Transcript oscfit
Beam Extrapolation Fit
Peter Litchfield
An update on the method I described at the September meeting
Objective;
To fit all data, nc and cc combined, with the minimum of cuts
To use the beam MC extrapolation parameters event by event to
produce a far detector prediction from the near detector data
Not to need beam, cross-section and/or reconstruction error fitting
Status
John Marshall is developing an independent program on the same
lines. John (Mark) is reporting his results in the cc session
I have used MDC MC both raw and tweaked to develop and verify
my program
I will show that it works, at least on MC data
Reminder of the method
GNuMI
Beam
particle
Near MC
truth
event
Near
MC reco
E - Es
Far MC
truth
event
E - y
Weight: Oscillation
Beam extrapolation
Gen/Extrapolated ratio
Far flattening weight
Xsec ratio
many
beam
particles
Predicted
Far reco
E - Es
distribution
Weight:
near data reco/
near MC reco
compare
Far MC
truth
event
weighted
Far data
reco
E - Es
distribution
Far MC
reco
event
E - Es
Data
All data is MC, I have not looked (for a long time) at any real data
MDC data, R18.2 reconstruction
Pure MC, no tweaking, far data oscillated (original MDC)
Near “data” 385 files : 0.03955 1020 pot
Near MC
382 files : 0.03934 1020 pot
Far “data”
100 files : 102.7 1020 pot
Far MC
177 files : 514.2 1020 pot
Tweaked MC, far data oscillated (MDC3)
Near “data” 396 files : 0.3996 1020 pot
Near MC
379 files : 0.3893 1020 pot
Far “data”
100 files : 103.2 1020 pot
Far MC
177 files : 514.2 1020 pot
Near detector E v Eshw weight
Untweaked MC
Plot reconstructed E v Eshw
Only cut is that the
reconstructed vertex should
be in the fiducial volume
No nc/cc separation
Sign of E is that of the
reconstructed
One bin for events with no
Bins of 1 GeV 0-10 Gev, 10
GeV 10-60 GeV
Tweaked “data”
Near detector E v Eshw weight
Weight the beam MC event by the ratio of near data to near mc in
the bin of E v Eshw
For untweaked MC should be 1, Could do with more statistics
Ratio near data/near mc
Eshw
(GeV)
-ve momentum
E (GeV)
+ve momentum
Tweaked Near E v Eshw weight
Tweaked MC, ratio different from 1
Ratio near data/near mc
Weights the near MC to allow for beam, cross-section and
reconstruction differences
Eshw
(GeV)
-ve momentum
E (GeV)
+ve momentum
Extrapolation to the far detector
Near-far extrapolation is done with only truth quantities
Each near detector mc event has a truth energy that a neutrino
hitting the far detector from the same beam particle decay would have,
together with the probabilities that the near and far detectors are hit.
Use far detector mc events with the same truth characteristics as
the extrapolated near detector event
Problem: the far detector energy is different from the near
therefore cannot use E and Eshw. Instead extrapolate in truth E
and y which should at least approximately scale.
Select events with the same truth initial state (nc,cc,qel,dis etc) and
in the same bin of E v y
Apply the far detector reconstructed fiducial volume cut and plot the
reconstructed E v Eshw distribution with the weights on the next slide
Again the only cut is on the reconstructed fiducial volume
Far detector extrapolation
Each selected far detector MC event has the following weights applied
The ratio of the probability of the neutrino hitting the far detector to
the probability of hitting the near detector
The ratio of the far to near fiducial volumes
The ratio of the pot in the far and near detector samples
The ratio of the cross section at the energy of the far detector
event to that at the energy of the near detector event
A weight to flatten the far detector events as a function of E and y.
Necessary to remove the cross-section dependence in the far MC
A weight to allow for the difference in truth distributions of accepted
events in the near and far detectors (see next slides)
The near detector data/MC weight
An oscillation weight, dependent on m2, sin22, fs
Far detector extrapolation
`Problem: the truth MC distributions in the far detector are not the
same as the extrapolated MC near detector spectrum
Truth E
Far MC
All events
-60.0
Extrapolated ND
0.0
E
60.0
`Due to split and superimposed events in the near detector
MC truth finder usually associates bigger MC event with the event
Split events, the MC event gets extrapolated twice
Superimposed events, the bigger event gets extrapolated twice, the
smaller event is lost
Far detector extrapolation
`Effect bigger for vertex selected events,
Differences in reconstruction efficiencies?
Non uniform vertex distribution in near detector + vertex resolution?
?
Weight events with the ratio far/near of events in the E-y bin
Selected
events
-60.0
Far MC
Extrapolated ND
0.0
E
60.0
Far detector weight
The extrapolation weight for the near to far truth should be close to
1.0
Far MC/Near MC projected
Could do with more statistics
y
E (Gev)
Raw MC fit
No oscillations
Fit to oscillated but
untweaked MC, test that
the program works.
Far data
Extrapolated
near data
nc
Use the MDC MC,
oscillated with parameters
m2=0.0238, sin22=0.93
Fitted to E v Eshw but
difficult to see effects,
project onto E
No cc/nc selection but
plot E for data divided into
nc/cc by Niki’s ann
cc
-60.0
0.0
E
60.0
Raw MC fit
True oscillated parameters within the 68% confidence contour
MC statistics is lacking, still contributions to likelihood from MC
nc
m2 0.0025
Oscillated
0.002
cc
-60.0
0.0
E
60.0
68 and 90% contours
0.9
0.95
▲ truth
sin22 1.0
* best fit point
Tweaked MC, Near data/MC
Note ratio now generally > 1.
Ratio near data/near mc
MDC3 data.
Eshw
(GeV)
-ve momentum
E (GeV)
+ve momentum
Tweaked MC , no oscillations
No oscillations
Far data
Extrapolated
near data
nc
Prediction from near
data includes correction
for tweaking
Truth oscillations have
different parameters
cc
-60.0
0.0
E
60.0
Tweaked MC, best fit
nc
cc
-60.0
0.0
E
60.0
0.0025
m2 0.003
Oscillated
0.75
▲ truth
0.80
sin22
0.85
* best fit point
Include sterile oscillations
Fits well with no sterile
component, therefore don’t
expect much in fit
▲
Summary and Conclusions
The beam event-by-event extrapolation works.
It works (on MC) without beam or cross-section fitting/adjustments
It works (on MC) without any cuts except a fiducial volume cut.
It works (on MC) for a fit to m2, sin22 and fs
It should work for a CPT separated and fit
Fitting to reconstructed E v Eshw includes the detector resolution in a
simple manner
I haven’t thought much about systematics but since it makes very few
assumptions and cuts, the systematic errors should be small
It will work as far as there are no effects unique to one detector which
are not represented by the MC
Need to compare far and near detector data to check that no such
effects are present.