Boost PC performance: How more available memory can improve productivity
Presentacion en Software Week, CERN, Clustering for vle particles in cbt
1. Clustering for VLE particles in
CombinedTB
Carmen Iglesias
IFIC-University of Valencia
(Dpt. FAMN)
TileCal Analysis and Combined Test Beam, 14 February, TileCal Week
2. Ntuples
Samples with electrons and pions from 1 to 9 GeV at eta=0.35, with
Calo info (LAr+Tile) and the tracks info from TRT only (pixels have
problems)
They have been generated by Vincent using 9.1.1 Release with the default values
of RecExTB. They can be located in:
castor/cern.ch/atlas/ctb/test/real_data/reconstruction/Combined/
Energy #Run Energy #Run
1 GeV 2101077 6 GeV 2101084
2 GeV 2101078 7 GeV 2101085
3 GeV 2101079 8 GeV 2101048
4 GeV 2101080 9 GeV 2101049
5 GeV 2101047
3. Cut to separate particles
• Selection of good tracks
• trk_nTracks==1Only 1 track
• trk_nTrtHits[0]≥20 More than 20 hits per track
• for electrons
• sADC_C2>650 Cherenkov2 counter cut
Pion-electron separation
• nHL>5 number of high-level hits Cherenkov counter cut isn’t enough.
• for pions and muons We use the future of low-level hit and
• sADC_C2<650 Cherenkov2 counter cut high-level trigger. A π track has less
• nHL≤2 number of high-level hits than 5 high-level hits. The e- tracks
have more than 2 high-level hits.
• Pion- Muon separation
• Pions: E<0.2 GeV in sample D (Pedestal in sample D,
we assume that only muons can reach D)
• Muons: E> 0.2 GeV in sample D (signal in sample D)
muons
pions
4. Clustering info in CBT ntuples
• Emcluster: clusters from the sliding window algorithm
• Tbemclusters: clusters from an algorithm used in previous test beam. It
has been added to allow comparison. It’s a window of 3x3 cells.
– Emclusters and tbemclusters use only cells from the LAr calorimeter.
• Cmbclusters: sliding window clusters but they are done on towers
(larg+tile) and not anymore on cells. It is not working for the moment
because of a coordinate problem between LAr and Tile.
– LAr is shifted with respect to Tile by "half module" :
• TileCal has just 3 modules -0.15 < eta < +0.15
• LAr has -0.2 < eta < 0.2,
i.e. there are 3 slices with ∆φ=0.1 in Tile and 4 slices in LAr, shifted by half of the slice
• Topo_EM cluster and Topo_Tile cluster: Finds a seed cell, then cluster
expands by checking energy in neighboring cells. Thresholds for defining
seed and including neighbors can be changed. The default values are:
– seed threshold is E/σnoise>6
– neighbor threshold is E/ σnoise>3
5. e- in Lar: Energy distribution
For electrons at 9 GeV
For electron it seems as the cuts on TRT
works good
6. e- in LAr: Number of Clusters
# clusters
SW SW_TB TOPO_EM
9 GeV 31585 31607 31614
8 GeV 30475 30506 30514
7 GeV 27252 27303 27292
6 GeV 26878 26969 26961
5 GeV 21689 22446 22464
4 GeV 10994 13751 13670
3 GeV 2292 (*) 11869 11472
2 GeV --- (*) 5853 4574
#clusters is very low
1 GeV --- (*) 1093 482
#clusters is very similar between them for each ET value.
#clusters defined increase with the energy.
(*) There is a cut (E>2 GeV) in this algorithm by definition
7. e- in LAr: Resolutions
In general, the E resolution is better when E increases
SW SW_TB TOPO_EM
9 GeV 7.57 8.92 10.48
8 GeV 8.51 10.04 11.64
7 GeV 7.85 6.93 8.51
6 GeV 8.83 7.81 9.62
5 GeV 13.07 15.47 17.34
4 GeV 11.04 11.47 14.78
3 GeV 9.59 (*) 14.38 20.39
2 GeV ---(*) 20.51 34.99
1 GeV ---(*) 80.75 48.38
E resolution slightly better
than it’s expected, WHY??
The best resolution is for SW, but all the algo present very similar results.
TOPO obtain the worst resolutions
maybe it will be needed to change the thresholds for seed and neighbor cells.
In my previous analysis of clustering in VLE in simulation: the best resolution with
seed threshold E/σnoise>4 and neighbor threshold E/ σnoise>2
(but they have been done without noise and pile-up)
(*) There is a cut (E>2 GeV) in this algorithm by definition
8. Cone algorithm (I)
Study in LAr the region ∆η=0.2 and ∆ϕ=0.2
Select cells in LAr comes from electron E depositions
sADC_C2>650 and nHL>5
Ecell >σnoise : Ecell > 100 MeV
Select the Cell with the longest E it will be the center of the cone
Define the radius of the cone: ∆R=√∆η2+∆φ2
where
∆η=ηcellEmax-ηcell ∆φ=φcellEmax-φcell
Strange results in DR distributions: it seems
as the most of the cells from electrons are far
from the cell with Emax ???
9. Cone algorithm(II)
Is due to the bad definition of ∆η or ∆φ??
Is due to the cell with Emax are in the wrong place?
If I repeat the analysis with an imaginary particle centred in η=0.35 and ϕ=0.0:
The same distribution appears
10. Pions and Muons in Lar (I)
The separation
between pions total
muons
and muons is not electrons
muons pions
so easy.
pions
There will be contamination on muon in the E distribution of Pions.
Pions
Muons
11. Pions and Muons in Lar (II)
A part of the problem could be from PEDESTAL…
Pedestal in SW cluster id up to 1 GeV, and in TOPOcluster is up to 2 GeV
maybe we must change the threshold for the min E inside cluster
But, if we compare with the pedestal
value per cell, the thresholds in TOPO
to avoid the noise MUST be enough:
cellThresholdOnAbsEt = 0.*MeV
neighborThresholdOnAbsEt = 100.*MeV
seedThresholdOnEt = 200.*MeV
12. Pions and Muons in Lar (III)
Problem with muons with E ≤2GeV
Very low muons don’t deposit in the last Tile sampling and are therefore
taken as pions!
-Using TRT and Tile information it’s possible to distinguish between
muons and pions only for E >3GeV, below it is difficult with this method
Improvement expected using the topological info in LAr and Tiles.
- Based on the fact that the muons leave their energy uniformly (normalizing to the
path length in a sampling) whereas pions should be giving most of their energy in
this sampling.
muons muons muons
pions pions pions
13. Problems in Tile
In Tile, we expect that the most of the ET comes from PIONS, and also a contribution
of muons. But the plots show that there are electrons in Tile!!
Ecell > 25 MeV Ecell > 70 MeV
total There are still electrons
electrons
for low energies
muons
pions
estimation of the noise for cell in Tile
from the pedestal for each cell
If we only select the
E in Tile > 1000 MeV Is possible to have electrons in Tile?
Are they really pions?
total
electrons
muons
At 9 GeV there is no e- in Tile (even at higher
pions energy it seems that e- cannot cross LAr)
It is possible to have pions that pass the cherenkov
cut. But in principle nHL>5 is a very efficient cut to select
electrons only.
14. Conclusions
About clustering algorithms:
• For electrons, the best results in E resolution comes from SW algo
•Anyway, it would be interesting to change the cut in Ecell>2 GeV to can
study the VLE runs at 3, 2 and 1 GeV
• Results from TopoCluster can be improved
• change the thresholds for seed and neighbor cells
• Needed to understand the behavior of Cone algorithm
About particles separation:
• In Lar
– For electrons it seems as the cuts on TRT works good.
– Using TRT and Tile information it’s possible to distinguish between
muons and pions only for E >3GeV. To improve:
• use topological information in LAr and Tiles
• Compare VLE runs with MC
• In Tile
– Pions are not well defined in Tile (some proportion of them are defined as e-)
TO GET PROGRESS IN CLUSTERING ANALYSIS, FIRST MUCH MORE
WORK IS NEEDED TO DO IN THE SEPARATION OF PARTICLES AT VLE!!