Dr. Paul Hovland von der Mathematics and Computer Science Division des Argonne National Laboratory wird am 15. Juni im Hoersaal LU in der Wuellnerstrasse die folgenden zwei Vortraege ueber die PETSC and TAO Toolkits for Scientific Computing (um 10.30) und Component Architectures for Scientific Computing (um 14.00) halten. Hierzu moechte ich alle Interessierten einladen.
Beste Gruesse, - C.B.
-------------------------------------------------------------------------= ---------------------
The talk at 10.30
PETSc and TAO: Toolkits for Scientific Computing
We provide an overview of two object-oriented toolkits for scientific computing, PETSc and TAO. The Portable, Extensible Toolkit for Scientific Computing (PETSc) is oriented toward the solution of PDEs and
provides nonlinear and linear solver objects, as well as fundamental datatypes such as vectors and (sparse) matrices. The Toolkit for Advanced Optimization (TAO) uses some of the PETSc solvers and data objects to solve optimization problems. We discuss the design of both toolkits, as well as mechanisms for extending them to, for example, incorporate a domain-specific preconditioner. We also present current research and development, including integration with automatic differentiation technology.
-------------------------------------------------------------------------= -----
The Talk at 14.00
Component Architectures for High-performance Scientific Computing
We present the motivation for and an overview of the current design of the Common Component Architecture (CCA). The CCA is a component model oriented toward high-performance scientific computing being developed by
researchers from several U.S. universities and national labs. The CCA is motivated by the fact that advanced computational science applications require a suite of numerical and computational tools, including mesh management, discretization, linear solvers, nonlinear solvers, optimization, eigensolvers, parallel I/O, visualization, and computational steering. Moreover, the physical models utilized by these applications are extremely complex, and often draw upon the expertise of scientists and engineers from a range of disciplines. The CCA seeks to address this complexity, providing the infrastructure required for the development of high-performance (parallel) applications using component-based software engineering methodologies . Among other benefits, this approach enables experts in a particular domain to focus on small components of a complex system, thus helping to ensure that these components encapsulate the state of the art in that domain. Furthermore, the dynamic composability offered by component-based approaches greatly facilitates algorithmic experimentation, allowing developers to defer decisions about numerical methods until runtime and to experiment with poly-algorithmic methods.