Closed tomchor closed 1 year ago
Leaving this for reference for future self: running this for longer still produces results that are kinda far from the correct ones, indicating that the formulation still isn't correct:
Furthermore, this is a diffusivity-dominated regime. It would be nice to test this in an advection-dominated one.
I improved the time-integration and results are now better. The in figure below, top panel, I'm comparing the KE, the new (supposedly conservative) formulation, the old (non-conservative formulation) of the time-integrated KE dissipation rate, and the whole diffusive term directly from Oceananigans.Models.NonhydrostaticModel
. The bottom panel shows time-derivatives of the same quantities:
The dissipation rate estimates all match up really well with each other. Given that one of those rates comes from straight from Oceananigans.Models.NonhydrostaticModel
I take them to be correct (at least for this example). It's a bit surprising that they match that well with the measured KE. Is that the limit of predictability due to the interpolations?
KineticEnergyDissipationRate
is now also implemented in conservative form. Tested with NonhydrostaticModel
up to eps()
:tada:
I also renamed ViscousDissipationRate
to KineticEnergyDissipationRate
, which is more informative.
So far in the current form the new version actually produces results that are very close to a non-conservative form and very close to the expected values based on the decrease of KE:
One thing to note is that the level of error between ΔKE and the integrated dissipation rate is about 3%, which seem high compared to what we got for the tracer variance dissipation rate (irrc it's less than 0.5% there).
This might mean there's something wrong with my formulation or that KE dissipation is simply more challenging to calculate numerically in conservative form.