1 of 10

SX simulations at CERN as part of IFAST-REX

F. M. Velotti, M. Fraser, P. Arrutia, M. Pari

2 of 10

Introduction

  • Very nice overview from Pablo at the last IFAST-REX collaboration meeting
  • Classify the usage of these different tools depending on application
  • Starting point for the WP3 for SX simulation in the IFAST-REX context

3 of 10

Classification

  • Depending on the usage needed for, different tools should be used
  • Very accurate phase-space description:
    • MADX thin tracking (or MADX batch)
    • PTC tracking
    • MapTrack at high order
  • Loss estimation:
    • Tracking code coupled with matter tracking (FLUKA, GEANT4)
    • Simple aperture cut (MADX or any aperture checks)
    • Tracking of primary protons (MapTrack or MADX + pyCollimate, SixTrack for collimation)
  • Spill quality:
    • Henon map tracking (2D or 4D)

4 of 10

Simulation accuracy

  • Crystal shadowing => estimate overall loss reduction
  • Phase-space folding with octupoles => estimate extracted beam emittance
  • Fast enough to run data-simulations numerical optimisation [1]

5 of 10

Loss estimation

  • Effect of beam intercepting devices (e.g. diffusers and crystals) on overall machine losses and activation
  • Fundamental to accurately describe particle-matter interaction:

SixTrack for collimation

Merlin

MapTrack + pycollimate

6 of 10

Spill quality

  • Spill quality simulations have rather different requirements than the other SX simulations
  • Here we need to track for many turns and we are not interested in reproducing very accurately the extracted beam phase-space
  • 2D and 4D Henon map tools proposed and successfully used to reproduce measured spill in the PS and SPS (M. Pari and P. Arrutia)

2D Henon map

2D Henon map

4D Henon map

7 of 10

Moving forward

  • SX is a very complete process, entangled with the whole accelerator beam dynamics
  • In some cases, even collective effects should be taken into account to be able to produce accurate simulations
  • Ideally, we would have a single tool to do it all => not suitable strategy for efficiency!
    • Rarely we will need to simulate phase-space at extraction, losses and spill quality all in the same run…
  • We believe that different tools optimised for different usages can guarantee the best efficiency of simulations

8 of 10

Moving forward

  • Accurate simulations: Xsuite => single particle dynamics tracking tool in python which can be used on GPUs
    • Developed by BE/ABP at CERN
    • Well documented and it seems the shining new star
    • Very well integrated with MADX and cpymad
    • It can also be extend to treat collective effects (e.g. space charge), collimators…and any “exotic” element
  • Extendible => For example, it can integrate pyCollimate easily (also for crystal treatment using probability density function from data)
    • pyCollimate not GPU ready…it should come soon
    • As very scriptable, it may be interfaced with other scattering routines? First examples where GEANT4 has been interfaced with Xsuite already available
  • Testing ongoing…it looks very promising!

9 of 10

Moving forward

  • To push even further speed for spill quality simulations, Henon map tracking on GPUs => it should be straightforward
    • Blond on GPUs?
  • What about Hamiltonian [2] and symplectic [3] neural networks?
    • The could represent a game change in speed for accurate simulations where the machine is static…
    • Or we could even parameterise the transformations with machine settings
    • We are testing these possibilities for TL tracking

10 of 10

Conclusions

  • SX simulations needed to look at many different aspects:
    • Phase space accuracy
    • Losses estimation
    • Spill quality…
  • Very difficult to have single-tool solution => computationally too expensive
    • Available tools are OK, but surely optimised for the machine for which have been developed for
  • Xsuite is a freshly developed tool at CERN which could help us significantly
    • GPU accelerated, written in python and easily extendible
    • Testing ongoing for SPS slow extraction
  • Hamiltonian and symplectic NN could further speed up simulations
    • First tests are ongoing, but more systematic studies needed to assess performance reach