HEP MC and Data Analysis Tutorial

Tutorial per il corso di Fisica Nucleare e Subnucleare, Dipartimento di Fisica, Università degli Studi di Trieste.

Slides: HepTutorialUniTS_Pinamonti_2015.pdf

References

MadGraph

• homepage: http://madgraph.hep.uiuc.edu/
• download page: https://launchpad.net/mg5amcnlo

Delphes

• homepage: https://cp3.irmp.ucl.ac.be/projects/delphes
• list of branches: https://cp3.irmp.ucl.ac.be/projects/delphes/wiki/WorkBook/RootTreeDescription
• to install it from MG5 shell: install Delphes

ROOT

• homepage: http://root.cern.ch/drupal/
• reference guide: http://root.cern.ch/root/html/
• to set it up on your ts-dip-phys pc: source setup.sh

Lecture 1

To start

Preliminary steps:

1. create a folder HiggsTutorial, which will be our "working directory"
2. dowload the latest stable version of MadGraph5 from this site: https://launchpad.net/mg5amcnlo (should be MG5_aMC_v2.3.3)
3. move the dowloaded file (MG5_aMC_v2.3.3.tar.gz) to our working directory and extract it (a new sub-directory MG5_aMC_v2_3_3 should have been created, which we will call "MG5 direcotry")
4. copy the script setup.sh (attached here: setup.sh) inside the working directory
5. from a terminal, from inside the working directory, execute the command source setup.sh: this will setup ROOT, gcc and python, and has to be done each time we open a new termnal (!)
6. test MadGraph: from inside the MG5 directory, from the terminal execute: ./bin/mg5_aMC; if it works you should be inside the MG5 shell
7. try to generate a physics process: from inside the MG5 shell, give the command generate p p > t t~ (this should prepare the generation of pp -> ttbar events)
8. exit MG5 shell, with CTRL+D

MadGraph preparation:

1. eventually modify the MG5 settings for the used browser and text editor: modify the text file input/mg_configuration.txt accordingly (suggested gedit or nano as text editor, firefox as browser)
2. enter the MG5 shell: from the MG5 directory, from a terminal (after the setup!) type ./bin/mg5_aMC
3. type install pythia-pgs to install Pythia6 (parton shower and hadronisation software)
4. type install Delphes to install the latest version of Delphes (detector simulation and object reconstruction fast simulation software)

Generate Processes with MadGraph

Let's start with the MG5 tutorial:

1. from the MG5 shell, type tutorial
2. follow the instructions on the screen

Then let's have one more exercise. What's the cross-section (at LO) for the process pp -> ttbar, with a CME of 7 TeV and a top mass of 172.5 GeV?

1. if you already have a ttbar directory, you can directly move to point 5
2. from the MG5 shell, type generate p p > t t~
3. then create the output directory, still from the MG5 shell, typing output ttbar
4. exit MG5 (CTRL+D)
5. enter the ttbar directory
6. enter the MadEvent shell: ./bin/generate_events
7. leave/set everything to FALSE (by typing 1, 2, 3, 4 or 5 and pressing ENTER each time)
8. press enter when ok
9. now we are asked to modify the cards: we want to change both the CME and the top mass, so we have to modify both run_card.dat and param_card.dat
10. to modify run_card.dat type 2, then ENTER
11. find the lines with the energies of the two beams and set them to 3500 GeV each
12. when finished, save the file and exit
13. now you can modify param_card.dat typing 1, then ENTER
14. find the line like 6 1.730000e+02 # MT, and set 1.725000e+2 instead of 1.730000e+02
15. when finished, save the file and exit
16. then press ENTER and the computation will start
17. wait few minutes you should get the result: you can read it from the screen or from the file ttbar/crossx.html

Process MadGraph outputs: Pythia, MadSpin and Delphes

We already installed Pythia. Now we can turn it on when generating events, in order to also simulate the parton shower and the hadronisation:

1. if you already have a ttbar directory, you can directly move to point 5
2. from the MG5 shell, generate the usual test process generate p p > t t~
3. create an output directory: output output ttbar
4. exit MG5 (CTRL+D)
5. enter the ttbar directory: cd ttbar
6. enter MadEvent shell: ./bin/generate_events
7. switch Pythia to ON
8. type 0, ENTER, and then again 0, ENTER
9. the events will be generated with Pythia switched ON
10. we can give a look at the text file called tag_1_pythia.log inside the directory ttbar/Events/run_01/: this log contains, among other info, the full record of the fort 10 generated events (you can see the initial protons, the inital partons, the intermediate tops and their decay products, as well as all the products of the subsequent parton shower and hadronisation - quite a long list...)

Final step, is to run detector simulation as well. Procees as before, but, before generating events:

1. switch/leave Pythia and Delphes ON

Read Delphes outputs

First exercise:

1. dowload the file MyReader.C from here: MyReader.C, and put it inside the directory Delphes/examples (note: it's inside MG5 directory)
2. open the file with a text editor
3. uncomment the proper lines in order to create, fill and show a histogram, then save the file
4. run the macro on the latest MG5 run output: from a terminal (again after the usual setup!), go to the directory Delphes and type root -l 'examples/MyReader.C("../ttbar/Events/run_XX/tag_1_delphes_events.root")'

Few more exercises:

1. plot the number of leptons (electron and muons) instead
2. plot the number of jets and the pT of the highest pT one in each event
3. print the fraction of events with at least two leptons (electrons or muons) and at least two jets
4. plot the number of b-tagged jets (hint: look at the Dephes list of branches, link on top of this page)
5. plot the number of jets with pT>40 GeV

Decay particles inside MadGraph

Now we want to also have tops (and Ws) decaying withing MG5 (so keeping the spin information, which is lost in Pythia!).

1. from the MG5 shell, generate the usual test process, but with some additional text: generate p p > t t~, (t > w+ b, w+ > l+ vl), (t~ > w- b~, w- > l- vl~)
2. create an output directory: output output ttbar_dilep
3. proceed as before
4. look inside the directory ttbar_dilep/SubPriocesses/P1_gg_ttx_t_wpb_wp_lvl_tx_wmbx_wm_lvl, at the ps files matrix1 and matrix2; compare with what you have generated in the previous example (directory ttbar)

Similarly, we can produce events where a Z boson is produced, together with two jets, and then decays to an electron-positron or muon-antimuon pair:

1. from the MG5 shell, first give the command define j = j b b~ (this will include b-jet production in association with Z)
2. generate the process generate p p > z j j, z > l+ l-
3. create an output directory: output output zjj_dilep
4. exit MG5 (CTRL+D)
5. enter the ttbar directory: cd zjj_dilep

Lecture 2 and 3

Useful Formulae

Cross section (x), integrated luminosity (Lumi) and number of events (N):

• 
• if xsec = 500pb, to simulate 100 of data, we expect events
• if we have N = 100 000 simulated events, for a process with , and we want to compare with 25 of data, we need to scale them by

Preliminary Exercise:

Goal:

• simulate Z->lep lep events
• analyse produced events:
  • select events
  • draw histogram for lep-lep invariant mass
  • fit it with a Gaussian to measure Z boson mass
• repeat the exercise for different mass values

Step by step:

• generate the events:
  • as usual, ./bin/mg5_aMC
  • in MG5 shell: generate p p > z, z > l+ l- (or generate p p > z > l+ l-, it's the same)
  • output zll
  • exit MG5, enter zll directory, enter MadEvent: ./bin/generate_events
  • set Pythia ON, Delphes ON, the rest OFF
  • check in param card the mass of the Z boson
  • keep 10'000 as number of events in run card
• Try the root macro MyReader.C on the ouput (see above)
• Try the root macro MyZmass.C
• Fit the output histogram to get the measured Z mass: see example macro MyFitZmassC
  • try both with a Gaussian and a BW
• Modify MyZmass.C in order to use also di-muon events
• Generate more samples with different values of Z mass, eg 85, 88, 90, 91, 92, 94, 97 GeV (you have to modify the param card in MG5)
• Fit each of them and build a table of generated mass vs fitted mass

Main Exercise:

Goal:

• select events coming from h->ZZ->4 lep and to produce a histogram with the reconstructed Higgs mass in these events
• compare "data" with MC simulation for Higgs signal and ZZ background
• everything should be done at CME = 8 TeV, considering the data integrated luminosity of 25

Where to get the data:

• find the Delphes output here: data_events.root

Generate the signal:

• as usual, ./bin/mg5_aMC
• IMPORTANT: before the generate command, type import model heft
• generate p p > h > l+ l- l+ l-
• output h4lep
• exit MG5, enter h4lep directory, enter MadEvent: ./bin/generate_events
• set Pythia ON, Delphes ON, the rest OFF
• when prompted, modify the run card:
  • set the CME to 8 TeV (4000 GeV each beam)
  • check that False  = cut_decays (should be by default in latest MG5)
• (eventually change the Higgs mass in the param card)
• IMPORTANT: multiply the signal x-sec given by MG5 by 10 (at LO both the x-sec and the BR are underestimated)

Generate the background (ZZ->4lep):

• as usual, ./bin/mg5_aMC
• generate p p > l+ l- l+ l- /h (the /h excludes the Higgs as intermediate state)
• output zz4lep
• exit MG5, enter h4lep directory, enter MadEvent: ./bin/generate_events
• set Pythia ON, Delphes ON, the rest OFF
• when prompted, modify the run card:
  • set the CME to 8 TeV (4000 GeV each beam)
  • check that False  = cut_decays (should be by default in latest MG5)

How to produce the final plot:

• look at the code MyHiggs4l.C: MyHiggs4l.C
  • use it as a template for your macro to create a histogram out of each delphes output files (signal, background, data)
  • the two inputs to the macro are the name of the root file with the delphes output and the output file where to store the histogram
  • example:
    • root -l -b -q 'examples/MyHiggs4l.C("../h4lep/Events/run_01/tag_1_delphes_events.root","signal.root")'
    • root -l -b -q 'examples/MyHiggs4l.C("../zz4lep/Events/run_01/tag_1_delphes_events.root","background.root")'
    • root -l -b -q 'examples/MyHiggs4l.C("data_events.root","data.root")'
• then look at the code MyHistMerger.C: MyHistMerger.C
  • use it as a template for your code to read the three inputs and merge them in a final plot
  • note the signal x-sec multiplied by 10

Additonal exercises:

• measure the signal x-section:
  • define an event selection (based on the Higgs mass?)
  • express the x-section as , where is the MG5 one x10 (see above)
• measure the Higgs mass:
  • try different mass values for the signal MC and compare them with data (chi2 test?)
  • try an anlystical fit with a gaussian after subtracting background from data
  • hint: you might consider to use a finer binning in the histogram

OLD STUFF - obsolete

Lecture 2

Useful Formulae

Cross section (x), integrated luminosity (Lumi) and number of events (N):

• 
• if xsec = 500pb, to simulate 100 of data, we expect events
• if we have N = 100 000 simulated events, for a process with , and we want to compare with 25 of data, we need to scale them by

Cross-section measurement:

• 
• N is observed in data after event selection
• B is obtained from background simulation scaled to Lumi and bkg x-sec (see above)
• eff is the signal selection efficiency: , from simulated data events

Ttbar cross section measurement

The problem:

• we have a "data" sample corresponding to 3 of data from the LHC at CME = 7 TeV
• we want to measure the cross-section for the total top-antitop production using this data
• we want to select events with 1 electron or muon, missing energy and jets (eventualy b-jets), the so called "single lepton channel"

What we will need:

• the "data": find them attached here: tag_1_delphes_events.root
• signal simulation
• background simulation (W+jets)
• the formula:

Event selection:

• we want to select electrons, muons and jets only if they have pT>25 GeV and |eta|<2.5
• we want to ask also for MET>25 GeV
• we can then play around with the request of how may jets, how many b-tagged jets, eventually higher pT threshold for jets...
• you can use / give a look at the macro here: MyEventCounter.C
• we want to minimize the relative statistical uncertainty:

How to generate Signal:

• generate p p > t t~
• IMPORTANT: we want all the decays of the tops, so set the decays of the W's from the tops to "all" in MadSpin card (or switch OFF MadSpin)
• don't forget to set the CME to 7 TeV
• number of events: you need to generate ~10 times more than what you expect in data
• to obtain eff in the formula, divide the number of events after and before the event selection (done automatically in MyEventCounter.C)

How to generate Background:

• define w = w+ w-
• define j = j b b~ (this includes the b's to our definition of j)
• generate p p > w j j
• to be more efficient, here you can set the W decays to e / mu / tau only (note: you need to include the tau!)
• don't forget to set the CME to 7 TeV
• number of events: you need to generate ~10 times more than what you expect in data
• to obtain B in the formula, you need to scale the number of selected events as explained above

How to generate pseudo-data:

• define w = w+ w-
• define j = j b b~ (this includes the b's to our definition of j)
• generate p p > t t~
• add process p p > w j j
• output data_ljets
• run a first test with Pythia ON and everything else OFF, 1000 events, to get the x-section
• get the number of events to generate for the real run as Lumi * xsec
• launch a second run with Delphes ON (MadSpin still OFF) with Nevents = Lumi * xsec

-- MichelePinamonti - 2014-12-15

-- MichelePinamonti - 2015-12-07