Post on 03-Jul-2018
Bs->mumu updateIn depth FTKSIM performance
analysis
F. Crescioli, M. Dell'Orso, P. Giannetti, G. Punzi, G. Volpi
13/3/2007 Francesco Crescioli 2
Bs->mumu status update● Produced background sample (generic bb with
one muon with Pt > 6 GeV)● Produced signal sample with both muons in the
barrel● Re-run reconstruction with 7 layers ftksim● Not analyzed yet:
Stopped to work on important FTKSIM issues
13/3/2007 Francesco Crescioli 3
FTKSIM tracking debugging● Monica highlighted some problems with WHuu/bb sample
– FTK worse than iPat on uu/bb separation– FTK IP distribution much different from iPat for high
Pt tracks● In our analysis for fake tracks, unmatched tracks with
some bad reconstructed parameter never really disappeared– Are them responsible for the behaviour observed by
Monica?
13/3/2007 Francesco Crescioli 4
Debugging sample● Produced special debug events
– ATHENA v12.0.31 (exactly the same version as training)– 10 tracks per event
● Fixed zero impact parameter● Other parameters with the same distributions as training sample● Vertex spread on
– Low noise digitization– Muon and pion samples– FTKSIM with 7 layers configuration
13/3/2007 Francesco Crescioli 5
Debug sample (muons) IP resolution
Looking at unmatched IP distribution (resolution because truth is zero IP) we observe a significant difference in IP resolution vs Pt
The difference correspond to 30um smearing on resolution
There's also an offset in the mean of the reconstructed IP
FTKiPat
13/3/2007 Francesco Crescioli 6
Looking at the matched (with truth, for both FTK and iPat) resolution shows some significant change in the sigma at low Pt
Mean offset is still there
Matched iPat tracks seems to have much less tail than unmatched. Smaller effect for FTK matched tracks.
Debug sample (muons) IP resolution - matched
FTKiPat
13/3/2007 Francesco Crescioli 7
Unmatched tracksVariance difference is almost constant for Pt > 3 GeV and increase for low Pt tracks
Debug sample (muons) IP resolution – variance difference
Matched tracks
Matching tracks grately reduce the increase of variance difference for low Pt tracks leaving the costant part almost the same
13/3/2007 Francesco Crescioli 8
Unmatched integrated tails shows that FTK and iPat have similar behaviour until tracks with Pt > 6 GeV are selected.Note that at Pt > 3 GeV, IP > 0.2 mm tail fraction is very similiar.
FTKiPat
Debug sample (muons) IP distribution integrated tails
13/3/2007 Francesco Crescioli 9
Matched tracks shows some improvement for FTK and big change for iPat tracks.Tails are almost gone for iPat, at every Pt cut.
FTKiPat
Debug sample (muons) IP distribution integrated tails - matched
13/3/2007 Francesco Crescioli 10
Debug sample (pions) IP resolution
Unmatched IP resolution shows no significant change in both mean and sigma from the muon case
FTKiPat
13/3/2007 Francesco Crescioli 11
Debug sample (pions) IP resolution - matched
For matched tracks mean don't change.Sigma is reduced for low Pt. There's some effect on the tails, but not as significant as in the muon case.
FTKiPat
13/3/2007 Francesco Crescioli 12
Unmatched tracksVariance difference is almost constant for Pt > 3 GeV and increase for low Pt tracks.Less difference than with muons.
Debug sample (pions) IP resolution – variance difference
Matched tracks
Matching further reduces difference.
13/3/2007 Francesco Crescioli 13
Debug sample (pions) IP tails
Tails seems better with FTK than with iPat, but there's some visible improvement of iPat with Pt which there isn't in FTK.
FTKiPat
13/3/2007 Francesco Crescioli 14
Debug sample (pions) IP tails - matched
As expected from scatterplot, matched tails are sligthly better but there isn't the big improvement for iPat as in the muon case.
Again at Pt > 3 GeV, IP > 0.2 mm our performance is quite similar to iPat.
FTKiPat
13/3/2007 Francesco Crescioli 15
Debug sample ~ Conclusions● Resolutions shows the same Pt dependence with FTK and
iPat due to multiple scattering at low Pt– At high Pt iPat goes asymptotically to a better
resolution than FTK (~20μm vs ~40μm)● FTK IP mean shows an offset of ~5μm● Integrated tail is reduced by Pt cuts for iPat, on both
muons and pions, much less for FTK● Performances at Pt > 3 GeV with 200μm IP cut should be
similar
13/3/2007 Francesco Crescioli 16
A real case: WHuu/bb● Reconstructed Monica's WHuu/bb sample with 7 layer
configuration
13/3/2007 Francesco Crescioli 17
WHuu/bb – IP distributionsAs predicted by debug sample we have slightly better performance with no Pt cut.
At Pt > 1.5 GeV iPat start to perform better.
We have a strange asymmetry.
No P
t C
ut
Pt
> 1
. 5 G
eV
FTK iPatWHuu
WHbb
13/3/2007 Francesco Crescioli 18
WHuu/bb – IP distributionsAs Pt cut increase iPat performances increase faster than ours.
Our strange asymetric second peak is more visible.
Pt
> 3
GeV
Pt
> 6
GeV
FTK iPatWHuu
WHbb
13/3/2007 Francesco Crescioli 19
WHuu – IP distributionsiPat tends to perform better increasing Pt cut, and the asymmetric second peak in FTK is more visible.
Is the peak an artifact from nonmatched tracks?
13/3/2007 Francesco Crescioli 20
WHuu/bb – 7 layers vs 6 layers
WHuu
WHbb
FTK 7 layers iPatFTK 6 layers
13/3/2007 Francesco Crescioli 21
WHuu/bb – 7 layers vs 6 layers
WHuu
WHbb
FTK 7 layers iPatFTK 6 layers
13/3/2007 Francesco Crescioli 22
WHuu/bb● As expected from debug our performance is similar to
iPat and greatly improved since 6 layer configuration– iPat improves more than FTK applying Pt cuts
● There is a strange second peak in the IP distribution of FTK– Not evident in 6 layer configuration– Under investigation
13/3/2007 Francesco Crescioli 23
Conclusions● 7 layer configuration give better resolution performance
than 6 layer● FTK is still slightly worse than iPat but tails are not
much different for low Pt (important for B-tagging)– There's some problem that generate the asymmetric
second peak– Next step: study and fix it
13/3/2007 Francesco Crescioli 24
WHuu – IP distributions (matched)With matched tracks iPat performs much better.
FTK doesn't improve much and the second peak is still there.
Under investigation.