Some complex systems have trouble converging. By that I mean that sometimes, the system is far away from singular solutions and the discrétisation level is correct (no need for refinement) and yet, convergence is very slow. This is especially the case for Target init states.
I suspect the linear solvers in the Newton Krylov loop to not be robust enough to handle these complex cases.
[x] Lgmres seems to perform better in these cases, but it is crashing in pyodide. Raise the issue with the pyodide team.
[ ] How can I help the linear solver? How to design a good preconditionner? Where should the preconditionner act (change of variables?)? Can the preconditionner be adaptive?
Some complex systems have trouble converging. By that I mean that sometimes, the system is far away from singular solutions and the discrétisation level is correct (no need for refinement) and yet, convergence is very slow. This is especially the case for Target init states.
I suspect the linear solvers in the Newton Krylov loop to not be robust enough to handle these complex cases.
[x] Lgmres seems to perform better in these cases, but it is crashing in pyodide. Raise the issue with the pyodide team.
[ ] How can I help the linear solver? How to design a good preconditionner? Where should the preconditionner act (change of variables?)? Can the preconditionner be adaptive?