Hello,
I would like to solve a nonlinear PDE together with an additional scalar constraint (involving some integral of the PDE solution).
I have a working serial code, where I assemble a block matrix and use a “hand-coded” Newton method (iteratively solving the linearized problem with something like x=A^-1*rhs).
Now I would like to go parallel. I see two methods and, in case both are possible, I am wondering which one would be easier and/or more efficient:
- use a “hand-coded” Newton method (with iterations of the linear solver
KSPSolve, somewhat similar to the constrained but linear example in laplace-lagrange-PETSc.edp), - use directly the nonlinear solver
SNESSolve(similar to the nonlinear but non-constrained example in navier-stokes-2d-PETSc.edp).
Is method 1 much worse than method 2?
For method 2, I have some trouble defining the arguments to be given to SNESSolve (the equivalent of J, funcJ and funcRes in navier-stokes-2d-PETSc.edp). Specifically, how should one properly resize, change numbering, and assemble/extract the blocks?
Is there by chance a working example somewhere?
Thank you in advance for your help!