02.02.2014 Views

SPEX User's Manual - SRON

SPEX User's Manual - SRON

SPEX User's Manual - SRON

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

5.5 The proposed response matrix 101<br />

The bin width constraint derived here depends upon the dimensionless curvature of the effective area<br />

A/Ej 2A′′ . In most parts of the energy range this will be a number of order unity or less. Since the<br />

second prefactor E j /FWHM is by definition the resolution of the instrument, we see by comparing (5.58)<br />

with (5.47) that in general (5.47) gives the most severe constraint upon the bin width, unless either the<br />

resolution gets small (what happens e.g. for the Rosat PSPC detector at low energies), or the effective<br />

area curvature gets large (what may happen e.g. near the exponential cut-offs caused by filters).<br />

Large effective area curvature due to the presence of exponential cut-offs is usually not a very serious<br />

problem, since these cut-offs also cause the count rate to be very low and hence weaken the binning<br />

requirements. Of course, discrete edges in the effective area should always be avoided in the sence that<br />

edges should always coincide with bin boundaries.<br />

In practice, it is a little complicated to estimate from e.g. a look-up table of the effective area its curvature,<br />

although this is not impossible. As a simplification for order of magnitude estimates we can use the case<br />

where A(E) = A 0 e bE locally, which after differentiation yields<br />

√<br />

8A<br />

Ej 2 = √ 8 d lnE<br />

A′′ d lnA . (5.59)<br />

Inserting this into (5.58), we obtain<br />

5.4.6 Final remarks<br />

∆E<br />

FWHM < √ 8<br />

( d lnE<br />

)(<br />

d lnA<br />

E j<br />

FWHM<br />

)<br />

λ(R) 0.5 N −0.25 . (5.60)<br />

In the previous two subsections we have given the constraints for determining the optimal model energy<br />

grid. Combining both requirements (5.48) and (5.60) we obtain the following optimum bin size:<br />

∆E<br />

FWHM = 1<br />

1<br />

w 1<br />

+ 1<br />

(5.61)<br />

w a<br />

where w 1 and w a are the values of ∆E/FWHM as calculated using (5.48) and (5.60), respectively.<br />

This choice of model binning ensures that no significant errors are made either due to inaccuracies in the<br />

model or effective area for flux distributions within the model bins that have E a ≠ E j .<br />

5.5 The proposed response matrix<br />

5.5.1 Dividing the response into components<br />

When a response matrix is stored on disk, usually only the non-zero matrix elements are stored and<br />

used. This is done in order to save both disk space and computational time. The procedure as used in<br />

XSPEC and the older versions of <strong>SPEX</strong> is then that for each model energy bin j the relevant column of<br />

the response matrix is subdivided into groups. The groups are the continuous pieces of the column where<br />

the response is non-zero. In general these groups are stored in a specific order: starting from the lowest<br />

energy, all groups of a single energy are given before turning to the next higher photon energy.<br />

This is not the optimal situation neither with respect to disk storage nor with respect to computational<br />

efficiency, as is illustrated by the following example. For the XMM/RGS, the response consists of a narrow<br />

gaussian-like core with in addition a broad scattering component due to the gratings. The FWHM of<br />

the scattering component is typically 10 times larger than that of the core of the response. As a result,<br />

if the response would be saved as a ”classical” matrix, we would end up with one response group per<br />

energy, namely the combined core and wings response, since these contributions have overlap, namely in<br />

the region of the gaussian-like core. As a result, the response becomes large, being a significant fraction<br />

of the product of the total number of model energy bins times the total number of data channels. This<br />

is not necessary, since the scattering contribution with its ten times larger width needs to be specified<br />

only on a model energy grid with ten times fewer bins, as compared to the gaussian-like core. Thus,<br />

by separating out the core and the scattering contribution, the total size of the response matrix can be

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!