ACIS Observer,
I learned from a colleague today two subtle points about spectral analysis
which is probably obvious to many of you, but might interest those with less
experience (like me).
Application of an energy filter, e.g. [energy=500:8000], while useful for
making images, is not in general a good idea for spectral analysis. The
problem is that spectra are of course computed in a channel space (binning the
PHA or PI columns). The first and last channels in a spectrum will be
underestimated if the data has been previously filtered by the ENERGY column,
since the low & high limits (e.g. 500eV and 8000eV) of that filter will, in
general, fall in the middle of a channel.
A similar problem occurs when you run dmgroup with the NUM_CTS option. Let's
say you tell dmgroup to group your spectral channels so there are at least 10
counts in each group, and the first channel in your spectrum with any data in
it is channel #35 (because you've filtered the spectrum below channel 35 ~=
500eV). If your spectrum actually starts at channel 1 (i.e. you ran dmextract
with the usual range [bin pi=1:1024:1]) then the first group produced by
dmgroup will cover 35 channels (1..35). This is of course not correct --
those counts all came in channel 35 and should be modeled as such. A similar
problem can occur at the high end. For example if your data ends at 8keV
(because you've filtered it) but your spectrum runs to channel #1024 (~15keV),
then your last group will span 7keV! At the high end, dmgroup will set the
QUALITY column to "2" to show that an incomplete group was found, and xspec
can be told to ignore this group.
I'm still trying to figure out how best to handle band-limiting the data. It
has been suggested that one refrain from applying _any_ energy filters
(restricting columns "energy", "pi", or "pha") prior to spectral extraction or
grouping. All band-limiting would then be done within xspec by ignoring
channels outside a desired range.
Alternatively, it has been suggested that one restrict the range of channels
used in your spectral file, e.g. 35:800 instead of 1:1024, so that dmgroup
will understand that the first group starts at channel 35 (~500eV) and the
last one ends at channel 800 (~8000eV). I don't yet know if the RMF would
have to be constructed with the same exact set of channels for xspec to
function properly.
Any thoughts you have, or better yet a document, on these points would be
appreciated by me at least, if not by many others.
Regards,
Patrick Broos
-- ==================================================================== Patrick S. Broos, Software Engineer patb@astro.psu.eduDepartment of Astronomy and Astrophysics My office 814-863-7947 Penn State University 525 Davey Lab FAX 814-863-8686 University Park, PA 16802-6305 http://www.astro.psu.edu Group office 863-9550 ====================================================================
This archive was generated by hypermail 2b29 : Thu Dec 13 2012 - 01:00:04 EST