ZERO DUALITY GAP FOR A LARGE CLASS OF SEPARABLE CONVEX PROBLEMS

Geometric and l_p-norm optimization are two classes of convex problems that
share a nice duality property : the duality gap between a feasible primal and
a feasible dual problem is always equal to zero.

We stress here the fact that this result holds even when these problems do not
admit Slater points, a condition that is known to be required in the general
convex case.

This result is known since the late sixties. However, its proof has been
recently revisited and simplified using the framework of conic optimization.
In this talk, we show how it is possible to generalize this approach to a large
class of separable convex problems.