This talk describes a heuristic method for mixed integer programming
(MIP), and its incorporation into PICO, a highly parallel
 branch-and-bound MIP solver.  The heuristic performs pivot-based
 rounding with the help of a concave integrality merit function.  The
 surrounding branch-and-bound environment begins with a synchronous
 "ramp-up" phase, and then transitions to asynchronous tree-parallel
 search.  During ramp-up, there are several ways of inducing
 parallelism: using multiple alternative merit functions, and
 partitioning the search space by prospective branching.  I discuss how
 to combine these two approaches and how to prospectively branch on
 multiple layers of variables, along with implementation and
 computational results for the PICO MIP solver.  The talk will also
 include basic information on the design and construction of PICO.

 Collaborators on the PICO MIP and MIP heuristic include: Cynthia
 Phillips (Sandia National Labs), William E. Hart (Sandia National
 Labs), Mikhail Nediak (Queen's University), and Konrad Borys (Rutgers
 University).