========================================
September 25th 2017 - Dilepton Z' search
========================================

Chaired by Cesar. Penultimate session on this paper.
Today's session was predominantly understanding the background estimation portion of the
paper: section 6, and the the system of equations that describe it (eqn. 1). 

----------
Topics: 
	- Background estimation when the fake rate is unknown
    - Extrapolating the estimate over the full mass range

----------
Details: 
    - In the Atlas convention: real efficiency from MC is the 'fake factor', fake efficiency from data is 'matrix method'
	- The task here is to estimate r,R,f,F from somewhere.
	- Generally, we need r to be large and f to be small. r should be on the order of 90%, f is usually 1e-3
	- For the most part, the 4x4 matrix is overkill, and we can get away with a 2x2 version of the matrix if f is small. 
	- A 2x2 version, for a single lepton rather than dilepton search is:
	  (NT)  =  (  r   f  )(NR) 
	  (NL)     ( 1-r 1-f )(NF)
	- Why don't we take f from MC? There is not necessarilly enough of a selection for dijets in MC. You won't get a 
	  good result becuase you're only measuring a small amount of the distribution. We can't trust that MC gives a good
	  measure of the jet fragmentation. It also depends on the type of fakes, which can be factors of 2 or 3 difference from truth.
	- Electrons coming from photon conversion are also not estimated properly: we never have enough material so we don't accurately 
	  predict the rate of them.
	- Fragmentation function also affects the heavy flavour phase. We can do reasonably well with the kinematics and multiplicity of 
	  the jets quark content, we cannot do as well at estimating what the jets look like around those decays as there we care not only 
	  about the jet itself but also its energy.
	- There are 3 different variables that matter in fakes for this analysis: relative contribution changes as a function of Et. For
	  heavy flavour, the probability of isolation also changes as a function of jet pT. Jets in different parts of the calorimeter 
	  will also look different (much of which is due to the physical construction of the calorimeter), so its also good to split
	  fakes from jets up in eta. We also have to consider whether a jet is coming from gluon of quark: different shapes. 
	- If there is a jet near a jet that has already passed the loose electron selection (and the loose selection does not contain isolation criteria)
	  then the fake rate will be higher becuase there is a greater chance of fake processes contamintaing the sample. 
	- When extrapolating your background estimate to the full mass range, you can decouple correlations by increasing some of your estimates by 1sigma and 
	  some by -1 sigma. Changing the whole background estimate by +/- 1 sigma is also a good way to propagate the uncertainties on the background.


**************
Next time: systematics and statistics, brief discussion on results, how do you get a limit?
**************