> On Apr 30, 2015, at 2:34 PM, "Jed Brown" <[email protected]> wrote: > > Lisandro Dalcin <[email protected]> writes: > >> Jed, I would like to make the following changes: >> >> *Add PetscErrorCode VecErrorNormWRMS(Vec U,Vec Y,PetscReal atol,Vec >> vatol,PetscReal rtol,Vec vrtol,PetscReal *norm). >> >> Any alternative name to suggest? > > Fine, but how would you deal with Shri's ts->is_diff? And Shri, why did > you implement that by creating these index sets instead of setting those > entries of vatol to +Infinity? Thanks Jed, I didn't think about that. It certainly seems cleaner to just set the entries of vatol to infinity rather than creating index sets.
> That's why we have vector weighted norms > after all. > >> * Implement TSErrorNormWRMS() in terms of VecErrorNormWRMS(). Please note that there is also a variation that uses infinity norm for error calculation. >> >> PS: These changes are preliminary ones, next I want to suggest a >> mechanism to let TS subtypes to decide how to compute the LTE error >> (or approximation to it). > > Cool; the current solution is more complicated and less flexible than > necessary, IMO. Shri
