Applied Math Seminar I Derek Steinmoeller, pyblitzdg: An Open-Source Python 3 Framework for Rapid Numerical Model Development

Microsoft Teams 

Speaker

Derek Steinmoeller | Aquanty

Title

pyblitzdg: An Open-Source Python 3 Framework for Rapid Numerical Model Development

Abstract

In this talk we introduce pyblitzdg, a free and open-source Python 3 module for numerical model development and unveil some of the power that it puts into the hands of the mathematical modeller. pyblitzdg is built off a robust and well-tested C++ code-base for the (discontinuous) Galerkin finite element method, which incorporates the blitz++ tensor arithmetic library. Pyblitzdg excels at carrying out fast simulations of wave dynamics in arbitrary geometries, but also includes support for heat- and diffusion-based problems. Object-oriented programming is not required to use pyblitzdg, and simple procedural-style simulation programs can usually be written in a single ~100 line Python 3 script. The syntax used relies on the NumPy library and would look familiar to users of wide-spread mathematical software like Matlab or GNU Octave. Worked examples that may be of interest to an applied math audience will be shown alongside scientific results visualized in ParaView. Extensions available to the end-user of pyblitzdg, such as CUDA acceleration (e.g., cupy or pycuda) and parallel cluster computing (mpi4py) will be discussed as well.