Generator fragments for MC production
The package includes the datacards used for various generators inclusing POWHEG, MG5_aMC@NLO, Sherpa, Phantom, Pythia...
Further details are reported in the twiki: https://twiki.cern.ch/twiki/bin/view/CMS/GeneratorMain#How_to_produce_gridpacks
Instructions on how to use the fragments are here https://twiki.cern.ch/twiki/bin/view/CMS/GitRepositoryForGenProduction
This repo contains centrall tools to produce gridpacks (genprocutions) along with custom EFT tools to produce EFT gridpacks with SMEFTsim.
A MG folder which has been optimised for batch-mode computation. The MadGraph "single-diagrams-enhanched multichannel" integration technique makes it possible to split the phase space integration into small bits that can be evaluated independently. The expolitation of a fully parallel compututational workflow allows to reduce the time required to obtain a MG folder by order of magnitudes. MG gridpack computation is split in two(three) steps: CODEGEN, INTEGRATE (MADSPIN), where madspin is not mandatory.
- CODEGEN: code generation step, finds processes / subprocesses and diagrams
- INTEGRATE: phase space integration
- MADSPIN: decay of particles accounting for spin, offshell eccets, ...
EFT contributions are simulated at LO via SMEFTsim (https://smeftsim.github.io/). A ready-to-go version of SMEFTsim UFO model with U35 flavour scheme and mW,mZ,Gf input scheme is publicly accessible at http://gboldrin.web.cern.ch/gboldrin/generators/SMEFTsim_U35_MwScheme_UFO.tar.gz This version is up to date as of 8th November 2021 and also contains restriction cards for 15 operators (cW,cHW,cHbox,cHDD,cHWB,cHl1,cHl3,cHq1,cHq3,cll,cll1,cqq1,cqq3,cqq11,cqq31) both for single insertion or for all the posssible pairs. The directory also contain custom restriction cards to activate all operators at the same time or a subset of them e.g. http://gboldrin.web.cern.ch/gboldrin/generators/restrict_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_massless.dat
If you need a more up to date version of SMEFTsim or different restriction cards, it is highly suggested to clone a copy of SMEFTsim, build the restrictions you need, move them under the UFO model, pack everything again and copy the compressed folder on your web area (e.g. /eos/user/y/yourusername/www on lxplus).
- Create a tar version of SMEFTsim with restriction cards
git clone [email protected]:SMEFTsim/SMEFTsim.git
mv SMEFTsim/UFO_models/SMEFTsim_MFV_alphaScheme_UFO . # choose the UFO model you need
cp SMEFTsim_MFV_alphaScheme_UFO/restrict_SMlimit_massless.dat restrict_your_set_of_op_massless.dat #copy SM restriction
#Modify restrict_your_set_of_op_massless.dat turning on and off operators under SMEFT or SMEFTFV blocks (on = 9.999999e-01, off = 0)
tar -zcvf SMEFTsim_MFV_alphaScheme_UFO.tar.gz SMEFTsim_MFV_alphaScheme_UFO
xrdcp SMEFTsim_MFV_alphaScheme_UFO.tar.gz /eos/user/y/yourusername/www
Examples of restriction cards (and script to automate the cards generation) can be found at https://github.com/UniMiBAnalyses/D6EFTStudies/tree/master/madgraph_model/v3_0 (https://github.com/UniMiBAnalyses/D6EFTStudies/blob/master/madgraph_model/buildRestrict_v3_0.py)
Gridpack generation can be done both on lxplus or cmsconnect as both have HTCondor to submit batch jobs. However, part of the gridpack will still run in local. A lxplus worker node has 10 cores while cmsconnect has 48. You will be 5 times quicker while working on cmsconnect than lxplus. Some comments:
- Never work fully local on cmsconnect. If you overload the system your jobs will be killed (e.g. if you submit multiple gridpacks creation in parallel).
- You can work fully local on lxplus, at least, you will hardly find resource issues.
Follow this tutorial on how to subscribe to cmsconnect: https://indico.cern.ch/event/615524/contributions/2520456/attachments/1430441/2197104/March20_2017_gen_meeting.pdf
Once you decided if you'll generate your MG gridpack on lxplus or cmsconnect, simply login, go to the desired working folder and clone this repo (check if this repo is up to date). If needed, change branch but scripts only work for master as of 8/11/2021:
git clone [email protected]:UniMiBAnalyses/genproductions.git && cd genproductions/bin/MadGraph5_aMCatNLO/
The main script is: gridpack_generation_EFT.sh which is a copy of gridpack_generation.sh with some lines added to it:
245 wget --no-check-certificate http://gboldrin.web.cern.ch/gboldrin/generators/SMEFTsim_U35_MwScheme_UFO.tar.gz
246 cd models
247 tar xavf ../SMEFTsim_U35_MwScheme_UFO.tar.gz
248 cd SMEFTsim_U35_MwScheme_UFO
249 # wget all restrictions
250 wget --no-check-certificate http://gboldrin.web.cern.ch/gboldrin/generators/restrict_cHWB_cHDD_cHl1_cHl3_cHq1_cHq3_cll_cll1_massless.dat
251 wget --no-check-certificate http://gboldrin.web.cern.ch/gboldrin/generators/restrict_cW_cHWB_cHDD_cHl1_cHl3_cHq1_cHq3_cll_cll1_massless.dat
252 wget --no-check-certificate http://gboldrin.web.cern.ch/gboldrin/generators/restrict_cW_cHWB_cHDD_cll1_cHl1_cHl3_cHq1_cHq3_massless.dat
253 wget --no-check-certificate http://gboldrin.web.cern.ch/gboldrin/generators/restrict_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_massless.dat
254 cd ../..
If you need to generate from a different branch than master as of 8/11/2021, simply copy gridpack_generation.sh and add those lines. If you genrated your version of compressed SMEFTsim UFO model, simply replace this paths with yours.
Accessory scripts needs to be modified in order to run gridpack_generation_EFT.sh on condor: submit_cmsconnect_gridpack_generation_EFT.sh, submit_condor_gridpack_generation_EFT.sh.
These are just copies of the same ones without _EFTsuffix and few lines modified. If you need to work on another branch and not master as of 8/11/2021, simply copy submit_cmsconnect_gridpack_generation.sh, submit_condor_gridpack_generation.sh and replace gridpack_generation.sh with gridpack_generation_EFT.sh everywhere.
Once you have all necessary scripts you are ready to issue the generation:
./gridpack_generation_EFT.sh card_name path/to/cards # this runs locally
Where path/to/cards contains the MG5 cards. Some are mandatory and must follow a naming convention: card_name_proc_card.dat, card_name_run_card.dat.
An accessory card will be needed for SMEFTsim: card_name_customizecards.dat, where you set the initial values for EFT coupling.
An additional card is mandatory if one want to use the reweight module: card_name_reweight_card.dat.
IMPORTANT: You do not need an extramodel card, the script will always download the SMEFTsim tarball even if you won't need it.
Example of cards can be found under:
cards/ZZ2e2mu/ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN
Where we generate a VBS ZZ process into 4 leptons of different flavour (2e, 2mu) with EFT contributions from 15 operators. We also use the reweight module to change hypothesis and turn off-on operators.
Following the ZZ example:
./gridpack_generation_EFT.sh ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN cards/ZZ2e2mu/ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN # this runs locally, should be avoided
./submit_cmsconnect_gridpack_generation.sh ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN cards/ZZ2e2mu/ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN (n_cores) ("memory")
./submit_condor_gridpack_generation.sh ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN cards/ZZ2e2mu/ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN (n_cores) ("memory")
Why is it highly suggested to generate through condor? You will be able to generate a gridpack ~ 10 times faster as the INTEGRATE step will be submitted to various worker node via condor.
As an example, a gridpack on cmsconnect running locally will be ablle to execute 48 jobs simultaneously (Idle means job waiting to be run)
INFO: Idle: 613, Running: 48, Completed: 535 [ 73h 49m ] # the INTEGRATE steps has been running for ~ 3 days, and 3 days approx. left
Different gridpack but with batch submission:
INFO: Idle: 0, Running: 423, Completed: 179 [ 3m 19s ]
You can see that condor submission can handle >10 times simultaneous jobs that local generation (and your jobs won't be centrally killed by cmsconnect mainteiners as you won't overload the machines).
Gridpacks usually take from O(h) to O(days) to be completed. It is unfeasible to stay connected for such a long time length. Screen (or tmux) are essential tools in order to issue commands and leave it running also after disconnecting from the ssh tunnel. Both lxplus and cmsconnect provide out of the box screen support:
screen # open a new scren session
screen -ls # list all open screen sessions on the worker node
screen -rD _ID_ # connect to a detached screen session with id = _ID_
After issuing the screen command a new shell will be opened. You can use it the same way as a normal terminal session. In order to detach from the screen session type Ctrl + A + D. You will be redirected to the shell in which the screencommand wa issued and you can inspect the opened screen session:
[gboldrin@lxplus731 MadGraph5_aMCatNLO]$ screen -ls
There is a screen on:
9358.pts-12.lxplus731 (Detached)
1 Socket in /var/run/screen/S-gboldrin.
9358 is the session ID. To reconnect:
screen -rD 9358
screen -rD 9358.pts-12.lxplus731
WARNING:
There some differences between lxplus and cmsconnect regarding screen. On lxplus you usually log into a worker node (from the example above 731 machine) while on cmsconnect there is only one machine. If you open a screen session on lxplus and then detach from it and log out from the ssh, you'll need to store the machine number and reconnect to the same machine in order to retrive the screen session. From the example above:
ssh [email protected]
screen -rD 9358
Furthermore, the screen session on lxplus works with kerberos tokens which allow the session to have permission to write/read from your afs area. These tickets are created with the command kinit and authorised with your lxplus login password. The tickets last for 25 hours after ssh disconnection and there is no way to change that. After 25h the commands won't be able to, for example, read or create files in your user area.
One way to create a persistent screen session on lxplus is to create a keytab which automatically refresh your expired tickets.
From a private folder on your afs (e.g. /afs/cern.ch/user/u/username/private) type:
mkdir ktokens && ktutil
A new prompt will appear. Type in the prompt the following lines:
add_entry -password -p [email protected] -k 1 -e arcfour-hmac-md5
add_entry -password -p [email protected] -k 1 -e aes256-cts
wkt USERNAME.keytab
Ctrl+D #close ktutil session
k5reauth can now be used, it will issue user defined commands in a child process and will renew your kerberos tokes after N seconds also for the child itself.
Using the ZZ VBS example, this command will create a permanent screen session with renewable tokens once every 3600 seconds:
cd /path/to/genproductions/bin/MadGraph5_aMCatNLO/
k5reauth -f -i 3600 -p USERNAME -k /afs/cern.ch/user/u/username/private/ktokens/USERNAME.keytab -- screen ./submit_cmsconnect_gridpack_generation.sh ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN cards/ZZ2e2mu/ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN
You can also create an alias for this command in your .bash_profile or .bashrc
kscreen(){
if [[ -z "$1" ]]; then #if no argument passed
echo "First argument must be the executable name"
else #pass the argument to screen
k5reauth -f -i 3600 -p USERNAME -k /afs/cern.ch/user/u/username/private/ktokens/USERNAME.keytab -- screen $1
fi
}
and use it as a normal screen session:
kscreen ./submit_cmsconnect_gridpack_generation.sh ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN cards/ZZ2e2mu/ZZ2e2mu_cW_cHWB_cHDD_cHbox_cHW_cHl1_cHl3_cHq1_cHq3_cqq1_cqq11_cqq31_cqq3_cll_cll1_SM_LI_QU_IN
Suppose your codegen sstep run fine but something broke at Integration. You do not need to run everything again. You'll have a folder named as your <card_name>
with two subfolders: <card_name>/<card_name>_gridpackand <card_name>/<card_name>_output.tar.xz . To resubmit some step both on local or condor:
source Utilities/cmsconnect_utils.sh
source Utilities/source_condor.sh
cd <card_name>
rm -rf <card_name>_gridpack
cmssw_setup <card_name>_output.tar.xz; cd ..
iscmsconnect=1 bash -x gridpack_generation_EFT.sh ${card_name} ${card_dir} ${workqueue} INTEGRATE ${scram_arch} ${cmssw_version}
For example the last line can be issued on condor as:
iscmsconnect=1 bash -x gridpack_generation_EFT.sh <card_name> <card_dir> condor INTEGRATE
It may happen that after the codegen step your EFT gridpack breaks at the pilot run with some cryptic compilation error. This may be due to an error problem as MGv2_6_5 is not design to exploit the full RAM of the node but it is limited. To unlock this limitation and resubmit successfully follow the steps of the chapter "Resubmitting Gridpacks" up to the line cmssw_setup <card_name>_output.tar.xz. This command will untar the codegen output and create a <card_name>_gridpack folder. Now issue
cd <card_name>_gridpack/work/<card_name>
you will see a MG folder e.g.
ls WmTo2JZto2L_dim6_ewk/WmTo2JZto2L_dim6_ewk_gridpack/work/WmTo2JZto2L_dim6_ewk
Cards Events HTML MGMEVersion.txt README README.systematics Source SubProcesses TemplateVersion.txt bin index.html lib madevent.tar.gz
Issue the command
sed -i.bak '/FFLAGS= -O -w -fbounds-check -fPIC/s/$/ -mcmodel=medium/' Source/make_opts
and resubmit the INTEGRATE step
cd -
iscmsconnect=1 bash -x gridpack_generation_all.sh <card_name> <card_dir> condor INTEGRATE