Title: Performance of Diffuse Interface Models Using a High-Level PDE Solver      

Speaker: Daniel Wheeler, University of Maryland/NIST

Date/Time: Monday, May 19, 2008, 1:00 – 2:00 pm

Location: CSRI Building/Room 90

Brief Abstract: This presentation will describe an object-oriented partial differential equation (PDE) solver, written in the Python scripting language, based on a standard finite volume (FV) approach (http://www.ctcms.nist.gov/fipy/). Numerous PDE solvers exist using a variety of languages and numerical approaches. Many are proprietary, expensive, and difficult to customize. They are generally not tailored to the specific needs of the materials researcher.  As a result, materials scientists spend considerable   resources repeatedly developing limited tools for specific problems.  FiPy's approach combines the FV method and Python, providing a tool that is extensible, powerful, and freely available.  FiPy allows materials scientists to quickly develop and solve models without specialized knowledge of numerical methods.
 
FiPy has already proven to be a very useful tool in our research providing a number of benefits over our previous approaches.  However, there are a number of persistent issues that remain outstanding. These issues mainly concern the inclusion of existing additional high performance numerical methods into the general FiPy framework.  This requires integrating freely available lower-level numerical libraries into Python and FiPy.  In this presentation, I will discuss performance data with reference to the lower-level libraries, such as Sandia's Trilinos, that are already partially incorporated into FiPy. The question of how to gain optimal benefit from these libraries is an important concern, particularly the packages in Trilinos, given the various constraints of Python and FiPy.  We will present a diffuse interface model of reactive wetting, an industrially relevant and computationally intensive phase transformation process, as a case study for the performance data.

CSRI POC: William Spotz, (505) 844-4241