Open landinjm opened 2 weeks ago
@david-montiel-t @fractalsbyx @wband This issue is bordering that line where we might start making major changes on the user end. How do we want to approach this?
Personally, I don't like the serial if statements that the user would have to make (I mean not necessarily because it we just recompute equations unnecessarily) since it feels unintuitive and sloppy. I feel like a better way to do this would be for the user to call some other function and do their math in that, calling it for each nonexplicit field.
I'm not exactly sure how it would work, but I want to hear your thoughts about how users should interact with PRISMS-PF
When you have 2+ linear solves that can be solved sequentially, PRISMS-PF forces them to be nonlinear.
For example, $\Delta u=f$ as equation one and $\Delta q = u$ as the second. If solved, sequentially, both should be linear solves. However, due to the architecture of
solveIncrement.cc
and the equation dependency parser, it forces these to be nonlinear.Part of reason for this is because
computeNonexplicitRHS
is done outside of the field index loop insolveIncrement.cc
, only recomputing the RHS for nonlinear equations. A simple fix would be moving it into the field index loop and recompute for all nonexplicit equations. This would change the waynonExplicitEquationRHS
works into something more likenonExplicitEquationLHS
.It would look something like this