2. Contenido
2.21 AUTORIZACION Y ENTREGA DE MEDICAMENTOS
2.2.5 Source Photometry
Once the source detection has run to completion the pipeline computes photometry for the X-ray point sources. The photometry calculations are described in detail in Laird et al (2009); what follows is a brief outline of the method. Photometry calculations are attempted in all four energy bands, regardless of whether they have been detected in all the energy bands. The only exceptions are sources with net source counts ≤ 0 in an energy band; these sources are assigned an upper flux limit based on the local background. To estimate the source flux, counts are extracted from the 90% EEF while the scaled background counts are calculated in identical fashion to those used in source detection. The scaled background counts are subtracted from the counts within the 90% EEF to give the net source counts. Sources whose 90% EEFs overlap are flagged as their source counts are likely to be contaminated by one another.
Source fluxes are calculated in 4 energy bands commonly used in X-ray analysis; 0.5- 10 keV (full band), 0.5-2.0 keV (soft band), 2.0-10.0 keV (hard band) and 4.0-10.0 keV (ultra-hard band). Counts are converted to flux estimates assuming a Γ = 1.4 power-law spectrum. For the full, hard and ultra-hard bands the spectrum is extrapolated from 7 keV to 10 keV. The Bayesian flux estimation method of Laird et al (2009) is used to correct for Eddington bias (Eddington, 1940).
P (S|N, B) ∝ L(N|S, B)π(S), (2.1)
where P (S|N, B) is the posterior distribution function for the source count rate, π(S) is the prior distribution function of source count rates and L(N |S, B) is the Poisson likelihood of obtaining the observed data given the source rate and background (see Equation 2.2).
L(N |S, B) = (S + B)
N
N ! e
−(S+B) (2.2)
The prior distribution of X-ray fluxes π(S) is modelled using a broken power-law (Geor- gakakis et al., 2008). This compensates for the Eddington bias by modelling the faint source distribution with a shallower slope than bright sources. This Bayesian method is much better at estimating faint source fluxes which can be overestimated by the classical method by up to a factor of 2 (see Laird et al. 2009).
2.3
X-ray Spectral Extraction
2.3.1 X-ray Spectral Extraction Overview
In this thesis, the nature of X-ray sources is studied in greater depth by fitting models to their X-ray spectra. In particular we use the spectra to study AGN demographics, estimate obscuring column densities and find better estimates for X-ray fluxes and luminosities. Spectral extraction is a non-trivial process as X-ray sources are observed with multiple
2.3 X-ray Spectral Extraction 54
overlapping obsIDs (with a variety of aim points and roll angles; see Section 2.2.2), with a single source possessing a variety of PSF profiles across the constituent obsIDs. To obtain the most accurate spectra possible the source counts must be extracted from each obsID using the respective PSFs. To facilitate this process, X-ray spectra were extracted using the ACIS EXTRACT IDL package (Broos et al 2010). Developed at Penn State University, the ACIS EXTRACT package is designed to deal with large numbers of sources across multiple overlapping obsIDs. The X-ray data products used as inputs by this package were created using the X-ray data reduction pipeline (see Section 2.2.2). The spectra are extracted directly from the level 2 event files without any binning according to event energy (unlike the reduced image data). Binning is only applied to the spectrum for the purpose of spectral fitting. Spectral extractions of sources in the CDFS 4Ms field were performed by myself, while the spectral extraction of EGS 800ks sources was carried out by Murray Brightman. All spectral fitting of X-ray detected sources in this thesis was carried out by myself.
2.3.2 X-ray Spectral Extraction Pipeline
Spectra were extracted following the recipe of Patrick Broos from the ACIS EXTRACT manual. First the ae make catalog tool uses the MARX ray-tracing tool to generate PSFs for each obsID. For the purposes of spectral extraction the 95% EEF is used to ensure the PSFs enclose and include as many of the source counts as possible (while maintaining a sensible extraction aperture size). The extraction region is iteratively reduced in size if the source is within a crowded region, to avoid contamination of the source counts by neighbouring sources. The PSFs are all created at a “primary” energy of 1.5 keV to minimise variation of PSF size relative to the high-end and low-end of the energy range. PSF ranges are pixelated as opposed to using a simple geometric ellipse to maximise their accuracy. ACIS EXTRACT uses the aspect solution files for each obsID to determine the orientation of the telescope as a function of time to ensure the source dithering is identical to the original observation. The FITS keyword RAND SKY is read to determine how much blurring ACIS EXTRACT should apply to the PSF images in order to account for pixel randomisation applied to the level 2 event file during the image data reduction (see Section 2.2.2). The ae make catalog tool also constructs circular masking regions of size 1.1 times the 99% source PSF; these are to be used later in the background extraction.
Next the ae standard extraction tool is used to extract the spectrum of each source. Using the 90% EEF extraction regions defined by ae make catalog, source counts are ex- tracted from the level 2 event files of each obsID and subsequently filtered by photon energy to produce the source spectrum (using the dmextract CIAO tool). To calculate the background spectrum, the masking regions calculated by ae make catalog are applied to the level 2 event file. Annuli are then applied around each source starting at the edge of the masked region and are increased incrementally until a minimum of 150 background counts have been enclosed. The background counts are then scaled according to the size