“Particle Identification Algorithms for the Medium Energy ∼ ( 1.5-8 GeV) MINERνA Test Beam Experiment” Tesista: Antonio Federico Zegarra Borrero Asesor: Dr. Carlos Javier Solano Salinas UNI, March 04, 2016 1
Contents ● (1)The MINERνA Experiment at Fermilab ● (2)Medium Energy MINERvA Test Beam experiment. ● (3)Tools for Data Analysis & Particle ID ● (4)Results on the composition (% p ± , π ± , μ ± , e ±) of the secondary beam for different energies & polarities ● (5)Efficiency-Purity analysis to find the optimum cuts to separate different species for the 2GeV sample ● (6)Conclusions 2
(1)The MINERνA Experiment at Fermilab ● Neutrino-Nucleon interactions & Neutrino Oscillation Experiments. ● Many Interaction-channels with different Cross-Sections at different Energies. ● Particles in the final state have a Specific-Pattern of depositing Energy inside the MINERνA main detector. ● Their Identification is important for the Reconstruction of the specific Event. ● Calculation of Cross-Sections. 3
The NuMI Beam & the MINERvA Main Detector 4
Data Acquisition (DAQ) 5
Energy Dependence of Neutrino Interactions 6
Some examples of Neutrino Interactions ● CCQE: 7
Looking them in Arachne ... 8
● RES production of a single π: 9
(2) Medium Energy MINERvA Test Beam experiment. ● Main Goal of the Medium Energy Test Beam experiment. ● Why ? ---> Test MC simulations ● How ? ---> Constructing up and analyzing a Beam ● Beamline elements: 10
11
Test Beam Detector Configurations ● Pion & Electron Data Samples (Folders) ● Advantage of the ECAL/HCAL configuration ● Different species –---> different behavior inside different Regions of the Detector (Energy deposition pattern, Number of Hits, etc...) 12
Time of Flight Device Elements making up the 2 Stations (Upstream & DOWNstream) ● How this device separates different species (different masses) ● Interpretation of the ToF measured-time histogram ● Limitations (Resolution: at E > 8GeV & particles inside the Pion-peak) ● 13
Early Result (ToF histogram) from its usage (Data from February 2015): ● 14
Limitations in the resolution of the ToF ● Resolution (ToF): 100-200 ps ● Considering a momentum as low as 1 GeV/c: ● At E>= 8GeV other process (DIS) dominates neutrino-Int. 15
(3) Tools for Data Analysis & Particle ID ● ROOT via C++ or python (concepts of DST, Chain, Tree, Branch) ● Ways to perform analysis: Monte Carlo simulations, Scatterplots, physical criteria ● Energy Deposition Patterns: Ionization (dE/dx), Electromagnetic Showers & Hadronic Showers ● Visualization of Events (for eye-scanning) via Arachne (a software developed by MINERvA) 16
Interaction of particles with matter 17
Ionization 18
Electromagnetic Showers 19
Hadronic Showers 20
Radiation & Interaction lengths in our Detector. 21
Tracks in the TB Detector seen in Arachne for a given view (XZ) 22
23
Mandatory Conditions to retain meaningful Events (==Particles) ● The Beam has to be ON: ● There is Activity in the Detector: ● All 6 PMTs of ToF Stations fire: ● The Veto does not fire: ● The Event occurs in the Triggered Slice: ● Relevant to know what Veto Branch to use (Veto Sanity Check) & an Analysis of Correlations between the ToF & the Veto -----> 1 Event == 1 Particle ------> Start Isolating Species ! 24
What is a Slice ? 25
Methodology for Particle Isolation ● In the ToF_measured_time histograms: – 1)Isolate the protons – 2) Eye-Scanning Events in Contamination-Intervals – 3)Isolate Events in the ToF Pion-peak (containing pions, muons & some electrons). – 4)To Isolate Species inside the ToF Pion-peak (definition of Detector-Variables) – For E >= 4GeV a cut in the histogram of Total-Energy deposited – For the 2GeV sample we need to cut on more than 1 Variable ! 26
● To separate Species (e, µ, π ) inside the ToF π-peak: – MC simulations of pure µ to find what Variable separates them better (µ are easier to locate). – Definition of many Detector Variables to see which one works better (via python dictionaries of dE/dx, PE & Hits per module). – How to look at any electron. 27
28
29
(4)Results on the composition (% p ± , π ± , μ ± , e ±) of the secondary beam for different energies & polarities ● Methodology used for the 8 GeV π + sample (the same for the 4 & 6 GeV for both + & - polarities ) 30
● Separating species in the ToF Pion-peak 31
● Patterns of Isolated species (from Data) & Pure species (Monte Carlo) 32
● The Relevant-Intervals in all histograms used for the isolation (in the ToF & Total-E) & how e+ are located (and counted) are detailed. The same criteria was used for the 6 & 4 GeV samples. Here Results for the 8GeV π+: 33
● Methodology used for the 2 GeV π + sample - More than 1 Var to separate species in the Pion-peak 34
● For separating µ from π in the ToF Pion-peak – Not possible to rely on 1 single variable to separate µ from π – 5 different kinds of cuts combining different variables – The initial Logic was to look at π by knowing where µ are located – These are the cuts that look at π (UNIONS) : 35
● Let's see how was made CUT-1 (for example) with the aid of MC simulations of µ (union of different cuts were used to look at π): 36
37
38
39
● So the particular cut called CUT-1 is a UNION of cuts of the Energy deposited in different regions of the detector & looks at π. 40
Results for All Energies & Polarities These results are estimations because there will never be perfect PID ● algorithms. For the 2GeV samples the specific CUT-i used are shown. ● 41
(5)Efficiency-Purity analysis to find the optimum cuts to separate different species for the 2GeV sample ● The Cuts used for the 2GeV samples: composed of a UNION of cuts over different variables (5 per kind of Var). ● For Data Analysis: Reduce Number of Cuts –-> Reduce Systematic Uncertainties. ● The IDEA: construct an OPTIMUM-CUT composed of only 2 cuts (among the 20 Variables) ● For this reason an Efficiency-Purity Analysis was reliable ● To analyze each of the cuts & the effect of one after another –--> CHANGE IN THE LOGIC needed. This will look at µ instead of π 42
Change in the Logic to look at µ 43
For the MC an extra condition for the PE for the Hits for any Event was needed... 44
The Analysis began while looking for the Best Variable to separate µ from π … 45
Concepts of Efficiency & Purity 46
47
48
49
● Best cuts to separate µ from π. Plot histograms of other Vars for events in the µ-interval (for the best cut) –---> To find the Best second cut. 50
Methodology followed ● Select the Best-Cut (let's call it in Variable ) to retain Events in the µ-Interval ( ). ● Fill histograms of the other variables with the previous Events to see which variable (let's call it ) separates better the µ & π present there. Select remaining µ in the new µ-interval ( ) of this new histogram. ● 8 candidates were selected as the second cut (to add to the one cited above) and the most efficient (in selecting µ) among them was chosen. 51
New-Logic of the “Cuts”: ● Cut_µ =={ } ● Cut_π == (*)=={ } == ~ Cut_µ ● Applying these cuts to the MC samples of pure µ & π we can find the efficiencies: – Fraction of µ looking as µ(pass the Cut_µ): – Fraction of µ looking as π: – The same for the case of π: ● The best cut was chosen as the one which maximizes 52
Candidate to be the Var 53
● In next slides: histograms of the variable (8 candidates) for events in which ● For each case: new µ-interval is shown, together with the four numbers: ● The 8 candidates for are: – Total_E – Total_E_HCAL – Total_PE – Total_PE_HCAL – <dE/dx>_Total – <dE/dx>_HCAL – Total_Hits_HCAL – Total_Hits_L8P 54
Due to its highest efficiency, the 2nd variable was chosen to be ● Total_PE_HCAL 55
56
Application (of this cut == composition of 2 cuts) to Data ● Relations : 57
A similar procedure for the Cut (only 1 cut in variables of type Var_i_ß is enough) that separate e from µ : (Here just results of the best cuts found) 58
Type of cut refers to any of the 20 sub-cuts to separate e from µ 59
60
61
Some notes about e-µ separation ● Many good cuts to separate e from µ. ● Many cuts in LP-Vars are almost perfect. We expect that electrons will almost never arrive at the LP so this is physically expected. ● I believe that the best-cut (Hits_L8P) is enough for a very good separation. ● The best cuts to separate any e that may be in a µ sample would be the ones with highest values of Eff*Pur: 62
Recommend
More recommend
