PARALLEL, ASYNCHRONOUS GLOBAL OPTIMIZATION TECHNIQUES FOR MEDIUM AND LARGE INVERSION PROBLEMS

WE HAVE BEEN DEVELOPING CODES FOR SEISMIC RAY TRACING AND TRAVEL TIME INVERSION IN THREE-DIMENSIONAL COMPLEX GEOLOGICAL MEDIA THAT REQUIRE SUPERCOMPUTERS FOR PRACTICAL APPLICATIONS. THIS INVERSION PROCEDURE, AS MOST OTHERS IN THE LITERATURE, SUFFERS FROM THE BUMPINESS OF THE RESPONSE SURFACE AND THUS GETS TRAPPED AT LOCAL SOLUTIONS. UNDER THEPROPOSED RESEARCH PROJECT WE WILL EXTEND THE INVERSION STRATEGIES TO INCLUDE COARSE GRAIN PARALLEL TECHNIQUES FOR ESCAPING FROM LOCAL SOLUTIONS. LOCAL OPTIMIZATION TECHNIQUES, SUCH AS NEWTON'S METHOD AND ITS VARIANTS, CAN GET TRAPPED AT LOCAL MINIMIZERS. GLOBAL OPTIMIZATION TECHNIQUES CAN BE USED TO OVERCOME THIS PROBLEMBUT ARE USUALLY RESTRICTED TO SMALL DIMENSIONAL PROBLEMS. THE PURPOSE OF THIS RESEARCH IS TO DEVELOP OPTIMIZATION ALGORITHMS THAT CAN ESCAPE FROM LOCAL MINIMIZERS AND WHICH CAN BE EFFECTIVELY APPLIED TO MEDIUM AND LARGE DIMENSIONAL PROBLEMS. THE ALGORITHMS WILL MAKE EFFICIENT USE OF COARSE GRAIN PARALLELISM AND WILL BE ASYNCHRONOUS. THE PROPOSED TECHNIQUES WILL BE DESIGNED IN RELATIONSHIP WITH MULTIOBJECTIVE OPTIMIZATION AND SEISMIC INVERSION RESEARCH CURRENTLY UNDER DEVELOPMENT.