Transcript oscfit_nc_dec07

Update on the nc analysis using
the beam extrapolation method
Peter Litchfield
A number of problems appeared when I tried to apply the beam
extrapolation method to the fits including the nc events. I believe
these are now fixed.
I have analysed the two MDC files and obtain the truth answers
I will show the problems which may or may not be relevant to
other methods
Quick summary of the method
 This is a test of the method and MC, it is done with NO data cuts other
than the fiducial volume cut (cc analysis). It remains to be seen whether the
real data has effects which cannot be compensated by the method.
Each near detector MC event in the PAN’s is extrapolated individually
according to the beam MC estimate of the energy at the far detector of a
neutrino from this beam particle decay and the relative probability of hitting
the two detectors.
Far detector events are selected with the same truth E and y and of the
same physics truth type (nc/cc, qel/dis/res). All near detector MC events
are used in the extrapolation.
The fits are done to the 2-D distribution of reconstructed E v Eshw There
is no separation of nc and cc events and no events are rejected.
For this analysis the near detector data is a second MC near detector file,
there is no systematic difference of the near MC and data
NC Problems
 Some problems surfaced in the nc analysis which were present but
probably insignificant in the cc analysis.
1) The difference between the generated far detector MC truth and
the truth extrapolated from the near detector
2) The difference between the number of reconstructed near
detector events with a vertex in the fiducial volume and the
number with truth vertices in the fiducial volume
3) The effect of MC statistics. I am using the full MDC statistics
(~80000 events) in order to see the fit biases. The MC only has
a factor ~2 more events.
Generated v Extrapolated Truth
 Note the excess of
generated events at low Eshw
in the nc sample
I believe that this is due to
the loss of small nc events in
the near detector due to
superimposition.
I have used near singles
MC for the extrapolation (see
insert) which will not have
this problem. Better but the
statistics is very limited, more
singles files have been
reconstructed, when the
PANs are run I will try them
nc events
singles
True Eshw
cc events
generated
extrapolated
Signed E
Fiducial volume cut
 Plots are the ratio of numbers of
events with truth vertices in the fiducial
volume to reconstructed vertices in the
fiducial volume as a function of truth
Eshw for the near detector (top) and far
detector (bottom)
Note that in the low energy region the
near detector has ~4% fewer truth
vertices than reconstructed.
I think this is due to showers being
reconstructed in the volume from
events produced outside (in the uninstrumented region?)
The far detector ratio  1.0
CC ratios are both  1.0
MC statistics
 The 2D plot of E v Eshw has many bins with small numbers of events,
particularly at high values of each parameter
The likelihood function takes account of bins with no data events
correctly but does not account for bins with zero MC events
Previously I was taking no account of MC statistics
Go to a 2 function with a limit on the number of data or MC events of
10 in each bin. Accumulate all the events from these bins in one
overflow bin which is used in the 2. These bins are almost all in the
region not affected by oscillations.
Use as the error in the 2 the statistical errors on the number of
events in that bin in the far data, far MC, near data and near MC added
in quadrature.
This probably slightly overestimates the error, the 2/number of bins is
 0.5 but is better than not correcting.
Results MDC3
 Shown is reconstructed E for events selected as nc or cc by Niki’s
ANN PID. For events with a track E is signed by the sign of the , events
with no track have positive E
Note that these are for illustration. In the fit the events are NOT
separated and the fit is to the 2-D E v Eshw distribution.
The predicted distribution is oscillated with the truth parameters
Contours MDC3
sin22
Best fit
Truth
.7
sin22
0.84
2
.7
3.5
m2x10-3eV2 3.9
0.84
Small remaining discrepances maybe due to incomplete compensation of
lost MC events
0
0.2
3
30
NC selected
CC selected
0
0
0
1
15
10
2
All events
0.4
18
1D fsterile 2 plots
fsterile
0
0.2 fsterile
0.4
0
0.2 fsterile
0.4
0
0.2 fsterile
0.4
MDC3 without corrections
0.84
Without corrections
sin22
sin22
0.84
With corrections
Best fit
.7
.7
Truth
0
0.2
fsterile
0.4
 Without the generation and
fid volume corrections too few
nc type events are predicted,
thus the sterile component is
reduced
0
0.2
fsterile
0.4
0.95
sin22
1.0
0.8
2.65
Truth
fsterile
CC selected
0
fsterile
0.1
0.1
2
NC selected
0
0
0
2
2
All events
0
0.7
Best fit
.95
4.0
sin22
m2 x 10-3
1.0
2.9
MDC1 contours
0
fsterile
0.1
0
fsterile
0.1
 The lack of nc type events drives the best fit to unphysical fsterile<0
Adding ’s
Contrary to what I thought at the collaboration meeting it was not the
lack of ’s which was causing me to find other than the truth
Adding  makes very little difference to the overall fit
data
 in nc
 in cc
0.84
0.84
pred
With 
.7
.7
sin22
sin22
No 
0
0.2
fsterile
0.4
0
0.2
fsterile
0.4
Conclusions
Extrapolation of all events, cc and nc works
Corrections which were insignificant for the cc analysis are
important when fitting for sterile components using the nc sample
I can add the electron MC in the same way I added the taus, but I
suspect it will be insignificant like the taus
I can fit any model and parameters required but I think the fit I
have done essentially encompasses all practical possibilities