For the fits discussed within this paper, we have used a combination of binning by signal-to-noise and uniform channel binning. Binning was employed to allow the use of statistics, which are faster to minimize (this is a non-trivial concern when using the warmabs model, which is an extremely slow to evaluate code). Furthermore, adding of data and binning of channels can serve to average over systematic calibration uncertainties. However, any non-uniform binning, e.g., a signal-to-noise criterion, can introduce biases in line fits.
For our data, the signal-to-noise per channel rapidly varies at energies keV, and there is no (small) set of uniform channel binnings that achieves adequate signal-to-noise in these channels. Thus, we choose mixed criteria (using the ISIS group function) that ensures both a minimum signal-to-noise and a minimum number of channels in the binned data. In practice for these particular data, only the minimum channel criterion applies above 1keV, as this is sufficient to ensure signal-to-noise in all of these channels.
Our fits indicating an outflowing wind with a velocity in the PG1211+143 frame of 17300 kms ( ) are primarily being driven by the lines of Helium-like and Hydrogen-like Ne, Mg, and Si. To ensure that these lines are not being biased by our binning (although five of the six primary lines in our fits are within the uniformly binned portion of the spectrum at energies keV), we have also performed a “blind line search” of the data. We grid the HEG data to the MEG bins, and combine all spectra, but otherwise do not perform any further channel binning. We use the same continuum model as discussed above, namely an absorbed disk plus powerlaw spectrum with exponential cutoff, and include line emission from the Fe region. We then loop through the spectra, adding one line at a time which is allowed to freely range between emission and absorption. The initial line fit is constrained to a narrow range of wavelengths (16 MEG channels, i.e., Å), but all possible wavelength bins are searched. The line with the greatest change in fit statistic is retained. After each step, all continuum and line parameters are refit (within the constraints of the existing wavelength region of the added line). This process is repeated (51 times in Figure 10, with the first 24 found lines listed in Table 7). As the goal is to identify candidate lines, we do not calculate confidence intervals for the final lines.
The putative NeX line is our single most significant residual, with the remaining 5 lines from H-like and He-like Ne, Mg, and Si all falling within the 17 most significant residuals. Additionally, there are three other residuals that might be associated with a outflow. These, however, fall within a very low signal-to-noise region of the spectrum. Several residuals are broad, and are undoubtedly modifying the continuum fit. Several could be spurious noise features. The rest remain unidentified. However, this blind search highlights those features that are driving the XSTAR and warmabs models to identify an outflowing component in the rest frame of PG1211+143 .
As a further diagnostic of possible absorber components, we take the 1st-order Chandra -HETGS counts, and using the 9 potential H- and He-like lines identified in Table 7, stack the data into cosmological rest frame velocity bins. The results of this stacking are also shown in Figure 10 (note that in this procedure, not every bin is statistically independent from one another, as counts can be reused for different ions). The strong feature at a velocity in the rest frame of PG1211+143 of 17300 kms ( ) is apparent. We also indicate the velocities of the absorber components suggested by the analysis of Pounds et al. (2016a). There is a feature near the ( ) velocity found by Pounds et al. (2016a), but we have not found any single line-like residual at such a velocity. It is possible, however, that such an absorption component, if real, only manifests significantly in a stacked analysis.