Simultaneous ASCA and Hubble Space Telescope/GHRS Observations of Cygnus X‐2/V1341 Cygni¹

Cygnus X‐2 is a low‐mass X‐ray binary (LMXB) with an orbital period of 9.843 days (Cowley, Crampton, & Hutchings 1979). Its optical companion V1341 Cygni is a 0.7 M_⊙ star of 15th magnitude with spectral type varying from A5 to F2. Rossi X‐Ray Timing Explorer (RXTE) observations have shown the presence of a long‐term period between 78 and 82 days in Cyg X‐2, which has since been confirmed with data from earlier missions (Wijnands, Kuulkers, & Smale 1996; Paul, Kitamoto, & Makino 2000).

Cyg X‐2 is a "Z‐source": an LMXB that shows three distinct X‐ray spectral states, referred to as the flaring, normal, and horizontal branches (FB, NB, and HB), which form a Z shape on an X‐ray color‐color diagram. While Cyg X‐2 displays no high‐Q pulsations, each state shows distinct modes of quasi‐periodic oscillations (QPOs) in X‐ray emission. Cyg X‐2 is the only Z‐source that regularly displays all three X‐ray spectral states (FB, NB, and HB) with their associated QPOs and is bright in the ultraviolet (Hasinger et al. 1990; Vrtilek et al. 1990, hereafter Paper I). It is also among the Z‐sources that show significant motion of the Z pattern on the color‐color plane (Wijnands et al. 1997).

Simultaneous ultraviolet and X‐ray observations of LMXBs show a unique relationship between the X‐ray spectral states and the ultraviolet flux. Sco X‐1 and Cyg X‐2, the two brightest Z‐sources, have ultraviolet flux that is directly correlated with the X‐ray spectral state of the source (Paper I; Vrtilek et al. 1991a); there is a factor of 3 increase in ultraviolet flux from the HB to the FB. Most of the optical and ultraviolet luminosity (L_opt∼L_UV∼10^-2L_X) likely comes from the accretion flow as it is illuminated by the X‐rays from the neutron star. The ultraviolet spectra of Cyg X‐2 and Sco X‐1 (Paper I; Vrtilek et al. 1991a) are dominated by strong emission lines of N v λ1240, C iv λ1550, and He ii λ1640. The ultraviolet line strengths, ratios, and profiles vary noticeably (∼20%) on the shortest timescales (∼30 minutes for Sco X‐1 and ∼120 minutes for Cyg X‐2) that can be probed with IUE. There is evidence that the X‐ray lines in Cyg X‐2 vary in the same way as the ultraviolet lines in that both are enhanced during the FB (Vrtilek et al. 1991a, 1991b).

Here we present simultaneous ultraviolet and X‐ray observations of Cyg X‐2 taken with the GHRS on the Hubble Space Telescope (HST) and the GIS and SIS detectors on ASCA. With HST, we can for the first time measure variability of the ultraviolet spectra of LMXBs on short (50 ms ≤  Δt ≤ 30 minutes) timescales. This enables a search for an ultraviolet manifestation of the QPOs observed in X‐rays. Finding QPOs in the ultraviolet would be important: whereas the X‐ray flux can be easily modulated by geometric effects, and emission in the visible can be confused with that of the noncollapsed secondary, the ultraviolet observations are most likely to yield direct evidence of phenomena related to the accretion disk. In addition, the HST spectral resolution ΔV∼19 km s^-1 (HRS G160M) enables the dissection of accretion flows according to velocity, a measurement of local physical conditions and dynamics that the X‐ray observations cannot do. From delays between X‐ray and ultraviolet line variations, the radial structure of velocity in the disk can be analyzed. This provides an unprecedented probe into the structure and dynamics of the accretion flow, enabling us to test models of the ultraviolet emission and constrain properties of the secondary star.

In this paper, we compare the simultaneous HST/GHRS and ASCA data as well as archival IUE data with models predicting ultraviolet continuum emission from the X‐ray–heated disk and star. We interpret the double‐peaked long‐term period of Cyg X‐2 in terms of a saddle‐shaped disk. The ultraviolet spectral features are tested with simple models for line emission from a disk and from an X‐ray–heated corona above the disk. A comparison is made of the ultraviolet spectra of Cyg X‐2 with those of similar binary systems taken with the GHRS. The observations and analysis are presented in § 2, and our interpretation is discussed in § 3.

Figure 1 shows the location of our observation in comparison with the 1 day averaged light curves obtained from the All Sky Monitor (ASM) on RXTE provided by the RXTE/ASM team. A description of RXTE can be found in Levine et al. (1996). Since the ASM data currently available cover a considerably longer time than those used by Wijnands et al. (1996), we redetermined the long‐term period using an analysis of variance (ANOVA) method with F‐statistic as described in Davies (1990, 1991). For the data shown in Figure 1, our most significant period is 82.5 ± 2.5 days; using the entire RXTE data set to date (through 2003 April) we find 81.7 ± 0.6 days; we use the 81.7 day period in this paper. The 82 day cycle has two maxima and two minima; we define phase zero as the center of the deepest minimum. Figure 1c shows the 82 day light curve we determine, and Figure 1b shows the light curve superposed on the data. This indicates that our simultaneous HST and ASCA observations occurred during one of the low states of the 82 day cycle.

Zoom In Zoom Out Reset image size

The ASCA light curves and HST coverage are plotted in Figure 2. The HST/GHRS took eight separate observations on 1995 December 7 from 3:26 UT (JD 2,450,058.643) to 12:49 UT (JD 2,450,059.034), which covers the orbital phases 0.70–0.74. ASCA observed Cyg X‐2 from 1995 December 6 22:15 UT (JD 2,450,058.427) to 1995 December 7 22:01 UT (JD 2,450,059.417) covering the orbital phase range 0.68–0.78. The orbital phases were calculated using an ephemeris of JD 2,443,161.68 for phase 0.0 and a period of 9.843 days (Cowley et al. 1979), where phase 0.0 refers to superior conjunction of the optical star.

Zoom In Zoom Out Reset image size

The first four HST observations were taken with the GHRS G160M (17 km s^-1) grating centered alternately on 1240 Å (N v) and 1640 Å (He ii) in the RAPID mode with a time resolution of 0.5 s (Fig. 3). A description of the GHRS and gratings is given by Cardelli, Savage, & Ebbets (1990). The GHRS observations were taken after the installation of COSTAR and the repairs to the faulty electronics of side 1. The last four observations used the GHRS/G140L (140 km s^-1) in the wavelength region 1220–1550 Å in order to test for ultraviolet continuum and line strength variability (Fig. 4). The G140L was used in RAPID mode with a time resolution of 50 ms to search for the QPOs so far seen only in the X‐rays. Our use of the GHRS in the RAPID mode prevented us from oversampling the spectrum, checking for bad counts, and correcting for the Doppler shift due to the spacecraft orbit. In addition, the continua for the G160M exposures are poorly determined because most of the counts (∼80%) are due to background. Hence, the standard products produced for counts per diode per time interval are not suited for temporal variability studies.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The X‐ray count rates from both the GIS and SIS detectors on ASCA (described in Tanaka, Inoue, & Holt 1994) varied by less than 10% during our observation. Consequently, the X‐ray color‐color diagram covers only a small segment of the Z (Fig. 5). The slope, when compared to the complete Z shown by Hasinger et al. (1990), suggests that we are on the HB, possibly at the turning point into the NB. No significant dips in X‐ray count rate were visible during our observation.

Zoom In Zoom Out Reset image size

Many of our SIS data were taken in FAST mode, which presents problems for spectral analysis. FAST mode complicates the removal of flickering pixels. Also, the response depends on position, and there is no position information in FAST mode. We obtained the FAST mode data for the timing analysis (search for QPOs). Our SIS BRIGHT mode data of Cyg X‐2 suffer from pileup because the count rate is so high. Pileup arises when more than 1 photon is collected in the same pixel (or neighboring pixels) in the SIS CCDs within an integration time (4 s for BRIGHT mode). The result of pileup is that multiple input photons are not detected separately, but a signal is recorded corresponding to the sum of the energies of the photons. Thus, pileup can falsely increase the high‐energy part of the spectrum while decreasing the low‐energy spectrum. (The latter effect is usually negligible, as the counts peak at low energies.)

Our treatment of pileup follows that of Ebisawa et al. (1996). For analysis of the SIS BRIGHT mode spectra, we discard the central circle of 13 pixel radius, as this is the region in which pileup is most severe. Then we construct a model of the high‐energy spectrum produced by piled‐up photons using Ebisawa's algorithm; this model is subtracted from the total counts before fitting. Piled‐up counts contribute 10% at 5 keV and 30% at 8 keV, according to Ebisawa's analysis. With the central 13 pixel radius region removed, the count rate is 86 counts s^-1 with a piled‐up rate of 1.5 counts s^-1 for A1–A3 (A4 had no BRIGHT mode data); without the central region removed, the count rate is 140 counts s^-1 and the piled‐up count rate is 4.6 counts s^-1. Once the model for the pileup is taken into account, the SIS BRIGHT mode spectra agree with the GIS models (detailed below) for E>1 keV. Below 1 keV, pileup effects are too strong to correct.

2.1. Continuum Fits

We used basic parameters for the source as found in Cowley et al. (1979), McClintock et al. (1984) and Paper I: a distance to the source of 8 kpc, an E(B−V) of 0.45, and a 0.7 M_⊙ F star companion with an effective temperature of 7000 K. We assume that the 82 day period is due to a warping and precession of the disk induced by the effects of uneven X‐ray irradiation on the disk. This is a commonly accepted means for producing superorbital periods in systems such as Her X‐1, LMC X‐4, and SMC X‐1 (Clarkson et al. 2003a, 2003b; Cheng, Vrtilek, & Raymond 1995; Vrtilek et al. 1997).

2.1.1. Ultraviolet: HST and IUE

2.1.1.1. Disk Model

We simulate the warp with a saddle shape for the inner disk of Cyg X‐2 (Fig. 6) in order to reproduce the double‐peaked shape of the long‐term X‐ray light curves. In effect, the shape of the disk is an "inversion" of the average X‐ray light curve. To determine the inversion, we loop through 28,800 points (120 points in latitude and 240 points in longitude) on the neutron star surface. This corresponds to a grid coarseness of 0.025r_NS; refining the grid by a factor of 2 in each dimension gives the same result within errors, so this number of grid points should be sufficient. For each point, the program computes analytically whether the line toward the viewer intersects the disk. We follow Howarth & Wilson (1983a, 1983b) in that the coordinates are fixed to the disk so that it is the line of sight that changes with the long‐term phase. This coordinate system allows us to specify the shape of the disk in a frame in which it does not vary over the long‐term phase. The equation of the saddle‐shaped disk is given by z_d = π θ_w(y²_d - x²_d)/180r_d for r_d<1.7r_NS; it is constrained to revert to a simple disk (z_d = 0) in the orbital plane for large radii (r_d≥1.7r_NS).

Zoom In Zoom Out Reset image size

The parameters describing the disk, listed in Table 1, are the radius of maximum disk warp R_mw, the disk thickness θ_d (in degrees), the tilt of the disk from the orbital plane α_d, and the line‐of‐nodes parameter Δψ (this parameter is "invisible" in the sense that it does not count against the χ² but sets the phase origin of the model relative to the observed long‐term X‐ray light curve). Our best‐fit parameters are a maximum saddle height of 1.3r_NS at a distance of 1.7r_NS radius with a tilt of the inner disk from the orbital plane of α_d = 41 ± 04 at long‐term phase 0.5 (Table 1). The tilt of the inner disk from the orbital plane allows for the uneven depths of the light curve: at long‐term phase 0.0, the disk is tilted away from the observer, causing a deeper dip than at long‐term phase 0.5. The X‐ray flux we see, as determined by the amount of X‐rays that are hidden by the saddle‐shaped inner disk, is compared to the RXTE ASM data in Figure 7. This is also the flux seen by the outer disk and the companion star.

Zoom In Zoom Out Reset image size

2.1.1.2. X‐Ray Reprocessing

We adapt our model for determining continuum flux from X‐ray reprocessing on the companion star and disk as described in Paper I in several ways. (1) We replace the spherical shape assumed for the companion star in Paper I with a tidally distorted shape that represents the filled Roche lobe of V1341 Cygni as was done for Her X‐1 by Cheng et al. (1995). (2) Instead of computing the temperature due to X‐ray heating on each of 40 points on the companion star surface, we computed the temperature at 7200 points (120 in longitude and 60 in latitude). (3) The disk is represented in three dimensions by using 1000 annular rings divided into viewing angles determined by a 15 × 29 array for each of eight "quadrants."

Given the shape of the disk and a distorted Roche lobe–filling companion star, we calculate the flux over the range 1150–7427 Å. At each point on the companion star and the disk, a temperature due to X‐ray heating is computed; the contribution to the flux is determined by summing spectra from the library of observed stellar spectra (Canizzo & Kenyon 1987) for a star at the computed temperature.

Predictions of the model from the ultraviolet and optical U and V bands are shown in Figure 8. The U and V optical predictions are compared to published photometry of Cowley et al. (1979) and Chevalier, Bonazzola, & Ilovaisky (1976). The extremes of U and V magnitudes reported are given as dotted lines in Figure 8. We have used archival IUE data for comparison with the ultraviolet predictions. Since the GHRS had a limited continuum coverage (Fig. 4), we integrated over the wavelength regions 1340–1380 and 1420–1480 Å to compare the UV fluxes. For the optical V band, we integrated over the range 5000–6000 Å and for the U band over 3510–3520 Å (although the U band is centered on 3500 Å, the stellar library [Canizzo & Kenyon 1987] that we use to determine fluxes has no values between 3200 and 3500 Å). We then converted the computed fluxes to magnitudes using formulae taken from Lang (1980).

Zoom In Zoom Out Reset image size

Because the 82 day period has a large uncertainty, the long‐term phase of the optical and IUE observations cannot be determined. The predicted U magnitudes fully encompass the range observed; the predicted V magnitudes overlap with, but are on average greater than, the reported extremes. The ultraviolet data—including the current observations and all archival IUE observations—fall within extremes given by a lower limit to M˙ of 3.0 × 10^-9 M_⊙ yr^-1 and an upper limit of 1.8 × 10^-8 M_⊙ yr^-1, consistent with the results of Paper I. There is uncertainty in the flux level for both the model and the data. The HST data are calibrated to only 10% in absolute flux, and there is a difference in resolution between IUE and GHRS. As for the model, uncertainties in reddening and stellar temperature can shift the flux (see the Appendix in Paper I for a discussion of these uncertainties). The relative errors in flux are likely to be much smaller than the uncertainty in overall flux normalization. The best fit to the HST continuum data (shown as a filled triangle) corresponds to an M˙ of 6.0 × 10^-9 M_⊙ yr^-1; this is consistent with a location in the HB (Paper I) (Fig. 4; the thin solid lines represent the models). The M˙ values computed by applying this model to the G160M data (observations H2 and H3; Fig. 3) were somewhat higher. Either we caught the source at the upper bend of the Z where the HB connects with the NB or the very small continuum band available in the G160M is not sufficient to constrain the model properly and the M˙ values are not reliable (here the background is much greater than the continuum).

2.1.2. X‐Ray: ASCA

The X‐ray spectra observed with the ASCA GIS were modeled with the sum of Comptonized bremsstrahlung (following Sunyaev & Titarchuk 1980), a blackbody, and an emission line at ≈1 keV (Table 2). In Figure 9, we show the spectra from the ASCA observations divided into three intervals: A1–A4, A6–A18, and A9–A11; an observation of Cyg X‐2 taken during ASCA's performance verification phase is also depicted (Smale et al. 1994). The region around 2.2 keV is excluded because of an instrumental effect not fully accounted for in the calibration (K. Mukai 1996, private communication). The spectral fits are consistent with interval A6–A8 located on the upper NB, the performance verification phase observation on the HB, and intervals A1–A4 and A9–A11 in a transition from the HB to the NB. Specifically, the blackbody component is reduced in the HB and line emission is enhanced in the NB; we note that a blackbody component is present in Cyg X‐2 at all phases of the Z with the exception of the upper HB (Kuulkers, van der Klis, & van Paradijs 1995). The luminosities implied by the above spectral fits for an assumed distance of 8 kpc are in the range (3–5) × 10³⁷ ergs s^-1, consistent with the HB and the upper NB as inferred by the ultraviolet continuum fits. The reality of the emission features around 1 keV has been questioned by Psaltis, Lamb, & Miller (1995, hereafter PLM95), who are able to fit ASCA data of Cygnus X‐2 with a Comptonized thermal emission model that does not require spectral features. They suggest that features near 1 keV may be due to known problems in Comptonization models. An approximation to the PLM95 model assumes that below a given energy (taken as a free parameter), the spectrum is a blackbody, while above that energy it has a Comptonized shape, with temperature equal to that of the blackbody. We find that fits of these models are not as good as the Comptonized+blackbody+line models; e.g., χ²_ν = 1.42 instead of 0.995 for A1–A4 and χ²_ν = 1.73 instead of 1.22 for A9–A11. The transition energy between the blackbody and Comptonized components occurs at a low energy (0.6 keV). Line features around 1 keV have been reported for Cyg X‐2 by several observers: Vrtilek et al. (1986, 1988) reported lines in Einstein OGS and SSS spectra of Cyg X‐2, attributing a line at 0.96 keV with Fe xx and Ni xx and a line at 1.12 keV with Fe xxiii–xxiv. The presence of these features was confirmed by Kuulkers et al. (1997) using the Low Energy Concentrator Spectrometer aboard BeppoSAX. In addition, Branduardi‐Raymont et al. (1984) using Ariel V Experiment C, Chiapetti et al. (1990) using the EXOSAT CMA, Lum et al. (1992) using the Einstein Focal Plane Crystal Spectrometer, and Smale et al. (1993, 1994) using the Broad Band X‐Ray Telescope and the ASCA SIS all report evidence for excess emission near 1 keV from Cyg X‐2 (cf. Kuulkers et al. 1997).

Zoom In Zoom Out Reset image size

2.2. Ultraviolet Spectral Features

Table 3 gives the line fluxes and velocities for the G160M and G140L observations shown in Figures 3 and 4. The velocities are determined from flux‐weighted wavelengths and are consistent with those observed in optical (He ii λ4686) by Cowley et al. (1979) for the orbital phase of our observations (0.70–0.74). The error in the fluxes includes both counting statistics and uncertainties introduced by choosing background and wavelength boundaries. No significant change is observed during the different observations, which is not surprising since they cover a very small fraction of the binary phase space. The N v doublet is clearly resolved with the doublet ratio close to 1.2:1 rather than the 2:1 expected from thin accretion disks. A similar result was found from GHRS observations of Sco X‐1 (an N v doublet ratio of 1.1:1; Kallman, Boroson, & Vrtilek 1998) and the narrow lines seen in Her X‐1 (ratio 1.2:1; Boroson et al. 1996). The Cyg X‐2 N v lines can also be fitted by including a broad emission component (Fig. 10) such as was observed in Her X‐1, but this is not required. For the C iv line, we were able to get only an upper limit to the line flux, as the line is cut off by the edge of the detector. For this line, the line velocity is found by flux‐weighting the wavelengths only near the peak, not over the entire line profile (where the flux is greater than the background) as for the other lines.

Zoom In Zoom Out Reset image size

In Figure 11, we show a composite of the four GHRS G140L spectra of Cyg X‐2 together with those of Sco X‐1 (Kallman et al. 1998) and Her X‐1 (Boroson et al. 1997). In each case, the solid line depicts the best‐fit X‐ray–heated accretion disk model for the continuum. (The slope in the Her X‐1 continuum spectrum is due to the strong contribution from the star; for Cyg X‐2 and Sco X‐1, the continuum is dominated by X‐ray heating of the disk). While the saddle‐shaped disk provides good fits to the continuum emission, the intensities of many of the spectral features remain unaccounted for by photoionization models, such as the XSTAR code (Kallman & McCray 1982; Kallman & Krolik 1993). For both Sco X‐1 and Her X‐1, the O v line at 1371 Å is surprisingly strong compared to N v, contrary to theoretical expectations (Raymond 1993; Ko & Kallman 1994), whereas Cyg X‐2 showed no O v λ1371 and relatively strong N iv λ1486. Since Sco X‐1 and Cyg X‐2 are both LMXBs of the same type, the differences in their ultraviolet spectra and the similarity of the Sco X‐1 and Her X‐1 ultraviolet spectra are surprising. The N v line in Her X‐1 (not shown) is much stronger than the C iv line, suggesting abundance anomalies (Raymond 1993). Cyg X‐2 and Her X‐1 have similar mass‐accretion rates when Cyg X‐2 is in the HB. The difference in flux can be attributed to reddening and distance effects. The presence of N iv], its strength relative to other lines, and the absence of O v in Cyg X‐2 are consistent with the predictions of Raymond (1993) model F. Since N iv] is an intercombination line that requires somewhat lower densities, it is produced preferentially at the outer edges of the disk where densities are lower. Cyg X‐2 has a larger disk (outer disk radius 3 × 10¹¹ cm; Paper I) than either Her X‐1 (outer disk radius ∼10¹¹ cm; Cheng et al. 1995) or Sco X‐1 (outer disk radius 3 × 10¹⁰ cm; Vrtilek et al. 1991a). Of the 10 models listed by Raymond (1993), none show O v that do not also show N iv] at a comparable or greater strength. Raymond extrapolates from his models that the density will be about ∼ 3 × 10¹³ cm^-3 at a radius of 3 × 10¹¹ cm; the lower density estimates of Paper I probably result from the assumption that the N v emission line is effectively optically thin. Kallman et al. (1998) point out that the O iv λ1340 and O v λ1370 lines are subordinate lines, and so require either that the emitting ions be in an excited state or that excitation occur from the ground state through a dipole‐forbidden transition. The former case requires a combination of high gas density and optical depth in the resonance line leading to the level from which excitation can occur. This can be expressed as nτ≥10¹⁶ cm^-3. Such densities and optical depths are predicted to occur in X‐ray–heated disk atmospheres (Ko & Kallman 1994).

Zoom In Zoom Out Reset image size

As first demonstrated in Paper I, X‐ray heating of a disk and companion star provides good fits to the continuum ultraviolet emission from Cyg X‐2/V1341 Cyg. We have refined the model described in Paper I by using a distorted Roche lobe filling surface for the optical companion. Owing to the low temperature of the companion star, the contribution from the unheated star to the ultraviolet flux is minor: the primary contribution to the UV continuum is X‐ray–heating effects on the disk. We further changed the model in Paper I by distorting the inner edge of the accretion into a saddle shape that reproduces the average X‐ray light curve. The reasons for changing only the inner disk are twofold: the distortion is due to X‐ray–heating effects, and these are maximum in the inner disk; the disk in Cyg X‐2 is rather large (3 × 10¹¹ cm), and even small changes in the outer disk would completely obscure the central X‐ray emission. The M˙ values necessary to fit the simultaneous GHRS/ASCA data we analyze here are well within the range obtained using simultaneous IUE/EXOSAT data of the system (Paper I). We have increased these limits here in order to enclose all IUE observations of the system.

While the highest flux predicted by the model in the V band is greater than the highest V magnitudes reported, the range in magnitude has the same excursion as that observed. The UV flux and U magnitude measurements lie well within the model predictions.

Van Kerkwijk et al. (1998) define a parameter F* that describes where in a disk warping occurs. When 0.1<F^*<0.15, the outer disk is warped; when 0.15<F^*<0.2, the inner disk can be warped also; and when F^*>0.2, the inner disk can be tilted by more than 90° and may behave chaotically; F* depends on the albedo of the disk, and the albedo can change as a function of angle of incidence. If we could prove that only warping of the inner disk provides a solution, then we have a limit on F* and from that a limit on albedo. We have not yet found a way to make the model work by warping only the outer disk.

The line profiles of He ii and N v obtained with the G160M grating do not show the double‐peaked structure that is predicted by simple models of line emission from disks. This suggests that most of the line emission observed is from the surface of the companion. The measured line velocities are consistent with this picture; however, the FWHMs are larger than expected. It is also possible that the line emission is from an accretion heated corona above the disk: the relative strengths of the features observed for Cygnus X‐2 are consistent with the predictions of Raymond (1993); however, the FWHMs are then smaller than expected.

The relatively low resolution GHRS observations are not able to separate many line components that STIS will be able to resolve. For example, if N iv] is preferentially produced at the outer region of the disk and the line widths are due primarily to Keplerian velocities, we would expect it to be narrower than the other lines (the outer disk rotates more slowly) that can also be formed in the inner disk. While there is some variation over the four G140L observations, the average values support this conclusion, with N iv] having an average FWHM of 398 km s^-1, whereas Si iv λ1393 averages to 600 km s^-1 and Si iv λ1402 averages to 459 km s^-1 (this is an ∼1 σ determination, given our velocity resolution of 140 km s^-1).

The HST/STIS extends in several significant ways the ultraviolet capabilities that were available with the GHRS: with the echelle grating, it is possible to sample continuously a broad region (600 Å) of the spectrum at greater spectral resolution than with the GHRS. STIS offers another improvement over the GHRS: a lower background (the STIS dark count rate is 7.0 × 10^-6 counts s^-1 pixel^-1, which is 50–100 times lower than that of the GHRS). For the GHRS, the background was not negligible compared with the source flux; in order of increasing severity, this hampered the line, continuum, and variability studies. STIS observations will allow us to refine these as well as to investigate the physical condition of the emitting gas through line and doublet ratios.

The X‐ray spectra are well fitted with the sum of a Comptonized bremsstrahlung, a blackbody, and an emission complex around 1 keV. We note that if we associate the Comptonized component with the neutron star and the blackbody+lines with emission from the disk, then the motion of the Z shape in the color‐color diagram could be attributed to the varying amounts of each component that are covered by the disk. This can be tested by looking at the source during the high and low states of the cycle: the motion of the Z should be correlated with the long‐term period. The correlation will not be absolute, since the X‐ray components also vary in strength depending on mass accretion rate.