Two Methods for Load Balanced Distributed Adaptive Integration

Parallel subregion-adaptive integration often exhibits sequential behavior when applied to functions with diicult local problems such as boundary singularities. To correct this, one needs to balance the loads of the processors by providing all of them with a reasonably diicult part of the problem. In this paper, we present two distributed adaptive integration methods: one based on a global heap priority queue (distributed over the processors) which remains load balanced and assures that the subregions are processed according to their (global) priority (size of error) and one based on a local heap priority queue enhanced by receiver-initiated load balancing. We report results of both methods.