Does 2x2 iterations work in the Semi-Implicit Quasi-Newton solver?

firedrakeproject / gusto

Three dimensional atmospheric dynamical core using the Gung Ho numerics.

http://firedrakeproject.org/gusto/

MIT License

11 stars 11 forks source link

Does 2x2 iterations work in the Semi-Implicit Quasi-Newton solver? #462

Closed tommbendall closed 6 months ago

tommbendall commented 10 months ago

I tried running 2x2 iterations in a mountain test case and it didn't get through a time step. I think we should investigate where this does and doesn't work!

One question is about whether the SemiImplicitQuasiNewton time stepper transports the correct fields (i.e. the fields from the start of the time step or the fields after the explict forcing step).

Another thought is whether the transporting velocity needs the semi-implicit off-centering alpha factor...

tommbendall commented 6 months ago

I had another look at this and wasn't able to replicate the failing test! In #480 we have therefore switched to using 2x2 as our default setting.