GSI Forum
GSI Helmholtzzentrum für Schwerionenforschung

Home » PANDA » PandaRoot » Analysis » Vertex fitters' problems
Vertex fitters' problems Thu, 06 June 2013 12:40
 StefanoSpataro Messages: 2736Registered: June 2005 Location: Torino first-grade participant From: *to.infn.it
Dear all,
Laura and me are checking the features of our vertex fitters,
with the trunk version 20077. In the latest trunk the function GetMcTruth() returns null pointer, than the trunk cannot be used.

We have simulated 10k ppbar->D*+D*-->D0pi+D0barpi-->k-pi+pi+k+pi-pi-, we have used MC truth index to identify the kaon- and the pion+ of the D0 (we have not touched the d0bar).
We have uploaded all the relevant macros, you have only to modify the path of the DEC file in the sim macro.

We run PndKinVtxFitter, PndVtxPoca and PndVtxPRG and compared the results. We were able to find a tutorial only for PndKinVtxFitter. The tutorial for PndVtxPoca and PndVtxPRG is missing then we hope we are using them in the proper way.

Vertex residuals reco - mc (in cm) and chi2 for PndKinVtxFitter:

Vertex residuals reco - mc (in cm) and chi2 for PndVtxPoca:

Vertex residuals reco - mc (in cm) and chi2 for PndVtxPRG:

These are the obtained resultion values:

```                |X res  | Y res | Z res                |
PndKinVtxFitter |   2um |   2um | wrong and asymmetric |
PndVtxPoca      | 400um | 400um | 400um                |
PndVtxPRG       |  60um |  60um | 100um                |
```

Then...

PndKinVtxFitter is obtaining very very very narrow distributions on X and Y. How is it possible? Z distribution is screwed. Chi2 is very low.
PndVtxPoca has too large resolution values, does not work. Decent chi2 maybe.
PndVtxPRG has more reasonable values, even if I believe they are a bit large. It is much slower than the others, most probably because of the 10 iterations, but w/o a tutorial we were not sure about which number to use. Chi2 too low.

Still not checked composite particles (i.e. D*)

Then...

What to use? Could you please take a look?

Laura and Stefano
Re: Vertex fitters' problems [message #14848 is a reply to message #14847] Thu, 06 June 2013 13:02
 Ralf Kliemt Messages: 507Registered: May 2007 Location: GSI, Darmstadt first-grade participant From: *gsi.de
Hi,

I'm on the Monte-Carlo list filling thing right now.

As for the fitters, I wonder how the 2um are possible.
The 400um of the POCA may be normal, as it is a coarse approximation.
The PRG fitter is limited in its precision as the algorithm itself is designed to be faster.

Ralf
Re: Vertex fitters' problems [message #14849 is a reply to message #14847] Thu, 06 June 2013 17:09
 Simon Reiter Messages: 44Registered: May 2013 continuous participant From: *zeu.uni-giessen.de
Hi,

I'm working on similar simulations and was also planning to compare these three fitter. But I'm wondering what
`RhoCandidate`
is. I'm using
`TCandidate`
. What are the differences? I could not even find it.

Greetings Simon

[Updated on: Thu, 06 June 2013 17:10]

Report message to a moderator

Re: Vertex fitters' problems [message #14850 is a reply to message #14849] Thu, 06 June 2013 17:11
 StefanoSpataro Messages: 2736Registered: June 2005 Location: Torino first-grade participant From: *to.infn.it
TCandidate if you are using release apr13 and before. RhoCandidate if you are using code after apr13 release.
Re: Vertex fitters' problems [message #14888 is a reply to message #14847] Thu, 20 June 2013 15:47
 Ralf Kliemt Messages: 507Registered: May 2007 Location: GSI, Darmstadt first-grade participant From: *gsi.de
Hi.

I did some fixes to Rho and the fitters. We had deeply hidden bugs...

Now, I ran the same simulations just as you did and modified the analysis macro to my liking (attached). The outcome is strange BUT:
A) ChiVtx and KinVtx give somewhat similar vertex resolutions. They should as these are the same (P.Avery) Algorithm programmed by different colleagues.
B) POCA is comparable to the fitters above - because it is the seed value for them. I looked it up and the KinVtx actually converges, so I blame the few excess energy available in your system. Both the kaon and pion tracks will almost move along with the CMS.
C) The PRG fitter (Billoir algorithm) produces better results. Wonderous to me but there we go; maybe its in the algorithm. Both fast and full fit give the same resolution where I would blame the kinematics again.

I hope I could solve the issues now.

Cheers
Ralf

PS: The overly good resolutions of the KinVtx came from the fact that the vertex vector was always zero and the negative MC coordinates were plotted.
Re: Vertex fitters' problems [message #14892 is a reply to message #14888] Fri, 21 June 2013 18:43
 L. Zotti Messages: 6Registered: December 2010 occasional visitor From: *to.infn.it
Dear Ralf,
Looking in your analysis macro, I found you just missed one "break" at line 273. Actually with this modification, I got some better results from the ChiVtx fitter.
Moreover, cleaning the sample looking at the FitStatus flag, I got the following resolution:

PndChiVtxFitter x:32um y:52um z:138
PndVtxPRG (full) x:61um y:60um z:103um
PndVtxPRG (fast) x:60um y:61um z:103um

Plots for PndChiVtxFitter:

Plots for PndKinVtxFitter:

Plots for PndVtxPoca

Plots for PndVtxPRG
1)fast:
2)full:

Still to be understood why resolution are different in x and y for the PndChiVtxFitter.
Do you think this results are reliables?
The KinVtxFitter and the Poca one still give strange results.
For what concern the kinematics, actually the kaons and the pions from the d0-meson are well separated in angle, how you can see here:
• Attachment: opening_angle.png
• Attachment: ChiVtx.png
• Attachment: FastVtx.png
• Attachment: FullVtx.png
• Attachment: KinVtx.png
• Attachment: PocaVtx.png
Re: Vertex fitters' problems [message #15019 is a reply to message #14892] Thu, 25 July 2013 16:22
 L. Zotti Messages: 6Registered: December 2010 occasional visitor From: *to.infn.it
Dear All,
in the detailed investigation that I'm carring out on the vertex-fit, I found an error in the class PndChiVtxFitter.cxx at line 255 in the definition of a TVector3.
Moreover, I found a lot of negative chi2..... Does somebody of you know how can this be possible and can have a look on this problem?

Regards
Laura
Re: Vertex fitters' problems [message #15023 is a reply to message #15019] Fri, 26 July 2013 13:26
 StefanoSpataro Messages: 2736Registered: June 2005 Location: Torino first-grade participant From: 2.235.190*
Very interesting...

```root [0] TVector3 e(1,2,3)
root [1] TVector3 f=(1,2,3)
root [2] e.Print()
TVector3 A 3D physics vector (x,y,z)=(1.000000,2.000000,3.000000) (rho,theta,phi)=(3.741657,36.699225,63.434949)
root [3] f.Print()
TVector3 A 3D physics vector (x,y,z)=(3.000000,0.000000,0.000000) (rho,theta,phi)=(3.000000,90.000000,0.000000)
root [4] TVector3 d
root [5] d=(1,2,3)
(class TVector3)41869040
root [6] d.Print()
TVector3 A 3D physics vector (x,y,z)=(3.000000,0.000000,0.000000) (rho,theta,phi)=(3.000000,90.000000,0.000000)

```

and without a single complain...

[Updated on: Fri, 26 July 2013 14:42]

Report message to a moderator

Re: Vertex fitters' problems [message #15024 is a reply to message #15023] Fri, 26 July 2013 14:48
 StefanoSpataro Messages: 2736Registered: June 2005 Location: Torino first-grade participant From: 2.235.190*
I have committed the modified class (there were a couples of such bad init).
Could you please check now if it produces "good" results?
Re: Vertex fitters' problems [message #15025 is a reply to message #14847] Fri, 26 July 2013 14:55
 Simon Reiter Messages: 44Registered: May 2013 continuous participant From: *zeu.uni-giessen.de
Without CINT's optimization it worked

I'm trying to look at that chi2, but will take me some time..
Re: Vertex fitters' problems [message #15028 is a reply to message #15025] Mon, 29 July 2013 11:11
 L. Zotti Messages: 6Registered: December 2010 occasional visitor From: *to.infn.it
Dear All,
after the last modification of the PndChiVtxFitter.cxx
these are the main results that I got.
Resolution on x: 32 micron

Resolution on y: 54 micron, why is it different from x?

Resolution on z: 113 micron.

Chi2 still some negative values: do you have suggestions on how to decide what is the best candidate in case of combinatorials?

Mass of the D0 before and after the fit: after the fit we got a low number of D0 under the central peak

Best regards,
Laura
• Attachment: x.png
• Attachment: z.png
• Attachment: y.png
• Attachment: chi2.png
• Attachment: mass_d0.png
 Previous Topic: [FIXED] PndAnalysis::McTruthMatch always returns false Next Topic: Updated Rho Tutorial
Goto Forum:

Current Time: Thu Nov 30 03:34:22 CET 2023

Total time taken to generate the page: 0.01111 seconds