Hayes 9504071 The object of this proposal is to test innovative variations of the IRAM approach so as to extend the range of systems that can be studied efficiently and accurately using the Miller's cumulative reaction probability method. Preliminary tests of these variations in the Miller approach, which is based on the use of absorbing boundary conditions, have demonstrated that it is possible to obtain accurate eigenvalues for the cumulative reaction probability matrix without assembling the full, dense matrix. As a result, the basic algorithm is both scalable and very efficient in an MPP environment because standard sparse-matrix concepts can be used to assemble the matrix-vector products required to execute the preconditioned IRAM approach.
