A Common Computational Infrastructure for Adaptive Algorithms for PDE Solutions

J.C. Browne, C. Edwards, R. van de Geijn, K. Klimkowski, M. Parashar and J. Singer
Department of Computer Sciences & TICAM
University of Texas at Austin
Austin, TX 78712

Technical Paper Submission (Postscript)

Corresponding & Presenting Author:

Abstract

This paper defines, describes and illustrates the use of an infrastructure for implementation of parallel adaptive algorithms for solution of sets of partial differential equations. The infrastructure is separated into two layers of abstraction: a layer which defines an implements a scalable dynamic distributed array (SDDA) and a layer in which several dynamic data abstractions are implemented in terms of the SDDA. The currently implemented data structures are: those needed for formulation of hp-adaptive finite element methods, those needed for hierarchical adaptive finite element methods and those needed for the fast multipole method for solution of linear systems. Implementation issues and performance measurements are given. The applications from which the data structure requirements were derived are described.

• Introduction
• Implementation Design for the SDDA
• Method Specific Data Abstractions
• Example Application Codes
• Conclusion and Future Work
• References
• About this document ...