Origins of X-Ray Line Emissions in Circinus X-1 at Very Low X-Ray Flux

The coadded spectrum in Figure 1 shows very strong H- and He-like lines for Mg, Si, and S with line counts between 150 and 320 cts line⁻¹. This is large enough to allow the determination of critical line parameters such as line shifts, widths, and flux. For this analysis we use Gaussian line functions. The H-like lines and the Fe K line are single Gaussians. We do not account for the spin–orbit split in the H-like lines, which produces a separation of the ${\alpha }_{1}$ and ${\alpha }_{2}$ lines by 0.0056 Å, and we use the average line location as was done in Schulz & Brandt (2002) and Schulz et al. (2008). This allows us to better compare the line measurements. As a consequence, we may slightly overestimate the line width in H-like lines. The He-like lines are triplets containing a resonance (r), an intercombination (i), and a forbidden (f) line with fixed line spacings. Here we froze the line spacings of the i and f lines relative to the r lines. We then divided the bandpass into five regions containing the Mg, Si, S, Ca + Ar, and Fe lines.

3.1.1. Line Fluxes

The results of the line fits to the periastron data in Table 1 are shown in Table 2. The first column lists the K-shell ion, the second column gives the theoretical line location as in Schulz et al. (2008), the third column is the measured wavelengths, the fourth column gives line fluxes in units of 10⁻⁵ photons s⁻¹ cm⁻², and the fifth column lists the sigma of line widths in km s⁻¹ (3 $\times {10}^{5}({\sigma }_{\mathrm{meas}}/{\lambda }_{\mathrm{meas}}$)). There are distinct differences in the line fluxes with respect to previous observations when the source was brighter. Figure 3 shows that the bulk of the H-like line fluxes were of the order of 20% of the ones reported in Schulz et al. (2008), while the He-like lines are up to 70% of the previously reported fluxes. Calcium lines are the exception as they were generally weak in observations III and IV (Schulz et al. 2008). The lines in the Fe K line region follow this picture: the H-like Fe xxvi is very weak and the region is dominated by the Fe xxv triplet. However, the Fe K line (Fe i–x) is very similar in flux and width to the one we observed in observation III. The line widths cluster around 550 km s⁻¹, which is very similar to previous detections. The line widths at shorter wavelengths appear higher, which is likely a consequence of the increasingly lower spectral resolution at shorter wavelengths. The size of an HETG resolution element of a dispersed grating order is constant in wavelength space, and thus resolution element sizes increase in velocity toward higher energies.

Zoom In Zoom Out Reset image size

Table 2.  X-Ray Line Properties from the Periastron Observations in Table 1

Ion	${\lambda }_{o}(1)$	${\lambda }_{\mathrm{maes}}$	Line Flux	σ
	(Å)	(Å)	(10−5 ph s−1 cm−2)	(km s−1)
Fe xxvi	1.780	1.782 ± 0.010	0.92 ± 0.99	<3300
Fe xxvr	1.850	1.845 ± 0.008	3.84 ± 2.23	1000 ± 820
Fe xxvi	1.859	1.858 ± 0.008	0.10 ± 8,38	1000 ± 820
Fe xxvf	1.868	1.870 ± 0.005	3.25 ± 3.60	1000 ± 820
Fe K	1.940	1.933 ± 0.001	2.49 ± 0.85	390 ± 430
Ca xx Lα	3.021	3.021 ± 0.008	0.59 ± 0.44	770 ± 640
Ca xixr	3.177	3.171 ± 0.007	0.30 ± 0.36	750 ± 510
Ca xixi	3.189	3.183 ± 0.007	0.74 ± 0.55	750 ± 510
Ca xixf	3.211	3.205 ± 0.007	< 0.1	750 ± 510
Ar xviii Lα	3.734	3.736 ± 0.003	0.88 ± 0.35	500 ± 460
Ar xviir	3.949	3.946 ± 0.004	1.19 ± 0.39	600 ± 310
Ar xviii	3.966	3.963 ± 0.004	0.14 ± 0.50	600 ± 310
Ar xviif	3.994	3.991 ± 0.004	1.32 ± 0.49	600 ± 310
S xvi Lα	4.730	4.722 ± 0.002	2.95 ± 0.55	450 ± 140
S xvr	5.039	5.036 ± 0.003	2.45 ± 0.54	470 ± 130
S xvi	5.067	5.064 ± 0.006	0.97 ± 0.44	470 ± 130
S xvf	5.102	5.093 ± 0.003	1.32 ± 0.43	470 ± 130
Si xiv Lα	6.183	6.175 ± 0.001	2.64 ± 0.28	420 ± 70
Si xiiir	6.650	6.640 ± 0.003	1.32 ± 0.24	500 ± 80
Si xiiii	6.669	6.678 ± 0.003	0.63 ± 0.34	500 ± 80
Si xiiif	6.740	6.734 ± 0.002	1.40 ± 0.24	500 ± 80
Mg xii Lα	8.422	8.409 ± 0.004	0.94 ± 0.18	530 ± 140
Mg xi(r)	9.169	9.179 ± 0.015	0.25 ± 0.14	< 480
Mg xi(i)	9.230	9.228 ± 0.012	< 0.1	< 480
Mg xi(f)	9.314	9.315 ± 0.013	0.22 ± 0.15	< 480

Note. (1) See Schulz et al. (2008).

Download table as:  ASCIITypeset image

3.1.2. Line Centroids

Figure 4 shows the line centroids with the expected rest positions in units of km s⁻¹. The filled black circles show the periastron observations from Table 1. The majority of the brightest lines concerning Mg, Si, and S clearly show a line shift to the blue of about 400 km s⁻¹. Higher Z element lines (Ar, Ca, and Fe) are fainter and uncertainties are larger.

Zoom In Zoom Out Reset image size

For comparison, we plot the results from observation IV from Schulz et al. (2008) for reference (squares). For this observation, we had to refit the He-like triplets (Fe, Ca, Ar, S, and Si) because in that previous analysis the widths and line spacings were free parameters and not tied according to their triplet properties. In some cases, specifically for the Si, Ar, and Ca triplets, this produced line centroids that are much more consistent with the rest of the sample. The line centroids from observation IV are consistent with the expected rest wavelengths.

The Si xiv ${L}_{\alpha }$ at the expected wavelengths is by far the most significant one in terms of total counts (>300 cts), and here we can look at the individual observations. The brightest contribution comes from observation VIIc (∼170 cts) and the faintest from observation V (∼35 cts), enough to allow centroid studies. Figure 5 shows the line counts of the unbinned data. The black line marks the expected rest wavelength at 6.183 Å, and the dashed line gives the fitted location of the coadded line data. It clearly shows that the blueshift is present in all observation segments and appears persistent over many years. The lines of observations V and VIIa,b are symmetric around the fitted centroid (dashed line).

Zoom In Zoom Out Reset image size

3.1.3. G and R Ratios

Figure 1 features strong He-like lines with especially strong r lines. The ionization parameter is defined as

$$\begin{eqnarray}&&\xi ={L}_{x}/({{nd}}^{2}),\end{eqnarray} \tag{ 1 }$$

where L_x is the source luminosity, n the plasma density, and d the distance from the X-ray source. A multiple plasma environment requires a range of ionization parameters to engage in a fit. This complicates the modeling process in terms of fitting time and parameter range. In that light, we try to limit the number of fit components to a very minimum. For the choice of continuum, we modeled a few cases including combinations of power laws as reported by Schulz et al. (2008) and Iaria et al. (2008), as well as blackbody spectra. For the fit itself, we use XSTAR's photemis function available in Xspec, which allows for a preset atomic-level population file to calculate a photoionized spectrum for a single ionization parameter. For the final fit setup, we used two different functions to account for two ionization parameters. Furthermore, in order to be able to fit the H- and He-like line morphology, the functions also needed individual absorption columns, for which we applied pcfabs functions. For the fits, we fixed the interstellar column to $1.8\times {10}^{22}$ cm⁻² as reported by Heinz et al. (2013). The fits generally converged to a one-component solution for the continuum, that is, either a power-law or a blackbody spectrum. The power-law solution was ruled out because it completely overshot the observed continuum above 2 Å for the Fe K line region, whereas the 1.6 keV blackbody provided the necessary steep decline above 2 Å (see Figure 6). The fit result is summarized in Table 3. The uncertainties are 90% confidence limits calculated by conf_loop in ISIS. In the fit, we also set the turbulent velocities to 600 km s⁻¹, and in the final stages we fixed some of the abundances we let float. The redshifts were also fixed to −400 km s⁻¹ (see dashed line in Figure 4) in one of the photemis components and allowed to float in the second. We need at least two ionization parameters because we observe Mg and Si lines as well as Ca and Fe lines, which cannot be modeled with a single ionization parameter (Kallman et al. 2004). The Mg abundance was slightly reduced to 0.93, which likely is a consequence of the statistically challenging Mg xi bandpass. The Si abundance was set to 1.73 and the S abundance to 1.68 for both photemis components.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 3.  Photoionization Fit Results

Parameter	Symbol	Units
Absorption column density	NH	1022 cm−2	1.8
Blackbody temperature	kT	keV	1.57 ± 0.04
Blackbody normalization	${\left({R}_{\mathrm{km}}/{D}_{10\mathrm{kpc}}\right)}^{2}$	⋯	0.392 ± 0.026
Pcf absorption column 1	NH1	1022 cm−2	2.4 ± 0.4
Photoionization normalization 1	K1	erg cm−1 s−1	121.5 ± 12.7
Ionization parameter 1	${\xi }_{1}$	erg cm s−1	340 ± 135
Redshift 1	z1	⋯	−0.0013334
Pcf absorption column 2	NH2	1022 cm−2	2.8 ± 1.1
Photoionization normalization 2	K2	erg cm−1 s−1	100 ± 65
Ionization parameter 2	${\xi }_{2}$	erg cm s−1	1500 ± 180
Redshift 2	z2	⋯	−0.00062
Fit statistic	Cash/dof	⋯	1.42

Download table as:  ASCIITypeset image

The photoionization model fit result is shown in Figure 7. The final ionization parameters obtained were ξ = 340 and 1500 erg cm s⁻¹, and both are associated with high column densities. The covering fractions were free parameters but resulted in 0.99 in both cases. The plasma with ionization parameter 340 erg cm s⁻¹ is responsible for the bulk of the line fluxes, while the one with ionization parameter 1500 erg cm s⁻¹ contributes to some of the H-like line fluxes but mostly to the Fe region.

The normalization of the photoionization model is defined by

$$\begin{eqnarray}&&K=\displaystyle \frac{\mathrm{EM}}{4\pi {D}^{2}}\times {10}^{-10},\end{eqnarray} \tag{ 2 }$$

where EM is the emission measure of the gas in units of cm⁻³ at the involved ionization parameter, and D is the distance to Cir X-1 in units of cm. For a distance of 9.4 kpc (Heinz et al. 2015), this leads to

$$\begin{eqnarray}&&\mathrm{EM}=K\times 1.06\times {10}^{56}.\end{eqnarray} \tag{ 3 }$$

For the plasma with log ${\xi }_{1}$ = 340 erg cm s⁻¹, we obtain EM${}_{1}=1.3\times {10}^{58}$ cm⁻³, and for log ${\xi }_{2}$=1500 erg cm s⁻¹ we obtain EM${}_{2}=1.1\times {10}^{57}$ cm⁻³. These emission measures are only slightly smaller than the one estimated by Schulz et al. (2008); the observed source luminosity, however, appears to be significantly lower.

This observed low X-ray luminosity from the continuum is inconsistent with the observed photoionized luminosity from the X-ray line analysis. One of the difficulties is maintaining the observed ionization parameters at such a low luminosity over such a large volume. While to some extent one could offset a larger source distance d with a lower density in the wind n, this eventually breaks down because the observed photoionized luminosity appears close to the actual source luminosity. Such a high ionization efficiency is nearly impossible to achieve under likely any circumstances. This points in the direction that part of the X-ray source is still obscured, something we encountered during the analysis of observations I and II (Schulz et al. 2008). To simply assume that some of the flux of our observed continuum is partially blocked and should be higher is possible but not a good solution because a 1.6 keV blackbody is inherently inefficient at ionizing high-Z atoms such as Fe and Ca at any possible luminosity. A more promising scenario relates back to our actual findings in Schulz et al. (2008). There we found that the observed photoionized lines are due to an accretion disk corona (ADC), and in order to sustain such a corona, the source X-ray luminosity needed to be significantly higher than observed, which leads to the conclusion that parts of the central X-ray source are obscured. This may still be the case. Figure 8 shows a recalculation of the content of Figure 5 in Schulz et al. (2008), including lower luminosity predictions and the locations of our photoionized data points in this picture. This indicates that for our results the source still is at a luminosity close to 10³⁷ erg s⁻¹, which allows for sustaining a static ADC. This then leaves the possibility that we do observe some highly absorbed and obscured parts of the ADC in the high-Z lines and the photoionized wind of the companion seen in the blueshifted lower-Z lines.

Zoom In Zoom Out Reset image size

The ionization parameter and emission measures then leave us with a wide range of viable plasma regimes. Two regimes that are opposite in character are of interest in the case of Cir X-1. One refers to the ADC emissions as described in Schulz et al. (2008), which are confined to the domain of the accretion disk. In that case, we can assume that d, the distance to the X-ray source, is of the same order of magnitude as r, the size scale of the photoemitting region. For distances of up to 10¹⁰ to 10¹¹ cm from the neutron star surface, we obtain ADC plasma densities between about 10¹² and 10¹⁴ g cm⁻³ in a compact volume within the accretion disk (see also Jimenez-Garate et al. 2002).

The observed blueshifts point to another regime. For very low densities as exist in weak stellar winds, that is, densities g cm⁻³, we also get viable solutions for values of d and r that are beyond the dimensions of the accretion disk. Weak, low-density winds are expected in mid-B-type stars. For such low plasma densities, the dimensional parameters are larger than 10¹² cm, consistent with ionized line emissions in a weak stellar wind.

The XSTAR:photemis function applies a model describing a purely photoionized plasma. The fact that this model cannot properly account for the strength of the resonance lines in some of the He-like triplets confirms our findings from the G ratios that there is a significant contribution from some recombination and resonance scattering. Future modeling of these effects should reveal the presence of such a scattering medium.

The final continuum spectrum in the photoionization fit turned out to be a 1.6 keV blackbody. We had fixed the interstellar column to 1.8 × 10²² cm⁻², which was found by Heinz et al. (2013) in the fit to the supernova remnant, consistent with the large amount of visible extinction toward Cir X-1, as well as a well-established distance of 9.4 kpc (Heinz et al. 2015). Such a distance is also consistent with Cir X-1 radiating at Eddington peak fluxes when it was brightest (Jonker & Nelemans 2004) and when it exhibited P Cygni X-ray lines from a radiation-driven wind (Brandt & Schulz 2000). The property of the blackbody fit is rather peculiar as it exhibits a very small emission radius. It is also notable that there is no or only very little contribution of the accretion disk to the observed X-ray spectrum. The only clear signature from the disk may come from the Fe K fluorescence line observed at 1.93 Å.

The big problem arises when we realize that our observed continuum is quite insufficient to photoionize the plasma at the level observed. This gives us reason to assume that emissions from the accretion disk are there but are highly absorbed and obscured. This provides quite some uncertainty to the observed luminosity and nature of the X-ray spectral continuum as we have to consider that the bulk of the emission is partly blocked and obscured.

Accreting neutron stars with magnetic fields significantly lower than 10¹⁰ Gauss are generally seen in low-mass X-ray binaries (LMXBs), which are considered to be older systems in which the original field had enough time to decay to such low values. With only a very few exceptions where we do not know the field through cyclotron lines, accreting neutron stars with high-mass companions have all high magnetic fields as they are considered very young. However, we point out that the question of the companion nature is fairly irrelevant here. Homan et al. (2010) showed that transient sources can morph through atoll and Z stages at various flux phases, making these spectral variability imprints more related to effective Roche-lobe overflow accretion rather than binary types. In the following, we discuss how the observed continuum relates to cases of low and high magnetic fields for the accreting neutron star.

4.1.1. Low-field Case

4.1.2. High-field Case

However, what is very common is the fact that young neutron stars, accreting or isolated, have high ($\gg {10}^{10}$ G) magnetic fields (Reig 2011; Özel & Freire 2016). In this case, we consider the possibility that this observed blackbody continuum does not contribute much to the photoionization process at all and the photoionizing continuum is entirely obscured. In a high-field accretion scenario, matter is predominantly funneled by the field onto poles of the neutron star, and the state of the accretion column depends on X-ray luminosity (see Becker & Wolff 2007 for an in-depth review). At a critical luminosity of ${L}_{\mathrm{crit}}=1.5\times {10}^{37}$ erg s⁻¹, the accretion flow becomes radiation pressure dominated, and X-ray emissions are defined by a complex mix of physical processes, resulting in something close to a power law with an exponential cutoff or similar as observed when the source was brighter (Schulz et al. 2008). However, in the subcritical case where the luminosity is erg s⁻¹, as is observed in the low state of Cir X-1, the emissions from the bare hot spot can be explained by such a hard blackbody.

In such a subcritical environment, we can expect that the X-rays are produced by impact on or very close to the neutron star surface. If this is the accretion hot spot, its radius r₀ should fulfill the condition (Lamb et al. 1973)

$$\begin{eqnarray}&&{r}_{0}\lt {r}_{\mathrm{ns}}{\left[\displaystyle \frac{{r}_{\mathrm{ns}}}{{r}_{A}}\right]}^{1/2},\end{eqnarray} \tag{ 4 }$$

where r_A is the Alfvén radius given by Becker & Wolff (2007) as

$$\begin{eqnarray}&&{r}_{{\rm{A}}}=2.6\times {10}^{8}{B}^{4/7}{r}_{\mathrm{ns}}^{10/7}{M}_{\mathrm{ns}}^{1/7}{L}_{x}^{-2/7},\end{eqnarray} \tag{ 5 }$$

where B is the magnetic field in units of 10¹² G, r_ns is the neutron star radius in units of 10 km, M_ns is the neutron star mass in units of M_⊙, and the X-ray luminosity L_x is in units of 10³⁷ erg s⁻¹. Figure 9 shows the evaluation of Equations (4) and (5) for two different neutron star radii, with 10 km (black) representing the lower end of most equations of state and 15 km (red) representing the higher end of most equations of state (Özel & Freire 2016). Equation (5) has only a weak dependence on neutron star mass, and here we used a canonical value of 1.4 M_⊙. For L_x we applied the measured X-ray luminosity in Table 2. For a distance of 9.4 kpc (Heinz et al. 2015), we can determine limits to the blackbody emission radius depending on the blackbody fits provided in Table 2. This then amounts to an r_bb = r_min = 0.512 km and an r_bb = r_max = 0.597 km. At a distance of 9.4 kpc, the observed emission radius in Table 3 is 0.589 ± 0.151 km, consistent with these values. A relativistic color correction would do little to the range of these values. From that we find a lower limit to the magnetic field strength of $4.0\times {10}^{10}$ G for a 15 km radius and $2.8\times {10}^{11}$ G for a 10 km radius. Note that these are only lower limits, and the actual field can still be higher than these values.

Zoom In Zoom Out Reset image size

A survey of about 20 confirmed cyclotron line energies in high-mass X-ray binaries (HMXBs; Caballero & Wilms 2012) shows that all but one magnetic surface field strengths are above 10¹² G, except for Swift J1626.6-5156 (DeCesar et al. 2009), which has a field slightly below 10¹² G. The result in Figure 9 of course depends on distance, while a cyclotron line measurement does not (Coburn et al. 2006). However, we deem a distance of 9.4 kpc quite robust. In order to have both the 10 and 15 km solutions above 10¹² G, the distance has to be below 6 kpc, which is implausible because of the amount of extinction along the line of sight, and it would take the X-ray luminosity in observations I and II (Brandt & Schulz 2000; Schulz & Brandt 2002) too far away from Eddington conditions to produce such a powerful disk wind. Higher distances do not change the result much unless they become unrealistically high. In order for the field values to drop below 10¹⁰ G, the blackbody radius has to be higher than 1 km and the source would reside outside the Milky Way. We therefore are confident that the surface magnetic field in Cir X-1 is somewhere between the values shown in Figure 9.

The moderately high $\gg {10}^{10}$ G magnetic field should still be consistent with a young system even though it is generally assumed that neutron stars are born with $\gt {10}^{12}$ G (see Kaspi 2010 and references therein). The neutron star in Kes 79 (Halpern & Gotthelf 2010) with a magnetic field of only $3.1\times {10}^{10}$ G is a good example for a younger neutron star with a moderate magnetic field (see also Shabaltas & Lai 2012). Other examples are PSR J1852 + 0040 and 1E1207.4-5209 (Halpern et al. 2007; Gotthelf & Halpern 2007), as well as PSR J0821-4300 in Puppis A (Gotthelf & Halpern 2009), even though these may not be quite as young as Cir X-1. Thus, young neutron stars with lower magnetic fields of the order of 10¹¹ G, also sometimes dubbed "antimagnetars," are not unusual anymore.

The observation of type I X-ray thermonuclear bursts (Tennant et al. 1986; Linares et al. 2010) also implies a field of G (Fujimoto et al. 1981). However, this criteria is more based on the lack of observations of type I bursts in X-ray pulsars rather than having a solid theoretical basis other than the suppression of convective motions to allow runaway burning in magnetic fields of about 10¹² G and higher (Gough & Tayler 1966; Bildsten 1995). Bildsten (1998) outlined the conditions for neutron star nuclear burning for the case of high mass accretion rates ($\dot{M}$ $\gt \,{10}^{10}$ M_⊙ yr⁻¹). At these rates, unstable nuclear burning is more easily realized. The luminosity of Cir X-1 during its outburst in 2010 was still below the stable burning criterion, and here the magnetic field may have been high enough to confine the accreted matter to the ignition pressure, but still low enough to allow convective motions. We therefore argue that in the case that the neutron star magnetic field in Cir X-1 is only moderately high, the two facts that the neutron star is very young and that we observe type I X-ray thermonuclear bursts are consistent with what we know today.

The most prominent difference in the observed line properties, besides the fact that the He-like resonance lines appear enhanced, is the blueshifts of about 400 km s⁻¹ in most of the observed lines. In Schulz et al. (2008), the lines were at rest and identified as ADC emissions from the accretion disk. In this low state, blueshifted line emissions now need a two order of magnitude higher ionizing luminosity to sustain such an ADC in the accretion disk. The flux of the lines and the amount of blueshift in the lines appear small and may not have significantly affected the line profiles in observations III and IV specifically because the shifts are of the order of a grating spectral resolution element. However, we note that Iaria et al. (2008) claim a faint blue component in their line profile analysis in observation III. This could mean that this blueshifted emission is always present but gets overpowered by ADC emissions when the source gets brighter.

There is no viable explanation for blueshifted lines within the accretion disk, suggesting emission regions outside the disk. While Iaria et al. (2008) proposed possible jet emissions, the determined ionization parameters as well as emission measures and the amount of the blueshift are also consistent with X-ray illumination of a massive companion wind. A wind velocity of 400 km s⁻¹ would point in the direction of a B5Ia supergiant (Prinja & Massa 1998) as a companion, confirming the identification from Jonker et al. (2007). There is not much room for later types because in this spectral type range terminal wind velocities decline rapidly with spectral type. We point out that B5 stellar winds themselves cannot produce X-rays through inner wind shocks, as we observe in more massive stars. In this case, the very-low-density outer terminal velocity zone of the wind gets illuminated by the X-ray source in Cir X-1. That we may observe the illumination of the stellar wind is also supported by an apastron observation taken during a more prominent outburst in 2010. Figure 2 shows a dominant continuum but no photoionized lines, which should have been detectable. Even with the brighter continuum, the three brightest lines in the periastron spectrum should still be detectable above a 5σ level. Yet the only line detected is the Fe K fluorescence line, which we commonly associate with the accretion disk. The absence of unshifted ADC lines at least with regards to S, Si, and Mg from the accretion disk is peculiar because if it is true that the X-ray emissions have not changed in luminosity and are further obscured than what we observed in observations III and IV, then these lines should still be there unless they are now also suppressed by heavy absorption. In that case, the photoionized emission from the stellar wind is all that is left for the observer, as well as some weak higher Z lines from the ADC.

The amount of the shift would also rule out much later types because terminal wind velocities decline rapidly with type (Lamers et al. 1995; Prinja & Massa 1998). Observations V and VIIa-c were taken between orbital phase 0.00 and 0.07 using the RXTE/ASM ephemeris (MJD = 50,082.04) and an orbital period of 16.54694 days from Shirey et al. (1998; see also Clarkson et al. 2004) for which the neutron star always illuminates the face of the star into a wind coming toward the observer, hence the blueshift. However, we point out that 400 km s⁻¹ is also near the known rotational velocities of any Be-star in an X-ray binary (Reig 2011). So if it is not the wind but the stellar surface that gets ionized, then we should see blueshifts and redshifts depending on whether the compact object comes toward the companion or moves away and none when the compact object is a zero phase. With all the distractions from disk emissions in the previous observations gone, we believe that this is the first time we might see the companion wind itself.

For this one can estimate the contribution of a spherical B-star wind with a constant velocity and a specific mass loss rate to the emission measure by integrating over the available volume from a B-star radius of 10¹² cm to infinity. With a spherical mass loss rate of 10⁻⁸ M_⊙ yr⁻¹ and a terminal wind velocity of 500 km s⁻¹, this results in emission measures of the order of 10⁵⁵ erg cm⁻³ s⁻¹, which is only a fraction of what we observe. Increased emission measures require increased mass loss rates as well as nonspherical geometries. This indicates that if this is indeed the companion wind, it has to be enhanced. One indicator that this is the case could be the existence of significant amounts of resonance scattering in the He-like lines.

The photoionized plasma requires a substantial absorber in the line of sight in addition to absorption from the interstellar medium (ISM). If the wind is indeed the emission line region, this absorber has to be able to cover the bulk of the stellar wind. Normal B5 supergiants are not well known to carry wind-produced stellar disks. The only plausible protagonists we know today are magnetic stars and Be-stars, in the latter case also mostly much earlier type B-stars. In both cases, a decretion disk fed by equatorial wind can be produced under various critical circumstances. In the case of Be stars, it may be a stellar rotation rate of more than 75% of the critical rate (see Rivinius et al. 2013 for a full review), while in the case of magnetic stars it is a strong magnetic field configuration confining the wind (see Gagné et al. 2005). Such disks can provide not only the additional line-of-sight absorber but also account for the resonance scattering we observe.

However, while this is still a lot of speculation, we can in the following consider what is observationally known for the case of Be-stars. Be-star X-ray binaries (BeXBs) are a sizable subgroup in the category of HMXBs. Perhaps the most famous in the class of BeXBs is GX 301-2. This system consists of an accreting magnetized neutron star in an eccentric orbit (e ∼ 0.46; Sato et al. 1986; Koh et al. 1997). This is a very similar system except its orbit is three times as long and periastron passages are not as violent as in Cir X-1. Its optical counterpart is a B1.5Ia supergiant with a lower mass limit of 39 M_⊙, which is much larger than what is considered for Cir X-1, and here the Roche-lobe overflow connection never breaks (Sato et al. 1986; Watanabe et al. 2003; Kaper et al. 2006). However, there are many similarities, such as an inclination larger than 44°, which does not produce an eclipse but just barely misses the face of the star, producing dips in the light curve by the dense stellar wind. Jonker et al. (2007) also concluded, even though the inclination is high (Brandt et al. 1996), that in the case of Cir X-1 the neutron star never crosses the star (see also Iaria et al. 2008). The X-ray spectrum in observation III (Schulz et al. 2008) is very similar to the one observed in GX 301-2, where the region above 5 Å is dominated by line emissions from the ionized wind and the region below is dominated by accretion disk activity (Sato et al. 1986; Koh et al. 1997; Watanabe et al. 2003; Fürst et al. 2011).

In Be-stars the regions above and below the disk are more or less equivalent to those surrounding normal B-stars (Rivinius et al. 2013). Fast rotation may even enhance the wind toward the polar regions (Puls et al. 2008). The fact that the companion in Cir X-1 is a supergiant does not make much of a difference. Most massive stars begin the main sequence as fast rotators and usually stay that way (Langer et al. 1997); during the supergiant phase, the star may see a reduction in the rotation rate due to the increasing stellar radius (Puls et al. 2008). Generally, there is no reason why a fast rotator would not retain its disk in the supergiant phase, and the case could be made that Cir X-1 might as well be the youngest known BeXB.

All BeXBs classified so far have orbital periods of 20 days and higher up to 300 days (see Reig 2011 for a review), and Cir X-1 with 16.5 days would mark the shortest period, which seems adequate since it would be the youngest in the sample. However, the list of known OBe stars mostly shows dwarfs (class V), subgiants (class IV), or giants (class III), and Cir X-1 would be the only other supergiant case (classes I and II) to GX 301-2. Supergiant X-ray binaries (SGXBs) tend to have shorter periods, and here Cir X-1 would have the longest of this class. It is also the case that almost all BeXBs show X-ray pulsations ranging from a few seconds to a few hundreds of seconds. To date, no pulsations have been reported for Cir X-1, and we also did not detect any in the range where we are sensitive, which is also above a few seconds. For the neutron star being that young, we should also expect much shorter periods, likely down to a few milliseconds. The lower magnetic field of 10¹¹ G should not brake its period as fast as the other cases in 4000 yr. The mere fact that we do not observe pulsations is somewhat of concern but not completely unusual. To date, no pulsations have been found in the SGXB 4U1700-37 (Seifina et al. 2016) and in a few BeXBs (Reig 2011). New model calculations also show that at a moderate magnetic field strength of the order of 10¹¹ G and a high inclination of the system (Brandt & Podsiadlowski 1995), pulsations are quite impossible to detect if the angle between the neutron star rotation and magnetic field axis is small (Falkner et al. 2019). We should also mention that Cumming (2008) proposed for the case of HETEJ 1900.1-2455 and possibly other accreting millisecond pulsars that B-fields can be partially buried, suppressing the observation of pulsations.

The fact that we know that the neutron star in Cir X-1 is very young combined with an identification of the companion as a B5Ia supergiant also can put some constraints on the evolutionary state of the entire binary. A formation scenario involving an AIC (Bhattacharya & van den Heuvel 1991) now seems highly unlikely. It also rules out that the progenitor star was a massive O-star as these only live less than 20 Myr, whereas B3 stars and later live more than 30 Myr (Behrend & Maeder 2001). Since the companion in Cir X-1 is in its supergiant phase, which the star is in during its very late main-sequence state, it must be much older than 20 Myr. More plausible is that the progenitor was quite similar to the companion, maybe one or two types earlier. This would put the progenitor into likely the lowest mass range (8–10 M_⊙) that can produce a neutron star. Today we know little about neutron star progenitor masses, and if this conclusion is true, it is quite extraordinary.

Last but not least, the classification of Cir X-1 as a BeXB might have the potential to at least partially explain the ∼30 yr transient flux behavior shown by Parkinson et al. (2003) by invoking a precession period for the Be-star–disk system. Such precession scenarios have already been suggested by Brandt & Podsiadlowski (1995) in terms of accretion disk precession and by Heinz et al. (2013) in terms of spin–orbit coupling effects between the neutron star spin and the binary orbit. Here we suggest a precession of a companion star disk. Superorbital periods in accretion disks are not unusual in X-ray binaries, as the examples of Her X-1, LMC X-4, and SMC X-1 show. Precessing Be-star disks are much more rare but not unheard of. Lau et al. (2016) recently reported on an apparent precessing helical outflow from the massive star WR102c. The interpretation is that the precessing outflow emerged from a previous evolutionary, rapidly rotating phase of the star and attributed the precession to an unseen compact companion. In a sense, this situation is not unsimilar to what we envision here. In the WR102c system, the period of the unseen companion was constrained to between 800 and 1400 days, which is much larger than the period in Cir X-1, but the precession period of $1.4\times {10}^{4}\,\mathrm{yr}$ is very close to the long-term variation cycle in Cir X-1. This sets a precedent, and it is no longer a question whether it can happen, but it calls for details of how it happens.

We thank all members of the Chandra X-Ray Center, especially D. P. Huenemoerder and M. N. Nowak for their assistance with the HETG data processing and implementation of fitting procedures. Support for this project was provided by NASA through the Chandra Award GO7-18035X to MIT (N.S.S.), issued by the Chandra X-Ray Observatory Center, which is operated by the Smithonian Astrophysical Observatory for and on behalf of NASA under contract NAS8-03060.