B6 to B9 stars exhibit a much larger / ratio than earlier types and also a somewhat larger scatter. This behaviour was already noticed by Meurs et al. (1992). Obviously these late B stars do not obey the same relation as hotter types. The range of HR2, however, is comparable to that observed in earlier B stars. Einstein observations of nearby A type stars by Schmitt et al. (1985) showed that none of these stars had any detectable X-ray emission. Nevertheless, the Einstein observatory detected several A type stars in the Pleiades cluster (Micela et al. 1990) and more recently several field A stars were detected in the ROSAT all-sky survey (Schmitt et al. 1993). These evidences led to the common assumption that the X-ray emission sometimes associated with late B or A stars was originating from an optically undetectable G-M type companion. However, ROSAT HRI observations seem to question this explanation and may point toward intrinsic X-ray emission at least in some late B stars (Berghöfer & Schmitt 1994). In spite of the high / ratio the actual un-absorbed X-ray luminosities all remain below 2 1032 erg s-1 and only four stars exhibit X-ray luminosity in excess of 6 1031 erg s-1. Inspection of the optical maps of these four stars suggests the presence of likelier optical counterparts in the ROSAT error circle. Therefore, considering the absence of good candidates displaying X-ray luminosities above the level at which an interpretation in terms of an optically unseen late type companion star becomes untenable and the rather large expected number of spurious matches resulting from the size of the entry catalogue, we decided not to investigate the late B stars for the moment. Consequently, in the following, we shall only consider stars earlier than B6 or those having the general 'OB' type designation totaling 15895 stars.
2.5. Rate of spurious matches
We show in Fig. 3 the histogram of X-ray to optical distances for the 128 stars earlier than B6 associated with an X-ray source. The shape of the distribution shows without ambiguity that several OB stars were indeed detected during the ROSAT survey. The average 95% confidence radius for all OB star matches in the original cross-correlation list is 34.5. With a total of 15895 OB stars earlier than B6 and a mean survey source density of one per square degree in the galactic plane (Motch et al. 1991b), roughly independent of galactic latitude and longitude (Voges 1992), we expect 5 spurious matches among the 108 spatial coincidences within r95 and 10 spurious matches among the 17 associations found in the range of 1 to 1.8 r95, this latter value corresponding to the limit. The estimated number of real OB/X-ray associations within and outside r95 is 103 and 7 respectively, thus confirming the validity of the statistics used for the computation of error radii.
Because the number of non X-ray detected OB stars is much larger than the number of OB stars detected in the survey, we expect most spurious OB/RGPS source associations to be with non detected early type stars and therefore to be preferentially found in the high / samples. Consequently, the recognition of a genuine accreting source requires additional information which may be the X-ray spectral or temporal behaviour and/or follow-up optical searches for alternative optical counterparts, essentially active coronae in the galactic plane. For the B6-B9 stars, we expect 6 spurious spatial coincidences among the 79 located within the 95% confidence radius implying that these random associations cannot explain the systematically higher / ratio exhibited by these late B stars.
2.6. Selection of the candidate stars
Using the computed luminosity ratio and optical / X-ray positional information we selected three different sets of candidate OB/X-ray binaries as defined in Table 4. The use of the 95% confidence radius was guided by the fact that the expected number of spurious matches within r95 ( 5) was about the same as the number of true associations expected to be located outside this radius. However, when referring to the accuracy of the X-ray positions we will mention the 90% confidence radius in order to be consistent with commonly used conventions in X-ray astronomy. The lower limit of / = 3 10-6 is the maximum ratio observed for normal OB stars with the Einstein satellite (Sciortino et al. 1990). As a final criterion we removed 3 stars displaying X-ray luminosities below 1031 erg s-1 since as already argued above, these luminosities may be radiated by the most active late type coronae. The 3 stars (HD 37272, HD 180939 and HD 68518) all have B5V spectral types and / in the range 4-6 10-6.
Table 4. Definition of the three groups of OB/X-ray binary candidates
For comparison, all but one known massive X-ray binaries detected by ROSAT fall into candidate group 1. None appears in group 2 and the only one listed in group 3 is 4U 1700-37 / HD 153919 which consists of a neutron star accreting in the wind of a hot and luminous O6.5Iaf+ star (Walborn 1973; Haberl et al. 1994).
We then searched the literature in order to check for possible errors in the spectral type listed in each SIMBAD object header and more generally with the aim to find evidences for an alternative explanation to the observed X-ray emission such as a referenced late type companion or a white dwarf.
As a final check we analyzed interactively each remaining source with the dedicated Extended Scientific Analysis System (EXSAS) developed at MPE (Zimmermann et al. 1992). Whereas SASS operated on distinct survey strips, EXSAS has the capability to use all photons detected from a given region of the sky (1 1 merged fields in our case) thus yielding improved background determination and source detection. For most sources, count rates, positions and hardness ratios derived from the interactive analysis were fully consistent with those given by the SASS output. We note that some of the positions computed by EXSAS were or more away from the SASS determinations but still compatible within the errors. This slight change of X-ray position together with the use of refined optical coordinates extracted from the Guide Star Catalogue (GSC; Lasker et al. 1990) for 'LS' stars explains the difference between X-ray to optical distances listed in Table 5 and those appearing in individual finding charts and later Tables.
Table 5. The list of candidate OB/X-ray binaries scheduled for optical and X-ray follow-up observations. This table summarizes the preliminary values of spectral type, magnitude and star distance used as input of the automatic process in order to estimate / and un-absorbed . The input information was extracted from the SIMBAD header and results of the SASS analysis of the survey (see Sects. 2.1 and 2.4). When a range of spectral types and luminosity classes was available, we used the average value listed here. d is the distance between the SASS position and that of the associated OB star as read from the SIMBAD header. Errors listed here only arise from the PSPC count rate and do not take into account uncertainties on the detailed spectral type, interstellar absorption and distance. The horizontal lines divide the three groups of candidate stars defined in Table 4
However, in a few instances, the EXSAS process derived count rates significantly lower than those given by SASS. In four cases (CPD -59 2854, LS III +58 47, HD 313343 and LS 4287), the SASS source was hardly or even not recovered at all by EXSAS and therefore dropped from the final list. These discrepancies could be related to the different data analyzed by the two processes. Survey spectra were accumulated for each source and light curves were systematically checked for variability and for the presence of features characteristic of a survey artifact.
The final list of SASS/SIMBAD OB/X-ray candidates containing 24 entries is printed in Table 5.
© European Southern Observatory (ESO) 1997
Online publication: May 26, 1998