Stop4Body

Repository for the analysis code of the stop four body decay analysis

Before running Stop4Body

First step is to get the nanoAOD samples at T3-cms: https://github.com/diogodebastos/cmgtools-lite/tree/104X_dev_nano_lip When this step is finished it's time to process samples locally.

To do so, connect to the lxplus machines and start the environment:

$. initS4B-10_4_0.sh

Then, you want to look into run_susyStop4Body_nanoAOD_cfg.py for the configurations, modules and skims on the nanoAOD samples. When you've choose which samples you want to get, to a test to assure everything is ok and as expected by:

$nanopy.py test18_WjetsPT_NLO run_susyStop4Body_nanoAOD_cfg.py -N 5000 -o year=2018 -o runWJets

At this step, is also good to look into how long each event takes to run, you're aiming for the ms mark. Then, you can send jobs to condor to get your samples by:

$nanopy_batch.py -o 18_WJetsPT_NLO run_susyStop4Body_nanoAOD_cfg.py --option year=2018 --option runWJets=True -b 'run_condor_simple.sh -t 480 ./batchScript.sh'

Now, that jobs are running, you can use submitFailedJobs.py to see jobs that have failed or have other errors by:

$python submitFailedJobs.py -i 18_WJetsPT_NLO/ -t 960

In case a large amount of jobs have failed, use submitList.py to submit failed jobs. At the same time, you want to open another lxplus session that checks every interval for jobs that have finished and their respective output root file and moves them to EOS:

$while true; do date; echo ""; source cpSamplesFromAFStoEOS.sh 16 16_WJetsHT; sleep 1800; done

TODO: save directly to EOS.

The final step is to move the samples to t3cms by:

$gfal-copy -rv nanoAOD/Stop4Body18/Autumn18_Prod2020Sep29/ srm://srm01.ncg.ingrid.pt:8444/srm/managerv2?SFN=/cmst3/store/user/dchasque/nanoAOD/Stop4Body18/Autumn18_Prod2020Sep29

Setup analysis and compiling

In this repo you find 2 setup scripts initNanoS4B.sh and initNanoS4B-forScram.sh. The first one to setup and run, the latter for compiling the framework. You should copy these 2 files to your home. This framework runs on SCRAM_ARCH=slc7_amd64_gcc820 so, if using ncg machines, you should connect to pauli. I suggest having 2 shell sessions in one you do $. initNanoS4B.sh and run your analysis and in the other one do $. initNanoS4B-forScram.sh and just use it to compile like $scram b.

Running the analysis

Now we are ready to start searching for stop 4 body decay!

Generating json files for signal, getting golden JSON, Pileup, (EWK)ISR and LHE weights

To generate signal json files run

$python generateJSONfiles.py

Get Data Pileup distributions

$. getPUJSONs.sh

Check setupJSONs.sh and setupPaths.sh to define the samples, the directories and the $YEAR you want to use. Then, for Pileup weights and (EWK)ISR weights run:

$./runStep0-preProcessSamples.sh

When jobs are done:

$./runStep0-preProcessSamples_Part2.sh

You should have preProcessSamples.root in $PREPROCESSSAMPLES_DIR.

For LHE weights do:

$./runStep0-LHEWeight.sh.

When jobs are done:

$./runStep0-LHEWeight_Part2.sh

You should have lheWeights.root in $LHESCALE_DIR.

Processing Samples

Now we are ready to process our samples. The script that does so is processSamples.cc. You should check this when updating weights, cleaning jets, create analysis object collections and add or remove HLTs. On buildAndSubmitJobs.py you choose the json files of the samples you want to process. Check the Templates folder for jsonTemplate.json and Step1_JobTemplate.sh and in the binaries buildJobs.cc for extra details in this step. To process your samples, run:

$./runStep1.sh

Data/MC agreement

This is an intermediate step you can run anytime during your analysis. But it's important to do it right after you processed your data and simulation samples. Which is, to plot your simulation distributions of multiple variables and be sure they are in accordance with data. This script is based on makePlots.cc. To get your Data/MC (and other plots) run:

$./StepN_makeDataMC.sh

The MVA method - Boosted Decision Trees

After having a good Data/MC agreement it's time to move to the next step of the analysis, the MVA method. First we have to split the samples in test and train:

$./runStep1_part1.5.sh

And then, add the signal samples and the background samples to be used to train the BDT:

$./runStep1_part2.sh

To train the BDTs use Signal full DM to use the kinematic similarity between all the signal points with the same DM and for background use TTbar and WJets only. From here, you can build your BDT (or other classification methods) on trainiMVA.cc. To start the training do:

$./runStep3_TrainMVA.sh

and apply it to your samples by:

$./runStep4.sh

After this step, if you want, you can compare the BDT (and other variables) shape distribution between different years. The idea is to compare between the samples of 2016 and if you want, compare with miniAOD 2017 too. To do so, you have to go to the pMacros directory and run root interactively from there:

$cd pMacros

$root -l

$.x ShapeProcVarYears.C(30,false,false,true,"BDT")

After training and applying the BDTs it's time to determinate the cut to apply on the BDT output in order so maximize the signal background separation. The first thing to be done, is to get the processes yields as a function ƒof the BDT score. To do so, run:

$./runStep5_Part1.sh

Then, get the upper limit cross-section as a function of the BDT cut:

$./runStep5_Part2.sh

And finally plot r from the ULXS, efficiencies and FOM from the datacards:

$./runStep5_Part3.sh

With step5, we have determined the BDT cut and therefore, the SR of our analysis. Now, we can get run our DD prediction methods of the background, namely, WJets, TTbar and Fakes. With this next step, we'll also look into the Validation Regions of our MVA, VR2 and VR3, more details in getDDEstimate.cc. To run this step do:

$./runStep6-GetDD.sh

$./runStep7-GetSys.sh

Datacards and preparation to get the Limits

$./runStep8-getFinalDataCards.sh

finalYieldsUncTable.py

Limits

runFinalStep_Part1.sh

runFinalStep_Part2.sh