Abstract
This thesis examines issues related to the paralielization of PESTI a parameter estimation code which was originally developed by Dr. A. T. Watson. PEST is based on the trust region implementation of Levenberg-Marquardt nonlinear least square algorithm including linear inequality constraints to the estimated parameters. It can be used with different simulator programs which solve the forward problem. We first study the design issues of parallelizing an existing serial code with focus on how to solve these issues with respect to PEST. We then propose a hierarchical parallel control structure for PEST based on the manager-worker parallel programming model. We also discuss in detail the implementation of the parallel version of PEST in an Intel Paragon supercomputer. We use a two-dimensional permeability estimation problem as the example to test and demonstrate the usage of the parallel PEST. An existing simulator program called US3D, which solves the three-dimensional groudwater flow and contaminant transport equations, is employed for the forward problem.
Li, Jie (1996). Nonlinear parameter estimation in parallel computing environments. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1996 -THESIS -L52.