Bad dealiasing of RK schemes
Runge-Kutta time schemes seem to be badly dealiased. Dealiasing happens at the end of the time-step only, allowing aliases to appear under the spectral truncation threshold during successive substeps.
This can be seen in the attached notebook:
by choosing RK2 for the time scheme and setting the dealiasing coefficient to 0.66. Even though aliases above the truncation limit are removed, some alias appear under this limit from the multiple evaluations of the nonlinear tendencies inside the time step.
To solve this, the method oper.dealiasing should be applied at the end of every substep in RK methods.
Edited by Jason Reneuve