ABSTRACT
We present detailed analyses of oxygen K absorption in the interstellar medium (ISM) using four high-resolution Chandra spectra toward the X-ray low-mass binary XTE J1817-330. The 11–25 Å broadband is described with a simple absorption model that takes into account the pile-up effect and results in an estimate of the hydrogen column density. The oxygen K-edge region (21–25 Å) is fitted with the physical warmabs model, which is based on a photoionization model grid generated with the xstar code with the most up-to-date atomic database. This approach allows a benchmark of the atomic data which involves wavelength shifts of both the K lines and photoionization cross sections in order to fit the observed spectra accurately. As a result we obtain a column density of NH = 1.38 ± 0.01 × 1021 cm−2; an ionization parameter of log ξ = −2.70 ± 0.023; an oxygen abundance of ; and ionization fractions of O i/O = 0.911, O ii/O = 0.077, and O iii/O = 0.012 that are in good agreement with results from previous studies. Since the oxygen abundance in warmabs is given relative to the solar standard of Grevesse & Sauval, a rescaling with the revision by Asplund et al. yields
, a value close to solar that reinforces the new standard. We identify several atomic absorption lines—Kα, Kβ, and Kγ in O i and O ii and Kα in O iii, O vi, and O vii—the last two probably residing in the neighborhood of the source rather than in the ISM. This is the first firm detection of oxygen K resonances with principal quantum numbers n > 2 associated with ISM cold absorption.
1. INTRODUCTION
X-ray studies of interstellar absorption have the potential to provide information that is not accessible by other techniques. This includes the relative abundances of a wide range of ion stages of interstellar cosmic elements (e.g., oxygen) and the abundances relative to H and He of elements with atomic number Z ⩾ 6. X-ray absorption can also provide signatures of the binding of these elements in molecules or solids; the inner-shell electronic transitions are key diagnostics since the ionization state or chemical binding shifts the line energies by a predictable amount. The ongoing effort to explore and exploit this effect in interstellar studies involves accumulation of synthetic spectra for inner-shell absorption of ions and neutrals, based on both atomic calculation and laboratory experiment, in addition to testing them against available observed interstellar X-ray absorption spectra.
As part of the effort to accumulate the required atomic data, the atomic database of the xstar photoionization modeling code (Bautista & Kallman 2001) has been systematically improved in the past 10 years in order to study the K lines and edges in the high-quality X-ray astronomical spectra obtained from the Chandra and XMM-Newton satellite-born observatories. This database includes K-vacancy levels, wavelengths, A values, radiative and Auger widths, and high-energy photoionization cross sections for complete isonuclear sequences with Z ⩽ 30 (see Palmeri et al. 2012 and references therein).
With this new database it has been possible, for instance, to estimate the model efficiency of iron K-line emission and absorption in terms of the ionization parameter and density (Kallman et al. 2004). It has been shown that the centroid of the Kα unresolved transition array, the Kβ energy, and the ratio of the Kα1 to Kα2 components are useful diagnostics of the ionization parameter. It was also found that the many strongly damped resonances below the K-ionization thresholds lead to edge smearing, which can certainly hamper the astrophysical interpretation of the absorption features. A synthetic spectrum has been modeled by Kallman et al. (2009) for the black hole X-ray transient GRO J1655-40 which constrains the odd Z element abundances (11 ⩽ Z ⩽ 27) and outflow parameters of the associated warm absorber. By analyzing the oxygen K-absorption structure in the interstellar medium (ISM) from the XMM-Newton spectrum of the low-mass X-ray binary Sco X-1, García et al. (2011) reproduced both the K edge and O i Kα absorption line, thus evaluating the impact of the atomic data on the interpretation of observations. In the present report we continue this work with a more stringent benchmark of the xstar atomic database, made possible by Chandra spectra of a low-mass binary displaying imprints of the ISM oxygen K absorption with higher resolution.
Juett et al. (2004) measured high-resolution spectra of the interstellar oxygen K shell absorption edge from seven X-ray binaries—Cygnus X-1, Cygnus X-2, 4U 1636-53, 4U 1735-44, GX 9+9, 4U 1543-624, and 4U 1820-30—using the Chandra High Energy Transmission Grating Spectrometer (HETGS). The oxygen column density NO was therein calculated using the optical depth at 21.7 Å and the cross section at this wavelength assuming the oxygen abundance from Wilms et al. (2000). With this value, the hydrogen column density NH was then determined. The 1s–2p transitions of O ii and O iii were identified, finding ISM abundances for O ii and O iii relative to O i of ≈0.1 and ⩽0.1, respectively. This became the first estimate of the O ii/O i and O iii/O i abundance ratios in the ISM.
By observing the oxygen K-shell edge toward galaxy clusters, Baumgartner & Mushotzky (2006) showed that the ISM hydrogen X-ray column density was in close agreement with the 21 cm radio value for columns less than approximately 0.5 × 1021 cm−2. For higher column densities, estimates by Baumgartner & Mushotzky were higher by as much as a factor of 2.5, indicating substantial absorption besides that due to neutral hydrogen (probably from clouds of molecular hydrogen). An average ISM oxygen abundance of 0.99 solar was found relative to the photospheric value (8.66 ± 0.05) by Asplund et al. (2004) which implied a high gas-to-dust ratio.
Pinto et al. (2010) reported a spectral model suggesting that, along the line of sight toward an X-ray source (GS 1826-238), the ISM was more complex than just simple neutral gas. Their gas model consisted of three components: cold gas (T ≈ 5.8–10 × 103 K), warm ionized gas (T ≈ 1–6 × 104 K), and hot ionized gas (T ≈ 2 × 106 K). It was also shown that the column densities NH of these three components span two orders of magnitude, where the cold gas contributed ∼90%–95%, the warm gas ∼5%–10%, and the hot gas ∼1% to the total . Costantini et al. (2012) analyzed the ISM using the bright X-ray binary 4U 1820-30, concentrating on the mildly ionized cold gas where the model consisted of mildly ionized gas, dust, and mildly ionized outflowing gas. In the oxygen edge region, dust compounds were included which did not show features below 23.7 Å. Finally, oxygen was found to be overabundant by a factor of 1.23 solar.
In order to model the oxygen K-absorption features in the ISM with xstar, we need a Chandra spectrum with high counts. In the catalog of low-mass X-ray binaries by Liu et al. (2007), we found in XTE J1817-330 a suitable source with an average count number of ≈8 × 106 counts s−1. Discovered by the Rossi X-Ray Timing Experiment (Remillard et al. 2006), XTE J1817-330 is a low-mass X-ray binary displaying a primary component which has an X-ray spectrum and variability behavior consistent with its identification as a black hole (Sala et al. 2007). Its location, l = 359.81 and b = −7.99 in Galactic coordinates, implies a line of sight in the Galactic center direction, a region that dominates the total mass of the Galaxy (Stasińska et al. 2012) providing a good opportunity to study the ISM absorption features. This source has been observed in different spectral ranges, namely the optical (Torres et al. 2006), near-infrared (D'Avanzo et al. 2006), radio (Rupen et al. 2006), and X-ray (Roy et al. 2011). Sala et al. (2007) fitted its X-ray spectrum finding an absorbing hydrogen column of 1.55 × 1021 cm−2 and identifying the following oxygen absorption lines: O i Kα(23.52 Å), O ii Kα(23.35 Å), O iii Kα(23.13 Å), O i Kβ(22.91 Å), and O vii Kβ(21.60 Å). Gierliński et al. (2009) estimated a hydrogen column density of NH = 1.08 ± 0.05 × 1021 cm−2, and Roy et al. (2011) investigated the timing characteristics in the spectra using a column of NH = 1.2 × 1021 cm−2.
The outline of the present paper is as follows. In Section 2, we describe the xstar warmabs physical model which is used to obtain the final spectrum fits, in particular the details concerning its atomic database. The data reduction aspects of the observations, e.g., the pile-up effect, are reviewed in Section 3, while the results of the broadband fits and resolved absorption features are given in Sections 4 and 5. Section 6 is dedicated to comparisons of the hydrogen column density that allow an estimate of its reliability. Section 7 describes the oxygen column densities derived from our fits. Finally, the discussion and conclusions are presented in Section 8.
2. XSTAR/WARMABS
xstar 8 is a code package designed to determine the physical conditions in partially ionized gases. It calculates the temperature, ionization, excitation, absorption, and emission due to neutral and all ionized species of elements with atomic number Z ⩽ 30. As many relevant physical processes as possible are taken into account, assuming a time stationary balance among them and a Maxwellian electron velocity distribution. Important processes for many problems relevant to X-ray astronomy include photoionization, electron impact collisional ionization and excitation, and radiative and dielectronic recombination. More complete descriptions of xstar and its atomic database are given by Kallman & Bautista (2001) and Bautista & Kallman (2001). In this work we employ the warmabs utility which allows for the calculation of xstar models within the widely used X-ray spectral fitting package xspec.9 It utilizes precalculated tabulations of the ion fractions and level populations in order to calculate synthetic emission and transmission spectra. It assumes that the gas responsible for emission or absorption has a uniform ionization and temperature throughout, although it is possible to superimpose two or more components and simulate a non-uniform situation.
2.1. Atomic Data
The K photoabsorption data sets for oxygen ions with electron numbers 2 ⩽ N ⩽ 8 contained in the xstar atomic database were generated by García et al. (2005). Energies for both valence and K-vacancy fine-structure levels, transition wavelengths, gf values, and radiative and Auger widths were computed with the atomic structure packages autostructure (Eissner et al. 1974; Badnell 1986, 1997) and hfr (Cowan 1981). Electron correlation effects were taken into account by means of configuration–interaction expansions within the n ⩽ 3 complexes, and relativistic corrections were included with a Breit–Pauli Hamiltonian. The accuracy of the theoretical level energies and wavelengths does not generally match that attained in experiments (see Table 1), and the scanty availability of measurements for the K shell of O ions with N > 3 limits the implementation of term-energy corrections and wavelength shifts that can be introduced in order to compensate for the theoretical shortcomings. Furthermore, as also shown in Table 1, experimental misassignments, such as that quoted by Gu et al. (2005) for O vi, are not uncommon. (The discrepancy in O iii is believed to be a typo.)
Table 1. Comparison of EBIT and Theoretical Wavelengths (Å) for K Lines
Ion | Lower Level (J) | Upper Level (J') | EBIT | Theory |
---|---|---|---|---|
O vi | 1s22s(1/2) | 1s2s2p(1/2, 3/2) | 22.0194 ± 0.0016a | 22.00c |
22.374 ± 0.008b | 22.05d | |||
22.03e | ||||
O v | 1s22s2(0) | 1s2s22p(1) | 22.374 ± 0.003a | 22.33c |
22.370 ± 0.010b | 22.35d | |||
22.37e | ||||
O iv | 1s22s22p(1/2, 3/2) | 1s2s22p2(1/2, 3/2) | 22.741 ± 0.005b | 22.78c |
22.73d | ||||
22.75e | ||||
O iii | 1s22s22p2(1, 2) | 1s2s22p3(1) | 22.071 ± 0.006b | 23.08c |
23.05d | ||||
23.07e |
Notes. aEBIT measurement (Schmidt et al. 2004). bEBIT measurement (Gu et al. 2005). chullac calculation (Behar & Netzer 2002). dR-matrix calculation (Pradhan et al. 2003). ehfr calculation (García et al. 2005).
Download table as: ASCIITypeset image
High-energy photoionization cross sections in the K-edge region for oxygen ions were obtained by García et al. (2005) with bprm, the Breit–Pauli R-matrix suite of codes (Scott & Burke 1980; Scott & Taylor 1982; Berrington et al. 1987; Seaton 1987). A conspicuous effect in the resonance structure of these cross sections that must be treated in detail is Auger damping. It is the result of the dominant spectator KLL Auger decay channels of a photoexcited K-vacancy state, say,



over the participator Auger KLn channels


which leads to edge smearing by resonances with symmetric profiles of nearly constant width. They are handled within the R-matrix package by means of an optical potential (Gorczyca & Robicheaux 1999; Gorczyca & McLaughlin 2000). Again, the accuracy of the resonance energy positions is limited by the absence of measurements that would enable empirical adjustments of the many K-vacancy target thresholds (i.e., the series limits) in the close-coupling expansion.
3. OBSERVATIONS AND DATA REDUCTION
In Table 2 we list the specifications of the four observations of XTE J1817-330 obtained by Chandra using the HETGS in combination with the Advanced CCD Imaging Spectrometer (ACIS). The oxygen edge is accessible with the HETGS exclusively via the Medium Energy Gratings (MEG). Given the large flux of this source (F = 1350 μJ, integrated in the 2–10 keV energy range), these observations were taken in continuous clocking mode which significantly increases the temporal resolution in order to minimize the pile-up effect (Cackett et al. 2008). Pile-up is an inherent feature of CCD detectors, such as those of the ACIS instrument that occurs when two or more photons are detected as a single event possibly causing a deformation in the level and shape of the continuum (Miller et al. 2006b). It is stronger in the MEG than in the HEG due to the lower dispersion and higher effective area of the former.10 Although pile-up is usually not an issue near the oxygen edge, it may be for the continuum which is used to establish the hydrogen column density by fitting the 11–25 Å interval; hence, we consider the pile-up effect in this range by applying the correction model simple_gpile2.sl (Hanke et al. 2009) before spectral fitting. Among the most popular computer packages for fitting X-ray data are xspec and isis, and we have used the latter (version 1.6.211) to include the pile-up model.
Table 2. Chandra Observations Used in This Paper
ObsID | Date | Exposure | Instrument | Grating | Label |
---|---|---|---|---|---|
(ks) | |||||
6615 | 2006 Feb 13 | 18 | HETG-ACIS | MEG | Observation 1 |
6616 | 2006 Feb 24 | 29 | HETG-ACIS | MEG | Observation 2 |
6617 | 2006 Mar 15 | 47 | HETG-ACIS | MEG | Observation 3 |
6618 | 2006 May 22 | 51 | HETG-ACIS | MEG | Observation 4 |
Download table as: ASCIITypeset image
Even though the spatial resolution is reduced to one dimension when the continuous clocking mode is used, there is essentially no difference with the timed exposure mode in the background extraction or in the data reduction process. Hence, we have reduced the data sets using the standard CIAO threads.12 In some cases, the zero-order data were not telemetered; for these observations the zero-order position was then estimated by finding the intersection of the grating arms using the findzo algorithm.13
Figure 1 shows the light curves of the four observations ObsID 6615, 6616, 6617, and 6618 in units of counts s−1. It can be seen that the average counts s−1 is approximately constant for all, and therefore we use the averaged spectra for this work. We have carried out the following spectral fitting procedure: first, each spectrum was rebinned with a 0.1 Å bin in order to fit the continuum in the 11–25 Å wavelength range so as to find the hydrogen column density using the TBnew model; then, the default spectral resolution was restored, and the oxygen edge region (21–24 Å) was analyzed with the warmabs model freezing the column density at the value obtained in the preceding step.
Figure 1. Light curves of the four observations, ObsIDs 6615, 6616, 6617, and 6618, in units of counts s−1. The approximately constant average counts s−1 that prevails is an indication of a low degree of variability.
Download figure:
Standard image High-resolution image4. BROADBAND FIT
Following Yao & Wang (2006), we rebinned each spectrum with a 0.1 Å bin size throughout the spectral range to decrease the number of absorption lines and improve the broadband fit. We fitted the four observations simultaneously in the interval 11–25 Å using the simplegpile2(TBnew(powerlaw)) model, where simplegpile2 is a convolution model to account for pile-up effects and TBnew is an X-ray absorption model that includes abundances for elements from H to Ni (Wilms et al. 2000). Also we added Gaussians to fit the remaining absorption lines after rebinning. We used abundances specified in Wilms et al. (2000) and cross sections by Verner et al. (1996). The O, Ne, and Fe abundances were handled as free parameters, and for the analysis χ2 statistics were employed. We fixed the absorption parameters in all the observations, and varied the power-law, pile-up, and Gaussian parameters for each case.
Figure 2 shows the fit in the interval 11–25 Å using the simplegpile2(TBnew(powerlaw)) model. Although the fit is carried out simultaneously for the four observations, they are plotted separately to improve clarity. Panels (a), (b), (c), and (d) correspond to observations ObsID 6615, 6616, 6617, and 6618, respectively (see Table 2). In each panel, black data points correspond to the observation, while the solid red lines represent the best-fit model for each case, and the base plots show the fit residuals in units of χ2. In each spectrum the K edges of Ne and O are, respectively, located at ≈14.3 Å and ≈23.2 Å. The O i and O ii 1s–2p absorption lines at 23.5 Å and 23.35 Å are also clearly observed. We improved the fit by including Gaussians to model the outstanding absorption lines: Ne ix Kα (≈13.43 Å), Ne iii Kα (≈14.50 Å), Ne ii Kα (≈14.60 Å), O viii Lyα (≈18.95 Å), O vii Kα (≈21.58 Å), and O ii Kα (≈23.34 Å). Residuals indicate that the model fits the data satisfactorily with deviations near the Ne edge (≈14.3 Å) due to background (i.e., clocking the whole chip) or an unaccounted for hot pixel. We also find high residuals in the oxygen absorption region; in particular, the O i Kα and O ii Kα are not well modeled.
Figure 2. Chandra MEG spectra of the X-ray binary XTE J1817-330 simultaneously fitted in the 11–25 Å region using the TBnew model. (a) ObsID 6615. (b) ObsID 6616. (c) ObsID 6617. (d) ObsID 6618. See Table 2 for further details pertaining to these observations. In each panel, the black data points are the observations, while the solid red lines correspond to the best-fit models. Residuals are shown as histograms for each case in units of χ2. Note the large residuals in the oxygen K-edge region, particularly in the neighborhood of the O i and O ii Kα lines at ≈23.50 Å and ≈23.35 Å, respectively.
Download figure:
Standard image High-resolution imageFit results for the individual free abundances, column density, power law, and pile-up are presented in Table 3. The fit statistics give a reduced chi-square of χ2 = 1.208. Differences in the power-law parameters are due to the count numbers in each observation. We obtain abundances for O, Ne, and Fe higher than solar (Wilms et al. 2000), especially for Ne (a factor of two), and a column density of cm−2 which is somewhat larger than those estimated in other spectral fits and by measurements of the 21 cm line. This will be further discussed in Section 6.
Table 3. Broadband Simultaneous Fit Parameters
Model | Parameter | Value |
---|---|---|
TBNew | NH (1021 cm−2) | ![]() |
TBNew | Neon abundance (ANe) | 2.258 ± 0.080 |
TBNew | Iron abundance (AFe) | 1.302 ± 0.050 |
TBNew | Oxygen abundance (AO) | 1.178 ± 0.022 |
Power law | Normalization | 9.416 ± 0.064a |
7.039 ± 0.041b | ||
4.613 ± 0.042c | ||
1.481 ± 0.001d | ||
Power law | Photon index | 1.545 ± 0.030a |
1.618 ± 0.030b | ||
1.735 ± 0.029c | ||
2.242 ± 0.038d | ||
SimpleGpile2 | β | 0.050 ± 0.001a |
0.051 ± 0.001b | ||
0.050 ± 0.001c | ||
0.049 ± 0.001d | ||
Reduced chi-square | χ2 | 1.208 |
Notes. Abundances relative to the solar values of Wilms et al. (2000). aObsID 6615. bObsID 6616. cObsID 6617. dObsID 6618.
Download table as: ASCIITypeset image
The nonlinear pile-up convolution model simple_gpile2.sl exponentially reduces the predicted count rate Rj(λ) (in units of counts s−1 Å−1) according to

where Rtot(λ) is the total spectral count rate and β is treated as a fit parameter. The model is only applicable to first-order grating spectra, although it employs information from the second- and third-order grating spectra.14 With this model we obtained an average β = 0.050 according to which the spectra have a pile-up degree greater than 25% for 14 ⩽ λ ⩽ 16 Å with the highest value of p = 42% at ≈14.3 Å for the observation with the largest number of counts (ObsID 6615). This value shows the need to include the pile-up effect in the 11–25 Å region to obtain a good fit. However, for all the observations, we have a pile-up degree lower than 5% in the 21–24 Å wavelength range which corresponds to the oxygen absorption region and, therefore, its effects were ignored.
5. ABSORPTION FEATURES
In order to identify all the absorption features present in the 21–25 Å wavelength region, we first applied a functional model consisting of a power law for the continuum and Gaussian profiles to describe the absorption lines. Figure 3 shows the spectral fit of the four Chandra MEG observations of XTE J1817-330 in the oxygen absorption region. Although shown in separate panels for clarity, the fit was performed simultaneously for the four observations. Nevertheless, the power-law parameters (photon index and normalization) are allowed to vary independently for each data set. Several absorption features are clearly observed, the most outstanding being the O i and O ii Kα lines. The base plot in each panel shows, in units of χ2, the fit residuals as a histogram. High residuals are found around 24.8 Å, and the model underestimates the number of photons around 23.2 Å.
Figure 3. Spectral fit of the Chandra MEG observations of XTE J1817-330 in the oxygen absorption region (21–25 Å) using a simple functional model (power law and several Gaussian profiles). (a) ObsID 6615. (b) ObsID 6616. (c) ObsID 6617. (d) ObsID 6618. The observed absorption features are interpreted as the neutral O K-edge (≈23.1 Å); the Kα, Kβ, and Kγ lines of O i (≈23.50 Å, ≈22.88 Å, and ≈22.68 Å, respectively); the Kα, Kβ, and Kγ lines of O ii (≈23.35 Å, ≈22.28 Å, and ≈22.10 Å, respectively); and the Kα lines of O iii, O vi, and O vii (≈23.10 Å, ≈22.02 Å, and ≈21.58 Å, respectively). See Table 4 for more details.
Download figure:
Standard image High-resolution imageTable 4 shows a list of the lines obtained with the functional fit including the centroid wavelengths derived from the Gaussian profile parameters. It also lists the wavelengths for the lines measured by Sala et al. (2007). Note that our wavelengths are within the error bars of the values reported by these authors, which are much larger given the lower spectral resolution of the XMM-Newton instruments. We obtain a reduced chi-square of χ2 = 1.434. The O i Kα is found at 23.502 ± 0.001 Å which has been previously declared to be of ISM origin (Sala et al. 2007). Furthermore, we have also detected four other O i absorption lines corresponding to resonances with higher principal quantum number n: two pairs of Kβ (n = 3) and Kγ (n = 4) resonances associated with the 4P and 2P K-hole core states. These are found at 22.884 ± 0.004 Å, 22.790 ± 0.001 Å, 22.686 ± 0.004 Å, and 22.609 ± 0.007 Å, respectively. In this set, only the first Kβ (belonging to the 4P core state) has been previously detected at 22.91 ± 0.03 Å in the XMM-Newton observation analyzed by Sala et al. (2007). This is then the first firm detection of oxygen K resonances with n > 2 associated with ISM cold absorption.
Table 4. Absorption Line Assignments
Ion | Transition | λa | λb |
---|---|---|---|
(Å) | (Å) | ||
O vii | 1s2 1S–[1s]2p 1Po | 21.589 ± 0.003 | 21.60 ± 0.06 |
O vi | 2s 2S–[1s]2s2p 2Po | 22.022 ± 0.003 | |
O ii | 2p3 4So–[1s]2p34p 4P | 22.101 ± 0.005 | |
O ii | 2p3 4So–[1s]2p33p 4P | 22.280 ± 0.003 | |
O i | 2p4 3P–[1s]2p4(2P)4p 3Po | 22.609 ± 0.007 | |
O i | 2p4 3P–[1s]2p4(2P)3p 3Po | 22.686 ± 0.004 | |
O i | 2p4 3P–[1s]2p4(4P)4p 3Po | 22.790 ± 0.001 | |
O i | 2p4 3P–[1s]2p4(4P)3p 3Po | 22.884 ± 0.004 | 22.91 ± 0.03 |
O iii | 2p2 3P–[1s]2p3 3Po | 23.054 ± 0.001 | 23.13 ± 0.09 |
O iii | 2p2 3P–[1s]2p3 3Do, 3So | 23.104 ± 0.005 | 23.13 ± 0.09 |
O ii | 2p3 4So–[1s]2p4 4P | 23.358 ± 0.002 | 23.35 ± 0.03 |
O i | 2p4 3P–[1s]2p5 3Po | 23.502 ± 0.001 | 23.52 ± 0.02 |
Notes. aPresent work. bSala et al. (2007).
Download table as: ASCIITypeset image
The line at 23.358 ± 0.002 Å is also of ISM origin (Juett et al. 2004) and corresponds to O ii Kα. As in the case of O i, resonances with n > 2 were also detected for O ii. These were found at 22.280 ± 0.003 Å and 22.101 ± 0.005 Å corresponding to Kβ and Kγ, respectively. Although the O iii Kα resonance is a triplet, only two absorption features were detected at 23.104 ± 0.005 Å and 23.054 ± 0.001 Å while Sala et al. (2007) only found one line at 23.13 ± 0.09 Å identified as O iii Kα. Finally, we have also located Kα transitions for both O vi (22.022 ± 0.003 Å) and O vii (21.589 ± 0.003 Å), from which only the strongest of the two (O vii) was reported by Sala et al. (2007) at 21.609 ± 0.06 Å.
The functional fit described above illustrates the complexity of oxygen K absorption. There are (at least) two different regimes: a cold component of mostly neutral gas and a hot component with highly ionized gas. Previous studies have suggested a multiple-phase ISM; however, it is not clear if the two high-ionization lines from O vi and O vii are intrinsic to the source or in fact arise in the ISM. Thus we direct our study to the cold phase. Furthermore, since cold absorption is not exclusively due to neutral oxygen, this means that a model such as TBnew will not completely represent the observed features. We have then simultaneously fitted the four exposures in the 21–25 Å wavelength range using the powerlaw*warmabs physical model. The advantage of using warmabs is that it contains the most recent atomic data for all the ions in the oxygen isonuclear sequence, thus enabling the fit of all the absorption lines from O i, O ii, and O iii. While implementing this model, the column density is held fixed at the value obtained in the broadband simultaneous fit (NH = 1.66 × 1021 cm−2). The ionization parameter and the oxygen abundance are treated as free parameters but the same for all observations, while the abundances of all the other chemical elements are held fixed at their solar values. The quantities describing the power law (photon index and normalization) are also taken as free parameters but independently for each observation.
Figure 4 shows the results of this fit. As before, we present for clarity the spectra with the best-fit model in separate panels, although the fit is performed simultaneously. Since most of the relevant absorption features in the spectra are due to neutral, singly, and doubly ionized O ions, the ionization parameter that describes the fit tends to be somewhat low: log ξ = −2.729. In this case, the warmabs model does not show any absorption due to O vii; therefore, this line is still represented with a Gaussian profile. We have found large residuals around most of the oxygen absorption features, the most prominent being around the O i and O ii Kα lines. The fit statistics give a poor reduced chi-square of χ2 = 1.769 which forced us to revise the atomic database.
Figure 4. Spectral fit of the Chandra MEG observations of XTE J1817-330 in the oxygen absorption region (21–25 Å) using a powerlaw*warmabs physical model. (a) ObsID 6615. (b) ObsID 6616. (c) ObsID 6617. (d) ObsID 6618.
Download figure:
Standard image High-resolution imageA close examination of the resonance positions from the photoabsorption cross sections used in warmabs in comparison with those obtained from the spectral fits with Gaussian profiles reveals the inaccuracies of the atomic data. Figure 5(a) shows with solid colored lines the atomic cross sections used in warmabs for O i, O ii, and O iii (García et al. 2005). Vertical dashed lines are placed at the positions of the absorption features found with the functional fit described above (see Table 4). A comparison shows that not only the positions of the O i and O ii Kα resonances are displaced with respect to the observed data but also those with n > 2. The solid black line represents the experimental O i photoabsorption cross section measured by Stolte et al. (1997; unfortunately, there are no laboratory measurements available for the other species) which displays a similar wavelength shift with respect to the observed resonance positions. This wavelength offset has been previously discussed by Juett et al. (2004) based on the Kα position observed in the Chandra spectra from several sources, all of which match the position we have found using Gaussian profiles. For O i, the comparison of the theoretical cross section with the laboratory measurement shows that, except for the Kα peak, all the other resonances agree very well. This implies that the relative position of the Kα with respect to the other resonances also needs to be adjusted. This is also the case for O ii since the wavelength difference of the theoretical and observed Kα positions is not the same as those for the rest of the resonances of the same ion. Consequently, we have decided on the one hand to adjust the positions of the Kα resonances (which are treated as lines in warmabs), and on the other hand, to shift all the cross sections for both species so as to obtain the best possible agreement with the observed lines. The new positions for the O i and O ii Kα lines are 23.502 Å and 23.343 Å, while the cross sections were shifted toward shorter wavelengths by 0.033 Å and 0.079 Å, respectively. No correction was applied to the O iii curve since the observed lines are weaker, in addition to the fact that in this case the Kα resonance is a triplet. Figure 5(b) shows the atomic absorption cross sections after the wavelength corrections are applied. In other words, the Chandra observed line positions suggest not only that the theoretical resonance positions from García et al. (2005) must be adjusted, but also that the wavelength scale of the only available experiment for O i (Stolte et al. 1997) needs to be shifted by 33 mÅ.
Figure 5. Comparison of the theoretical photoabsorption cross sections for O i (red), O ii (green), and O iii (blue) computed by García et al. (2005) that are implemented in the warmabs model. The black solid line is the laboratory measurement by Stolte et al. (1997) showing peaks of O i and O2. The vertical dashed lines are placed at the wavelengths of the observed absorption lines (see Table 4). Panel (a) displays the original cross sections, while panel (b) shows the same curves after the wavelength shifts are applied to both O i and O ii (see the text for details).
Download figure:
Standard image High-resolution imagePossible effects due to the instrument wavelength calibration, model uncertainties, and Doppler shifts due to motion of the gas in the ISM are important for accurate line identification. The instrumental resolution for the first-order spectra from MEG is Δλ ∼ 23 mÅ, which is better than the largest shift required for our theoretical cross sections (33 mÅ). In fact, according to García et al. (2005), the uncertainties in the theoretical cross sections for O ions with electron occupancies N ⩽ 4 were estimated to be ∼50 mÅ, comparable with the wavelength uncertainties in the laboratory measurements of Stolte et al. (1997). Assuming a velocity dispersion for the ISM of ⩽200 km s−1 (Juett et al. 2004), the largest wavelength shift due to Doppler effects for the O i Kα is Δλ = 14 mÅ, which is smaller than the instrumental resolution. Hence, we can safely ignore the effects of ISM gas motion in the line of sight.
Figure 6 shows the fit results with warmabs after the theoretical line positions have been corrected, where the residuals are now significantly reduced and more evenly distributed. This is reflected in better fit statistics (reduced χ2 = 1.245) and a well-constrained ionization parameter (log ξ = −2.699 ± 0.023) which is consistent with the low-ionization oxygen lines observed. As before, the O vii Kα resonance at 21.589 Å is not well represented with this model and must be fitted with a Gaussian profile. We have also noticed that the absorption feature at 22.022 Å is not entirely due to the O iii Kδ resonance and is thus fitted with a Gaussian profile, confirming a blend of both O iii Kδ and O vi Kα. In this final analysis, we also treated the column density value as a free parameter in order to increase the accuracy of the fit. We find a column density of NH = 1.38 ± 0.1 × 1021 cm−2 that is lower than the value previously mentioned in connection with the broadband fit (see Section 4). The oxygen abundance obtained with the warmabs model is relative to solar, also lower than the value for the best fit with TBnew (AO = 1.178 ± 0.22). All the best-fit parameters of the warmabs are summarized in Table 5.
Figure 6. Spectral fit of the Chandra MEG observations of XTE J1817-330 in the oxygen absorption region (21–25 Å) using a powerlaw*warmabs physical model, after the wavelength corrections in the O i and O ii atomic cross sections are introduced. (a) ObsID 6615. (b) ObsID 6616. (c) ObsID 6617. (d) ObsID 6618.
Download figure:
Standard image High-resolution imageTable 5. warmabs Simultaneous Fit Parameters
Model | Parameter | Value |
---|---|---|
Warmabs | NH (1021 cm−2) | 1.38 ± 0.01 |
Warmabs | Log ionization parameter (log ξ) | −2.699 ± 0.023 |
Warmabs | Oxygen abundance (AO) | ![]() |
Power law | Normalization | 2.384a |
2.544b | ||
1.511c | ||
0.685d | ||
Power law | Photon index | 3.044a |
2.479b | ||
2.720c | ||
2.617d | ||
Reduced chi-square | χ2 | 1.245 |
Notes. Oxygen abundance relative to the solar value of Grevesse & Sauval (1998). aObsID 6615. bObsID 6616. cObsID 6617. dObsID 6618.
Download table as: ASCIITypeset image
6. HYDROGEN COLUMN DENSITY
Table 6 shows a comparison of the hydrogen column density obtained in this work with previous estimates including measurements using the 21 cm line. By means of the simultaneous broadband fit described in Section 4, we have obtained a column density of cm−2 which is somewhat larger than those quoted in previous spectral fits. Miller et al. (2006a) reported a hydrogen column density of NH = 8.8–9.7 × 1020 cm−2 using a 50 ks Chandra spectrum, a value smaller than those obtained from Galactic H i surveys, e.g., 1.58 × 1021 cm−2 (Dickey & Lockman 1990) and 1.39 × 1021 cm−2 (Kalberla et al. 2005). Using the equivalent widths (EWs) of several interstellar bands in the optical spectrum, Torres et al. (2006) found the column density to be in the range of NH = 1–3 × 1021 cm−2. Moreover, both Rykoff et al. (2007) and Gierliński et al. (2008) gathered acceptable fits of several Swift X-Ray Telescope (XRT) spectra with a fixed Galactic column density of NH = 1.2 × 1021 cm−2. Gierliński et al. (2009) used the same XRT spectra combined with data from UVOT (the Ultraviolet and Optical Telescope) to derive a column density of NH = 1.08 ± 0.05 × 1021 cm−2 from broadband spectral fits.
Table 6. Hydrogen Column Density Comparison
Method | NH (× 1021 cm−2) |
---|---|
TBnew fita | ![]() |
warmabs fita | 1.38 ± 0.1 |
21 cm surveyb | 1.58 |
21 cm surveyc | 1.39 |
Spectral model fitd | 1.55 ± 0.05 |
Spectral model fite | 1.2 |
Spectral model fitf | 0.88–0.97 |
Spectral model fitg | 1.0–3.0 |
Spectral model fith | 1.08 ± 0.05 |
Notes. aPresent work. bDickey & Lockman (1990). cKalberla et al. (2005). dSala et al. (2007). eRykoff et al. (2007), Gierliński et al. (2008), Roy et al. (2011). fMiller et al. (2006a). gTorres et al. (2006). hGierliński et al. (2009).
Download table as: ASCIITypeset image
The agreement between TBnew and Sala et al. (2007) is around 7% despite the fact that their fits include data from XMM-Newton EPIC-Pn (0.6–10.0 keV), RGS1 (0.3–2.0 keV), and OM covering a much wider energy range; furthermore, they model the foreground absorption with TBabs, an older version of TBnew (Wilms et al. 2000). On the other hand, the present fits of the oxygen K region (21–25 Å) with the warmabs physical model yield a lower column density (NH = 1.38 ± 0.1 × 1021 cm−2) than that derived from our broadband fit and TBnew. This discrepancy is expected due to the different atomic data sets involved, particularly since TBnew only includes the photoabsorption cross section for neutral oxygen. Taking into account the scatter of the previous column densities in the literature and the fact that we have adequately modeled the absorption features from O i, O ii, and O iii, we would expect the value derived from warmabs to be more reliable than that from TBnew.
7. OXYGEN COLUMN DENSITY
In order to derive the O column density along the line of sight of XTE J1817-330, we have calculated the curve of growth for each of the oxygen Kα transitions using the atomic data from García et al. (2005), and EWs have been obtained using the functional fit described in Section 5. Table 7 shows a comparison between the present Kα EWs for O i, O ii, O iii, O vi, and O vii and those by Sala et al. (2007) for XTE J1817-330. EWs listed in Juett et al. (2004) for seven other sources are also included as well as those by Yao et al. (2009) for Cygnus X-2. For O i Kα, our EW (51 ± 5 mÅ) is around the lower limits in Sala et al. (2007) and Juett et al. (2004), and for O ii Kα, the present value of 52 ± 5 mÅ is within the range that Juett et al. (2004) associated with the ISM. However, we find the Kα lines in both O i and O ii to be saturated which could influence our derived EWs. Furthermore, the Kα EWs in O iii, O vi, and O vii are in good agreement with the values obtained by Juett et al. (2004), Sala et al. (2007), and Yao et al. (2009).
Table 7. Equivalent Width Comparison for the Oxygen Kα Transitions in the ISM
Source | O i Kα | O ii Kα | O iii Kα | O vi Kα | O vii Kα |
---|---|---|---|---|---|
(mÅ) | (mÅ) | (mÅ) | (mÅ) | (mÅ) | |
XTE J1817-330a | 51 ± 5 | 52 ± 5 | 16 ± 5 | 14 ± 5 | 54 ± 4 |
XTE J1817-330b | 70 ± 20 | 40 ± 20 | 30 ± 30 | 40 ± 30 | |
Cyg X-2c | 67 ± 11 | 26 ± 17 | 9 ± 8 | ||
Cyg X-2d | 50 ± 3 | 28 ± 6 | 11 ± 2 | ||
4U 1543-624c | 68 ± 11 | 41 ± 11 | 18 ± 5 | ||
4U 1820-30c | 70 ± 20 | 41 ± 17 | 20 ± 10 | ||
4U 1735-44c | 80 ± 30 | 40 ± 20 | 24 ± 10 | ||
GX 9+9c | 80 ± 30 | 50 ± 19 | 27 ± 13 | ||
4U 1636-53c | 90 ± 30 | 60 ± 20 | 33 ± 13 | ||
Cyg X-1 (ObsID 3407) c | 95 ± 17 | 20 ± 11 | 35 ± 15 |
Notes. aPresent work. bSala et al. (2007). cJuett et al. (2004). dYao et al. (2009).
Download table as: ASCIITypeset image
Table 8 shows a comparison of the oxygen column densities obtained from the EWs and the warmabs model fit. As previously mentioned in Section 5, for the ionization parameter found in our best warmabs fit, only O i, O ii, and O iii are taken into account in the model. As a reference, we have also included the oxygen column density for O vi from Savage et al. (2003), obtained from Far-Ultraviolet Spectroscopic Explorer (FUSE) data for four sources located near the same Galactic latitude as XTE J1817-330. The curve-of-growth values for O i, O ii, and O iii result in a total column density of cm−2, in relatively good agreement with that derived from the ionic fractions in the warmabs fit, namely NO = 6.41 × 1017 cm−2. However, the ratios O i/O ii are significantly different: 1.4 and 11.9, respectively. Due to line saturation, the measured EWs are likely to be underestimated, in particular for the O i Kα line. This leads to a higher uncertainty in the column and ion fractions estimated from the EW measurements. On the other hand, since the parameters in the warmabs fit are constrained by all the lines as well as the K edge, we are more confident about their reliability. In the case of the O vi and O vii Kα transitions, because the EWs are on the flat section of the curve of growth, they strongly depend on velocity dispersion. For O vi Kα, the oxygen column density spans from
cm−2 to
cm−2 when using dispersion velocities of v = 20 km s−1 and v = 200 km s−1, respectively. The smallest value is still about an order of magnitude larger than that reported by Savage et al. (2003), making it difficult to constrain the velocity dispersion. Also, no real improvement is achieved by considering larger dispersions, as the column density is hardly reduced once the velocities are larger than 200 km s−1. This may be an indication that the observed O vi line is likely to originate from the neighborhood of XTE J1817-330 rather than from the ISM. In the case of O vii, the oxygen column density varies from
cm−2 (v = 20 km s−1) to
cm−2 (v = 200 km s−1).
Table 8. Oxygen Column Density Comparison
Source | O i | O ii | O iii | O vi | O vii |
---|---|---|---|---|---|
(× 1017 cm−2) | (× 1017 cm−2) | (× 1017 cm−2) | (× 1015 cm−2) | (× 1015 cm−2) | |
XTE J1817-330a | 3.18 ± 0.79 | 2.18 ± 0.54 | 0.35 ± 0.08 | 5.06 ± 2.53d | 45 ± 18d |
5.99 ± 1.9e | 988 ± 405e | ||||
71.12 ± 29.15f | 4484 ± 1838f | ||||
XTE J1817-330b | 5.85 ± 1.75 | 0.49 ± 0.12 | 0.07 ± 0.02 | ||
PG 1302-102c | 0.16 ± 0.03 | ||||
Mrk 1383c | 0.38 ± 0.07 | ||||
ESO 141-G55c | 0.31 ± 0.06 | ||||
PKS 2005-489c | 0.60 ± 0.01 |
Notes. aCalculated from the EWs and the curves of growth (see Section 7). bDerivated from the warmabs model fit using the solar value of Grevesse & Sauval (1998, see Section 5). cObtained from FUSE data by Savage et al. (2003). dUsing a velocity dispersion v = 200 km s−1. eUsing a velocity dispersion v = 100 km s−1. fUsing a velocity dispersion v = 20 km s−1.
Download table as: ASCIITypeset image
8. DISCUSSION AND CONCLUSIONS
There are several interesting conclusions emerging from this work that configure a set of useful guidelines in the analysis of ISM absorption features in Chandra and XMM-Newton spectra. In order to resolve such spectral signatures, bright sources with high count rates are desirable. In the present XTE J1817-330 study case, we were fortunate to find four good quality spectra that were fitted simultaneously so as to obtain improved statistics. However, it was realized that the pile-up effect must be taken into account in order to obtain reasonably reliable hydrogen column densities. This effect was treated adequately with the nonlinear pile-up convolution model simple_gpile2.sl, whereby the pile-up was estimated by fitting the continuum to be on average around 25%. The resulting hydrogen column density is cm−2, which is somewhat (≲ 7%) higher than those from Kalberla et al. (2005) and Sala et al. (2007), the former obtained from 21 cm data.
The absorption features in the oxygen region (21–25 Å) were first picked up with a functional model (power law and Gaussian profiles) which yielded Kα lines of O i, O ii, O iii, O vi, and O vii previously observed by Sala et al. (2007), but for the first time, also K resonances of O i and O ii with principal quantum number n > 2. These line identifications at least confirm a two-phase plasma with both neutral (mostly) and highly ionized components. In the present work we have been mainly concerned with the former. In this respect, since neutral oxygen is not the sole charge state responsible for the cold oxygen absorption, the four exposures are then simultaneously fitted with a representative physical model referred to as warmabs, which contains the atomic data for the oxygen isonuclear sequence computed by García et al. (2005). In this step, the hydrogen column density is held fixed at the previously determined value of NH = 1.66 × 1021 cm−2, while the ionization parameters and oxygen abundance are treated as free parameters. A detailed comparison of this model with observations brings forth the inaccuracies of the atomic data, which led us to introduce wavelength shifts in both the Kα line positions and photoabsorption cross sections of O i and O ii.
The scope of this corrective procedure is severely limited by the general poor availability of laboratory data for the oxygen inner shell, but it certainly leads to improved fits and statistics, in particular, to a well-constrained ionization parameter of log ξ = −2.699 ± 0.023 consistent with the absorption lines observed. In order to further improve the fit, the column density is allowed to vary resulting in a more reliable value (NH = 1.38 ± 0.1 × 1021 cm−2) which is within 13% of previous estimates (Kalberla et al. 2005; Sala et al. 2007). Furthermore, the fit yields an oxygen abundance of and ionization fractions of O i/O = 0.911, O ii/O = 0.077, and O iii/O = 0.012. Regarding the oxygen abundance, it may be clarified that the quoted value is given relative to the solar standard (8.83 ± 0.06) of Grevesse & Sauval (1998). If it is rescaled to the recently revised standard by Asplund et al. (2009) of 8.69 ± 0.05, then an ISM abundance (
) close to solar is obtained which is in excellent accord with the averaged value by Baumgartner & Mushotzky (2006), and can be included as further support for the solar abundance revision. Moreover, our ionization fractions are in good agreement with those found by Juett et al. (2004) of O ii/O i ≈0.1 and O iii/O i ⩽0.1.
The functional fit allows us to derive oxygen column densities by means of a curve-of-growth method using the EWs for each of the observed Kα absorption lines, which can then be compared with the values obtained from the warmabs fit. Even though we find by both methods a relatively consistent total column density (5.71 × 1017 cm−2 in the curve of growth and N = 6.41 × 1017 cm−2 in the warmabs model fit), the O i/O ii ratio for the Kα lines is significantly lower (an order of magnitude) in the former. This is attributed to the high saturation level in both lines which leads to underestimates of the measured EWs, and we are therefore confident that the results from the warmabs model fit are more reliable. In the case of O vi, the width of the line strongly depends on the velocity dispersion of the gas. For the velocity dispersion values expected in the ISM (v ∼ 20–200 km s−1), the EWs obtained with our functional fit yield oxygen column densities far above from the FUSE values by Savage et al. (2003). This may be an indication that the O vi observed absorption in the XTE J1817-330 spectra is bound to occur near the source rather than in the ISM.
We find acceptable fits to the observed oxygen K-edge spectrum with a model consisting of atomic oxygen ions if the energies are slightly adjusted. However, it is also of great interest to detect or set limits on features due to other phases of interstellar oxygen and other elements, namely molecules or solids (dust). It is expected that these phases comprise a significant fraction of the total oxygen column in the ISM. Evidence for this includes observed depletions of atomic oxygen along many lines of sight relative to certain elements (Jenkins 2009; Yao et al. 2009; Whittet 2010), together with observations of molecular features in other wavelength bands. An example of a potentially detectable signature of such material is the K resonance of the CO molecule expected at a rest wavelength of 23.21 Å (Barrus et al. 1979). This line is separated from the nearest feature we clearly detect by approximately 100 mÅ, which can be compared with the magnitude of the wavelength shifts we adopt in order to get acceptable fits, the largest of which is 33 mÅ. Therefore, we conclude that the existing atomic absorption data are not sufficiently uncertain to mask the presence of this molecular signature. The search for molecular and solid features in X-ray spectra remains of great interest, both as tests of the atomic data and of models for the ISM.
We believe that the present benchmark has produced an improved version of the warmabs model that can now be used for a more extensive and reliable study of oxygen K photoabsorption toward other X-ray sources. This will be the next step of the current study where the oxygen abundance variations in the ISM are certainly of interest. In a similar fashion, we also intend to evaluate the Ne and Fe edge regions for ongoing physical model refinement.
Part of this work was carried out by Efraín Gatuzz during attendance of the Committee On Space Research (COSPAR) Capacity Building Workshop in San Juan, Argentina in 2011 July, and visits in 2011 August to the Laboratory of High Energy Astrophysics, NASA Goddard Space Flight Center, Greenbelt, Maryland, USA, and in 2012 February–March to the European Space Astronomy Centre (ESAC), Madrid, Spain, the latter funded by the COSPAR Fellowship Program. Warm hospitality and tutelage at these institutions are kindly acknowledged, in particular, from Andy Pollock and Carlos Gabriel at ESAC. We also thank Michael Nowak from MIT for useful discussions relating to the pile-up model and isis implementation.
Footnotes
- 8
- 9
- 10
- 11
- 12
- 13
- 14