First look at Xsuite for Slow Extraction

Pablo Andreas Arrutia Sota

• A small example with slow extraction with SPS
• Very easy to get started with the documentation and examples
• Xsuite is the new player in the zoo of simulation software

Key Features

• GPU support
• Easy for time-dependent
• Active community
• Easy to combine with particle-matter software

Benchmarking Test

• Very simple SloEx sim with
  • MAD-X Thintrack
  • Xsuite direct import from MAD-X
  • Henontrack (linear transport + virtual sextupole)
• Looking at small amplitude phase-space portait: very similar charachteristics
• Separatrix arm: important for loss studies
• Good agreement with Xsuite and MAD-X
• Henontrack cannot reproduce amplitude detuning so some errors are expected
• Great agreement with MAD-X and Xsuite when looking closer

Transit Time

• Important for time structure and spill quality studies
• 100 particles simulated, plotted the time taken to leave the machine
• Perfect agreement with Xsuite and MAD-X
• Henontrack also performs well for small transit times

Ripple Mockup study

• Add a 50hz pertubation
• Over 400 quads in the machine but can be trimmed together

Conclusion and next Steps

• Easy importing from MAD-X to Xsuite
• Xsuite and MAD-X are in great agreement
• Time-dependent parameters easily implemented
• Looking at implementing pyCollimate

Questions

• Slide 11: time dependent studies. Other parameters like RFKO can be handled similarly.
  • Longitudinal coordinates are accessible
  • Depending on what the change is - if it is a fast change, it is not equivalent
  • If it is a slow change, it is the same idea
  • A possibility to code it yourself, take voltage as a function of z
• Do you have RF cavities that you can track longitudinally? Can you track outside the bucket?
  • Tracking can be done for a few turns outside the bucket
  • One possibility is to introduce a periodicity
  • These SloEx simulations were done debunched
• RFKO: typically frequencies are lower, arrival time doesn’t matter, as a practical simplification
• How many turns have we tried?
  • No real limitation
  • ~30 million have been done for LHC
• Question: tracking on transfer lines?
  • Discussing this currently
  • Twiss can be given as initial conditions
  • Slowly getting there
  • Putting together a use-case would be useful for example
• How do you know that after a million turns things are correct?
  • Check for symplectic response matrix
  • Maps are symplectic by construction
  • Compare to other codes (SixTrack)
  • Convergence checks by changing the number of slices
• Is it possible to have a 3D field map?
  • We already have 2D time dependent map, implemented in a symplectic way
  • Ideas of 3D static map