Home » PANDA » PandaRoot » General » Psi3770 study with run981 tpc mode
| Psi3770 study with run981 tpc mode [message #12305] |
Wed, 20 July 2011 15:01  |
donghee
Messages: 385
Registered: January 2009
Location: Germnay
|
first-grade participant |
From: *dip.t-dialin.net
|
|
Hi,
I'm now testing latest production of psi3770 at Gridka.
The data corresponds to run981 with tpc mode and just look 84500 events from totally produced 0.1M evt. PndKinVtxFitter failed and cannot go through whole 500 events.
I require following cuts.
At least 3 positive particles and also 3 negative particles since psi3770 will decay with 6 final states.
E/p < 0.85 due to electron tracks (Energy from EMC info.)
Rejects bad reconstructed track with checking GetMicroCandidate().GetChiSquared() == -1 in the track lists.
In figure tracks.gif, total charged and reconstructed tracks from 84500 events for psi3770 data is shown.
In D_mass.gif, you can find D_mass distribution with three different categories.
First column is raw D+ and D- distribution, and second column shows D+ and D- mass distributions after applying a PndKinVtxFitter.
Third one is the plots matched with MC true information, if 6 reconstruced(candidate tracks) tracks are associated with MC true tracks IDs and without overlapping and double counting, then assigned as a MC true matched events.
The mean position of D+ and D- masses are slightly shifted from norminal position.
The resolution of D+ and D- vertex are shown in Resolution.gif.
In order to get precise vertex resolution, one need to apply sidebin subtraction for signal region, because bg+sig cannot be separated.
So, I took the data only region within 1.8693 +/-0.05 from fitted D+ and D- mass distribution, and then got the vertex position for that.
Subtraction of true with reco for vertex position X, Y, Z of D+ and D- are plotted in this plot.
I have still 0.13 cm resolution for x, y direction.
Even if I take only MC true matched events, the resolution doesn't change so much.
Psi_distribution.gif shows psi3770 reconstruction with using signal region of D+ and D- mass. Here once again vertex fits for psi3770 have been applied in second, and in third coloumn psi3770 with MC true matched events are additionally plotted to get the feeling for estimation of Psi3770 efficiency.
When I required the MC true matching to reconstructed track, some events have double or even more counting at MC true matching, since sometimes two or three reconstructed tracks are induced essencially same origin of one MC track.
If I have those event, I throw away those events. I need still some tests to reduce those multi-counting problem in my analysis.
PID would be real solution.
So, efficiency of Psi3770 is extremely low with this approach.
If I try same analysis with 4CFitter, I couldn't achieve until finishing complete analysis with 500 events, simply run crash with 4Cfitter for psi3770.
If you have some other suggestions and items, which we need to check in psi3770 data, please don't hesitate make an advice.
Best wishes,
Donghee
I found D_mass
|
|
|
|
| Re: Psi3770 study with run981 tpc mode [message #12313 is a reply to message #12305] |
Thu, 21 July 2011 10:14  |
StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
|
first-grade participant |
From: *to.infn.it
|
|
Hi,
| Donghee Kang wrote on Wed, 20 July 2011 15:01 |
Hi,
I'm now testing latest production of psi3770 at Gridka.
|
For the nth time, you are not using gridka, you are using our pandagrid (I suppose).
| Quote: |
The data corresponds to run981 with tpc mode and just look 84500 events from totally produced 0.1M evt. PndKinVtxFitter failed and cannot go through whole 500 events.
|
What do you mean by "failed"? Has it crashed, or simply not produced good results?
| Quote: |
E/p < 0.85 due to electron tracks (Energy from EMC info.)
|
You don't have electron tracks in the data sample, then you can take out this selection which decreases the efficiency. Once you used ideal pid you use mc to select your particles.
| Quote: |
Rejects bad reconstructed track with checking GetMicroCandidate().GetChiSquared() == -1 in the track lists.
|
These are tracks with problem in the kalman, but maybe they are not bad reconstructed, I would use them.
| Quote: |
In D_mass.gif, you can find D_mass distribution with three different categories.
First column is raw D+ and D- distribution, and second column shows D+ and D- mass distributions after applying a PndKinVtxFitter.
|
How are you calculating prob(chi2)?
| Quote: |
The mean position of D+ and D- masses are slightly shifted from norminal position.
|
Have you fitted the peak? How much is the center and the sigma?
| Quote: |
The resolution of D+ and D- vertex are shown in Resolution.gif.
In order to get precise vertex resolution, one need to apply sidebin subtraction for signal region, because bg+sig cannot be separated.
|
If you use mc pid you don't have physical background but only the one from reconstruction tails. You should not take out such bg.
| Quote: |
I have still 0.13 cm resolution for x, y direction.
Even if I take only MC true matched events, the resolution doesn't change so much.
|
I have not yet checked vertex resolution fir psi, for for eta_c it is on the order of 100ub.
| Quote: |
Psi_distribution.gif shows psi3770 reconstruction with using signal region of D+ and D- mass. Here once again vertex fits for psi3770 have been applied in second, and in third coloumn psi3770 with MC true matched events are additionally plotted to get the feeling for estimation of Psi3770 efficiency.
|
Have you tried with stt? I have a much higher efficiency there.
| Quote: |
When I required the MC true matching to reconstructed track, some events have double or even more counting at MC true matching, since sometimes two or three reconstructed tracks are induced essencially same origin of one MC track.
If I have those event, I throw away those events. I need still some tests to reduce those multi-counting problem in my analysis.
|
You should not reject nthose events, but simply take the best cadidate, with momentum closer to the mc one. This would save you a lot of tracks.
| Quote: |
PID would be real solution.
|
You have already mc pid, then you have already the solution 
| Quote: |
So, efficiency of Psi3770 is extremely low with this approach.
If I try same analysis with 4CFitter, I couldn't achieve until finishing complete analysis with 500 events, simply run crash with 4Cfitter for psi3770.
|
Maybe you have some bug in your code. I was able to run both the fitters with eta_c and w/o any problems/crash. Maybe you have some problems with indices.
Good luck!
[Updated on: Thu, 21 July 2011 10:15] Report message to a moderator |
|
|
|
|
|
| Re: Psi3770 study with run981 tpc mode [message #12320 is a reply to message #12313] |
Thu, 21 July 2011 14:44 |
donghee
Messages: 385
Registered: January 2009
Location: Germnay
|
first-grade participant |
From: *kph.uni-mainz.de
|
|
Dear Stefano,
I'm using vertex fitter for D- in following procedure.
kip is positive kaon candidate and pim is negative pion candidate, respectively.
Then I do
| Quote: |
Dminus.Combine(kap, pim, pim);
PndKinVtxFitter Dmvtxfitter(Dminus[j]);
Dmvtxfitter.Fit();
double chi2 = Dmvtxfitter.GlobalChi2();
unsigned ndf = 1;
TChisqConsistency cons(chi2,ndf);
double prob = cons.Likelihood();
|
Concerning resolution issue, if I have wrongly used the vertex fit, then I can understand bad resolution of D+ and D-.
but if I was using correctly the fit function, then I have to have oder of ~100 micrometer = 0.1mm = 0.01cm
I have still ~1.4mm(=0.14cm) after subtracting value between
| Quote: |
TVector3 Dvertex = (TVector3)mckaontrack->GetStartVertex();
TVector3 DpVtx = Dpvtxfit->Pos();
|
Unit is here cm... because the smearing is visible with same unit in the D vertex distribution.
Upper plots are generated true vertex of D, and belows are reconstructed one, as I show you already few times.
I will check some bug..
Best wishes,
Donghee
|
|
|
|
|
|
|
|
|
|
| Re: Psi3770 study with run981 tpc mode [message #12333 is a reply to message #12332] |
Mon, 25 July 2011 11:43 |
donghee
Messages: 385
Registered: January 2009
Location: Germnay
|
first-grade participant |
From: *kph.uni-mainz.de
|
|
Hi Stefano,
| Quote: |
a) Are you identifying pions and kaons from mc? Have you checked that they are primaries? And are yiou sure the the mccorrelation is done properly? (i.e., you could have misplaced trees, in particular if you are summing upo files)
|
I made running my analysis code with one produced run file (contain 500 events), and then I added up produced all histograms (typically 200 runs) with simple merging process.
Here is the way to get MC true information in the analysis code.
For both STT and TPC, I can have correct values of MC momentum and the theta and all other info about true MC from below lines.
| Quote: |
for(Int_t mc=0;mc<mc_array->GetEntriesFast();mc++)
{
PndMCTrack *mctrack=(PndMCTrack*)mc_array->At(mc);
double mom=mctrack->GetMomentum().Mag();
double the=mctrack->GetMomentum().Theta()*TMath::RadToDeg();
TVector3 primary = (TVector3)mctrack->GetStartVertex();
if(!(mctrack->IsGeneratorCreated())) continue;
if(mc==0 && mctrack->GetPdgCode()== -321){ ... //stuff for histogram}
if(mc==1 && mctrack->GetPdgCode()== 211){ ... //stuff for histogram}
if(mc==2 && mctrack->GetPdgCode()== 211){ ... //stuff for histogram}
if(mc==3 && mctrack->GetPdgCode()== 321){ ... //stuff for histogram}
if(mc==4 && mctrack->GetPdgCode()== -211){ ... //stuff for histogram}
if(mc==5 && mctrack->GetPdgCode()== -211){ ... //stuff for histogram}
}
|
| Quote: |
b) Have you applied a selection around D mass? To take away bad combinations with bad momenta
|
I used selector with the range of mass 1.8693 +/- 0.05 GeV/c, in panda word,
TPidMassSelector *DMassSel=new TPidMassSelector("Dselector", 1.8693, 0.1);
When Psi reconstruction is made, I require mass window of D+ and D- according 1.8193 < M_(D+ or D-) < 1.9193 GeV.
| Quote: |
c) Why you assume the ndf is 1? I think at this stage one should not use any chi2 cut, considering that we have to understand the errors before.
|
The probability must be 0 < prob(chi2,ndf) < 1.
When I use ndf getting from the fitter for D+ and D-, ndf is always 4.
From that values, prob can only have 0 < prob(chi2,ndf) < 0.25. That should be wrong.
For Psi case, I used ndf directly from PndKinVtxfitter.
| Quote: |
double chi2 = vtxfitter.GlobalChi2();
double pull = vtxfitter.GetPull();
unsigned ndf = vtxfitter.Getdof();
TChisqConsistency cons(chi2,ndf);
double prob = cons.Likelihood();
|
I understand that the meaning of probability bigger than 0.001 is that D or Psi events simply would be taken only with successfully fitted events.
You suggested that I don't need to reject any events from probabilities or even chi2s cut.
In this analysis, I have actually 4 different data samples like
So, I can take the data without requirement of chi2 fit as you want to do.
Best wishes,
Donghee
[Updated on: Mon, 25 July 2011 11:46] Report message to a moderator |
|
|
|
| Re: Psi3770 study with run981 tpc mode [message #12342 is a reply to message #12331] |
Tue, 26 July 2011 11:14 |
donghee
Messages: 385
Registered: January 2009
Location: Germnay
|
first-grade participant |
From: *kph.uni-mainz.de
|
|
Dear
Here is some new study of psi3770 channel.
I have tried to test the effect of pion and kaon hypothesis in the psi3770 reconstruction.
Please remind the psi(D+D-)-> decay into 6 final state with K-pi+pi+K+pi-pi-.
In the STT reconstruction, I have prepared an extra reco and pid file with kaon hypothesis, which has an assumption of kaon mass in PndRecoKalmanTask during the kalman fit.
So, I have two different pid files according only pion hypothesis and only kaon hypothesis, respectively.
Then I tried to build D+, for instance, with negative Kaon candidate from kaon pid file and positive pion candidates from pion pid file. That is quite simple approach to test the influence on D+ and D- reconstruction and vertex resolution.
In attached files, the D+ and D- mass distributions are compared.
Dmeson_mass_distribution plot show the results on the pion hypothesis, which is our standard approach. Tracking has been done only with pion hypothesis in this case.
In Dmeson2_mass_distribution, kaon hypothesis and pion hypothesis are used according D+(K-pi+pi+) or D-(K+pi-pi-) decay topology. As I said, I made kaon pid file with kaon hypothesis as an extra step.
What we can learn from both plots is the number of reconstructed events are increased due to correct mass hypothesis.
But vertex resolution of D+ and D- are practically unchanged.
I (or you) could expect some improvement of D+ and D- vertex resolution, but unfortunately it doesn't so much help in the reconstruction level.
welcome any comments.
Best wishes,
Donghee
|
|
|
|
|
|
|
|
|
|
Goto Forum:
Current Time: Sun Dec 22 04:01:34 CET 2024
Total time taken to generate the page: 0.00837 seconds
|
GSI Forum
Search
Help
Members
Login
Home
