Jets in the AOD

This page is under heavy construction!


This page collects information related to the discussion of the AOD jet and cluster content for release 12 and the transition period towards release 13. The motivation here is to preserve or at most improve the jet performance with respect to the tower-based jets with the cell signal weighting calibration. Physics studies should not be affected by the transition beyond a good reason, i.e. performance improvements.

Algorithm Considerations

The standard algorithms used for jet finding in Athena up to and including release 12.0.0 are the seeded fixed cone algorithm with cone sizes ΔR = 0.7 and ΔR = 0.4, and the Kt algorithm with a distance parameter R = 1.0.
For the reconstruction of the CSC physics simulation we should produce the same jets, only replacing the present Kt implementation with the much faster one provided recently (future link to Pierre-Antoine Delsart's note, in preparation). It has been shown that at least for CaloCluster based jet finding exactly the same jets can be reconstructed with both algorithms. The small and very insignificant differences for CaloTower based jets can easily be ignored (they are due to slight differences in the ordering of towers of equal geometrical distances). The table below shows a summary of the default jet finders we should use for physics studies with release 12.0.X reconstruction.

Summary of Default Jet Algorithms

Jet Finding Algorithm Basic Parameters Comments
Fixed Cone ΔR = 0.4 classical fixed cone jet built around seeds
Et(seed) > 2 GeV
Et(jet) > 10 GeV
Fixed Cone ΔR = 0.7 classical fixed cone jet built around seeds
Et(seed) > 2 GeV
Et(jet) > 10 GeV
Kt clustering R = 1.0 fast implementation based on Cacciari/Salam hep-ph/0512210
Et(jet) > 10 GeV

Release 12.0.X Physics Production

For the default physics production we need jets with the expected performance for general physics studies. The reference are jets made from CaloTowers with electromagnetic energy scale signals, and calibrated with the default cell signal weighting calibration (so-called H1-style).
Even though the following statement is likely valid, a more conservative approach has now been suggested, see following section. Frank Paige's recent studies (May 23, 2006) show strong indication that first replacing CaloTowers with uncalibrated topological CaloClusters, and then applying the same cell based calibration with parameters derived from the CaloTower based jets, with an additional Et and η depending correction function to establish a flat (linear) response, yields significantly better jet reconstruction efficiencies at low Et, and a slightly better resolution, again especially in the lower Et regime (improved noise suppression due to the clustering). This indicates at least the same if not an improved performance with respect to the CaloTower based jets.

Standard AOD Content

The standard AOD for most if not all physics studies (excluding maybe b-physics - they probably produce their own AOD anyway) should contain the following jet collections and associated data objects, assuming that the jets from topological clusters perform as well as the tower-based jets, after calibration and corrections:

  1.  ConeClusterJets    cone jets with R = 0.7; built from uncalibrated topological clusters CaloTowers with default cell calibration. and the new corrections applied in jet reconstruction.
  2.  Cone4ClusterJets   cone jets with R = 0.4; as above built from uncalibrated topological clusters CaloTowers and with the default cell weighting calibration applied, but probably with additional corrections for out of cone effects (needs coordination with the top group!).
  3.  KtClusterJets   Kt jets with distance parameter R = 1.0, again from uncalibrated topological clusters with default cell-based calibration CaloTowers and (maybe?) appropriate correction applied; the issue of the default value for the distance parameter R is actually physics driven (QCD, jet cross-section), just tuning it to make the jets similar to cone jet without any other qualification is probably not the correct way to go! For the moment we should stay with R = 1.0. The newly implemented fast Kt algorithm should be used, though, as this avoids the pre-clustering step and gets the jets closer to the native Kt implementation without pre-clustering.

This means that basically all jet collections which are used heavily by physics groups stay as they are, while the little used Kt jets improve a little bit.

AOD for Jet Calibration (needs work!)

The idea of calibration AODs is to provide a (sub)set of the physics production with more than the default information for calibration studies. These are different AODs, i.e. for a given ESD there should be two AODs: the standard AOD as discussed above in any case, and the calibration AOD if the correponding data set is supposed to be included into calibration studies. The important thing is that everybody doing jet calibration studies should use the same calibration AOD! This increases the quality of the compoarisons significantly, and avoids confusion with lower level reconstruction issues, as they are all the same for everybody in this case.

Calibration AOD Content

Jet Collections Cluster Collections Comments
ConeClusterJets CaloTopoCluster same jet collections as for standard physics production; input to jet finding is the same collection of uncalibrated topological clusters in all three cases;
ConeCalClusterJets CaloCalTopoCluster input to jet finding is the same collection of calibrated topological clusters in all three cases; projected to be the 13.0.0 default;
Optional Extension
ConeHitClusterJets CaloHitTopoCluster input to jet finding is the same collection of topological clusters from calibration hits in all three cases; potentially useful for selected data sets?
only for Data Sets with Calibration Hits

Note that the cell content of an uncalibrated cluster and the corresponding calibrated cluster is identical. Having the corresponding uncalibrated and calibrated clusters in the same order in their specific collections also helps with access and comparisons.
Similarly, the clusters from calibration hits have the same cell content as the corresponding uncalibrated and calibrated clusters. They should also be ordered such that corresponding clusters have the same index in the collection. The specific truth representation provided by these clusters should not only include the calibration hit energies in cells, but also the associated dead material losses, if any. This allows to compare direcly to the calibrated clusters, for which a dead material correction is performed.
Backnavigation to ESD should be turned on for these special AODs, because it might be necessary to retrieve CaloCells in certain calibration studies.
All collection names used here are very preliminary and should be replaced by more appropriate names, if needed!

Major updates:
-- Main.lochp - 24 May 2006 -- Main.lochp - 30 May 2006

%RESPONSIBLE% PeterLochSecondary

Topic attachments
I Attachment History Action Size Date Who Comment
PDFpdf 0512210.pdf r1 manage 133.8 K 2006-05-24 - 20:24 PeterLochSecondary  
Edit | Attach | Watch | Print version | History: r10 < r9 < r8 < r7 < r6 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r10 - 2008-06-24 - PierreAntoineDelsart
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Atlas All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2020 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback