25 April Antineutrino selection for constraining the e beam Goal: extract component of rate from + decays Requirement: High purity at low neutrino energy Pedro Ochoa(CalTech) & David Jaffe(BNL) This is what we are trying to measure Result: Development of preliminary ‘standard’ cuts
25 April First tried to reproduce Jeff’s cuts described in his talk at Oxford (minos-doc 1409): For this, used tracks in fiducial volume (1m<vtxz<5m & vtxr < 1.0m), and: 1) q/p > 0 2) Fit.pass + chi2<ndf <10 + UVasym < 6 3)|(q/p)/(σ q/p)|<0.3 4) Prob(chi2,ndf)>0.1 5) Petyt PID > 0.4 Starting point: Jeff Hartnell’s cuts Jeff’s cuts of Oxford Using powerpoint, a miracle of modern technology, the current results are compared to Jeff’s Oxford results on next slide…
25 April Jeff Background Comparable results achieved All neutrinos Selected as antineutrinos Background Pedro Overall efficiency: 52.5% Overall purity: 98.2% Pedro Background composition
25 April Overall efficiency: 52.5% Overall purity: 98.2% Reconstructed neutrino energy (GeV) This is what we are trying to measure All antineutrinos Selected as antineutrinos Background
25 April Jeff’s cuts work well but for our analysis we want lower background at low energy. Worked on improving the NuBarPID ! The first improvement came out by noticing that separation is better for longer events (all distributions normalized to unit area) : Our selection (q/p) / (σ q/p) 0 < Planes < 3030 <= Planes < 6060 <= Planes < <= Planes < <= Planes < 153
25 April So tried the following 2D PDFs for the NuBarPID (in addition to number of planes, y, and dcosz) neutrinosantineutrinos Note: Every “row”, or slice of planes (for instance from 0 to 30) is normalized to unity, as seen in previous slide. This reduces the energy dependence of these 2D PDFs and keeps them independent of the PDF of the number of planes. (q/p) / (σ q/p) Event length (planes)
25 April An improvement is observed ! After Before After Purity Efficiency Some (probably very long) events are really well separated ! Here the efficiency does not include the basic cuts. Before
25 April Now for something slightly different: Scan 30 events with looser Petyt PID cut to try to increase low E acceptance Cuts: 1m<vtxz<5m vtxr<1.0m q/p > 0 UVasym < 6 Prob(chi2,ndf)>0.1 Petyt PID > <Rnear<2m Rnear (m) Rnear = smallest radius on track. Small radius: near coil hole, higher Bfield. Large radius: lower Bfield Accept
25 April Scan results for non-CC that passed cuts on previous page identity
25 April Conclusion from scan: Comparison of momentum from range and curvature can reject some 1)protons because conversion of range to momentum assumes muon mass and 2) - because range of kinked tracks is unchanged. Effect of cut on (p(curve)-p(range))/p(range) for investigated on following pages. Would it be useful for CC analysis? Alternative might be to compare expected and measured dE/dx for strips on track.
25 April Based on scan result, Pedro tried adding an extra cut on (p(curvature)-p(range))/p(range), only for tracks that stopped in the detector, to the NuBarPID: Used NuBarPID with 4PDFs: 1) 2D q/p/(σ q/p) vs. planes 2) planes 3) y 4) cosz The pdfs were made with with following basic cuts applied: 1 < Zvtx < 5m Rvtx < 1m At least 1 track Trk.fit.pass==1 U-V asym < 6 /ndf < 20 Plots of Purity vs. Efficiency were made. The efficiency now includes all cuts (including a cut on ). In other words, efficiency is measured with respect to all CC nubar events.
25 April NuBarPID and - No extra cut - x=1.0 - x=0.5 - x=0.3 - x=0.15 A small improvement, but it’s not enough !
25 April NuBarPID and: - No extra cut - x=0.15 cut - Prob(chi2,ndf)>0.1 cut Combination of NuBarPID with one of Jeff’s cuts, Prob(,ndf) > 0.1 = “fit significance cut” gave the best performance : BINGO !
25 April Interesting ! Separation looks different when calculating doing the PDFs with and without the fit significance cut: In both cases the fit significance cut is applied. The difference is whether or not the PDFs were calculated with it or not. At the end, not much difference in separation even if shape above is so different PDFs done with fit sig. cut PDFs done without fit sig. cut NuBarPID PDFs done without fit sig. cutPDFs done with fit sig. cut Purity Efficiency
25 April Tried combining NuBarPID + fit significance cut + cut: Purity Efficiency NuBarPID NuBarPID + fit sig. + prange cut NuBarPID + fit sig. No improvement. Will stick to NuBarPID + fit significance. Note: PDFs were calculated with all corresponding cuts included.
25 April From now on always included fit significance cut (among all others) when calculating the PDFs. Now, need to see what happens as a function of energy. Make a NuBarPID cut at 0.7 and see what happens: NuBarPID > 0.7 puts you here Purity Efficiency
25 April Compare NuBarPID> 0.7 to purity & eff’y with Jeff’s cuts: PurityEfficiency Jeff’s cuts NuBarPID>0.7 and Prob(chi2,ndf)> NuBarPID>0.7 and Prob(chi2,ndf)>0.1 This is what we trying to measure Overall efficiency: 50.2% Overall purity: 99.5% Low energy purity improved but with some loss of efficiency Some increase in higher energy efficiency
25 April NuBarPID>0.7 Eff’y 50.2% Purity 99.5% NuBarPID>0.75 Eff’y 48.5% Purity 99.6% NuBarPID>0.80 Eff’y 46.7% Purity 99.7% Further tightening of NuBarPID cut Characteristics of remaining events? Current study probably suffers from lack of stats
25 April Conclusions Jeff Hartnell did a good job. NuBarPID >0.7 improves purity at low energy with some loss of efficiency. We now have preliminary selection criteria for a sample to constrain the e flux. May still investigate possible further improvements in low energy selection Any of this applicable to other analyses?
25 April
25 April If make cut at NuBarPID>0.7 find: Overall efficiency: 50.21% Overall purity: 99.48% All neutrinos Selected as antineutrinos Background
25 April If make cut at NuBarPID=0.75 find: Overall efficiency: 48.52% Overall purity: 99.63% All neutrinos Selected as antineutrinos Background
25 April If make cut at NuBarPID=0.80 find: Overall efficiency: 46.67% Overall purity: 99.73% All neutrinos Selected as antineutrinos Background