[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Tracks with no hits.
Hi Richard,
Great! Thanks for jumping on this so quick. I will take a look at it
and let you know if it looks fixed for me too.
Regards,
-David
Richard Jones wrote:
> Dave and all,
>
> This seems to be an optimization in the core tracking code of geant
> that causes mother volumes to appear empty of daughters under certain
> special conditions (floating-point rounding errors) that are usually
> rare, but apparently not so rare when all of the interactions are
> turned off and you are tracking with a magnetic field! Bravo for
> finding such an easy way to reproduce it. I have commented out the
> optimization line in gtrack.F that causes the fault, and the tracking
> seems to run at essentially the same speed, without the errors. I
> think we should adopt the fix permanently, but please check it first
> and make sure it completely fixes the problems you found. The patched
> gtrack.F can be found in the HDGeant source directory under the main
> trunk in svn. I have incorporated it into the Makefile.orig that I
> use to build the hdgeant executables. Dave, you will have to
> incorporate it into the BMS make system.
>
> -Richard Jones
>
>
> David Lawrence wrote:
>>
>> Hi Richard,
>>
>> I have run across a mystery that I think you may be the only
>> person who can solve it. Recently (~ 1 month ago), I ran a simulation
>> to map out the efficiency and resolution for charged particle
>> tracking using the latest geometry and tracking code. One of the
>> things that showed up was an odd structure in the efficiency map (in
>> ptot and theta) where there was a several percent loss in efficiency
>> in a thin band that appears to be for a constant transverse momentum
>> (though now, I'm sure that is actually the case). Mark Ito repeated
>> the exercise with his independent fitter and found the same result.
>>
>> I ran a few events with the particle gun aimed at the
>> low-efficiency area and I found that for some tracks (several
>> percent) there were no hits at all. I turned on the
>> DMCTrajectoryPoints which record the track parameters at every call
>> to gustep, including the radiation length of the material they are
>> passing through. I could see that the particles were going through
>> the detector, but for the problem events, the radiation length was
>> always very large (30k cm) indicating that it never left the gas. For
>> good events, I could see it getting smaller occasionally, presumably
>> as it passed through the straw tube walls. I'm wondering if this is a
>> MANY/ONLY problem, but I'm not sure where to begin in order to figure
>> it out. Oh, and I also saw this same thing for some forward going
>> tracks that were passing through the FDC.
>>
>> I've attached a control.in file that shows the problem on events
>> 6, 25, 32, 46, ... I have verified that I see the problem on event 6
>> on both Mac OS 10.5 and Fedora 8. I am using the source code tagged
>> for a special study that can be found here:
>>
>> https://halldsvn.jlab.org/repos/trunk/Studies/2009.02.track_resolutions/src
>>
>>
>> This contains the complete source tree including hdds and HDGeometry
>> and HDDM so you should be able to reproduce exactly what I'm doing.
>>
>> Could you maybe look into this when you get a chance?
>>
>> Regards,
>> -David
>>
>
--
------------------------------------------------------------------------
David Lawrence Ph.D.
Staff Scientist Office: (757)269-5567 [[[ [ [ [
Jefferson Lab Pager: (757)584-5567 [ [ [ [ [ [
http://www.jlab.org/~davidl davidl@jlab.org [[[ [[ [[ [[[
------------------------------------------------------------------------