Long-time problems running CBC on Marconi/Archer
I see some weird long-time behaviour running CBC with and without flow shear on both Archer and Marconi:
After a little while, the heat flux (grabbed from the ASCII output) dramatically increases. This happens on both Archer and Marconi. I’ve attached one of my input files where this happens. Can anybody else reproduce this?
Comments (16)
-
-
reporter I’ll take a look at the spectra and report back. I was using the latest version of the code off the main branch.
-
Just to confirm I’ve reproduced this locally. Looking at the spectra suggests there’s a couple of kx/ky values which dominate the flux, with the lowest ky becoming more significant towards the end of the simulation. Perhaps larger ny may help? No obvious signs of things building up at the edge of the domain at least.
-
I’m not sure if this is intentional or not but I notice that you’ve got a
/
in front of the&hyper_knobs
line which I think will mean that this section is ignored and as such you’ll end up with no hyper terms enabled. -
reporter Aaah shoot. Let me try this again. I’m sure this is the problem.
-
I’m not too familiar with these terms but the documentation suggests
const_amp = .false.
is recommended for nonlinear simulations (https://gyrokinetics.gitlab.io/gs2/page/namelists/index.html#hyper_knobs-const_amp) -
reporter - marked as minor
-
reporter That didn’t fix it… I’m going to run without const_amp and see what happens. Bumping the priority back to major.
-
reporter - marked as major
-
I’ve run the case with const_amp = false (and without flow shear) and it does seem to be well behaved as shown below. (sorry for the lack of labels, this is es_heat_flux_tot vs time).
-
reporter Hmm… I don’t see why running with a bog-standard hyperviscosity wouldn’t be enough to stop this growth at late times. Maybe I’ll bump up D_hypervisc and see if that helps.
-
reporter Update: upping D_hypervisc to 0.2, using const_amp and g_exb = 0.005 results in this:
Now the heat flux gets clobbered at late times. Let me see if the same happens with flow shear when const_amp = .false.
-
@Denis St-Onge are you still investigating this or can the issue be closed?
-
reporter I haven’t really taken a closer look at this issue since I’m not a regular GS2. That being said, this ticket shouldn’t be closed because the issue has yet to be resolved. If I find some time down the line, I can take a closer look at it.
-
reporter I’ve run this again but now with D_hyper = 5.0. This now works like it should. I think this was just a matter of running with insufficient dissipation, since nx was so high and the const_amp hyperviscosity is scaled to max(kperp2).
-
reporter - changed status to resolved
If one increases the resolution using const_amp dissipation, one may also need to increase D_hyper accordingly.
- Log in to comment
@Denis St-Onge I don’t think I’ve seen this in the cyclone before but I’ve seen it in some other cases. Do you know which version of the code you’re using? I can try to reproduce this locally. Have you looked at the full spectra to see if it’s one region just taking off?