alfven test noise

[nicolasaunai]

nightly build test "alfven wave" shows a very large and unusual noise in build of the branch syncmoments of PR #697 .
Other tests do not show issues. The alfven wave seems to still propagate OK since the wave speed is correct, but a large noise arises on the refined level.

A priori this seems to be the "typical" issue of the Niquyst mode piling up onto the fine level that is usually fixed with hyperresistive dissipation. What's weird is that for this test the hyperresistivity is 1e-3, which is very weak for dx=1 (L0) and dx=0.5 (L1) and so we should have seen this issue already before.

[nicolasaunai]

Here is what I get for master (3947023) :

and for the branch levelghostinterp (7db2c9d) :

It looks like on levelghostinterp there is more noise than on master on the refined level. To what extent this is just a random deviation that could also happen on master or related to the diff introduced (fixing time interpolation of level ghost particles) is unclear.

[nicolasaunai]

on branch syncmoments: Increasing hyper-resistvity from 0.001 to 0.01 surprisingly does not help

[nicolasaunai]

Increasing hyper-resistivity to 0.02 or 0.05 does not help. Probably need to smooth and revive #511. Testing harris with dynamic refinement now. Given that tdtagged1d works ok, it is possible that the fixed refinement boundaries here do not help and moving boundaries prevent pile up of grid scale fluctuations.

[nicolasaunai]

Tested Harris with dynamic refinement (from the Paper's test case) and it seems OK:

What probably happens in the case of the Aflven test, is that the fixed refinement area leaves time for the high freq. noise to pile up and grow. With dynamic refinement, the motion and removal of refined patches also removes high freq. modes and effectively acts as a dynamical filter.

I let this issue open until smoothing is implemented #511 to see if that helps. Considering fixed refinement for a million time step is not the main use case and Harris still works, this issue is probably not so major.

[nicolasaunai]

PR #867 fixed the straight lines by changing the way 1D plots are made