2. Spectroscopy of the pulse-phase averaged data

The source was nearly centered in the A0 pixel. However, because of the 0.8 arcminute offset between the two telescopes the source fell further off axis in the B0 pixel. Thus B0 detected only half as many events as A0, yielding poorer counting statistics. Because systematic calibration uncertainties can affect results from a high quality spectrum more than those from a low quality one, the A0 and B0 data were treated separately. Thus, in the preliminary analysis fits were performed to the A0 data only. The pulse-phase averaged spectrum detected with the A0 pixel is shown in Figure 2.2.

Figure 5: (a) The phase average spectrum of Cen X-3 folded through the telescope response. The A0 data have been fit with an absorbed power law and a Gaussian emission feature at 6.7 keV. (b) The residuals to the fit. Features between 1 and 3 keV are due to instrumental effects. The excesses around 6.4 keV and 7.5 keV may be due to the presence of Fe  K_$\alpha$ and Ni  K_$\alpha$.

The spectra of most X-ray binaries may be described by a power law with an exponential roll off above some cutoff energy E_c (e.g. White et al. (1983)). The cutoff energy is generally in the range 10-20 keV (Nagase et al. (1992)). The present data do not require such a roll off. This indicates that E_c is greater than the upper limit of BBXRT's energy range (12 keV), in agreement with previous observations (Nagase et al. (1992)). The best fit model is an absorbed power law continuum, typical of high mass X-ray binaries, and emission at 6.67$\pm$0.07 keV (modeled using a Gaussian). The centroid of the emission line corresponds to transitions of helium-like iron. The photon index is 1.10$\pm$0.02, the absorption column N_H = (1.28$\pm$0.04) x 10²² cm^-2, and the iron emission line has equivalent width 198$\pm$45 keV and FWHM 740$\pm$210 keV. Fits to the A0 data are summarized in Table 1.

If the continuum is modeled by a simple absorbed power law there are residuals of about 10% between 1 and 3 keV. They lie close in energy to the silicon K and gold M_IV and M_V edges, respectively. Similar residuals are present, but are less significant, in the B0 data because of the poorer counting statistics. It was found that these could be removed from the A0 data by including two more components in the model: Gaussian absorption at 1.5$\pm$0.1 keV and gaussian emission at 2.24$\pm$0.04 keV. These have equivalent widths of 69$\pm$43 and 16$\pm$9 keV respectively. Taken together, these features are significant at the 84% confidence level. When they are included in the model the value obtained for the column density, N_H, is slightly lower, changing from 1.28 to 1.17 x 10²² cm^-2. A similar correction to the BBXRT response was necessary for another bright source, Cyg X-2 (Smale et al. (1993)). Fits to spectra of the Crab obtained with BBXRT have residuals around 1.7 and 2.2 keV due to uncertainties in modeling the Au and Si edges (Weaver et al. (1995)). Owens et al. (1997) have shown that these features in the Crab spectrum are due to X-ray absorption fine structure and are a consequence of the high spectral resolution of BBXRT combined with the brightness of the source. These features in the A0 data were thus interpreted as instrumental effects. It must be emphasized that these residuals only appear in data from bright sources such as the Crab. In the majority of BBXRT observations the counting rates were much lower and these residuals were not detectable. In their analysis of the spectrum of Cyg X-1, Marshall et al. (1993) dealt with these instrumental features by ignoring a narrow band of channels around 1.7 and 2.2 keV. It was decided that the same approach would be appropriate for the Cen X-3 data. Ignoring these channels lowered $\chi^{2}_{}$ significantly and did not appreciably change the continuum parameters. This enabled us to obtain acceptable fits to the A0 data.

Analysis of the B0 data yielded similar results. The iron K emission line energy obtained from the B0 data is slightly lower than that obtained with the A0 data as shown in Table 1. However, the model which was fit to A0 gives an acceptable $\chi^{2}_{}$ when it is fit to the B0 data and only the overall normalization is allowed to vary. The residuals around the silicon and gold edges are not as large as in the A0 data and it is not necessary to remove them to obtain a satisfactory fit. A simultaneous fit to the A0 and B0 data yields an emission line whose FWHM is 0.78$\pm$0.18 keV with a centroid energy of 6.63$\pm$0.06 keV and equivalent width 200$\pm$40 keV.

The presence of residuals around 6.4 keV led us to include a second emission line in the model fit to the A0 data. The best-fit centroid energy for this line is 6.40$\pm$0.03 keV. This would correspond to fluorescence of iron in low ionization stages. Its equivalent width is 42$\pm$18 keV. The best fit value for its physical width is zero with a 90% confidence upper limit for one interesting parameter of 120 keV FWHM which is less than the FWHM energy resolution. Inclusion of this line in the fit pushes the energy of the broad line to 6.77$\pm$0.05 keV and changes its physical and equivalent widths to 420$\pm$180 keV FWHM and 128$\pm$37 keV respectively. The fit is shown in Figure 2.2. A two line simultaneous fit to the A0 and B0 data yields similar results as shown in Table 1. Although the 6.4 keV line is certainly physically plausible, it is only significant at the 97.2% or 2.2$\sigma$ level in the A0 data. A second line is not required to obtain a satisfactory fit to the B0 data. This is consistent with the 50% lower counting rate in the B0 pixel as compared with the A0 pixel. When this model is fit to the B0 data the two lines become degenerate.

Figure 6: (a) Folded A0 data fit with narrow and broad iron lines at 6.4 and 6.7 keV. (b) Residuals to the fit.

Table 1: Spectral parameters of fits to phase average data. All errors are for 90% confidence in one interesting parameter.

Detector	A0	B0	A0 and B0
Model 1: Absorbed power law
NH (1022 cm-2)	1.28$\pm$0.04	1.26$\pm$0.06	1.27$\pm$0.03
$\alpha$	1.10$\pm$0.02	1.06$\pm$0.03	1.09$\pm$0.02
$\chi^{2}_{\nu}$/$\nu$	1.326/393	1.151/384	1.241/778
Model 2: Absorbed Power Law with one emission line
NH (1022 cm-2)	1.13$\pm$0.04	1.29$\pm$0.06	1.30$\pm$0.03
$\alpha$	1.14$\pm$0.03	1.10$\pm$0.04	1.13$\pm$0.02
E (keV)	6.67$\pm$0.07	6.5$\pm$0.1	6.63$\pm$0.06
$\sigma$ (keV)	0.32$\pm$0.09	0.4$\pm$0.2	0.34$\pm$0.08
EW (keV)	198$\pm$45	210$\pm$71	200$\pm$40
$\chi^{2}_{\nu}$/$\nu$	1.086/389	1.049/381	1.071/775
Model 3: Absorbed power law with two emission lines 2
NH (1022 cm-2)	1.31$\pm$0.04	...	1.30$\pm$0.03
$\alpha$	1.13$\pm$0.02	...	1.12$\pm$0.02
E1 (keV)	6.77$\pm$0.05	...	6.73$\pm$0.09
$\sigma_{1}^{}$ (keV)	0.18$\pm$0.07	...	0.25$\pm$0.13
EW1 (keV)	128$\pm$37	...	142$\pm$54
E2 (keV)	6.40$\pm$0.03	...	6.40$\pm$0.03
$\sigma_{2}^{}$ (keV)	< 0.052	...	< 0.055
EW2 (keV)	42$\pm$18	...	36$\pm$23
$\chi^{2}_{\nu}$/$\nu$	1.069/386	...	1.066/770
Model 4: Absorbed power law with narrow lines fixed at
NH (1022 cm-2)	1.31$\pm$0.04	1.28$\pm$0.06	1.30$\pm$0.03
$\alpha$	1.13$\pm$0.02	1.08$\pm$0.03	1.12$\pm$0.01
EW6.4 (keV)	53$\pm$14	54$\pm$20	53$\pm$11
EW6.7 (keV)	60$\pm$15	41$\pm$21	54$\pm$12
EW6.93 (keV)	41$\pm$15	< 38	33$\pm$12
EW7.53 (keV)	26$\pm$17	< 22	17$\pm$14
$\chi^{2}_{\nu}$/$\nu$	1.063/388	1.076/380	1.074/774

There is also a feature at 7.5 keV which corresponds to K_$\alpha$ emission from neutral nickel. It is doubtful that this is astrophysical in origin because there are residuals around 7.5 keV in both central pixels when the Crab spectrum is fit (Weaver et al. (1995)). It is possible that this line is due to fluorescence in the nickel mesh of the detector's blocking filter. When this feature is fit by a narrow line in the A0 data it is significant at the 2$\sigma$ level. It is not required by the B0 data and when it is included in a fit to B0 an equivalent width consistent with zero results.

Fitting a blend of three narrow lines to the iron emission feature instead of a broad and a narrow line yields a $\chi^{2}_{}$ which is not significantly different. Such a fit to the A0 data is shown in Figure 3.4. It includes another narrow line which is significant at the 2$\sigma$ confidence level to fit the suspected nickel feature. The energies of these lines were fixed at 6.4, 6.7, 6.93, and 7.53 keV, corresponding to emission by un-ionized iron, helium- and hydrogen-like iron and un-ionized nickel. Thus the data are consistent with either a broad line or a blend of narrow lines with physically plausible energies. In an attempt to distinguish between the two line and four line models, the energy range was restricted to 5.5-8 keV and the data were fitted with the continuum parameters held fixed. The continuum parameters were those obtained from fits to the 1-12 keV data. F-tests (Bevington (1969)) were still unable to distinguish between the two models.

An attempt was made to improve the counting statistics and uncertainties by relaxing the data quality filter. Fits were made to data in which the only events rejected were those that had the VLE, pulse-pulse, or pixel-pixel flags set. The resulting light curve was flat as the telescope passed through the SAA, indicating that contaminating events were being rejected without significant loss of good data. This is because most of the background events from the SAA were at low energies and had already been rejected by restricting the analysis to energies above 1 keV. The spectral parameters agreed with those obtained from fits to the data which had all flagged events removed. The uncertainties in the derived parameters are slightly smaller with the relaxed quality filter but all of the fits have unacceptably high values of $\chi^{2}_{}$.

Footnotes

... lines²

When this model is fit to the B0 data the two lines become degenerate.