CORSIKA Cosmic Ray Simulation Workshop [virtual]

Jun 22, 2020, 2:00 PM → Jun 25, 2020, 5:55 PM Europe/Berlin

virtual

General and open workshop to discuss technology and science of air shower simulations, and meeting of the CORSIKA 8 Project.

ZOOM link: https://zoom.us/j/786000579

WorkshopPhoto2020.jpg

Monday, June 22

1 Welcome

Speaker: Ralf Ulrich (KIT)

Welcome.pdf

CORSIKA Workshop: Project overview

Convener: Ralf Ulrich (KIT)

Minutes of questions and discussions

Session on Monday 14:10

• To Maximilian Reininghaus: why are e.g. on slide 9 the differences for muons so small and for other particles much larger?
  
  One aspect is that e.g. kaons are in general much less frequent and even larger discrepancies will not generate large effects.
   
• To Augusto Alves: does the current refactoring is in conflict with the mantra of small individual, incremental changes?
  
  Yes, it is. It is like a "prototype version2 framework" which corrects and changes some of the original choices and concepts. It is necessary to allow further development to become more clear, and more easy.
   
• To Ralf Ulrich: Isn't one of the important next work packages "e.m." cascades?
  Absolutely yes. But this is basically a physics contribution and not part of "work on the framework". We want to include e.m. showers asap. 

2 CORSIKA8 project status

Speaker: Maximilian Reininghaus (KIT / IKP)

corsika8-project.pdf

Q: Why are e.g. on slide 9 the differences for muons so small and for other particles much larger?

A: Two aspects: 1) Keep in mind that some particles are rarer than others, e.g. Kaons. So a relativly large difference for them is rather small in absolute numbers. 2) The energy spectra a steep. What counts most for the longitudinal profiles in the end are just a few low-energy bins.

Q: What are the differences for UrQMD in CORSIKA 7 vs 8?

A: 1) Impact parameter range: hardcoded values in CORSIKA 7, default given by UrQMD itself in CORSIKA 8. 2) Intertial system of the event: lab frame in CORSIKA 8, sth. else in CORSIKA 7

3 Status of CORSIKA7

Speaker: Tanguy Pierog (IKP, KIT)

corsika7.pdf

4 CORSIKA8 framework structure and design

Speaker: Dr Antonio Augusto Alves Junior (KIT)

COSIKA_Workshop_2020.pdf

5 Combined discussion, Q&A

Speaker: Ralf Ulrich (KIT)

Discussion.pdf

3:20 PM Coffee

CORSIKA Workshop: Air shower physics

Convener: Tanguy Pierog (IKP, KIT)

Minutes of questions and discussions

Session "Air shower physics" on Monday 15:35

• To Ken Ohashi: how was diffraction defined in the study?
  The event generator definition was used.
   
• It may be a great extension of the study to include RMS(Xmax) and/or fluctuations of Nmu into the observables. This would complement the study in another important way.
   
• To Oliver Fischer: are full shower simulations needed?
  Sometimes it may be needed. There may be also dynamic involved in BSM, this may be visible in an unknown and unexpected way in showers.
   
• Why not use dedicated cascade equations as input and the just "fold" with BSM?
  This is possible, but requires to go back to square one for every new approach, idea, correction. It would be great to have a standard interface with documentation to insert BSM into shower observables and products. Both, for CE and full MC. There may be many theorists interested to work with such tools, if they are available.
   
• To Álvaro Pastor Gutiérrez: was available accelerator data used to compare some of the results to? There is CRMC that can be used to generate such data sets.
  Not yet, but sounds like a good idea.
   
• To Hans Dembinski: Why was oxygen in LHC shifted back in time? Because of corona virus?
  No, there were other reasons.
   
• Is remaining SMOG data going to be analyses and published?
  Unknown right now.
   
• To Jan Ebr: what is event statistics on slide 6?
  Probably 500 events, but statistical errors needs to be checked. They may be small?
   
• Also on slide 6: going to f19=5 is probably too extreme. One should keep in mind that the entire approach is useful only for ~small changes.

Session "Theory and models" on Tuesday 14:00

• To Klaus Werner: What is the plan on EPOS4, and the difference to EPOS3?
  Main feature of EPOS3 is the micro-canonnical hadronization, which will allow to efficiently be applied to small and large systems. There maybe needs to be a dedicated "cr" version of EPOS3 for highest performance. In EPOS4, the overall treatment of multiple scattering and hydrodynamic is going to be unified in a fundamental way. This should result in the first fully self-consistent model to treat small and large systems at small or high multiplicities.
   
• To Marcus Bleicher: Which version of UrQMD was used? 
  Different versions,  3.4 is the latest and should resemble what was shown.
  In CORSIKA we still use 1.6, mainly because of xs tables: not difficult, but to update it needs to be done carefully ~1 month.
   
• How much difference is between 1.6 and 3.4?
  Pions, Kaons, Protons don't differ a lot, but forward physics not so clear (important for CR, but not important for HI). What changed more than a few percent is Xi, Omega, etc. but can be important.
   
• Switching energy between low and high E model?
  One should use the best model to describe the physics as long as possible. The range for UrQMD is actually very broad. It goes at least to 100GeV, even to 1TeV (but not perfect any more). Current default in Corsika is 80GeV. EPOS may go down to very low energies. SIBYLL is limited.
   
• How does UrQMD include elastic scattering which apparently becomes relevant at higher energies (see e.g. slide 5)? Is there a theory behind it (optical theorem)? Is it a fit? Or is it ignored?
   
• Is there anything resembling diffraction. If not: is this a relevant limitation of the model or does this limit the potential phase-space of [best] applicability?
   
• In the comparisons to ALICE data on slide 17: is the hybrid actually absolutely needed, or does the "pure" UrQMD not also provide a good description? It seems the figures suggest this.
   
• To Lev Kheyn: Protons have high elasticity, they are not much modified?
  yes.
   
• MC is not ok at large delta-eta in central detector (slide 10) seems ok.  but it becomes bad after all corrections: The key difference is the inclusion of HF and extra cuts on ND events.
   
• How were gamma-hadron collision simulated?
  Were not simulated and are not included in study. (Audio connection not perfect to understand everything)
   
• Very nice propaganda for proton-oxygen beams! None linear effects that cannot be extrapolated. LHCb can also measure this.

 

• General question to everybody: why does EPOS and Sibyll have so similar CR predictions while being so different internally?
  
  Some parameters play an important role. Many details are in fact not so critical. There is no hydro in SIBYLL. The latest change in Sibyll to 2.3d was "undoing" a bug in sibyll that was imitating the effects of hydro to some extend. Thus, if there was data: the models can be fit also to such data. The question is: what data is available, what data is relevant, how well is it described.

 

• How important are the lower energy interactions in showers relevant to high energy models.
  
  Important is also the number of "child generations". Thus, high energy collisions are much more important because the effects accumulate. At lower energies it is just a single "effect", however, the number of particles is extremely high at lower energies. Systematic studies are needed to learn more.

6 A simulation study of the effects of diffractive collisions on observables of air showers experiments

A hadronic interaction plays an important role in air shower development, and the detailed understanding is a key to understand the mass composition.
The diffractive collision is one of the proposed sources of the uncertainties of air shower predictions. In this work, we investigate how the difference in the diffractive collision treatment between models affects the depth of the maximum of the air shower developments $X_{max}$, the depth of the maximum of muon productions $X^{\mu}_{max}$ and the number of muons on the ground $N_{\mu}$ with focusing on detail characteristics of diffractive collisions; the fraction of diffractive collisions and the modeling of the diffractive collisions. We demonstrate the fraction of diffractive collisions to inelastic collisions affects 8.9 $g/cm^2$ and 9.5 $g/cm^2$ on $\langle X_{max} \rangle$ and $\langle X^{\mu}_{max} \rangle$, respectively. On the other hand, the effects of and the modeling of diffractive collisions are relatively small.

Speaker: Ken Ohashi (Nagoya University)

corsika2020_ohashi_diffraction.pdf

7 Opportunities for BSM physics in Cosmic Rays

Speaker: Oliver Fischer (Institut für Kernphysik, Karlsruher Institut für Technologie)

BSM_CR_Fischer.pdf

8 Sub-TeV hadronic interaction model differences and their impact on air-shower development

In the sub-TeV regime, popular hadronic interaction models disagree in their predictions for post-first interaction and ground-level particle spectra. These model differences generate a significant source of inherent uncertainty in their experimental utilization.
We investigate the nature and impact of such differences through a simultaneous analysis of ground level particles and first interaction scenarios. We focus on initialized events at energies close to the transition between high and low energy hadronic interaction models, where the discrepancies have been shown to be maximal. We find the models to diverge as more concrete shower scenarios are compared, pointing to characteristic differences in the models phenomenology. Finally, we discuss an argumentation for the scaling of such differences and their decrement at higher initial energies.

Speaker: Álvaro Pastor Gutiérrez (MPIK)

HadIntDiff_AlvaroPastor.pdf

9 Input from LHCb towards understanding air showers

Speaker: Hans Dembinski (Max Planck Institute for Nuclear Physics, Heidelberg)

dembinski_lhcb.pdf

10 Modifying interactions in CORSIKA with MOCHI

Speaker: Jan Ebr (Institute of Physics, Prague)

mochi-corsika-workshop.pdf

11 Combined discussion, Q&A

Speaker: Tanguy Pierog (IKP, KIT)

Tuesday, June 23

12 Note

Speaker: Ralf Ulrich (KIT)

Times.pdf

CORSIKA Workshop: Theory and models

Convener: Hans Dembinski (Max Planck Institute for Nuclear Physics, Heidelberg)

Minutes of questions and discussions

Session "Air shower physics" on Monday 15:35

• To Ken Ohashi: how was diffraction defined in the study?
  The event generator definition was used.
   
• It may be a great extension of the study to include RMS(Xmax) and/or fluctuations of Nmu into the observables. This would complement the study in another important way.
   
• To Oliver Fischer: are full shower simulations needed?
  Sometimes it may be needed. There may be also dynamic involved in BSM, this may be visible in an unknown and unexpected way in showers.
   
• Why not use dedicated cascade equations as input and the just "fold" with BSM?
  This is possible, but requires to go back to square one for every new approach, idea, correction. It would be great to have a standard interface with documentation to insert BSM into shower observables and products. Both, for CE and full MC. There may be many theorists interested to work with such tools, if they are available.
   
• To Álvaro Pastor Gutiérrez: was available accelerator data used to compare some of the results to? There is CRMC that can be used to generate such data sets.
  Not yet, but sounds like a good idea.
   
• To Hans Dembinski: Why was oxygen in LHC shifted back in time? Because of corona virus?
  No, there were other reasons.
   
• Is remaining SMOG data going to be analyses and published?
  Unknown right now.
   
• To Jan Ebr: what is event statistics on slide 6?
  Probably 500 events, but statistical errors needs to be checked. They may be small?
   
• Also on slide 6: going to f19=5 is probably too extreme. One should keep in mind that the entire approach is useful only for ~small changes.

Session "Theory and models" on Tuesday 14:00

• To Klaus Werner: What is the plan on EPOS4, and the difference to EPOS3?
  Main feature of EPOS3 is the micro-canonnical hadronization, which will allow to efficiently be applied to small and large systems. There maybe needs to be a dedicated "cr" version of EPOS3 for highest performance. In EPOS4, the overall treatment of multiple scattering and hydrodynamic is going to be unified in a fundamental way. This should result in the first fully self-consistent model to treat small and large systems at small or high multiplicities.
   
• To Marcus Bleicher: Which version of UrQMD was used? 
  Different versions,  3.4 is the latest and should resemble what was shown.
  In CORSIKA we still use 1.6, mainly because of xs tables: not difficult, but to update it needs to be done carefully ~1 month.
   
• How much difference is between 1.6 and 3.4?
  Pions, Kaons, Protons don't differ a lot, but forward physics not so clear (important for CR, but not important for HI). What changed more than a few percent is Xi, Omega, etc. but can be important.
   
• Switching energy between low and high E model?
  One should use the best model to describe the physics as long as possible. The range for UrQMD is actually very broad. It goes at least to 100GeV, even to 1TeV (but not perfect any more). Current default in Corsika is 80GeV. EPOS may go down to very low energies. SIBYLL is limited.
   
• How does UrQMD include elastic scattering which apparently becomes relevant at higher energies (see e.g. slide 5)? Is there a theory behind it (optical theorem)? Is it a fit? Or is it ignored?
   
• Is there anything resembling diffraction. If not: is this a relevant limitation of the model or does this limit the potential phase-space of [best] applicability?
   
• In the comparisons to ALICE data on slide 17: is the hybrid actually absolutely needed, or does the "pure" UrQMD not also provide a good description? It seems the figures suggest this.
   
• To Lev Kheyn: Protons have high elasticity, they are not much modified?
  yes.
   
• MC is not ok at large delta-eta in central detector (slide 10) seems ok.  but it becomes bad after all corrections: The key difference is the inclusion of HF and extra cuts on ND events.
   
• How were gamma-hadron collision simulated?
  Were not simulated and are not included in study. (Audio connection not perfect to understand everything)
   
• Very nice propaganda for proton-oxygen beams! None linear effects that cannot be extrapolated. LHCb can also measure this.

 

• General question to everybody: why does EPOS and Sibyll have so similar CR predictions while being so different internally?
  
  Some parameters play an important role. Many details are in fact not so critical. There is no hydro in SIBYLL. The latest change in Sibyll to 2.3d was "undoing" a bug in sibyll that was imitating the effects of hydro to some extend. Thus, if there was data: the models can be fit also to such data. The question is: what data is available, what data is relevant, how well is it described.

 

• How important are the lower energy interactions in showers relevant to high energy models.
  
  Important is also the number of "child generations". Thus, high energy collisions are much more important because the effects accumulate. At lower energies it is just a single "effect", however, the number of particles is extremely high at lower energies. Systematic studies are needed to learn more.

13 EPOS and EPOS3

Speaker: Klaus Werner

corsika.pdf

14 SIBYLL2.3d

Speaker: felix riehn (LIP, Lisbon)

sibyll-slides-c8-workshop-2020-v2.pdf

15 UrQMD

Speaker: Marcus Bleicher

2020_Bleicher_CCRSW.pdf

16 Diffraction in pPb collisions measured with CMS

Speaker: Lev Kheyn

CORSIKA2020_lk.pdf

17 Combined discussion, Q&A

Speaker: Hans Dembinski (Max Planck Institute for Nuclear Physics, Heidelberg)

3:25 PM Coffee

CORSIKA Workshop: Connection to accelerators

Convener: Remy Prechelt (University of Hawai'i at Manoa)

Minutes of questions and discussions

Session "Connection to accelerators" on Tuesday 15:50

• To Matias Perlin: why is the "remnants" flat in slide 5? Can this be in pseudorapidity?
  Yes, since this is just the "contribution" to energy, not the energy.
   
• To Anna Schaefer: what kind of physics can be switched off?
  SMASH is very flexbile, but it remains to be investigated to what is most relevant for air shower applications, where e.g. it is not very relevant what happens to the nucleus at rest.
   
• What is the afterburner and how is it related to hybrid mode?
  There is not yet a full hybrid mode. The afterburner is just parts of it.
   
• Seconds per event is actually VERY slow for CR, we need ms, as little as possible.
  This really has to be benchmarked. There is no good information here. The initialization is what makes it slow, if this can be factored-out or tabularized, it might be MUCH faster. 
   
• To Andrii Tykhonov: why is FLUKA and CRMS_DPMJET different?
  This is hard to answer at this moment. The FLUKA geometry is much simpler than the G4, also DPMJET is not the same. Also the lower energy part is treated differently.
   
• The differences between CRMC-DPMJET and FLUKA should be minimal (Anatoli), also the low energy part is not so important at such high energy showers. Thus, a better understanding would be good, geometry may be the key.
   
• What is the runtime difference between FTFP_BERT and CRMC at high energies?
  Not yet precise estimation, but difference is small -- if noticeable. The difference is just in a few single high-energy collisions.
   
• To Felix Kling:  Will FASERnu also see muons?
  Yes, it will record ALL muons. It is not clear where they all will come from. Not fully studied yet. There will also be contributions of collisions with the beam pipe etc.
   
• What is the reason for large uncertainties on flux e.g. in slide 5?
  This is mainly flux uncertainty. Very model dependent.
   
• LHCf is even much smaller in acceptance, thus, observed differences will not translate 1:1 to FASER. In particular, the not-extremely-forward LHCf data show different/better agreement with MC.
   
• To Mike Albrow: What is possible for timing in FMS? Timing precision of <100ps, maybe down to 20ps.
   
• Radiation hardness of all components is good enough to leave it always there. Only the transition radiation will have to be removable.
   
• There is complementarity with different aspects of both, FASER and also LHCb.

18 Core-Corona approach and the muon deficit

Speaker: Matias Perlin

Perlin-CORSIKA_Workshop.pdf

19 Lower energy hadron collisions with SMASH

Speaker: Anna Schäfer

SMASH_CORSIKA.pdf

20 CRMC in Geant4

Speaker: Andrii Tykhonov

CRMC_in_Geant4_V2_CORSIKA_workshop.pdf

21 FASER: very-forward neutrinos

Speaker: Felix Kling (SLAC)

FASER_Corsika.pdf

22 A Forward Multiparticle Spectrometer for CMS

I describe a Forward Multiparticle Spectrometer (FMS) that could be installed as a new subsystem for CMS in Run 4, with p+p collisions at s = 14 TeV and with p + O and O + O collisions. It uses a new superconducting dipole as a spectrometer magnet to measure multi-TeV charged hadron spectra behind a large radius beam pipe. The tracking detectors and calorimeters, between z = 116 m and 126 m, are clones of the planned CMS Endcap upgrade, supplemented by transition radiation detectors for hadron identification. In addition to measuring the spectra of  , K , p, pbar and light antinuclei, charmed hadrons,
J/ and other decaying particles can be measured at high Feynman-x. At high luminosity the FMS in a different mode can search for penetrating long-lived neutral particles. An Expression of Interest is in preparation.

Speaker: Mike Albrow

Albrow-corsika20200623.pptx

23 Combined discussion, Q&A

Speaker: Remy Prechelt (University of Hawai'i at Manoa)

Wednesday, June 24

24 Announcement

Speaker: Ralf Ulrich (KIT)

Announcement.pdf

CORSIKA Workshop: Cherenkov photons

Conveners: Gernot Maier, Konrad Bernloehr, Luisa Arrabito

Minutes of questions and discussions

Session "Air shower physics" on Monday 15:35

• To Ken Ohashi: how was diffraction defined in the study?
  The event generator definition was used.
   
• It may be a great extension of the study to include RMS(Xmax) and/or fluctuations of Nmu into the observables. This would complement the study in another important way.
   
• To Oliver Fischer: are full shower simulations needed?
  Sometimes it may be needed. There may be also dynamic involved in BSM, this may be visible in an unknown and unexpected way in showers.
   
• Why not use dedicated cascade equations as input and the just "fold" with BSM?
  This is possible, but requires to go back to square one for every new approach, idea, correction. It would be great to have a standard interface with documentation to insert BSM into shower observables and products. Both, for CE and full MC. There may be many theorists interested to work with such tools, if they are available.
   
• To Álvaro Pastor Gutiérrez: was available accelerator data used to compare some of the results to? There is CRMC that can be used to generate such data sets.
  Not yet, but sounds like a good idea.
   
• To Hans Dembinski: Why was oxygen in LHC shifted back in time? Because of corona virus?
  No, there were other reasons.
   
• Is remaining SMOG data going to be analyses and published?
  Unknown right now.
   
• To Jan Ebr: what is event statistics on slide 6?
  Probably 500 events, but statistical errors needs to be checked. They may be small?
   
• Also on slide 6: going to f19=5 is probably too extreme. One should keep in mind that the entire approach is useful only for ~small changes.

Session "Theory and models" on Tuesday 14:00

• To Klaus Werner: What is the plan on EPOS4, and the difference to EPOS3?
  Main feature of EPOS3 is the micro-canonnical hadronization, which will allow to efficiently be applied to small and large systems. There maybe needs to be a dedicated "cr" version of EPOS3 for highest performance. In EPOS4, the overall treatment of multiple scattering and hydrodynamic is going to be unified in a fundamental way. This should result in the first fully self-consistent model to treat small and large systems at small or high multiplicities.
   
• To Marcus Bleicher: Which version of UrQMD was used? 
  Different versions,  3.4 is the latest and should resemble what was shown.
  In CORSIKA we still use 1.6, mainly because of xs tables: not difficult, but to update it needs to be done carefully ~1 month.
   
• How much difference is between 1.6 and 3.4?
  Pions, Kaons, Protons don't differ a lot, but forward physics not so clear (important for CR, but not important for HI). What changed more than a few percent is Xi, Omega, etc. but can be important.
   
• Switching energy between low and high E model?
  One should use the best model to describe the physics as long as possible. The range for UrQMD is actually very broad. It goes at least to 100GeV, even to 1TeV (but not perfect any more). Current default in Corsika is 80GeV. EPOS may go down to very low energies. SIBYLL is limited.
   
• How does UrQMD include elastic scattering which apparently becomes relevant at higher energies (see e.g. slide 5)? Is there a theory behind it (optical theorem)? Is it a fit? Or is it ignored?
   
• Is there anything resembling diffraction. If not: is this a relevant limitation of the model or does this limit the potential phase-space of [best] applicability?
   
• In the comparisons to ALICE data on slide 17: is the hybrid actually absolutely needed, or does the "pure" UrQMD not also provide a good description? It seems the figures suggest this.
   
• To Lev Kheyn: Protons have high elasticity, they are not much modified?
  yes.
   
• MC is not ok at large delta-eta in central detector (slide 10) seems ok.  but it becomes bad after all corrections: The key difference is the inclusion of HF and extra cuts on ND events.
   
• How were gamma-hadron collision simulated?
  Were not simulated and are not included in study. (Audio connection not perfect to understand everything)
   
• Very nice propaganda for proton-oxygen beams! None linear effects that cannot be extrapolated. LHCb can also measure this.

 

• General question to everybody: why does EPOS and Sibyll have so similar CR predictions while being so different internally?
  
  Some parameters play an important role. Many details are in fact not so critical. There is no hydro in SIBYLL. The latest change in Sibyll to 2.3d was "undoing" a bug in sibyll that was imitating the effects of hydro to some extend. Thus, if there was data: the models can be fit also to such data. The question is: what data is available, what data is relevant, how well is it described.

 

• How important are the lower energy interactions in showers relevant to high energy models.
  
  Important is also the number of "child generations". Thus, high energy collisions are much more important because the effects accumulate. At lower energies it is just a single "effect", however, the number of particles is extremely high at lower energies. Systematic studies are needed to learn more.

25 Introduction & CTAO

Speaker: Gernot Maier

GMaier_Cherenkov.pdf

26 Hadronic interaction models & IACTs

Speaker: Michiko Ohishi (ICRR, University of Tokyo)

CORSIKA_workshop_2020_MO_v2.pdf

Questions and discussions

• Very impressive talk, with a lot of very interesting content.
   
• Atmospheric conditions, refractive index and their time dependence. How much does it affect the simulations and the performance of CTA?
  
  The magnitude of changes is not known right now (at the meeting), but there is an extended atmospheric monitoring program (LIDARs, etc.) to keep very detailed time-dependent information as input for the analysis. So this is included in all studies.
   
• Ratio of gamma-ray sensistivity e.g. side 19, and does it depend on observation time.
  
  Yes, it does. It may be smaller for smaller observation time, since it also depends on cuts.
   
• I have not seen this during the talk, maybe you have to point me to it: The fraction of proton showers misidentified as photons, how does it depend on parameters like E_e.m. (as shown e.g. on slide 10)?

          Probability of  E_e.m./E_primary > 0.80 roughly corresponds to the fraction of proton showers misidentified as photons. But this "0.80" depends on the gamma-hadron discrimination capability of the system, 0.8 is not so bad for CTA, and the value may be lower for the current IACT systems.

27 Cherenkov propagation optimization in Corsika 7 for the CTA use case

We will report about the optimization work done so far on CORSIKA 7 for the use case of CTA. We will present in particular how we have applied vectorization techniques to the Cherenkov module obtaining a speed-up of almost a factor 2.

Speaker: Luisa Arrabito

Corsika_2020.pdf

Questions and discussions

• Is random number generation a candidate for vectorization?
  • I am not an expert but I know that in GeantIV they were able to vectorize random number generation in VecMath lib.
  • MKL provides vectorized RNG for Intel processors that is *significantly* faster than standard generators. The documentation provides some benchmarks that might interest you.
     
• Why not going to FP32? Any experience already on Threadripper or ARM?
  • Only limited tests with FP32, but it's clearly a path we want to explore. The difficulty with MC codes is that intermediate different numerical results determine completely different execution paths and it's difficult to compare results and perf. I agree that we should try out FP32 and validate the results based on statistical distributions.
  • No experience yet with Thraedripper and ARM.
     
• Does "horizontal vectorization" require similar concepts as GPU code?
  • Yes, it needs data to be accumulated before-hand to be processed in parallel.
  • Vectorization and GPUs have similar requirements.
     
• Is "inner loop vectorization" a necessity for "horizontal vectorization" or not at all?
  • No, they can be implemented independentely and gains can be cumulated.
     
• Does anyone know an example, where RNG generation was actually the bottleneck? One can accelerate a lot of things, but only when the bottleneck is accelerated the program gets faster.
  • function rmmard (rng) represents 7% of total cycles for information in corsika 7 (nshowers=5000). We can maybe win some cycles with vectorize version
  • The challenge is whether vectorizing RNG is taking away AVX cycles from other code paths. If you are saturating the AVX pipeline with Cherenkov production, it might hurt but I agree that it should be benchmarked and explored.
  • There are several newer rng that have better Performance with similar periodicity. There is alot of work that can be applied and tested there
  • yes and the problem is the frequency. Intel cpu loose 100-300mhz in general when we use avx instruction. It was the problem we saw with avx512. Vectorization is not always the good way
  • I think the newest Generation comsumer ones (intel) does not reduce the frequency but comsumes absurd amounts of power

28 Cherenkov photons on GPU

Speaker: Dominik Baack (TU Dortmund)

2020 Vortrag.pdf

Questions and discussions

• Is there a fundamental reason why GPU are more efficient on 3D than on 4D? Is it "just" memory?
   
• When you write "you can calculate 1e6 photons per second", what hardware does this refer to?

29 Fluorescence emission & IACTS

Speakers: Daniel Morcuende (Universidad Complutense de Madrid), Jose Luis Vazquez-Poletti (Universidad Complutense de Madrid (Spain))

CORSIKA2020_fluorescence_DMorcuende.pdf

Questions and discussion

• Will there be an implementation for C8 too?
  • yes, this is the goal.
  • In fact, in C8 the actual work to implement a similar physics should be considerably less.
     
• For the fluorescence, adaptation to GPU might also be very good.
  • Photon propagation is very simple and very fast without losing accuracy.

30 Combined discussion, Q&A

Speaker: Konrad Bernlöhr (MPI für Kernphysik)

cherenkov_discussion.pdf

Questions and discussions

• Magnetic fields in Corsika8, will it be limited in a similar way as so far?
  • No, the modelling of environment is very modular and to a large extend open to the users. A position dependent magnetic fields will be there in Corsika8. Standard sets of fields, homogeneous, but also Earth-wide, will be distributed with Corsika8.
• The interface for Cherenkov photon production in terms of programming. Will there be a "track segment" class?
  • The point is really about the clean interface and that all needed information is easily accessible. If there will be "track segment" class is not clear, since it might not be needed as-is.
  • The current C8 design is already very close to what was suggested by Konrad. There is a "process" interface that get basically all important information. However, this will be extended and refined whenever needed and useful!
• For the fluoresence light emission, further geometric cuts on the "viewing angle" of observers might be able to significantly reduce the computing time.

 

3:40 PM Coffee

CORSIKA Workshop: Radio emission

Convener: Tim Huege (KIT)

Minutes of questions and discussions

Session "Connection to accelerators" on Tuesday 15:50

• To Matias Perlin: why is the "remnants" flat in slide 5? Can this be in pseudorapidity?
  Yes, since this is just the "contribution" to energy, not the energy.
   
• To Anna Schaefer: what kind of physics can be switched off?
  SMASH is very flexbile, but it remains to be investigated to what is most relevant for air shower applications, where e.g. it is not very relevant what happens to the nucleus at rest.
   
• What is the afterburner and how is it related to hybrid mode?
  There is not yet a full hybrid mode. The afterburner is just parts of it.
   
• Seconds per event is actually VERY slow for CR, we need ms, as little as possible.
  This really has to be benchmarked. There is no good information here. The initialization is what makes it slow, if this can be factored-out or tabularized, it might be MUCH faster. 
   
• To Andrii Tykhonov: why is FLUKA and CRMS_DPMJET different?
  This is hard to answer at this moment. The FLUKA geometry is much simpler than the G4, also DPMJET is not the same. Also the lower energy part is treated differently.
   
• The differences between CRMC-DPMJET and FLUKA should be minimal (Anatoli), also the low energy part is not so important at such high energy showers. Thus, a better understanding would be good, geometry may be the key.
   
• What is the runtime difference between FTFP_BERT and CRMC at high energies?
  Not yet precise estimation, but difference is small -- if noticeable. The difference is just in a few single high-energy collisions.
   
• To Felix Kling:  Will FASERnu also see muons?
  Yes, it will record ALL muons. It is not clear where they all will come from. Not fully studied yet. There will also be contributions of collisions with the beam pipe etc.
   
• What is the reason for large uncertainties on flux e.g. in slide 5?
  This is mainly flux uncertainty. Very model dependent.
   
• LHCf is even much smaller in acceptance, thus, observed differences will not translate 1:1 to FASER. In particular, the not-extremely-forward LHCf data show different/better agreement with MC.
   
• To Mike Albrow: What is possible for timing in FMS? Timing precision of <100ps, maybe down to 20ps.
   
• Radiation hardness of all components is good enough to leave it always there. Only the transition radiation will have to be removable.
   
• There is complementarity with different aspects of both, FASER and also LHCb.

31 General thoughts on radio in CORSIKA 8

Speaker: Tim Huege (KIT)

huege-corsika8-radio-v1.pdf

Questions, Discussions

• Will it be possible to have more complicated implementation of refractive index than just density-dependent in corsika8?
  • Yes, this is possible and should be made available (e.g. humidity dependence,...). However, many typical applications will probably not need this.
  • In C7/CoREAS it si basically already discoupled since it is tabulated? The table is just filled according to the density.
  • That is correct. If you use GDAS atmospheres, the humidity effects are already included.
• It is a good time for the CORSIKA8 project to "change gears" and move more focused towards full application to real problems (radio).

32 Suggestion for a radio interface in CORSIKA 8

Speaker: Remy Prechelt (University of Hawai'i at Manoa)

radio_interface.pdf

Questions, Discussions

• Why separating (templating) radio emission and antenna response.
  • The goal is to make it clear and easy for users in special applications to modify any parameter of the emission and recording physics. If people want non-standard output (antenna designs, response) they can easily plug this into C8.
• It would be good to have time-dependent Poynting vector at the receiver points, to i.e. even record magnetic field components at antenna locations - if needed.
  • This is straightforward to extend, may be already possible in current C8 radio interface ideas.
• Now is the right time to voice your wishes in respect to radio simulations. Later this may become more difficult.
• Further technical discussion should move to the regular corsika phone meetings, and/or https://gitlab.ikp.kit.edu/AirShowerPhysics/corsika/merge_requests/202

33 Simulations of radio emission for air showers penetrating from air to ice

Speaker: Krijn de Vries

Corsika8_KdV.pdf

Questions, Discussions

• The shower on slide 11, does it has a maximum already in air?
  • At high altitudes and high energies may showers have their maximum in-ice. This may be one of those examples.
• Is transition radiation emitted at the air-ice interface?
  • This is not yet included in those simulations. But yes, it is this interface.
• Has this been studied in a laboratory for coherent radio emission?
  • Yes, references are available for this.
• What index of refraction should be used, and how can it be parameterized. See slide 13. Can this be calculated in a fundamental way, without dt/dtprime?
  • No, this happens in based on first principles.
     
• On coherent transition radiation:
  • So coherent transition radiation has been observed in several beam test experiments. Work to which I was directly involved was the emission from the Telescope Array Electron Light Source beam which was directed into air to mimic a cosmic-ray induced particle cascade: https://arxiv.org/abs/1902.02737
  • Coherent transition radiation has also been observed and simulated in the microwave regime (5-8GHz) in the context of accelerator detector design as well.
  • In this work there are several references to earlier works at SLAC and ARGONNE where this emission has been observed from beams moving between different media, and not just from the accelerator into air. This formalism is calculated / applied to airshowers in these works: https://arxiv.org/abs/1503.02808 ; https://arxiv.org/abs/1811.11003 ; https://arxiv.org/abs/1903.08750. The proces of an air shower hitting Earth has also been detected and investigated by the EXTASIS experiment: https://arxiv.org/abs/1903.02792
    Likely I am missing many references, where most can probably be found in these papers though

34 Simulations for in-ice radio experiments

Speaker: Christian Glaser

inice_neutrino.pdf

Questions, Discussions

• On slide 8, there is also a second solution to the red line which is bounced from the ice. This may have considerable time delay and may lead to big problems in simulations.
  • This was also further discussed in the "combined discussion session".

• There were plans shown for an "emit" interface, but where is the "propagate" function?
  • This becomes very technical, see links to other discussion platforms where we can continue that.

35 Combined discussion, Q&A

Speaker: Tim Huege (KIT)

Discussion, Questions

there was a long disucssion, but I was mostly involved actively and could not write down many of the details. If someone has: please add here.

• About the question of antenna directionality, including "bounced" signals.
  • Zenith-angle dependence at the antenna might not be neccessary once we have the Poynting vector implemented. The reflected signal would have a drastically different Poynting vector
  • So this would be automatically taken into account when applying the antenna model time-dependent in the direction of the Poynting vector.
  • I am not so sure. If you had two signals from different directions arriving simultaneously, it is not the same as one signal arriving from an "average direction"!?
  • Is it? There only can be one electric field at a single time and place, which would be the superposition. Would be intersting to investigate wether it makes a difference.
  • So if one is located inside the Cherenkov cone, you get signals from two different heights at the same time. That might be the exception. For the rest I think I agree (up to very good approximation) with Frank, although this is a bit tricky and I need to think on it.
  • If you have a directional antenna, for example, it's not clear how you can go from the single Poynting Vector to detector response. Note that this is not only a problem in ice. Multipath is possible in the atmosphere as well especially in grazing geometries.
  • of course, if you have the time-dependent poynting vector everywhere in the volume, you have everything you need, but just having it at the detector I think is not enough.
  • Why do you need it in the medium? The (time-dependent) measurement of an antenna should only depend on the electric and magnetic fields at the position of the antenna.
  • I guess because there is no such thing as an ideal antenna, so any real antenna depends on the electric field in the neigborhood
     
• One of the main points is, how can the radio community be connected to Corsika8, how long will it take to implement the remaining critical parts of the framework?
  • https://gitlab.ikp.kit.edu/AirShowerPhysics/corsika
  • and within ~months Corsika8 should be able to to full fledged radio emissions ready for full comparisons to other codes.
  • slack: https://join.slack.com/t/corsika/shared_invite/zt-f61qw29z-9vALF1VcY9wJNEPtqtXDdw
  • corsika.slack.com

 

Thursday, June 25

36 CORSIKA in Heidelberg

Speaker: Jim Hinton

CORSIKAHinton.pdf

37 Workshop photo

WorkshopPhoto2020.jpg

CORSIKA Workshop: GPU and acceleration

Convener: Dominik Baack (TU Dortmund)

Minutes of questions and discussions

Session "Connection to accelerators" on Tuesday 15:50

• To Matias Perlin: why is the "remnants" flat in slide 5? Can this be in pseudorapidity?
  Yes, since this is just the "contribution" to energy, not the energy.
   
• To Anna Schaefer: what kind of physics can be switched off?
  SMASH is very flexbile, but it remains to be investigated to what is most relevant for air shower applications, where e.g. it is not very relevant what happens to the nucleus at rest.
   
• What is the afterburner and how is it related to hybrid mode?
  There is not yet a full hybrid mode. The afterburner is just parts of it.
   
• Seconds per event is actually VERY slow for CR, we need ms, as little as possible.
  This really has to be benchmarked. There is no good information here. The initialization is what makes it slow, if this can be factored-out or tabularized, it might be MUCH faster. 
   
• To Andrii Tykhonov: why is FLUKA and CRMS_DPMJET different?
  This is hard to answer at this moment. The FLUKA geometry is much simpler than the G4, also DPMJET is not the same. Also the lower energy part is treated differently.
   
• The differences between CRMC-DPMJET and FLUKA should be minimal (Anatoli), also the low energy part is not so important at such high energy showers. Thus, a better understanding would be good, geometry may be the key.
   
• What is the runtime difference between FTFP_BERT and CRMC at high energies?
  Not yet precise estimation, but difference is small -- if noticeable. The difference is just in a few single high-energy collisions.
   
• To Felix Kling:  Will FASERnu also see muons?
  Yes, it will record ALL muons. It is not clear where they all will come from. Not fully studied yet. There will also be contributions of collisions with the beam pipe etc.
   
• What is the reason for large uncertainties on flux e.g. in slide 5?
  This is mainly flux uncertainty. Very model dependent.
   
• LHCf is even much smaller in acceptance, thus, observed differences will not translate 1:1 to FASER. In particular, the not-extremely-forward LHCf data show different/better agreement with MC.
   
• To Mike Albrow: What is possible for timing in FMS? Timing precision of <100ps, maybe down to 20ps.
   
• Radiation hardness of all components is good enough to leave it always there. Only the transition radiation will have to be removable.
   
• There is complementarity with different aspects of both, FASER and also LHCb.

38 Cascade equations and accelerated methods

Speaker: Anatoli Fedynitch (DESY Zeuthen)

200625_CKA8_WS_Anatoli.pdf

CKA8_WS_demo.ipynb

Questions, Discussions

• It is possible to couple python code to C8 in the future, if data exchange is all there.

  • But it is actually very simple to implement the inner part of MCEq in C++.

• What about e.m. cascades?
  • Stephan will talk about this. But it works in the same way.
     
• Use MCEq in a similar way to CONEX in C7: calculate sub-showers with weight.
  • Yes, but validation of results is tedious and is only partly done so far. So people prefer much more time-consumimg solutions, like e.g. unthinned showers.
  • But from theory CE are at least as precise as full MC, so this is worth to be further explored on.

39 Random number generation in multi-threaded programs and recent developments in numpy | Efficient static dispatch mechanisms in C++

Speaker: Hans Dembinski (Max Planck Institute for Nuclear Physics, Heidelberg)

dembinski_rng.html

dembinski_static_dispatch.html

Questions, Discussion

• The shown statistical test for RNGs may not be ideal for the example. Is it known why and how this was done?
  • This is not known, but chosing another test suite will probably look even better for PCG RNGs.
     
• What about C++20?
  • We need excellent compiler support on all relevant systems. For C++20 this is still premature. 
     
• The use of exact type matching:
  • It is faster than enable_if, in terms of compilation. Compilation time can be a relevant parameter.
  • This is complementary to enable_if, also the code maintenance needs to be factored in. It requires investigation, and is another trick in the collection.

 

 

 

40 EM cascades

Speaker: Stephan Meighen-Berger

EM_Showers.pdf

Questions, Discussion

• The LPM energy threshold is only slightly density dependent. It is always in the 10TeV range.
   
• But dielectric effects are much more material dependent.

41 Electromagnetic shower simulation using PROPOSAL

Speaker: Maximilian Sackel

em_shower_using_proposal.pdf

 

• Where is the problem with inhomgeneous LPM effects?
  • Maybe just technical. One issue is that integrals are solved in energy, so mapping to position is a critical question.

 

42 Combined discussion, Q&A

Speaker: Dominik Baack (TU Dortmund)

Discussion, Questions

• Is there a MC version of the EmCa model.
  • Only rudimentary yet.

 

• Material related properties are interesting. Should integrals be calculated on the fly? Or tabularized? But this has consequences on the environment model: layers in the atmosphere?
  • Besides tabularization, also step-lengths are important.
  • Extra problems are LPM in CE: the dimension of tables required (also angles) becomes problematic.

 

• How is LPM implement in C7:
  • This is following the AIRES approach. It is a MC method using rejection / importance-sampling.
  • It is only a local-density estimation.
  • No LPM in CONEX.
     
• The LPM approach of C7 is a smart and good approach, but it has to be checked.
   
• From radio perspective (also Cherenkov) the most important is to have ANY e.m. shower in C8, if something is incomplete, this is not relevant.
• Also: for CTA the LPM effects is not relevant.
• Are there any plans to include EGS4 in c8 as a reference?
  • There may be an option to run EGS4 via CONEX. CONEX is already linked to C8. This should be OK for for testing and validation, but very unlikely for a "production mode". 
  • Also EmCa can be configured in a way to resemble EGS4 in a physical way, however, EGS4 in C8 is NOT the original EGS4... so there are unknown deviations.
     

3:45 PM Coffee

CORSIKA Workshop: Summary, Outlook

Convener: Lukas Nellen (I de CIencias Nucleares, Universidad Nacional Autonoma de Mexico)

Minutes of questions and discussions

Session on Monday 14:10

• To Maximilian Reininghaus: why are e.g. on slide 9 the differences for muons so small and for other particles much larger?
  
  One aspect is that e.g. kaons are in general much less frequent and even larger discrepancies will not generate large effects.
   
• To Augusto Alves: does the current refactoring is in conflict with the mantra of small individual, incremental changes?
  
  Yes, it is. It is like a "prototype version2 framework" which corrects and changes some of the original choices and concepts. It is necessary to allow further development to become more clear, and more easy.
   
• To Ralf Ulrich: Isn't one of the important next work packages "e.m." cascades?
  Absolutely yes. But this is basically a physics contribution and not part of "work on the framework". We want to include e.m. showers asap. 

43 Recipes for primordial soups (Astrobiology)

Speakers: Dominik Schwarz, Petra Lutter

LutterSchwarz.pdf

Discussion, Questions

• This BAPS approach is mostly sensitive to atmospheric processes, CR may be more relevant in interstellar materials.
  • This is debated. Everything has to be considered. There is complex chemistry in protostellar clouds, but how can it cluster in planets. There may be many/several origins of complex molecules.
  • It may also be a "fine-tuning" problem and depend on planets and environments.
• There is a large number of unknowns, what is potentially the most interesting effects of CR showers. There is a interesting problem to understand atmospheric profiles in the early stages of planets. What are probable interesting properties of EXO air showers, what are the uncertainties?
  • Very little is known. This is basically interesting to investigate. At least we known quite a bit about the Earth atmosphere also in the past, we already known significantly less about Venus/Mars...
  • Titan is in the focus of many roadmaps, maybe as laboratory for "early Earth".
  • Energies deposited by cosmic rays in the atmospheres are a key quantity.

 

 

44 CORSIKA: exotic physics, and pandemic research

Speaker: Stephan Meighen-Berger

CORSIKA_Exotic.pdf

Discussion, Questions

• On p6, why does the BSM signal shape follows so much the "mu-nu" and not the "mu-had" part.
  • This is a feature of attenuation. It is not energy spectra, it is basically signal almost not attenuated.

45 End and final discussion

Speaker: Ralf Ulrich (KIT)