This page will summarize the work theta I did on the likelihood to separate signal form background in the context of long baseline neutrino experiments.

Index:

.

---

Current work

070514

work is in: /work/fdufour/t2kk/0705_like-10gev

I updated the plot for the cut which keeps 40% of signal (now I don't use the overflow bins anymore, which gives more sensible results).
I create the likelihood that use the cuts that maximizes S/sqrt(B).

Here are the plots:

Cut which keeps about 40% of signal
Cut which maximizes S/sqrt(B)

Comparison of both method


Plot of S/sqrt(B) (to check that the cut chosen is really the one which maximize S/sqrt(B):
(NB properly normalized for T2K)

for bin 1 - 4
for bin 5 - 8

and the cut value was chosen as in this file





Number of events summary in the T2K MC that I use:

E rec	Signal	Background
	raw -  normalized to T2K	raw    -  normalized to T2K
0-0.35 GeV	1390 -  9.18	1003 -  56.52
0.35-0.85 GeV	3045 -  20.11	920 -   47.77
0.85-1.5 GeV	1768 -   11.67	417 -  21.97
1.5-2 GeV	878 -   5.79	139 -   8.02
2-3 GeV	1280 -   8.45	166 -   10.48
3-4 GeV	557 -   3.67	90 -  5.71
4-5 GeV	191 -  1.26	24 -   1.739
5-10 GeV	282 -   1.86	34 -  2.51

070513

work is in: /work/fdufour/t2kk/0705_like-10gev

Up until now I always separate background from signal by placing the likelihood cut at 0, but the FNAL group place the likelihood cut such that they keep about 40% or signal in each energy bin.
This seem to be giving better results.

Here are a comparison of the 2 methods:

Cut at 0
Cut which keeps about 40% of signal

My next step will be to create a likelihood that uses the cut that maximizes s/sqrt(b)