[These notes are unedited and freely paraphrase the discussion; corrections are welcome. SD]
T Burnett, B Berenji, M Hirayama, J Chiang, D Band, D Davis, S Digel, J
Ballet, J Peachey, E Winter, C Shrader, T Stephens
SD - Some GSSC members will be coming to SLAC for discussions and work on the
Science Tools on April 18-20. This will not be in parallel or before or
after any other meeting, and Toby will be able to come down for it.
No news
JC - 3 developments since we last met - one is implementation of the
hierarchical summing scheme inspired by Andy. No difference in TS is seen
for small number of events but yes for large numbers of events. Fixed bug in
PyLikelihood related to TS. [The 3rd was related to gtfindsrc, I think. I
didn't get it in my notes.]
JB - Serious bug in PyLikelihood?
JC - Happens in binned analysis in PyLikelihood. If you recall I implemented a
scheme to speed up the implementation - after subtracting a source from a model
renormalized the counts in all of the other bins - the scheme currently works
correctly only for unbinned analysis. The member functions in Pylikelihood were
not working correctly - just crashing.
JB - That is related to the fact that in binned likelihood you cannot access the
number of events predicted for each source
JC - That is right - you can do it but not through the swig-exposed interface,
which was part of the problem
JB - Up to now we have not used the TS with binned likelihood but certainly we
would have one day
SD - Binned Healpix likelihood is something the Diffuse group has been
discussing. Andy has pointed out that binned analysis is the way to go for
large numbers of photons and HEALPix is the only practical way to handle
convolutions with large area/small pixel maps
TB - I've been working on that...
JC - Pat has been working on that...
SD - Will make sure Andy is in touch with them - thanks
DD - Related to gtbin - [audio problem]
JP - We have a newcomer here in the Science Tools meeting, Eric Winter, who will
be starting on cleaning up things related to binning in gtbin (Bayesian block
binning)
EW - Hello to everybody that I didn't meet yesterday - got the Science Tools to
compile yesterday - should be working soon on contributing to the development
MH - Nothing very new - we are still working on the A4 tool, which is now
gtpspec. At this point, the central engine is working. We are trying to make
the output right; the FFT tends to have lots of bins; we cannot do everything as
before in terms of the plot window - want to have a good summary and
representative graphical output
SD - Execution times for typical data sets?
MH - Don't know yet - will do soon
SD - PulsarSpectrum update - Max has got a model of timing noise implemented
as an option in PulsarSpectrum; the time-out-of-range problem for simulations
that use a pointing history file has not actually been solved, but he is working
on it.
[unrelated note]
RD - Anyone write up their symposium contribution yet? It barfs on the
makeTitle macro
DB - Yes, I have; I had to edit some lines; AIP macros have given me problems in
the past. I don't worry about the details
RD - I got the body of the thing to print but you have to emit the page without
the title.
SD - Richard has been implementing a source to simulate microquasars.
He's joining us today to provide some background and details.
RD - Many people in this virtual room know more about microquasars than I do -
won't get into details here - Galactic XRBs with radio jets - may be related to
quasar phenomenon, with intrinsically much faster time scales than AGN
variations. In the Data Challenge and SC1 run we had just modeled sources like
LS I 61+303 and LS5039 as periodic sources - the intensity of the stellar
radiation field at the compact object varies periodically - implemented via
Jim's PeriodicSource. These are not looking like typical microquasars LS I
61+303 is a binary pulsar actually and LS5039 is 'under attack'. No other
microquasars have been seen in gamma rays, so simulation is going out on a limb.
The model is a disk-jet system - outburst cycles - implemented as something like
PeriodicSource but also episodic for outbursts - turning the jet on and off.
The features of the model include on/off cycle for the jet. Also for LS5039
HESS found spectral variations with phase, and this is allowed for in this
model.
I made this a compiled class in celestialSources and as such it follows the
properties of spectrum classes - it supplies the energy and interval to the next
photon. I thank Toby lots and lots for advice about implementation - including
the on-off cycles.
JP - Is that the one on the agenda page?
[Confluence access problem - James not yet in the group that has access]
RD - Idea is to make cuumulative distribution - and draw from that
distribution. You take the difference - amounts to finding the zero of a
function. Toby worked this out. In essence, Jim had implemented this tabular-ly
in PeriodicSource - this is an extension that can handle on/off periods. Toby
had the idea there that if the next interval was going to take it to a clock off
period just advance the interval to the next one that has clock on.
I tested it and it is not a dog - a year takes about 30 CPU seconds. I have
been working with Robin Corbet on modulation stuff - he has tools for power
spectrum evaluation - sent him light curves and he looked for orbital
modulations and whether the jet-on cycles were visible (they weren't). I did
put a fair number of fluctuations in this - you can smear when a jet turns on
and for how long, so it is not strictly periodic. If I turn off the
fluctuations, I see it well in the phase plots. In working with Robin I figured
that one of the things we could do was compare with DC2 sources - see link. The
comparison is good - maybe suspiciously too good - like difference in number of
events of only 1 in 4000. But the model seems to be going well in that sense.
I set up sources for the next obsSim run; I have a Confluence page linked into
this page. The way that you use this thing is that you have an XML file driven
by the sources. You give it a name and a long list of parameters, like 15. At
first I took sort of a standard here-are-15-numbers scheme. That was awkward.
I ended up swiping the code from the sun and moon sources - key-value pairs.
The housekeeping that I have to do is clean up the documentation - has old
string of numbers approach. The doxygen page did not seem to come out - I think
it is ok, though.
TB - I can't take credit for the root finder, which comes from Numerical
Recipes.
RD - For the obssim model, this source has proved handy for modeling outbursts -
we'll have the standard steady sources - we'll have like GX 339-4, which has a
~1-year outburst period. I implemented it with 3 ten-day bursts. Another
source has a 1-day jet cycle.
TB - About doxygen, I think you are missing release.notes, which Windows doxygen
looks for and expects to find
RD - I have one now - and microQuasar has not been in a release since I added
release.notes. So we'll see. One of the things that I would have liked while
putting several sources together I added debug output. It would be nice for it
to know who it was. So the event knows the name. Bascially, I name the sources
in the XML file. Is there any way that the instance of the source can know the
name that it was given in the XML file
TB - I think that is outside its realm but if that is useful we could figure out
how to make it possible.
RD - I would get messages when rtsafe (rootfinder) was not converging but did
not know (without binary search) which microquasar it was complaining about.
SD - Workbook page will derive from the doxygen page?
RD - Yes, although that is a leading question
JC - No obs sim news on my own but Toby has been interacting with Nicola about
the GRB source
TB - It is a situation that has been a little vexing for me because Nicola wants
to test GRBs when he is running only GRBs and may have intervals for years or
more - not a situation that we normally have in sky models. We have tracked
down errors related to handling long intervals. Other issue that I might bring
up is solar-centered frame. I think that I finally have it right. We take a
MapSource source and transform (rotate) events to sun centered system. There
are several ways of making transformations to do that but only one is right -
and it took me a while to get there.
SD - DataMind does not have support yet for developing the GUI that Marco and
Riccardo have described; there's no news that I'm aware of about the GUI.
RD - Francesco's proposal was to have them develop a prototype and then pursue
funding for a full-up model. Also, Riccardo was worried about the general scope
of this thing - much more comfortable with the idea of supporting something for
the collaboration, a finite, bounded set of people -vs a polished product for
the world
JP - No infrastructure news
SD - Does the GSSC have a set of user support tools for Cycle 1 proposals?
DB - GLAST-spec (XSPEC Web interface) will become public in a week or so. Also
have sensitivity estimator. The tool behind that is Toby's but for server
reasons we are not able right now to hook up the tool to the interface. Steve
and Julie are working through what we want to present to the public. May post
maps that show sensitivity to different time scales. A lot of text is getting
ready to go out - Seth has been sending comments about the handbook.
SD - I'm not sure when we will meet next; most likely not until after the collaboration meeting.