Skip to content

tauRecTools: enable cells for 0-prong taus and update 0p MVA TES (try3)


This MR is the 3rd resubmission of the same MR (!45290 (merged) , !45294 (closed) ) where we saw random count variations in the digest due to concurrent muon MRs. Sorry if 3 identical MRs makes no sense, but I don't understand how the CI and digest work. At least this time I did not get fake muon changes, so it should work fine :-) Original description pasted below.

This MR is hopefully the last big MR for taus for the R22 reprocessing! It implements two features.

First, it is revising the input variables used for the MVA Tau Energy Scale calibration for 0-prong taus. The new variables improve slightly the resolution. Also, true 3-prong taus reconstructed as 0p were given increased weight in the BRT training, as the resolution for 3p is natively quite bad with this calo-only energy calibration.

Second, it enables the pi0 and photon shot reconstruction for 0p taus. So far, the reconstruction of these tau constituents was limited to 1-5 prong, because tau substructure reconstruction is not run for 0p nor >5p. However, we anticipate to rerun a dedicated LLP->tau reconstruction with Large Radius Tracks on AODs (ATLTAU-1754). Therefore we need to build pi0s and shots in the RAWtoESD step, and save the associated cells in the AOD, to make substructure reconstruction possible for 0p at AOD level (ATLTAU-1770). In order to limit the AOD size increase to ~5 kb/event (ATLASRECTS-6505), the tau pt cut applied to 0p taus when writing to AOD was raised from 7.5 to 9 GeV. To save more space, tau tracks for 0p candidates are also no longer written to disk, and will have to be re-associated (together with LRTs) and re-classified in the dedicated LLP reconstruction. Edit: this was actually not done because it was causing trouble downstream (monitoring).

I have verified that if doPi0andShots has the old behaviour (true for 1-5 prong, false otherwise), the tau reconstruction is unchanged (when ignoring TES + thinning cut changes), in particular the neutral PFO correction is identical.

Pi0s and shots are built for 0p, but are not used in the substructure reconstruction, where 0p are basically still ignored. This is why I moved the vertex correction of neutral PFOs from TauPi0ClusterScaler (still run only for 1-5p, not 0p) to TauPi0ClusterCreator (now called for 0-5p, anyway a much better place to do that), so that the 0p neutral PF0s are properly corrected.

Adding the 'urgent' flag as we'd like to have this for the R22 reprocessing.

Cheers, Bertrand

Edited by Bertrand Martin Dit Latour

Merge request reports