Thanks. Note that us removing the “simple minded” AD stuff from PETSc doesn’t mean we think AD is a bad idea. It can be an effective way to generate Jacobians.
Barry On Oct 1, 2014, at 4:09 PM, Gaetan Kenway <[email protected]> wrote: > HI Xiangdong > > We have use the Tapenade AD package to compute sparse matrix jacobians for > CFD problems. We use the result from the AD to populate a PETSc sparse matrix > and then use PETSc to solve the resulting linear system. > > Hope that helps. > > Gaetan > > On Wed, Oct 1, 2014 at 4:23 PM, Barry Smith <[email protected]> wrote: > > On Oct 1, 2014, at 3:16 PM, Jed Brown <[email protected]> wrote: > > > Xiangdong <[email protected]> writes: > > > >> Is there any reason that AD is not supported in PETSc? Is it because the > >> ADIFOR/ADIC library itself? > > > > Those libraries are not maintained and have license problems. You're > > welcome to use an AD tool yourself. The support in PETSc was mostly > > just making simple examples cleaner; it wasn't widely used in real > > applications. > > Wasn’t used at all in anything but trivial codes. > > > >
