A&A 371, 1056-1064 (2001)
DOI: 10.1051/0004-6361:20010476
I. Negueruela 1,2,3,4 - P. Reig5,6
1 - Observatoire de Strasbourg, 11 rue de l'Université,
67000 Strasbourg, France
2 - SAX SDC, Agenzia Spaziale Italiana, c/o Telespazio, via Corcolle
19, 00131 Rome, Italy
3 - Astrophysics Research Group, Liverpool John Moores University,
Byrom St., Liverpool, L3 3AF, UK
4 - Physics and Astronomy Department, Southampton University,
Southampton, SO17 1BJ, UK
5 - Foundation for Research and Technology-Hellas, 711 10, Heraklion,
Crete, Greece
6 - Physics Department, University of Crete, 710 03 Heraklion, Crete, Greece
Received 8 December 2000 /Accepted 26 March 2001
Abstract
The recent discovery of a
period in the X-ray lightcurve
of the massive X-ray binary 4U 2206+54 has opened the
possibility that it is
a Be/X-ray binary with an unusually close orbit, which, together with
its low intrinsic luminosity, suggests that the system is actually
a Be + WD binary, in which a white dwarf accretes material from
the dense circumstellar disc surrounding a classical Be star. In this
paper we present new X-ray observations and for the first time
high-resolution optical spectroscopy of the source. We show that
both the X-ray behaviour and the characteristics of the optical
counterpart, BD
2790, are more consistent
with a neutron star accreting from the wind of an early-type star. The
X-ray lightcurve shows irregular flaring and no indications of
pulsations, while the very high hydrogen column density supports
accretion from a dense wind. BD
2790 is
shown not to be a classical Be star, as believed until now, but rather
a very peculiar late O-type active star, exhibiting emission
components in the He II lines, complex spectral variability
and strong wind resonance lines in the ultraviolet. Though many of
the characteristics of the spectrum resemble those of the He-rich
stars, the absence of He I variability makes a connection
unlikely. The spectrum is compatible with a composite of two stars of
similar spectral type, though circumstantial evidence points to a
single very peculiar active early-type star. This adds weight to the
growing evidence that the traditional subdivisions of supergiant and
Be/X-ray binaries fail to cover the whole phenomenology of massive
X-ray binaries.
Key words: stars: binaries: general - stars: early-type - X-rays: stars
High Mass X-ray Binaries (HMXBs) are X-ray sources composed of an early-type massive star and an accreting compact object (generally a neutron star, but occasionally a black hole and, at least theoretically, possibly a white dwarf). HMXBs are traditionally divided (see Corbet 1986) into Classical or Supergiant X-ray binaries (SXBs) in which the compact object accretes from the stellar wind (sometimes directly from the atmosphere through localized Roche-lobe overflow) of an OB supergiant and Be/X-ray binaries (BeXBs), in which a neutron star orbits an unevolved OB star surrounded by a dense equatorial disc. Almost all known HMXBs fit well into one of these two categories (with a majority of systems being BeXBs), though a few systems, such as LMC X-4 (Hutchings et al. 1978) or RX J1826.2-1450 (Motch et al. 1997), seem to contain a compact object accreting from a "normal'' main-sequence O-type star.
The hard X-ray source 4U 2206+54 was first detected by
the Uhuru satellite (Giacconi et al. 1972). It appeared
in the Ariel V catalogue as 3A 2206+543 (Warwick et al. 1981).
Steiner et al. (1984; hereafter S84) used the refined position from
the HEAO-1 Scanning Modulation Collimator to identify the optical
counterpart with the early-type star
BD
2790. S84 reported that the
H
line was in emission, showing two distinctly separated peaks
with
.
From their
photometry, they estimated that the counterpart was a B0-2e main sequence
star, and therefore concluded that the system was a Be/X-ray binary.
In this subclass of HMXBs,
the X-ray emission is due to accretion of matter from a Be star
by a compact companion (see Bildsten et al. 1997; Negueruela 1998). The
name "Be star'' is used as a general term describing an early-type luminosity
class III-V star, which at some time has shown emission in the Balmer series
lines (Slettebak 1988, for a review). Both the emission
lines and the characteristic strong infrared excess when compared to
normal stars of the same spectral types are attributed to the presence
of circumstellar material in the shape of a decretion quasi-Keplerian
disc (see Negueruela & Okazaki 2000 for a recent discussion).
Assuming a distance to 4U 2206+54 of 2.5 kpc, S84 calculate
an average luminosity for the source of
erg s-1 between 1974 November and 1981 October.
Saraswat & Apparao (1992, henceforth SA92) presented X-ray
observations of 4U 2206+54 made with the EXOSAT
satellite at different epochs between 1983-1985. The source was always
detected, though in different states. In August 1983 and June 1985, the
source was active, with a low-level
luminosity of
5 1034 ergs-1 and
aperiodic flaring phases (a few hundred seconds long) in which the
overall X-ray flux increased by a factor 3-5 and the X-ray spectrum
changed, becoming harder. In December 1984, the source was in
quiescence, and the X-ray flux was weak
(
ergs-1)
and stable. SA92 also announced the possible detection of a spin period for
the compact object which would be in the range 390-400 s and suggested
that the accreting object was a white dwarf.
The source appears in the ROSAT All Sky Survey (Voges et al. 1999) as
1RXJ220755+543111 and has been consistently detected by
the All Sky Monitor on board RXTE according to the quick-look
results provided by the ASM/RXTE team.
Corbet et al. (2000) have announced the detection
of a
periodicity in the X-ray lightcurve. If
this is the binary period, then it would be the shortest known for a
BeXB - unless the
periodicity in the optical
lightcurve of RX
J0050.7-7316 reflects its orbital period
(Coe & Orosz 2000).
BD
(= LSIII +54
16
= Hilt 1086) is
included in several catalogues of bright stars. Measurements of its optical
magnitudes are reported since the work of Hiltner & Johnson (1956). In spite
of this, very little previous work on this source has been reported. We have
undertaken a major multi-wavelength monitoring campaign on this source, the
results of which will be presented in a subsequent paper. Here
we concentrate only on observations that provide information on the
nature of the system.
![]() |
Figure 1: Background subtracted RossiXTE lightcurves in three energy ranges. The original lightcurves have been rebinned into 8-s bins. The light-curve, clearly dominated by flaring activity, is similar to those of low X-ray luminosity supergiant binaries |
Open with DEXTER |
We have analysed X-ray data taken with the Proportional Counter Array
(PCA) onboard the Rossi X-ray Timing Explorer (RXTE).
The data were
retrieved from the RXTE archive and correspond to an observation made on
March 11-13, 1997. After the screening and filtering of data, i.e.,
ensuring
that all five PCA units were functioning and removing data taken at low
Earth elevation angle (< 10)
and during times of high particle
background, we were left with
9000 s of on-source clean data.
Also in order to improve signal-to-noise we selected only events from the
top layer (the PCUs have three Xenon layers, each consisting of two anode
chains; see Jahoda et al. 1996 for a technical description of the instrument).
Figure 1 shows the background subtracted light curves of
4U 2206+54 in three
different energy ranges. The temporal variability is characterised by
erratic flaring activity on short timescales. The intensity shows changes
by a factor of 3 in less than 2 min. The source becomes increasingly
variable as the energy increases. The
rms of the light curves varies from 40% for the energy range
2.5-5 keV to
45% for 5-10 keV and
50% for
10-30 keV. This flaring
and erratic behaviour was also reported by SA92 during their observations
(which covered the 2-10 keV range) on the two occasions in which
the source was active. Similar lightcurves are observed in SXB
binaries in which a neutron star accretes from the radiative wind of
an evolved star, such as
2S 0114+65 (Yamauchi et al. 1990) or Vela X-1
(Kreykenbohm et al. 1999), which have typical
-
(from now on, low-luminosity SXBs).
![]() |
Figure 2: Hardness ratios as a function of 2.5-30 keV count rate. The X-ray spectrum tends to become harder as the count rate increases |
Open with DEXTER |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
normalization |
![]() |
![]() |
0.9(56) |
The correlation between the hardness ratio 5-10 keV/2.5-10 keV and the count rate (see Fig. 2) indicates that the X-ray spectrum becomes harder during the peak of the flares.
Figure 3 shows the power spectrum of 4U 2206+54. No evidence for the 390-400 s pulse period reported by SA92 was found. The power spectrum is dominated by a strong red noise component and no periodicity is detected at any significative level.
![]() |
Figure 3: Power spectrum for the RXTE/PCA observation of 4U 2206+54 |
Open with DEXTER |
Spectral analysis was performed on Standard2 data, in the energy
range 2.5-30.0 keV. The best-fit model was an absorbed power-law and
a high
energy cutoff yielding an unabsorbed X-ray flux of
.
The best-fit parameters and their
68% confidence
errors are given in Table 1, while
Fig. 4 shows the photon distribution. No evidence
for an iron line at around 6.4 keV was found. This line is seen in the
spectra of the low-luminosity SXBs which display similar X-ray
lightcurves. We can set an upper limit on the equivalent width of such
line at < 10 eV.
![]() |
Figure 4: X-ray spectrum of 4U 2206+54 (circles) and the best-fit model (solid line). The continuum is represented by a power-law plus a cutoff at 7.4 keV (see Table 1). No iron line is required |
Open with DEXTER |
Ultraviolet observations of BD
2790 were
retrieved from the International Ultraviolet Explorer archive at
Rutherford Appleton Laboratory. The
database contains pre-processed spectra, which were subsequently reduced and
analysed using the Starlink packages IUEDR (Giddins et al. 1996)
and DIPSO (Howarth et al. 1997). The low resolution
spectra from the Uniform Low Dispersion Archive (ULDA) LWP18128 and
SWP39111 did not provide enough detail to
allow accurate line identification. The high resolution spectrum
SWP39112, taken with the short-wavelength camera in the large
aperture mode on June 18, 1990, is displayed in
Fig. 5. The original resolution of the spectrum is
approximately 0.05 Å, but it has been rebinned to 0.4
Å for display. The wavelength calibration, which has been checked
with different interstellar lines, is accurate to a few km s-1.
The most remarkable features are the strong P-Cygni profiles of the
resonance wind doublets C IV
1548.2, 1550.8
Å and N V
1239, 1243 Å. They are
stronger than those reported for Be stars of different spectral types
by Prinja (1989) and resemble those typical of O-type main
sequence stars (Walborn et al. 1985). The
subordinate wind line N IV
1718 Å, which generally
follows the behaviour of the resonance doublets (Walborn & Panek 1984a), does not
show clear evidence for a P-Cygni profile, but it
could be masked by the blend of two strong metallic lines just
shortwards of it. The Si IV
1394, 1403 Å doublet shows only a moderate wind effect, with narrow absorption troughs.
![]() |
Figure 5:
The ultraviolet spectrum of
BD
![]() ![]() ![]() ![]() ![]() |
Open with DEXTER |
All the data have been reduced using the Starlink software packages CCDPACK (Draper 1998) and FIGARO (Shortridge et al. 1997) and analysed using DIPSO.
The spectra of BD
2790 in the classification
region do not readily correspond to any spectral type. This fact was
recognised by Hiltner & Johnson (1956), who classified the star as
O9IIIp from photographic plates (apparently not detecting any emission
in the blue at the time). An immediate conclusion of our monitoring
(see Fig. 6) is that the spectrum is also variable. The presence of
strong He II lines, specially He II
4200 Å, would
indicate an O-type classification. Though the spectrum from August 1998
is relatively close to that of a normal O9 star with high
,
most other blue spectra of the source (such as those from July 1995 and
July 2000) display abundant and strong O II lines, together with
a relatively strong Si III triplet. These lines do not correspond
to an O-type star, but are typical of early B-type stars.
From our spectra, we can deduce the following information:
![]() |
Figure 6:
High-resolution blue spectra of BD
![]() |
Open with DEXTER |
The photospheric lines are those characteristic of a late-O/early-B
main-sequence star, with Fe V lines dominating the spectrum
shortwards of 1500 Å. The Al III
1854.7, 1862.8 Å lines, although dominated by the sharp
interstellar features seem to have shallow broader
photospheric components, which
would indicate a spectral type not much earlier than B0. The
photospheric lines seem, in general, to be broader than those of the
standards listed by Walborn et al. (1985).
![]() |
Figure 7:
Red spectrum of BD+
![]() ![]() ![]() ![]() ![]() |
Open with DEXTER |
In a normal star, the condition
He II 4545 Å
Si III
4552 Å
would indicate a spectral type O9.5. The ratio between
He I
4471 Å and He II
4545 Å, which
remains constant (He lines do not change intensity), also
indicates a spectral type O9.5 (using the values from Mathys
1988). On the other hand, the presence of a variable spectral component,
corresponding to a lower effective temperature, cannot be readily
reconciled with the idea of a normal single O-type star. The following
possibilities are open:
The shapes of the H
and He I lines in the red are, at first
sight, typical of a shell star. Shell lines are believed to be formed when
the line of sight to the observer is intercepted by the outer cooler parts of
the envelope of a Be star, which absorb the photospheric continuum
(see Hanuschik 1995, 1996). Therefore Be stars seen very close to edge-on
show deep absorption cores going down below the continuum level on top
of their emission lines and are referred to as shell stars.
The spectra of Be shell stars are characterised by an absorption
spectrum corresponding to a lower temperature (generally displaying
many weak lines corresponding to Fe II and other singly-ionised
metals) superimposed on the photospheric spectrum. To the best of our
knowledge, the spectrum of an Oe shell star
has never been described in the literature. It is therefore possible to
speculate that the envelope surrounding an Oe shell star could produce
the O II and Si III lines, which correspond to a
temperature of
.
This would not explain the emission component almost certainly seen in
He II 4686 Å and would imply that either
BD
2790 is both an Of and an Oe star (the first ever
identified) or that the He II emission originates in an
accretion disc around the compact object. The presence of an accretion
disc in the system, though not directly ruled out by observations, is
difficult to reconcile with the absence of photometric variability.
For example, Hiltner & Johnson (1956) give U=9.40, B=10.11,
V=9.86. Barbier et al. (1973) report U=9.46, B=10.12, V=9.82,
while S84 give
U=9.45, B=10.05, V=9.85, R=9.61, I=9.41 measured with the 0.91-m
telescope at Kitt Peak National Observatory on October 5, 1981. Given the
expected errors (for instance, St84 estimate their errors at 0.04 mag), the
values of UBV are consistent with no variation at all. The Tycho
catalogue gives magnitudes
and
,
which transforms into B=10.13and V=9.87, again compatible with no changes at all.
A more obvious complication for this model is that the metallic lines
do not seem to be any narrower than other presumably photospheric lines.
As a matter of fact, it is difficult to estimate the rotational velocity
of the star. Using the linear relationships for the width of He I
lines developed by Steele et al. (1999) for Be stars observed with the
same configuration, we obtain
from the lines on the
August 1998 spectrum (i.e., that with the weakest metallic spectrum).
This measurement clearly confirms that the star is a fast rotator, but
there is a large
dispersion in the values estimated from different lines.
Another problem faced by this interpretation comes from the asymmetry
of the apparent shell lines. It is well known that in Be stars, symmetric
lines arise from a quasi-Keplerian decretion disc, while asymmetric profiles
correspond to perturbed configurations of such a disc (Hanuschik et al. 1988,
1995, 1996). Whenever asymmetric lines are seen, the ratio between the
Violet and Red peaks (V/R ratio) varies quasi-cyclically, due to the
development of global one-armed density oscillations (Okazaki 2000). In
contrast, the V/R ratio in the H
line of
BD
2790 has been <1 during 15 years, in spite
of the large changes in the overall shape and strength of the line. At
the same time, the symmetry of H
and H I
6678 Å has changed. This behaviour has not been observed in
other Be stars and makes it unlikely that
BD
2790 might be a classical Be star.
Further arguments against the interpretation of
BD
2790 as a classical Be star will be
discussed in a forthcoming paper.
One further complication is the detection of a likely
period. It is difficult to see how an extended envelope could form with such
a close binary companion. The only possibility would be that, due to
tidal truncation by the companion (Negueruela & Okazaki 2001), a very
dense and small shell would form around the Oe star.
One obvious possibility to explain the presence of a lower-temperature spectral component is that the star is in reality an unresolved spectroscopic binary, containing an O9 star and a slightly later companion. If this was the case, the companion should be of spectral type close to B1 in order to display the O II spectrum. Since the UV spectrum is clearly dominated by a main-sequence O-star, this companion cannot have a high luminosity and should be fainter than the O star in the UV and have a comparable magnitude in the B band.
Such a combination could be obtained with, for example, an
O9V((f))+B1III binary. The O9V star, due to its Of nature, would contribute
the emission in He II and an asymmetric component to H.
The
B1 giant could be a Be star and contribute the Balmer and He I
line emission. It is clear that such configuration could never
produce the observed X-ray luminosity because of colliding winds, since
both stars would have relatively weak winds and even binaries with
very strong winds produce lower luminosities. Moreover, the X-ray spectrum
observed is very different from those of colliding wind systems, which
are generally soft (with little signal above 10 KeV) and interpreted as
coming from hot,
optically thin plasma with fixed or variable solar element abundances
(Skinner et al. 1998; Stevens et al. 1996). None of such models or a
combination of them (MEKAL, VMEKAL, RAYMOND in XSPEC terminology)
gave acceptable fits to the X-ray spectrum of 4U 2206+54.
Therefore, the presence of a compact object is necessary. Given the
period, the compact object would be orbiting
one of the components (presumably the Of star, since the Be star would
not have a wind that could explain the observed X-ray lightcurve) in
a close orbit and accreting from its wind. The other component
would then be in a much wider orbit.
Abt & Bautz (1963) found no evidence for binarity in
BD
2790 in their study of radial velocities of
early-type stars. The reported value
is typical for a
member of the Perseus arm. However, this does not rule out the binary
model, since the system could be very wide. Accurate radial velocity
measurements of the weak metallic lines in order to check whether they
are consistent with the strongest lines, should be a test of this
hypothesis.
The spectral variability of BD
2790 resembles
in many aspects that of He-rich stars, though these objects have later
spectral types (clustered tightly around B2). Some of these systems
are known to display spectral
variability in their Si III lines in antiphase with their He I
lines. These variations are strictly cyclical and correspond to
the rotational period of the star, which is
(Walborn 1982). Some of the He-rich
stars (allegedly, those which are fast rotators) display H
in
emission, but the line profile does not look shell-like (Zboril et al. 1997).
Only the peculiar He-rich star
Ori E displays an
H
line profile similar to that of
BD
2790, but in this star the V/R ratio varies
cyclically with the same period as the He I lines. In any case,
BD
2790 is unlikely to be related to He-rich
stars (say, as an early-spectral type relative), because all the
values of EWs measured for He I lines in
BD
2790 are compatible with absolutely no
changes in their strength or the He I/H I ratio.
In spite of this, the possibility that the peculiar changes seen in the
spectrum of BD
2790 can be attributed to some
unknown physical mechanism operating in a single star remains.
Whatever the model adopted for BD
2790, it is
clear that its ultraviolet and blue spectrum is dominated by a star of
approximate spectral type O9.5V. For such a star, the intrinsic colour
would be close to
(B-V)0 = -0.3. Then, using the value from St84,
(B-V)=0.2, we have
and, assuming standard reddening,
.
An O9.5V star has an absolute magnitude
MV = -4.3 (Vacca et al. 1996)
and therefore the measured V=9.85 implies
kpc.
At this distance, the average X-ray luminosity of 4U 2206+54 is
.
The observational history of 4U 2206+54, which
has been detected by all satellites that
have pointed at it and has never been observed to undergo an
outburst, is notably different from that of Be/X-ray transients,
such as 4U0115+63 or A0535+26 (see Negueruela & Okazaki 2000).
There is a second
subclass of Be/X-ray binaries characterised by low-luminosity,
persistent X-ray emission with little variation, e.g., X
Persei (Haberl et al. 1998), but they are believed to have large orbital
periods (Reig & Roche 1999) - which is certainly the case for X
Persei (Delgado-Marti et al. 2001). The 9.5-d period of 4U 2206+54 makes a
connection unlikely, though the low X-ray luminosity Be/X-ray binary
3A 0726-26 (
)
could have a
short period
(Corbet & Peele 1997). Moreover, the
decrease in
by more than one order of magnitude
reported by SA92 is also atypical for these systems.
The X-ray luminosity and behaviour, with short erratic flares, point then
to wind accretion as the mechanism producing the X-rays, if the compact
object is a neutron star. The strong wind
profiles seen in the UV spectrum indicate a large mass-loss rate which
would fuel the X-ray system. Because of this, we can expect a
similarity with low-luminosity SXBs, at least as far as the
accretion process is concerned. Wind accreting supergiants with
orbital periods similar to 4U 2206+54 generally show
rather higher luminosities. This is certainly the case of
Vela X-1 (
)
with
(Kreykenbohm et al. 1999) and likely 2S 0114+65 (
)
if the large distance estimates are correct (e.g., Reig et al. 1996). Such differences would be due to the
much weaker wind of the Of star compared to a supergiant.
In this respect, it must be noted that the absorption column to
4U 2206+54 derived from our X-ray spectral fitting
atomscm-2 (not very different
from the values found by SA92) is much larger than that corresponding to the
interstellar absorption. The standard relation from Bohlin et al. (1978)
indicates that
translates into
atomscm-2, i.e., one order
of magnitude less than observed. This would indicate the presence
of very optically thick material in the vicinity of the
compact object, though it is not clear how this interpretation
can be reconciled with
the absence of an iron line.
It is also worth mentioning that the X-ray source RX J1826.2-1450, whose optical counterpart is also a main-sequence O star, could harbour a black hole, since it contains a microquasar (Paredes et al. 2000), and has an X-ray luminosity similar to or lower than 4U 2206+54. Therefore we cannot rule out the possibility of a black hole companion in 4U 2206+54, though the presence of a high-energy cutoff in the X-ray spectrum (which is typical of X-ray pulsars) favours a neutron star companion.
Even though we have no explanation for the spectral changes shown by
BD
2790, a very close similarity to
2S 0114+65 is suggested.
The optical counterpart to this system, V662 Cas, apparently
is a normal B1 supergiant (Reig et al. 1996), in spite of the fact
that some authors have claimed that the strength of the Balmer
lines is not as high as expected for such a star. However, van Kerkwijk
& Waters (1989) report an instance of spectral change in this source that
occurred on November 4th, 1986. On this occasion, the complete metallic
spectrum (i.e., O II + Si III) of the source - which
is typical of a B1 supergiant - disappeared,
leaving behind what looked like a normal B2-3III spectrum.
It is interesting that the same set of lines that are variable in
BD
2790 also varied in V662 Cas,
though in the latter case, their
disappearance seems to leave behind a cooler stellar spectrum. We take
this as a suggestion that some stars in binaries with close compact
object companions may be structurally unstable, perhaps due to
their previous history, though at the moment we are unable to
propose any physical mechanism for this variability.
Moreover Guarnieri et al. (1991) and Minarini et al. (1994) report the
occurrence of optical outbursts in both BD
2790
and V662 Cas,
during which some lines which are generally not seen or in absorption
go strongly into emission, and argue that these events suggest that both
stars are Be stars. The classification spectra of both objects show that
they are not classical Be stars, after all, but such episodes point to
a further connection between the two systems. In this respect, it must
be noted that Hall et al. (2000) have presented evidence strongly
suggesting that the 2.7-h periodicity observed in the X-ray lightcurve
of 2S 0114+65
should correspond to the spin period of a very slowly rotating neutron star.
If 4U 2206+54 exhibits periodic behaviour on a similar
timescale, it is very unlikely that our observations could have
detected it.
Whichever of the scenarios proposed turns out to be closer to reality,
the X-ray emission from 4U 2206+54 seems certain to be due to
direct accretion from the wind of an active O-type star. This adds
to increasing evidence that the traditional divisions of Supergiant
X-ray binaries and Be/X-ray binaries are not enough to describe the
whole set of Massive X-ray binaries. While the case of some objects, like
2S 0114+65 and 4U 1907+09, which seem to share
characteristics of both groups, had always been shown to be problematic, it
is clear now that there are a variety of objects, such as
RXJ1826.2-1450,
RX
J0050.7-7316 and 4U 2206+54, which simply
do not belong to any of those categories.
Acknowledgements
The INT is operated on the island of La Palma by the Royal Greenwich Observatory in the Spanish Observatorio del Roque de Los Muchachos of the Instituto de Astrofísica de Canarias. The G. D. Cassini telescope is operated at the Loiano Observatory by the Osservatorio Astronomico di Bologna. This research has made use of data obtained through the High Energy Astrophysics Science Archive Research Center Online Service, provided by the NASA/Goddard Space Flight Center and of the Simbad data base, operated at CDS, Strasbourg, France.
We would like to thank the referee, Frank Haberl, for his helpful comments. IN would like to thank Prof. Nolan Walborn for useful discussions on the spectral classification and Dr. Manfred Pakull for valuable comments on the manuscript. The August 1998 spectrum was taken by Dr. I. A. Steele. During part of this work IN was supported by a PPARC fellowship and later by an ESA external fellowship. PR acknowledges support from the European Union through the Training and Mobility of Researchers Network Grant ERBFMRX/CT98/0195.