Research Homepage of Erik Schnetter

Welcome!

I am an assistant research professor at the Center for Computation & Technology and in the Department of Physics & Astronomy at Louisiana State University in Baton Rouge.

My scientific interests lie in high performance computing (HPC), in harnessing the computing power of current (and future) HPC systems and making it available to end users and programmers, so that these systems can be applied towards solving scientific and engineering problems. Such systems are notoriously difficult to use, and their architecture and programming models change as hardware advances and becomes more powerful.

Using HPC systems requires not only correctly and efficiently implemented compute kernels, but requires also tools to build (complex) applications around these kernels. I am in particular interested in
   - efficient computational infrastructure for HPC architectures,
   - tools to ensure that the modules comprising an application interact correctly, and
   - interacting with and educating end users about high performance computing. The latter is especially important since it is the end users who choose run-time parameters for HPC applications, and correctness and efficiency of real-world applications thus depends on these end users' choices.

I use software frameworks as vehicle to implement ideas, test them in realistic environments, and ensure they work together. Frameworks allow application scientists to create large, complex multi-physics applications by coupling independently developed modules. It is important to find abstractions which lead to an overall modular structure while permitting efficient couplings between modules, and to have clear boundaries between application science parts and high performance computing parts.

In addition to the above, I have a long-standing interest in relativistic astrophysics, and I maintain close collaborations with researchers at Caltech and the Albert-Einstein-Institut in Germany. In these collaborations we study compact objects such as black holes or neutron stars and their interactions.


Projects and Collaborations

Carpet is and mesh refinement infrastructure for the Cactus framework. Carpet supports adaptive mesh refinement (AMR), multiple grid patches, is parallelised using MPI, and runs on most existing computer architectures. It supports hybrid parallelisation combining MPI and OpenMP in support of the modern multi-core architectures. Carpet is mature and scales to more than 10k processors.

The Alpaca project develops high-level tools that help develop and maintain large, complex applications. These tools will help both developers and end-users validate the correctness of an application, and aid them in improving its performance. Alpaca is based on the Cactus framework, and these tools are components themselves, built into the application and interacting with it, different from traditional debuggers and profilers which examine a programme only from the outside.

Kranc is an automated code generation system that creates complete Cactus modules from Mathematica equations. It expands of tensorial expressions, discretises derivatives with high-order finite differences, and generates the code. Automated code generation is in a certain sense the equivalent to using libraries of efficient solvers, since no such libraries exist for explicit, stencil-based codes.

XiRel is developing a highly scalable, efficient and accurate adaptive mesh refinement layer for the Cactus Framework, based on the Carpet driver, and optimized and supported for numerical relativitists studying the physics of black holes, neutron stars, and gravitational waves.
In a collaboration with the company Decisive Analytics Corporation (DAC) and the NASA Goddard Space Flight Center (GSFC), we are developing ParCa, a new Cactus AMR infrastructure based on the Paramesh library. This project will also create a general relativistic magneto-hydrodynamics (GRMHD) code.

The Einstein Toolkit is a collection of Cactus thorns for numerical relativity. The Cactus group at AEI and CCT maintains a set of core thorns (the CactusEinstein arrangement) ensuring interoperability. Many people and groups worldwide have contributed to the Einstein Toolkit over the years, which contains today a set of high-quality evolution methods, initial data solvers, apparent and event horizon finders, and wave extraction methods.


Recent Publications
Recent Talks