On Mon, Jul 30, 2012 at 5:40 PM, Todd Munson <tmunson@mcs.anl.gov> wrote:

> 1. change SNESVI to support general complementarity constraints because only doing box constraints on state variables is lame

Okay; and how do you want to do this? You need to write out the corresponding KT system
and then you just have a box constrained problem. One can make it easier to write the
KT system, but then you have to precondition the KT system.

Yes, but if the user created the augmented system, they have to inform the linear solver of that structure. I think it is much more convenient for the user to not have to manage those things.

There are some reformulations for polyhedral constraints, but they are, in my opinion,
a bit unwieldy.

Why unwieldy?

Note that changing the size of the constrained size is also important, and box constraints strike me as even more confusing in that context.

> 2. have a mode to solve the system in the full space, instead of eliminating the semi-smooth variables (which produces the bad conditioning)

There are all these options in the TAO methods.