Things to know, or to look out for, when analysing XRT data. See also the SDC XRT Digest Page which links to various useful documents and the XRT threads page for step-by-step guides to analysing the data.
See also the XRT Science analysis digest page
Sources which are heavily absorbed (columns of about 1022 cm-2 and above) can show a 'bump' in the spectra at low energies (between 0.4-1 keV) and a turn up at the very low energy end. The bump is only present in event grades 1 and above (although the redistribution tail - where the data turn up - is always present). If a source is in WT mode and absorbed, we recommend extracting both grade 0 and grades 0-2 spectra and comparing them. If the bump is present, then grade 0 only should be used, since the current grade 0-2 RMF does not model the bump well.
Even moderately absorbed source (few 1021 cm-2) can show a turn-up in the data/model ratio plot below ~0.4 keV. Such features are likely not intrinsic to the source, but also due to slight RMF redistribution modelling issues, compounded by traps (see below), in this mode.
Charge traps are faults in the Si crystalline structure of the CCD, caused by radiation damage, which can hold onto ("trap") some of the charge released by X-ray photon interactions. These traps lead to a broadening of the measured width of lines, possible residuals around the oxygen edge (0.545 keV) and a turn-up or -down of the data/model ratio for relatively unabsorbed source below ~0.35 keV.
Version 3.8 of the software (HEASoft 6.11) and later use a trap-corrected gain file to process the data, which improves the spectral resolution. The corresponding RMF/ARF files (see table) are included in the CALDB.
If the data/model ratio rapidly turns up or down below ~0.35-0.4 keV (depending on the epoch of the observation), this region of the spectrum should be ignored.
Power spectra created from light-curves extracted at the WT mode time resolution of 1.766 ms show a drop-off at high frequencies (above about 50 Hz; see Figure 4 below), which is related to the read-out method. WT data are clocked out into the serial register 10 rows at a time. Events which would be classed as grades 1, 3 and 5-12 in PC mode and which happen to fall across this 10-row boundary will be split into 2 events of lower energy; this can affect about 2% of the events from a typical powerlaw-like source. Because WT grade zero events include some vertically split events (those classified as grades 1 and 3 in PC mode), this drop-off cannot be avoided by grade selection.
This fall off in power has been confirmed by detailed simulations of the WT mode readout scheme. The simulations include a complete description of the charge-cloud spreading associated with X-ray interactions and resultant 10 row binning induced event splitting. As Figure 4 (below) shows, the effect is energy dependent, tracking the linear absorption depth of X-rays in the Silicon (as shallow interactions form predominantly single pixel events, while deep interations form multi-pixel events which are susceptible to splitting at the 10 row boundaries).
In addition, if a source suffers from pile-up, the power at all frequencies is suppressed and the overall level falls below a mean value of 2 (as expected for Poisson noise when the powers are normalised following Leahy et al), as shown below.
Occasionally, if a cosmic ray hits the baseline, it corrupts a line in the bias map. When baseline subtraction occurs, this row becomes apparently "hot". All the events will be rejected during the pipeline processing, so the only time when this row becomes visible is in an exposure map. An example is shown below.
The X-ray background detected is a combination of the Cosmic X-ray background, electronic noise and particles (cosmic rays, solar protons etc.) The WT background spectrum can be quite variable, sometimes showing a strong Nickel line at 7.5 keV and sometimes a dip between 2-3 keV. Examples are shown below.
On 2007-08-30, the XRT CCD substrate voltage was raised from 0V to 6V as part of a plan to improve its performance at the relatively high temperatures at which it has to operate (up to -50C).
Experimental data taken with the flight spare CCD at the Leicester calibration facility had shown that by raising the substrate to 6V, the CCD dark current level would be reduced. This has been confirmed in orbit: Figure 7 below shows a measure of the thermally-induced dark current (calculated as the mean of the noise peak minus the mean of the bias level) as a function of CCD temperature, with the CCD substrate voltage at both 0V (blue circles) and 6V (red triangles). This shows the benefit of raising the substrate voltage: for the same level of dark current, the CCD can operate 3 to 4C higher with the substrate voltage at 6V compared with 0V, and thus can provide useful scientific data at hotter temperatures.
However, by raising the substrate voltage in this way, we expect the CCD depletion depth to decrease slightly, which will cause a small drop in Quantum Efficientcy (QE) at high energies. Observations of calibration sources before and after the substrate voltage change reveal a drop in QE of ~8% at 6 keV (Godet et al. 2007), consistent with Monte-Carlo simulations of the RMF.
Residuals of about 10% around the Si and Au edges (around 2 keV) were previously apparent in high signal-to-noise spectra. Improved calibration files have noticeably decreased such residuals. Note that, if the energy scale is off by 10 eV or more, residuals may still be seen.
See this short calibration document on energy scale offsets.
Please note that the bias maps obtained between days 200 and 214 (19 July - 2nd August) of 2006 were incorrect. This means that there is a potential that the bias map values were too low, meaning that photon energies could appear too high - i.e., a gain offset. It also caused a temporary upsurge in hot pixels.
Before version 2.4 of the software (release date of 2006 April 26), there were two versions of the Ancillary Response File (ARF) which could be generated for the Photon Counting (PC) and Windowed Timing (WT) modes: empirical (the default when running xrtmkarf) and physical (obtained by typing xrtmkarf inarffile=none) files. Note that this is no longer the case for more recent versions of the software. The empirical ARFs were produced by comparing the observed Crab off-pulse spectrum (between phase 0.5 and 0.8, chosen to limit pile-up effects) with standard parameter values (Γ = 2.1; normalisation = 9.2 photon keV-1 cm-2 s-1 at 1 keV) and "tweaking" the ARFs until these match as closely as possible.
There are certain differences between the ARFs which users should be aware of. In general, the physical ARFs do not model the gold edge (i.e. between 2-4 keV) very well. The empirical ARFs should not be completely trusted below about 0.6 keV.
The uncertainties at lower energies can manifest themselves as an apparent excess absorbing column of up to 1021 cm-2.
Examples are shown below.
Work on both the ARFs and the corresponding response matrices (RMFs) is ongoing. We caution users not to apportion high significance to observed spectral features unless a thorough investigation (using different ARFs) has been performed.
The latest documentation relating to the ARFs and RMFs can be found here.
On 2007-08-11, Swift experienced an attitude control problem, which led to the observatory being offline for a number of weeks (see GCN Circ. 6760, 6797 and 6946 for more details. Because of this problem, the accuracy of Swift's slewing was poor during September and October 2007. Versions 2.8 onwards of the Swift software filter out times of particularly bad attitude. However, earlier versions of the software do not, meaning that there can be apparently multiple images of the source in sky coordinates (see Figure 11 below). Data obtained between 2007-08-11 and 2008-10-19 should be treated with caution.