Start Date: 12/01/2005
End Date: 11/01/2008

The proposed work addresses the theory and practice of terascale parallel computing for large scale simulations involving nonlinear optimization or solution of nonlinear systems of equations. Specifically, the research problems to be addressed are: (1) theoretically rigorous decompositions and algorithms for multidisciplinary design optimization problems; (2) a scalable parallel implementation of the DIRECT algorithm for global optimization; (3) a homotopy algorithm for large scale polynomial systems of equations that exploits the general linear product structure theory for polynomial systems; (4) a linear version of Shepard's algorithm for constructing surrogates for sparse data in high dimensions; (5) production quality parallel mathematical software from (2)-(4) tuned to DOE terascale supercomputers. Because of the presence of the terascale System X, Virginia Tech is uniquely positioned to develop parallel mathematical software that is directly transferable to DOE's leadership class machines. The common theme in the apparently diverse topics (1)-(4) is parallel mathematical software based on rigorous mathematics.

Grant Institution: Department of Energy

Amount: $346,255

People associated with this grant:

Layne Watson