Speaker: Carol S. Woodward

Center for Applied Scientific Computing

Lawrence Livermore National Laboratory

Title: 

Applications of Anderson Accelerated Fixed Methods for DOE Applications

Time / Date:   4:00 - 5:00 PM Monday, March 7

Location: Blocker 220


Abstract: 

Developing efficient and accurate simulations of complex physical systems present numerous challenges.  Numerical methods, software development, and computer science must come together with the driving science or engineering application in order to produce a truly usable simulation tool.  In this presentation, I will discuss the mission of the US Department of Energy and overview a typical mathematician's job at a DOE laboratory.  I will then discuss work in subsurface flow and dislocation dynamics where we have applied Anderson accelerated fixed point methods as the nonlinear solver finishing with some comments on taking this method to a GPU-based machine.

Slides of talk




Speaker: Tim Davis

Department of Computer Science and Engineering, TAMU

Title: 

Mathematical Software Development in Academic, Industry and Government Labs

Time / Date:   4:00 - 5:00 PM Monday, February 22

Location: Blocker 220


Abstract: 

Why would an academic write high-quality software?  Why not just publish the paper, post a rough prototype (if anything at all), and let someone else do the rest?  There are many facets to the answers to these questions: (1) Academic Tradition: Dating back to Wilkinson and Forsythe, creating and publishing high-quality peer-reviewed software has long been a highly-regarded hallmark of scholarly work in applied mathematics.  (2) Impact:  rather than let your method sit on a shelf, you can see it in action in 100's of applications in academia, government labs, and industry.  (3) Rigor:  a "theorem" without a proof is just a conjecture.  Likewise, it's not a real algorithm unless it is embodied in a working code.  (4) Reproducible research: an increasingly important aspect of academic research is giving others the ability to replicate your results;  publishing the software behind your research is one important aspect of this.

The speaker will relate his insights and experience in creating algorithms and writing software, in academia, and in collaboration with government labs and industry.  His solvers are used in a wide range of applications, including x=A\b in MATLAB when A is sparse, Google StreetView, and a vast range of other applications in academia, government labs, and industry.

Slides of talk





Speaker: Matthew Farthing

Computational Hydraulics Laboratory

US Army Engineer Research Development Center

Title: 

Computational Modeling for Coastal and Hydraulic Applications in the
Corps of Engineers

Time / Date:   4:00 - 5:00 PM Monday, February 8

Location: Blocker 220


Abstract: 

Computational models are increasingly important for achieving the U.S. Army  Corps of Engineer's mission of delivering vital public and military engineering services. The problems we are asked to solve often involve many coupled physical processes that may exhibit strong variation at a range of spatial and temporal scales. Such multiphysics and multiscale problems rarely have analytical solutions and typically require sophisticated, model-specific numerical methods, which may in practice be developed in tandem with the mathematical models.

In this talk, we will review some of these problems and the development of computational tools for supporting research on new models for coastal and hydraulic processes and improvements in numerical methods. The range of models of interest to the Corps includes multiphase flow in porous media, shallow water flow, turbulent free surface flow, as well as flow-driven processes like species transport.

Here, the focus will be on stabilized finite element methods for free-surface flows. We will look at challenges associated with both building accurate, high-fidelity approximations as well as lower fidelity or reduced models for applications involving risk assessment or ensemble forecasting that can require thousands of forward simulations. We will discuss issues that arise in the formulation of the schemes themselves as well as computational performance and software design.

Slides of talk