Based on Phys.Rev.D83:074023,2011 (arXiv:1012.1207) by Yanou Cui, Zhenyu Han and Matthew D. Schwartz.

If you just want the code, here is the C++ implementation of our W-tagger.


Efficient W-tagging can save signals from background

Before W-tagging -- new physisc (Z') is lost in background

before

After W-tagging -- new physics (Z') is clearly visible with high significance

 

How do we identify highly-boosted hadronically-deacying W bosons?

The key is a multivariate approach using jet substructure. We use the following 7 variables

  1. Jet mass with R=0.5 Cambridge/Aachen Jets
  2. Jet mass with R=0.4 Cambridge/Aachen Jets
  3. Filtered Jet mass
  4. Mass of hardest subjet, from filtering
  5. Mass of second hardest subjet, from filtering
  6. ratioof the pT's of the two hardest subjets
  7. Planar Flow

The variables are combined using Boosted Decision Trees. The result can increase S/sqrt(B) by larger than a factor a 5, as compared to preious approaches (filtering + mass drop) which acheived slightly larger than a factor of 2, as shown in this figure

Wtagefficiencies


The C++ CODE impliementating the W-tagging algorithm can be found here.