Use existing CaloDigit-Track matching and Brem recovery information to filter calo clusters. First attempt at addressing #484.
With the current implementation, a veto for track-matched clusters is applied when constructing the NeutralBasicParticle objects, while a veto for Bremsstrahlung-matched clusters is applied in the TwoCaloClusters selection algorithm (which is the underlying algorithm of Hlt1Pi02GammaGamma).
This MR also includes a Parallel Reduction algorithm (device/utils/reduction) to perform parallel sum of arrays.
This is implementation is based on these examples, and it is used to replace the sequential sum of ECAL cell energies performed in TotalECALEnergy.cu.
Some preliminary results based on the Hlt1Pi02GammaGamma monitoring line can be found below.
The line is run with no prescaling and without the nPV cut on the upgrade_mc_minbias_scifi_v5_retinacluster_000_v1_newLHCbID.mdf sample (10k events).
Mass fits are performed on the diphoton mass distrubution using a Gaussian to describe the \pi^0 signal and a power law to describe background.
With the default requirements, 16652 cadidate diphotons are selected, out of which 980 +- 68 are signal (see fit below).
Removing diclusters for which at least one seed cell has been matched to a track, the number of selected candidates falls to 13140, out of which 945 +- 114 are signal (see fit below).
Removing diclusters for which at least one seed cell has been matched to Bremsstrahlung, the number of selected candidates is 14445, out of which 910 +- 134 are signal (see fit below).
Finally, applying both vetoes we get 11351 candiates and 922 +- 59 signal events (see fit below).
Below you can alse find the distribution of vetoed diphotons fitted with background only.
In summary,
Candidates
Variation
S/B
Default
16652
-
(5.9 +- 0.4)%
TM-only
13140
-21.1%
(7.2 +- 0.9)%
BM-only
14445
-13.3%
(6.3 +- 0.9)%
BM+TM
11351
-31.2%
(8.1 +- 0.5)%
Edit: Updating comparison after modifications.
New comparison done with the same HLT1 configuration: running Hlt1Pi02GammaGamma on upgrade_mc_minbias_scifi_v5_retinacluster_000_v1_newLHCbID.mdf with no prescaling and no n_pvs cuts.
With the current reconstruction and selection of 2024-patches, 21162 cadidate diphotons are selected, out of which 1014 +- 73 are signal (see fit below).
With this MR, 14238 cadidate diphotons are selected, out of which 910 +- 63 are signal (see fit below).
