tracer particles

[BenWibking]

It would be useful to have tracer particles, following the Parthenon example.

We could even use the marker particle advection algorithm using the time-centered, face-centered velocities: https://ui.adsabs.harvard.edu/abs/1965PhFl....8.2182H/abstract

[pgrete]

This is actually on my list right are OpenPMD/Adios output (as we need those for another project).
I already went a bit through the literature as there are some gotchas, see https://doi.org/10.1093/mnras/stt1383 and https://doi.org/10.1111/j.1365-2966.2010.16810.x with follow up work by https://doi.org/10.1051/0004-6361/201834496 and potentially also of interest (as more simple mitigation strategy compared to MC particles): https://doi.org/10.1016/j.jcp.2019.109054 and https://doi.org/10.3847/1538-4357/ac6c2b

[BenWibking]

Yes, I'm aware of this issue. I think it is solved by using the face-centered velocity from the Riemann solver (see equation 7b: https://ui.adsabs.harvard.edu/abs/2008ASPC..385..273L/abstract). I implemented this in Quokka, and it seems to work well.

[pgrete]

Code todo list (ping @mfournier01 )

• MC tracer movement
  • start from Add support for tracer particles #102 and create branch derived from there
  • for now use vl2 integrator (for fluid)
  • because fluxes at the end of the driver (when the operator split particle pusher is called), contains the full dt fluxes
• dynamic injection mechanism
• require globally unique ids over the entire simulation (time)
• tracking (additional/other) fields
• separate tracer feature from pgen specific features

[pgrete]

Yes, I'm aware of this issue. I think it is solved by using the face-centered velocity from the Riemann solver (see equation 7b: https://ui.adsabs.harvard.edu/abs/2008ASPC..385..273L/abstract). I implemented this in Quokka, and it seems to work well.

Is this the right reference?

[BenWibking]

Yes. They happen to use the face-centered velocity for the dual energy implementation. The face-centered velocity is actually their F^2 in this equation:

So you can get the face-centered velocity from any Riemann solver by taking the rho component of the Riemann flux (i.e., the mass flux) and dividing it by either rho_L or rho_R, depending on the sign of the mass flux.

[pgrete]

Gotcha, so this effectively boils down to the strategy used in https://iopscience.iop.org/article/10.3847/1538-4357/ac6c2b (see eq 9)

[BenWibking]

Yes, I think so. I was not aware of this paper, so I am not 100% sure the interpolation is the same, but the basic strategy is the same.