Home » PANDA » PandaRoot » Analysis » Bug fixed in PndKinFitter read-in covariance matrix
Bug fixed in PndKinFitter read-in covariance matrix [message #19588] |
Thu, 29 September 2016 16:33 ![Go to next message Go to previous message](/theme/gsi/images/down.png) |
Lu Cao
Messages: 77 Registered: February 2013
|
continuous participant |
From: *ext.kfa-juelich.de
|
|
Dear all,
I just found a bug in RhoKinFitter::ReadMatrix(), when the original cov matrix is re-arranged from (X, Y, Z, Px, Py, Pz, E) basis to (Px, Py, Pz, E, X, Y, Z).
For example, the origianal cov matrix of a photon in (X, Y, Z, Px, Py, Pz, E) is
| 0 | 1 | 2 | 3 | 4 | 5 | 6 |
---------------------------------------------------------------------------------------------
0 | 0.2111 0.008221 -0.02881 0.0002914 1.135e-05 -3.977e-05 0
1 | 0.008221 0.2217 0.02912 1.135e-05 0.0003061 4.02e-05 0
2 | -0.02881 0.02912 0.008063 -3.977e-05 4.02e-05 1.113e-05 0
3 | 0.0002914 1.135e-05 -3.977e-05 3.203e-06 -2.683e-06 1.97e-05 2.013e-05
4 | 1.135e-05 0.0003061 4.02e-05 -2.683e-06 3.022e-06 -1.897e-05 -1.939e-05
5 | -3.977e-05 4.02e-05 1.113e-05 1.97e-05 -1.897e-05 0.0001393 0.000142
6 | 0 0 0 2.013e-05 -1.939e-05 0.000142 0.0001447
but the re-arranged one in (Px, Py, Pz, E, X, Y, Z) is
| 0 | 1 | 2 | 3 | 4 | 5 | 6 |
----------------------------------------------------------------------------------------------
0 | 3.203e-06 -2.683e-06 1.97e-05 2.013e-05 1.135e-05 -3.977e-05 0
1 | -2.683e-06 3.022e-06 -1.897e-05 -1.939e-05 0.0003061 4.02e-05 0
2 | 1.97e-05 -1.897e-05 0.0001393 0.000142 4.02e-05 1.113e-05 0
3 | 2.013e-05 -1.939e-05 0.000142 0.0001447 0 0 0
4 | 0.0002914 1.135e-05 -3.977e-05 0 0.2111 0.008221 -0.02881
5 | 1.135e-05 0.0003061 4.02e-05 0 0.008221 0.2217 0.02912
6 | -3.977e-05 4.02e-05 1.113e-05 0 -0.02881 0.02912 0.008063
.
One can notice that some elements in the new cov matrix are wrongly filled, eg. the new (0,4) should be equel to the old (3,0)=0.0002914 rather than 1.135e-05. Since the new cov matrix is not symmetric, the whole fitting procedure will be badly influenced. This is caused by a bug in the code:
272 //Change to px,py,pz,E,x,y,z
273 for(int i=0; i<7; i++) {
274 for(int j=0; j<7; j++) {
275 if(i>=3) {
276 if(j>=3) {
277 p4Cov[i-3][j-3] = p3Cov[i][j];
278 } else { p4Cov[i-3][j+3] = p3Cov[i][j]; }
279 } else {
280 if(j>=3) {
281 p4Cov[i+4][j-3] = p3Cov[i][j];
282 } else { p4Cov[i+4][j+4] = p3Cov[i][j]; }
283 }
284 }
285 }
.
The line #278,
Quote: p4Cov[i-3][j+3] = p3Cov[i][j] should be p4Cov[i-3][j+4] = p3Cov[i][j].
I've studied the pull distributions of befor & after fixing, combining with another bug fixed in this fitter recetly (trk 29480), with my Ds semileptonic decay chain. The result shows several issues regarding to the covariane matrix itself and the fitter, and seems may evoke a long and detailed discussion. For a more efficent and direct communication, I would prefer to present the plots in the upcoming software meeting on 4th Oct..
BTW, the fixed code is attached. (No right to commit... )
Best regards,
Lu
|
|
|
Goto Forum:
Current Time: Wed Aug 28 17:30:27 CEST 2024
Total time taken to generate the page: 0.00789 seconds
|