[petsc-dev] perhaps where the custom adapativity perversity began

Barry Smith bsmith at mcs.anl.gov
Sat Apr 29 22:57:08 CDT 2017

> On Apr 29, 2017, at 10:49 PM, Jed Brown <jed at jedbrown.org> wrote:
> Barry Smith <bsmith at mcs.anl.gov> writes:
>>    ierr = PetscOptionsBool("-ts_theta_adapt","Use time-step adaptivity with the Theta method","",th->adapt,&th->adapt,NULL);CHKERRQ(ierr);
> https://bitbucket.org/petsc/petsc/commits/3b1890cda3a41f3c81730c41d1c1f55cb4d78b93
>  Added time-step adaptivity for the theta method. It's currently
>  switched off by default till the error estimation is fixed.
> And this is why workarounds are so much worse than fixing the root of the problem.

 But why does theta need its own flag for controlling adaptivity when TS also has a flag; new types are usually/always made by copying old implementations and then modifying them, thus bad code gets propagated to new methods. Why could this workaround just use the general adapt instead of adding its own option?

>>   Really, even a method so simple minded I can understand it, requires its own flags for adaptivity?
>>   Based on this model we'll have -ksp_cg_max_its and -ksp_gmres_max_its pretty soon and where is -ts_arkimex_max_steps, we'd better add it soon.

More information about the petsc-dev mailing list