Date: Mon, 09 Aug 2004 17:37:38 +0200 (MEST)
Subject: Re: SDAST meeting


  Hi Stefan,

  sorry to hear that unpleasant virus caught up with you again!
Evidently, we were lacking your input on several areas, but there
are three topics which I can remember affecting your work specificly:

(1) Niels Joergen uses normally the selection "STATUS<256" which
    actually leads to 5% higher Crab normalizations than the
    default empty, translated by your program to "STATUS==0".

    In my opinion, we need to straighten out STATUS selection and
    define clearly which selections we use by default and recommend.
    If I'm not mistaken, Niels Joergens's selection would include
    events with doubtful gain, for example. Could you look again at
    the list of STATUS values and formulate an opinion which ones
    you consider dangerous for spectral/lc extraction?

(2) Niels pointed out, that at the extremes of the energy range,
    the detector is actually very non-uniform, since at the lower
    end events falling onto an anode area of low gain will fall below
    the cutoff and vice versa at the high end. This could in principle
    be modeled - but we'd need to dicuss the details carefully.

(3) We should probably define our estimate of systematic unertainties
    and write that to the spectra you produce. It's not clear to me
    yet if our informatioon should be in an IC table, in hidden parameters
    or in both (use IC if hidden parameters say "default").

  Cheers,
  Peter

P.S. Your answers will probably be read tomorrow ...



To: Stefan Larsson <stefan@astro.su.se>
Subject: Re: SDAST meeting 
Date: Tue, 10 Aug 2004 11:27:44 +0200
From: Tim Oosterbroek <toosterb@rssd.esa.int>

Dear all,

I saw that in your email exchange on of my own pet topics was brought up:
the issue of spectral rebinning.

Since it is my 'favorite' topic I wish to comment on it (although I
do not have many references handy). It is indeed a fact that a lot
of authors in X-ray astronomy use too fine bins when doing spectral
fitting and, as a result of that, get too low reduced chi-squares.
And especially comparing models based on these chi-square values is
rather meaningless.
This has already been investigated quite some time ago (Davelaar 1979 ??
comes to mind) by using Monte-Carlo simulations and the conclusion from
this is that one should not oversample the resolution of the instrument
too much when doing spectral fits. The exact value of the oversampling
can be debated a bit, but roughly one should oversample the resolution
by a factor 2-3 (3 is my own personal favorite).

I thought of bringing up this point when doing the Crab fits, but at the
time I thought it was one step too far (since it does not [significantly]
affect obtained parameter, but only comparisons between different models).

I therefore recommend that one should keep this in mind. In an ideal
world one should provide a recipe in the analysis manual to obtain this
oversampling of the resolution by a factor 2-3. (or even better a tool
to do this).


Regards,

Tim

Date: Tue, 10 Aug 2004 12:57:13 +0200 (MEST)
From: Stefan Larsson <stefan@astro.su.se>
Subject: Re: SDAST meeting
To: toosterb@rssd.esa.int

Hi Tim,

Thanks a lot for your comments. I am eager to learn why this is.
I looked briefly at some Develaar references from that time (GSPC) 
but could not find this point. Please send me the references when
you have time to look them up.

Cheers,

Stefan


Date: Tue, 10 Aug 2004 18:14:09 +0200 (MEST)
From: KRETSCHMAR Peter <Peter.Kretschmar@obs.unige.ch>
Subject: Re: SDAST meeting
X-X-Sender: pkretsch@crab
To: Tim Oosterbroek <toosterb@rssd.esa.int>

  Dear Tim,

  thank you very much for your very valuable input.
I'm embarrassed to have been caught without a good
answer to this very basic question of our trade.
If you can supply a few literature leads, I'd be
happy.

  We should definitely say something about this
in the User Manual. For a start, grppha can work on
the output of spe_pick. But j_src_spectra could
probably do a default binning based on such a requirement
and statistics. SInce in the OGIP standard binning
is only a flag, a user could always change our choices
again.

  Regards,
  Peter