Absolute step tolerance is not passed to petsc nonlinear solver

[GiudGiud]

I think

in newton_solver.C you see it being used by that solver. Upstream you can see the parameter being set in the system, and then passed to solver.

but in the petsc non linear solver class it appears unused unlike the other tolerance parameters, that are used below for example:

  // Set the tolerances for the non-linear solver.
  LibmeshPetscCall(SNESSetTolerances(_snes,
                                     this->absolute_residual_tolerance,
                                     this->relative_residual_tolerance,
                                     this->relative_step_tolerance,
                                     this->max_nonlinear_iterations,
                                     this->max_function_evaluations));

[roystgnr]

I think you're correct. I must have added absolute_step_tolerance for NewtonSolver, pulled it up to NonlinearSolver at some point since it was applicable to any iterative nonlinear solver, but never figured out (or never bothered to figure out) how to get it into PETSc. Does PETSc support an absolute_step_tolerance? It may be that the best we can do is emit a warning or an error when the default has been overridden and we're in a PETSc subclass.

[GiudGiud]

I dont see an option in PETSc, see https://petsc.org/main/manualpages/SNES/SNESSetTolerances.

I ll make a PR adding a warning

[roystgnr]

Should we close this? Unless PETSc adds an option I don't think there's much we can do beyond #4339.

[GiudGiud]

works for me