Login

SCIENCE BRINGS NATIONS TOGETHER
Discussion on MPD MC productions: issues, QA, knowledge sharing

→ Europe/Moscow
Description

THIS IS A LOCAL MEETING (in Russian). Results of discussions may be presented later to a broader audience. 

Join vidyo

Hide

Participants 

SPbSU
Andrey Seryakov, Daria Prokhorova, Evgeny Andronov, Igor Altsybeev, Vladimir Kovalenko, Vladislav Sandul

JINR
Alexander Mudrokh, Alexey Aparin, Pavel Batyuk 

MePhI
Peter Parfenov 

INR
Marina Golubeva, Alexander Ivashkin

+ unknown: Akerke, Elena

 

In blue: important issues.

In red: important issues, we need to write to responsible people / to collaboration management.

 

slide 2

  • Do we have more MC datasets than shown here? 

-  Almost all MC datasets are in the list.

slide 4

  • Events in UrQMD are written even if there were no interaction. As a consequence, if no interaction, minimum number of particles in the output is nNucleons*2.

slide 5

  • By default, reaction plane angle (Psi_RP) in UrQMD is random in the range 0-30 deg. --> should be 0-360 deg., it’s important to monitor this as a part of QA!

(this can be changed in runMC.C: https://git.jinr.ru/nica/mpdroot/-/blob/dev/macro/mpd/runMC.C#L96)

  • On reconstruction level, track->GetPhi() returns values in interval (-2pi, 2pi), while in principle it should be converted in the (0, 2pi) range automatically. Who to contact?    - A.Zinchenko, O. Rogachevskiy.

slide 6

  • Proper cuts on DCA? - it makes sense to fit DCA distributions in eta and pT bins by Gauss, and establish some cut on nSigma in each bin, say, 2 or 3.

  • motherId == -1 is assigned for all particles that come from event generator (before GEANT), so it’s difficult (impossible?) to figure out if a particle is a decay product of some resonance.

slide 8 

  • Efficiencies seem to be not the same in different MC datasets. One way to investigate this is to compare the number of global tracks (MpdTrack) with the number of MpdTpcKalmanTrack tracks. To be checked in detail.

slide 9 

  • Mis-identification of charge? This issue is observed by many analyzers. Can be due to TPC-TOF mis-matching for global tracks? Track duplication (due to multi-matching with TOF)? One way to check is to compare the charge sign, pt, eta, etc. between sim, global tracks and corresponding MpdTpcKalmanTrack tracks. 

slide 10

  • No decay channels hadron->e+e- in UrQMD. Can be an issue for PID analysis (with nSigma cuts method, Identity method...). At some stage we may contact the UrQMD developers.

slide 11

  • FHCal: 

    • Info from M. Golubeva: correct geometry of FHCal is committed to /dev, but there is no digitization. One should be careful with a step parameter in Geant. P.Parfenov’s UrQMD productions have already been done with the correct geometry.

    • No fragments in UrQMD, therefore the LAQGSM generator is used for FHCal performance studies.

    • It would be good to test and use other generators, in particular, DCM-QGSM-SMM that is fine-tuned and used in CBM. 

    • In general, need better coordination with other people and groups in Dubna.

slide 12

  • Three different farms @hydra, @nc, @lxpub => three different administrators, weak contacts between them.

  • HydraLIT is more stable, while NICA cluster in some cases is not: in order to work with /eos directories at @nc, one should use commands like ‘eos ls’, ‘eos mkdir’ etc. to avoid problems with synchronization. 

  • Resources for MPD (currently): 1000 cores at @hydra; up to 400 parallel jobs at @nc, currently the cluster is almost empty.

  • If a problem with wrong OS version appears (when one runs MpdRoot jobs on a cluster), it could be solved by commenting several lines where the check is done in the config.h file in MpdRoot.

slide 13

  • For job submission, the mpd-scheduler is available at both @hydra and @nc. But standard batch tools like slurm at @hydra and SGE at @nc are more flexible and preferred by experienced users.

slide 14

  • Wrappers for many event generators exist in MpdRoot, all should work.

  • It would be nice to add a wrapper for SMASH since this generator seems to be of high interest for many groups (it is a “successor” of UrQMD).

    • should not be a big problem, one can take code of Mpd3fdGenerator as an example. Will be tried (SPbSU?).

    • other generators? AMPT? iEBE-VISHNU? MUSIC? etc

  • Geant3 was a default in MpdRoot for many years, several productions with Geant4 appeared only recently. Much more config options in G4 than in G3.

  • Caution! PID framework in MpdRoot is “tuned” to Geant3, could give incorrect results in G4-based simulations.

    • How is it possible? With real data, we won’t know which Geant version Nature is using. --> The PID framework will be tuned using some ‘marker’ info, i.e. when m of a particle is known (weak decays?..).

slide 15

  • Some event generators are build-in in MpdRoot: ION BOX HADGEN (also SHIELD), for others - only wrappers exist.

  • It would be good if all generators give their output in the same (preferably .root) format.

 

slide 16

  • There is no database for MPD MC productions yet.

    • PB: the same was in BM@N until the first real data came, and suddenly things (large-scale productions, databases, web-interfaces) started to emerge very quickly. see http://bmn-web.jinr.ru/#/home 

  • It is essential to know (= to keep in a database) which MpdRoot version was used for a particular MC production.

    • There are (still) no releases of MpdRoot, so every produced output (sim, reco) could be ‘marked’ only with a commit number (tag).

 

Summary:

  • Before making any MC production request, we need to clearly determine which physics we want to study. This motivation is required when we make request to PWG conveners, which should transfer our requests further.

  • Configurations of some generators (for instance, vHLLE+UrQMD) are needed to be done by experts/developers (a trivial blind change of configs leads to a garbage-like outcome).

  • Usually, it’s a good practice to make QA of a dataset using an established list of checks. The rule of thumb is to do it on 10% of statistics. Could be automated in future (already done in BM@N), now could be done “by hand”.

    • Petr Parfenov can provide a list of QA checks that are used for flow studies in his group, to be used as a starting point.

 

Next steps:

  • An e-mail to A.Zinchenko and O.Rogachevski will be sent: 

               (1) about an issue with rec. phi in (-2pi, 2pi), 

               (2) about mis-identification of charge.

 

Potential questions to be raised at MPD collaboration meeting in April:

  • About the three different farms @hydra, @nc, @lxpub => three different administrators, weak contacts between them.

  • FHCal coordination?..

There are minutes attached to this event. Show them.
Powered by Indico v3.3.5