Parallel Programming Methodologies for Non-Uniform Structured Problems in Materials Science
Scott B. Baden · Defense Technical Information Center (DTIC) · 1996
The purpose of this investigation is to explore software techniques for reducing the development costs of scientific applications running on distributed memory MIMD parallel computers. Our objective is to develop a domain specific software development tool called LPAR. LPAR is targeted to non-uniform numerical methods that employ elaborate representations--such as adaptive and multilevel finite difference meshes or particles--to accurately capture irregular phenomena such as turbulence and shocks. We are using LPAR to develop new parallel software techniques and numerical algorithms for the predictive modeling of technologically important materials using first principles simulations.