| [FIXED] Double tracks/double MC match issue [message #15819] |
Fri, 31 January 2014 08:16  |
Klaus Götzen
Messages: 293
Registered: June 2006
Location: GSI
|
first-grade participant |
From: *gsi.de
|
|
Hi,
I took another look to the track doublers and the multiple MC match (multiple tracks point to the same MC truth object) issue for release dec13 and a current trunk (rev 23404).
Therefore I ran the standard macros in macro/run and afterwards a modified ana_complete.C
The subroutine to count the doublers is essentially a nested loop to compare each track pair in the charged list and looks like this:
double d=0.0001;
for (int i=0;i<l.GetLength()-1;++i)
{
for (int j=i+1;j<l.GetLength();++j)
{
TLorentzVector dl = l[i]->P4() - l[j]->P4();
bool chkmc = (l[i]->GetMcTruth()==l[j]->GetMcTruth());
bool chktrk = (fabs(dl.X())<d) && (fabs(dl.Y())<d) && (fabs(dl.Z())<d) && (fabs(dl.E())<d);
if (chkmc) n_smc++; // double MC reference
if (chktrk) n_strk++; // double track
if (chktrk && chkmc) n_both++; // both at the same time
}
}
I summed these counters for 100 psi' -> J/psi (mu+ mu-) pi+ pi- events (expected total primary #tracks = 400) and the results were:
dec13 -> Trk:470 Dbl trk:66 Dbl MC:170 Both:66
trunk -> Trk:324 Dbl trk:0 Dbl MC:27 Both:0
So for release dec13 we see many more tracks (+150 compared to current trunk), which obviously cannot be explained purely with the track doublers, which are only counted to be 66. The number of double MC references is also much higher. All double tracks seem to point to the same MC object (at least this is consistent...)
Concerning MC truth tree matching, for my J/psi example, I got the following number of events with more than 1 truth matched composite (J/psi and psi')
dec13 -> #ev(mult J/psi):62/100 #ev(mult psi'):27/100
trunk -> #ev(mult J/psi):11/100 #ev(mult psi'):2/100
My conclusion is (maybe limited due to the low number of events):
- The double track issue seems to be solved, since no double tracks are present in the current trunk (in this example).
- The MC doublers issue also is improved, but it still might spoil efficiency calculations due to too large number of truth match composites, perhaps faking a too good efficiency.
- The dec13 release basically seems unusable for anything with this huge problem. It might be a good idea to either remove it or patch it in the sense, that the track doublers vanish.
Best regards,
Klaus
[Updated on: Fri, 07 February 2014 19:48] by Moderator Report message to a moderator |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Re: Double tracks/double MC match issue [message #15828 is a reply to message #15821] |
Fri, 31 January 2014 19:36  |
Gianluigi Boca
Messages: 177
Registered: March 2004
|
first-grade participant |
From: *pv.infn.it
|
|
Lieber Klaus (und Stefano)
the version of the Pattern Recognition offline presently in the trunk (say, version 23652) is the one I recommend to use.
The story is that before the last meeting, during a period of modification of my code, I left by mistake an "experimental" version of the PR producing ghost tracks (thanks Maria for telling me the problem).
So, as soon as svn at GSI has worked again (last January 10th, approximately) I put everything in order again in svn.
THEREFORE :
if you (Klaus) are analysing J/Psi data with the present trunk version and have still problems BITTE tell me ASAP. In that case send me you SIM, DIGI, RECO macro and I will certify if the problem is the PatternRecognition ODER NICHT.
Danke Schoen und Tschüß
Gianluigi
| Klaus Goetzen wrote on Fri, 31 January 2014 16:37 |
Hi,
I just did an 'svn up', but I didn't see new code in tracking except in macro/gem and geometry. I think my trunk was sufficiently up-to-date.
Best,
Klaus
|
|
|
|
|
|
|
GSI Forum
Search
Help
Members
Login
Home
