TheProject

The ClusterID package contains all the necessary functionality to do physics analysis and evaluate calorimeter designs. However, improvements to its resolution are possible in many areas. One can always try to develop additional discriminators and algorithms to increase the efficiency of identification, decrease the fake rates and improve the energy resolution. Regarding overlapped clusters which contain hits from more than one particle, there are techniques available to try to separate the cluster into two or more clusters. Regarding a particle which created a main cluster and a separate fragment cluster, there are techniques available to try to associate fragments with the primary cluster. And there are other areas for further development as well.

In the course of LCD studies several people have made studies of some of these specific issues outside of any framework for calorimeter reconstruction. The ClusterID project provides a way for people to incorporate these techniques into the ClusterID package and use the various tools for evaluating and improving the efficacy of their particular technique. Also, the project has a long list of ToDos that have arisen as it was being developed. For people who are interested in getting involved in calorimetry software selecting a project off of the ToDo list is a handy way to get familiar with the package, which is quite complex, as well as to make a contribution to the overall LCD studies.

ClusterID project sub-projects list

Why are high energy photons mis-IDed as neutral hadrons in the single particle samples with tt distributions in SDnocalgaps?

Can we identify and separate overlapped photons?

Pi0 finder

Can we point hadron fragments back at the beginning of a hadron shower?

For neutral hadron showers the beginning is the first hit layer area.

For charged hadrons we will have to find the shower beginning point perhaps where the # hits/layer increases or energy/layer increases or both.

Why are some pion showers not connected as it seems they should be from looking at Wired and the hits in single particle events? This is a low percent effect.

Do we need a MakeSignalEventTrainingDataSets?

Why are there Simple clusters with adjacency 1 with significant energy and #hits all lying in one layer?

Add muon and electron ID.

Add overlapped clusters ID.

Why do we get different answers for hemisphere and jet masses each time we run the program over the same data?

Does Ron’s CID code output match the ClusterID input type and probs?

Adding stages to Particle Reconstruction:

add( new MCFast() );

add( new SimpleClusterBuilder() );

add( new ApplyClusterIDToSioFile(net));

add( new ClusterSplitter() );

add( new FragmentAssociator() );

add( new Dirc&dE/dxParticleID() );

add( new TrackClusterAssociator() );

add( new TPCGammaConversionFinder() );

add( new FinalStateParticleReconstructor() );

add( new Pi-ZeroFinder() );

add( new K0,Rho,etcFinder() );

add( new ZVTop() );

add( new W,Z,t,Higgs,etcFinder() );

add( new JetMassAnalysis() );

Contributors:

Gary Bower, SLAC

Ron Cassell, SLAC

Mark Donsillman, SLAC

Ayanah George, SLAC summer student

Tony Johnson, SLAC

Ayodele Onibokun, SLAC summer student

Saurav Pathak, U Penn

LINKS to contributed projects.

List specific projects and outline how the might be done. Include tutorials to try first and tutorials to evaluate results and outline code to do the project.

Cluster Association: Start with list of frags ordered by energy and list of hads ordered by energy. Take perhaps top 10 on each list and calculate the 100 pairs worth of DOCAs

Cluster seperation: 1) increase min hit energy until it splits into two and apply CLusterID to each. First, find a way to add low energy hits back to the clusters in a rational way, or if C% of energy is cut and C1% and C2% is energy remaining in the two clusters redistribute the energy proportionally if using the reduce clusters gives better pointing resolution, otherwise add the hits back. 2) Maciel approach 3) Vishnu approach. Maybe with all three approaches start with NxN adjacency and do some kind of smoothing of energy over hits, ie, each hit gets the energy of hit + its 26 nieghbors divided by 27.