Initial condition doesn't cleanly initialise unstable mode.

[benmcmillanwarwick]

The standard testcase is a two-stream instability, which should be cleanly initialised by a simple sinusoidal density or velocity perturbation, but it seem like these runs are very sensitive to numerical parameters, as if they were starting from numerical noise.

[TomGoffrey]

Tagging the relevant code sections here:

    REAL(num), PARAMETER :: v_drift = 0.2_num * c
    REAL(num), PARAMETER :: v_therm = 0.01_num * c
    REAL(num), PARAMETER :: v_pert = 0.1_num * v_therm

    ...

    DO ix = 1-ng, nx+ng
      current_species%drift(ix,:,:,1) = gamma_drift * m0 &
          * (v_drift + v_pert * SIN(3.0_num * x(ix)))
    END DO

Not sure if (1) there's an error in the above code, or (2) the perturbation is too small (I believe it matches that described in Lapenta 2017) or (3) a density perturbation is need as well/would work better.

[benmcmillanwarwick]

Looks like the perturbation is the same sign in velocity for each species: I think it should be the opposite sign to make phase space whorls?

[TomGoffrey]

I think that's a very sensible suggestion. So much so I've already started testing it.

[TomGoffrey]

Doesn't seem to have done anything to fix the differences between different core counts, which I think is a good test for this.

[benmcmillanwarwick]

Results of the equivalent input in EPOCH (1d)

input.deck.gz

[TomGoffrey]

What's the variation like with core count in EPOCH?

[benmcmillanwarwick]

Haven't tried yet, but since it doesn't seem to be initialising the unstable mode, doubt it is very robust...

[benmcmillanwarwick]

With initial perturbation modified to be longest mode in box, and more markers.

Still a bit too close to the noise level to get really consistent results: think if I increase initial pertubation x4 this will be better.

[benmcmillanwarwick]

Deck for last case. About 50 second runtime.

input_m1.deck.gz

[benmcmillanwarwick]

4x initial perturbation level

Which is pretty consistent between core counts, but not a very good way of evaluating the true linear growth rate.