The Exanauts Team

The Exanauts Team Webpage

Exanauts

Who Are We?

Originating from the Mathematical and Computational Science Division at Argonne National Laboratory, we are the Exanauts—a dynamic group united by our shared mission to tackle complex energy systems challenges. Our journey began within the Exascale Computing Project ExaSGD, where we harness the power of cutting-edge DOE computing resources. Driven by our enthusiasm for the Julia programming language, we embark on developing innovative solutions for today’s energy problems. Discover more about our ECP demonstration project.

Our team covers a broad spectrum of expertise, allowing us to oversee the entire software design, consisting of application, modeling, optimization methods, and linear algebra.

Our Vision

At the core of our vision is the transformative potential of applying mathematical principles through the use of flexible, high-performance numerical algorithms optimized for contemporary specialized hardware. We champion the Julia programming language and its innovative frontends for intermediate representations (such as LLVM IR and MLIR) as pivotal enablers of this transformation. Our ultimate goal? To usher in a new era of scientific discovery, including the rapid prototyping of digital twins.

Our Mission

Under the auspices of the DOE Office of Science, our endeavors are directed toward contributing significantly to a multitude of projects. Each project is carefully selected and designed to further our overarching goal: advancing DOE’s scientific simulations to address the pressing challenges associated with current energy infrastructures.

Current Research

Energy systems modeling

In response to the escalating demand for renewable energy sources and the seamless integration of electric vehicles, we develop sophisticated models of energy systems. Our work facilitates the transition to a more sustainable and efficient energy landscape.

Software Contributions

Differentiable Sparse Linear Solvers on GPUs

We develop and enhance differentiable sparse linear solvers on GPUs available in Julia. This innovation is pivotal for solving large-scale optimization problems prevalent in the energy sector with unprecedented efficiency.

Software Contributions

Nonlinear Optimization on GPUs

Tackling the omnipresent nonlinear optimization challenges within energy systems, we are at the forefront of formulating and implementing novel algorithms and software. Our focus is on harnessing GPU capabilities to revolutionize how these problems are approached and solved.

Software Contributions

ECP Demonstration

Our ECP demonstration project involved a multiperiod security-constrained optimal power flow simulation modeled entirely in Julia. We used pure Julia numerical solvers that leveraged Julia’s flexibility to run on GPU architectures and ECP systems. This included Summit at OLCF, Frontier, and Aurora. To implement our methods, we heavily relied on the Julia packages KernelAbstractions.jl, CUDA.jl, AMDGPU.jl, and oneAPI.jl.

The entire code base of our ECP demonstration Milepost7 can be found at Milepost7.jl.

Our adventure was published in SIAG/OPT Views and News and is summarized by the following highlights.

Highlights

Team Members

Ongoing Collaborations

Past Team Members

References