Experimental Particle Physics Experimental Particle Physics - PowerPoint PPT Presentation

Experimental Particle Physics Experimental Particle Physics

Detector by function • Position: – Beam Tracker – Vertex Telescope – Multi-wire Proportional Chambers (MWPCs) • Energy: – Zero Degree Calorimeter (ZDC) • Charge: – Quartz Blade ∑ z ∑ 2 s s z • PID: PID: ∝ – Hadron Absorber and Iron wall 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 2

From position o track • That is the job for … reconstruction 1. Chose start and finish point 2. Try to fit track to targets targets 3. Add middle points 4. Check that all the groups have groups have points in the track 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 3

Reconstructed event 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 4

Experimental Particle Physics • Chose a particle and a particular decay channel channel. (PDG) • From that it will depend what is more i important for you in terms of detector, and t t f i t f d t t d tracks • For this presentation you’re going to see: 0 0 + + − K → π π s 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 5

Choice of good events • You need to make sure that all the detectors you depend for your study were detectors you depend for your study were working correctly at the data taking time. 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 6

First mass spectrum 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 7

Cuts This is 90% of the Work… What these cuts are: IV PCA Track distance Δ z They make sense because: • Daughter particles of a V0 decay originate at the same point in space originate at the same point in space • The particles have decay lengths of 2.7 cm (becomes 72 cm in the laboratory frame) 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 8

After the Cuts 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 9

Background Subtraction This is the other 90% of the Work… Combinatorial Fit The idea: • Take a “particle” that could be real but that you could be real, but that you are sure it is not. – Each track is from a different collision – The ditracks characteristics are according to the real ones • Take enough of them • Subtract their mass to • Subtract their mass to your histogram 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 10

Acceptances • The result you “see” has been biased by the detector and by the analysis steps the detector and by the analysis steps. • Now you must “unbias” so that you can publish a result comparable with other bli h lt bl ith th results. • This is again… 90% of the work • But after this you are done… You just y j have to write the thesis/article ☺ 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 11

Pause for questions 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 12

Multivariate analysis • Multivariate statistical analysis is a collection of procedures which involve observation and procedures which involve observation and analysis of more than one statistical variable at a time. • Some Classification Methods : – Fisher Linear Discriminant – Gaussian Discriminant Gaussian Discriminant – Random Grid Search – Naïve Bayes (Likelihood Discriminant) – Kernel Density Estimation e e e s y s a o – Support Vector Machines – Genetic Algorithms – Binary Decision Trees – Neural Networks 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 13

Decision Trees Node A decision tree is a sequence of cuts. f Choose cuts that h h partition the data into bins of increasing i bi f i i purity. Key idea: do so Leaf recursively. i l MiniBoone, Byron Roe 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 14

TMVA, what is it? • Toolkit for Multivariate Analysis – software framework implementing several ft f k i l ti l MVA techniques – common processing of input data common processing of input data (decorrelation, cuts,...) – training, testing and evaluation (plots, log-file) training testing and evaluation (plots log file) – reusable output of obtained models (C++ codelets text files) codelets, text files) 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 15

Implemented methods • Rectangular cut optimisation • Likelihood estimator Likelihood estimator • Multi-dimensional likelihood estimator and k- nearest neighbor (kNN) g ( ) • Fisher discriminant and H-Matrix • Artificial Neural Network (3 different implementations) • Boosted/bagged Decision Trees • Rule ensemble R l bl • Support Vector Machine (SVM) • Function Discriminant Analysis (FDA) F ti Di i i t A l i (FDA) 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 16

Advantages of TMVA • Distributed with ROOT • several methods under one 'roof‘ – easy to systematically compare many classifiers, – and find the best one for the problem at hand – common input/output interfaces – common evaluation of all classifiers in an objective way l ti f ll l ifi i bj ti – plugin as many classifiers as possible • a GUI provides a set of performance plots • the final model(s) are saved as simple text files and th fi l d l( ) d i l t t fil d reusable through a reader class • also, the models may be saved as C++ classes (package independent) (package independent), which can be inserted into any hich can be inserted into an application • it’s easy to use and flexible • easy t to implement i l t th the chosen h classifier l ifi i in user applications 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 17

Logical Flow 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 18

Plots 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 19

Correlation Plots 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 20

Comparison of all the methods Comparison of all the methods • In this plot we can see how good see how good each of the methods is for our problem. • The best method seems to be the BDT (boosted decision trees) that is basically a method that expands the usual cut method to more dimensions 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 21

Methods output Methods output All the methods output a number (the output classifier) that represents All the methods output a number (the output classifier) that represents how well the given event matches the background. Here we can see the distributions of this value for two chosen methods (the best: BDT and the worst: Function Discriminant Analysis). This plots can help us to pinpoint the cut value to chose for our study. s to pinpoint the c t al e to chose for o r st d 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 22

Where to cut Where to cut • Th The TMVA produces this kind of plots, which are very useful to help TMVA d thi ki d f l t hi h f l t h l deciding how pure the selected signal can be 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 23

Eye Candy 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 24

Eye Candy II 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 25

End 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 26

Backup 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 27

PID in NA60 This is the “muon part of NA60”: After the hadron absorber only muons survive and are tracked in the MWPCS After the hadron absorber, only muons survive, and are tracked in the MWPCS 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 28

Decision Trees 200 f(x) = 0 f(x) = 1 Geometrically, a decision y, tree is an n- dimensional B = 10 B = 10 B = 1 B = 1 histogram whose bins histogram whose bins S = 9 S = 39 are constructed 100 recursively y f(x) = 0 f(x) = 0 MT Hits Each bin is associated B = 37 B 37 PM with some value of the S = 4 desired function f(x) 0 E Energy (GeV) (G V) 0 0.4 MiniBoone, Byron Roe 29 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008

Decision Trees 200 f(x) = 0 f(x) = 1 For each variable find For each variable find the best cut: B = 10 B = 10 B = 1 B = 1 S = 9 S = 39 Decrease in impurity 100 f(x) = 0 f(x) = 0 = Impurity (parent) = Impurity (parent) MT Hits - Impurity (leftChild) -Impurity (rightChild) it ( i htChild) B B = 37 37 I PM S = 4 0 and partition and partition using the sing the E Energy (GeV) (G V) 0 0.4 best of the best 30 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008

Decision Trees 200 f(x) = 0 f(x) = 1 A common impurity A common impurity measure is (Gini): B = 10 B = 10 B = 1 B = 1 S = 9 S = 39 Impurity = N * p*(1-p) 100 f(x) = 0 f(x) = 0 MT Hits where p = S / (S+B) p = S / (S+B) B = 37 B 37 PM S = 4 0 N = S + B N = S + B E Energy (GeV) (G V) 0 0.4 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 31

How to use TMVA 13 Feb. 2008 Pedro Parracho - MEFT@CERN 2008 32