Data and Monte Carlo Processing

The entire 1997 data set has been processed. Table 1 shows the number of events that remain after the various levels of processing the experimental data and Monte Carlo. The efficiencies are computed by dividing the number of events at a given level by the number of triggered events.

 

Table 1: The efficiency is relative to the number of events that triggered the array. The Monte Carlo is processed the same as the experimental data. Note values shown for MC signal are unweighted (see section 2.3 for details).

	Data (138.2 days)		Background MC		MC Signal
Process Level	Events	Efficiency	Events	Efficiency	Events	Efficiency
Generated	--	--	1.675x107	--	3x106	--
Triggered	1.234x109	1	879716	1	459907	1
LBNL Filter	23.05x106	0.01868	13649	0.01552	243791	0.5301
UCI Filter	2.11x106	1.71x10-3	1284	1.46x10-3	174989	0.3805
Optimized Cuts	1269	1.028x10-6	0	0	67897	.1476

The raw data was processed through the LBNL filter [2] giving $\sim$6.5 Gbytes of data. To reduce CPU time and lower disk space requirements by a factor of $\sim$10, this data set was further reduced by additional restrictions on the LBNL filter procedure, termed UCI filter in table 1. The additional filtering was implemented by MUFF [3], before processing it with MUNT [3] to produce HBOOK files. The HBOOK files are then read in as ntuples into PAW to analyze the data. The UCI filter reduces the signal efficiency to roughly 75% relative to the LBNL filter efficiency, but it reduces the background data on disk to a more manageable quantity.

The cuts used in the UCI filter are: 1) zenith(2)><>100 degrees, 2) zenith(1)><>90 degrees, and 3) jkrchi(2)<<>8.5. The zenith angle cuts are simply a stronger requirement on the upward direction of the reconstructed track. This cut improves the signal to noise for upward going neutrinos (zenith><>100) by reducing the data by 30% and passing $\sim$85% of the signal events. Since the events near the horizon are background dominated by down-going muons and thus will be very difficult, if not impossible, to search for high energy neutrinos. Earlier studies of the jkrchi(2) (likelihood parameter for the full reconstructed fit) demonstrated that signal to noise improvefor high energy neutrinos over a large interval of zenith angles. (figure 37 in appendix C). In addition, this variable exhibited fair agreement between experimental data and background MC. These two factors generated confidence in the use of this variable for filtering. It is the most effective cut used, reducing the data by 85% while keeping $\sim$85% of the signal.