Will commit on Friday if I don't hear from anyone. Thanks.
On Thu, May 24, 2012 at 4:28 PM, Paul T. Bauman <ptbauman@...> wrote:
> FEMSystem users,
> I've attached a patch that fixes a problem when numerical Jacobians are
> forced on the user side and verify_analytic_jacobians = 0.0. In particular,
> elem_jacobian got zeroed without the stored residual getting added back in
> during the side loop. A private review by Roy also picked up the case for
> when element Jacobians are computed, but side Jacobians are not. This patch
> should fix both cases. I also updated the adjoint examples to test this. In
> particular, I set verify_analytic_jacobians = 0.0 and analytic_jacobians =
> false in adjoints_ex1 and then enabled verify_analtyic_jacobians in
> adjoints_ex2 to still exercise that capability. However, none of the
> examples currently exercise the case where element Jacobians are
> analytically computed while the side Jacobians are not.
> Feedback before committing?