Hi Carole and Ozana,
this is just a brief message to outline the tag and probe strategy and its specialized usage for the electron identification stuff.
We have developed an algorithm which, given an object reconstructed as an "electron" in the CMS detect, returns the probability for it to be a real electron or a fake (typically pions or kaons inside a QCD jet).
This algorithm has been tuned and studied on MC samples of "pure" electrons and "pure" QCD jets used as background.
It is necessary to setup a working strategy to control the setup and the performances of this algorithm with the data (when we will have them). So you need to find what is called a "control sample", i.e. a sample on data which is made of electrons (for signal category) or pions (fo background category) with high purity where check that the output of your algorithm is what you expect from Monte Carlo simulation.
While it is quite easy to have control sample for QCD jets (you have a lot of them from minimum bias events which contains a negligible fraction of leptons), more difficult is find a pure sample of electrons.
A method has been studied, called tag & probe.
It is based o the fact that you will produce a lot of Z bosons which decay in e+e-. This will produce a lot of electrons also in the early stage of the experiment.
Since you have two electrons, you define your signal (an electron) as follows:
a) you look for a well reconstructed electrons which satisfy a number of quality requests.
You define this "electron" as "tag".
b) then you look simply for a cluster in the electromagnetic calorimeter and this is your
"probe" electron.
You don't require any electron identification request on this because you want to test the likelihood algorithm on this.
The only request you do on this is uncorrelated with the "electron properties" of the e.m. cluster: i.e. you require that combining it with the "tag" electron its invariant mass is consistent with Z mass.
This should reduce a lot the background.
What you could do is to apply the recently developed likelihood algorithm on the "probe" candidates and look at the performances of the algorithm.
This is a fundamental test because it is a test you do on data in order not to rely on Monte Carlo simulation which could not reproduce perfectly the data.
There are people already working on it and they defined the criteria to select the "tag" and "probe" electron candidates.
Take a look to this presentation:
http://indico.cern.ch/getFile.py/access?contribId=5&resId=1&materialId=slides&confId=12396
you could try to reproduce their selections to define the "tag" & "probe" samples.
Then a further step could be this.
After having selected the "probe" samples, you have a hopely very high purity. But still there will be background events which smear the distributions of the discriminating variables which are the inputs of the likelihood algorithm.
One idea could be to do a background subtraction based on the di-electron invariant mass.
This is a statistical technique, and could be done in different ways. A smart way of doing it could be to fit the di-electron invariant mass with a model for the signal and a model for the combinatorial background. This will provide a event-by-event probability for the event to be "signal". You could then weight the event with the likelihood to be signal and this will provide
automatically a "background-subtracted" distribution for the discriminating variables.
A reference for this can be this:
http://arxiv.org/abs/physics/0402083
One target for us can be to set up an automatic tool to do this. This is complicated by the fact that
we have to do things for the different classes, for barrel and endcap, for different pt's, etc.
For spin reasons, for example, you can find very few events of Z decays at eta~0 and low Pt's,
so there are a number of challenges to face.
You could enjoy them ;)
We will talk about this when we are all together at CERN (next days).
Ciao!
emanuele
