#22 Wakis team meeting minutes

Present: Antonia, Clara, Carlo, Elena, Jakob

• Latest updates and developments (E. de la Fuente):
  • New contributions to IDDEFIX: calculation of uncertainties for each resonator parameter & updates in the parameters display table w/ S. Joly (Helmholtz) and A. Gamelin (Soleil)
  • New contributions to Wakis: Bernd on Panofsky wenzel implementation vs transverse lorentz force, bug fix on non-uniform transverse wake potential. S. Lauber on pre-commit hooks, that automatize formatting across all files before commiting. ACTION add to documentation to be adopted by the dev team.
  • Discussion: Carlo shared the needs for FCC studies & bugs discovered in the impedance model. Hannes starting in May can work on streamlining the parameter retrieval with IDDEFIX from partially decayed impedance. Valentina will test it with the MKI. Dora started testing impact of using wake potential instead of wake function in Beam dynamics simulations to see how far can we push the bunch length.
    • ACTION: ensure the plane (longidutinal/transverse) is a required parameter to avoid sign mistakes or accidentally swaping real/imaginary parts

• Importing of thin and non-manifold surfaces (C. Ruby Wimmelman):

  • Clara shared the results of the compute_implicit_distance() and voxelize_rectilinear() on the thin, non-mainfold surface of the fingers of the 6L2 LHC module.
    • Impact of the grid resolution & impact of the threshold value to retrieve the simulation grid cells inside/on the surface of the solid. Promising results for voxelize_rectilinear ACTION next step is to implement the different backends in Wakis GridFIT3D and test it on all the 6L2 solids.
  • Discussion: Carlo asked how to check if the fingers are or not in contact between them? One could identify the triangles belonging to each finger and have separate collision masks per finger. Clara asked how to find the right value for the resolution & the threshold. It was agreed that the suggested resolution should be coming from the smallest feature to model. In the case of the fingers, it would be the spacing between them. TODO Implement a feature to analyze the imported geometry and suggest the resolution needed (based on snap points?)  

• First implementation of CPML (A. Huber):

  • Antonia showed the equations needed for CPML and her first implementation using split fields TODO convert to FIT matrix formalism, possible by splitting the Px, Py, Pz matrices into Pxy, Pzy, (...). Antonia explained the physical meaning of each parameter needed to optimize the CPML.
    • Results for a single gaussian wavepacket show a reflection of ~E-9 compared to previous ~E-4. Tested also in a waveguide port however the input introduces reflections in the longitudinal field. ACTION To use soft source instead. 
    • Antonia showed the optimization heatmaps for each parameter pair (kappa vs alpha) for different number of layers. Seems that 8 layers is too low. Also comparison of different profiles (geomspaced vs cubic) and the obtained sigma_max. ACTION To decide on which profile to use.
  • Discussion: the sigma max is constrained by the reduction of timestep -> TODO this could be solved by implementing the exponential time differencing (ETD) methods. Clara pointed out that the optimization could be done by ML. TODO one could use IDDEFIX logic having the objective function to minimize the reflection. It would require 1 simulation per generation which, depending on the search space, can be very heavy. ACTION Before diving into a parameter optimization, test the current implementation on a broadband source i.e. Beam.