Parallel custom Newton solver

[dokken]

@federicobetti

apply_lifting(b, [J_form], [bcs], x0=[dvp_["n"].vector], scale=1.0)
        set_bc(b, bcs, dvp_["n"].vector, 1.0)
        b.ghostUpdate(addv=PETSc.InsertMode.INSERT_VALUES, mode=PETSc.ScatterMode.FORWARD
        )

Ghost updates is not done correctly, as for instance covered in:

Residual norm different in parallel and serial

PETScVec.scale should be done after ghost updating. I.e. I guess something along the lines of: with L.localForm() as loc_L: loc_L.set(0) A.zeroEntries() assemble_matrix(A, J_form, bcs=bcs) A.assemble() assemble_vector(L, res_form) apply_lifting(L, [J_form], [bcs], x0=[u.x.petsc_vec], alpha=-1.0) L.ghostUpdate(addv=PETSc.InsertMode.ADD_VALUES, mode=PETSc.ScatterMode.REVERSE) set_bc(L, bcs, u.x.petsc_vec, alpha=-1.0) …

I’ll update the tutorial:

bc - Custom Newton solvers — FEniCSx tutorial

to reflect this.