Covering everything from wind tunnels and cardiovascular electrodes to the futuristic world of exascale computing, and with a few fantastic beasts thrown in for good measure, the common denominator in the career of LLNL computer scientist and math programmer Brian Gunney has been finding solutions for unsolvable problems.
As a kid who spent a lot of time around planes, Gunney gravitated towards all things aerospace. He would go on to receive his PhD in Aerospace Engineering and Scientific Computing from the University of Michigan in Ann Arbor and work at some of America’s leading research institutions. One summer as an undergrad, he designed the piping layout for NASA Ames Research Center’s high pressure air storage facility, which powered the center’s high speed blow-down wind tunnel. Among the challenges were thermal stress (temperature drops 100 degrees F), foundation settlement, earthquake survivability, and the safety of nearby buildings in case of a catastrophic failure. Gunney developed a promising design (slinging the pipes under the enormous tanks), but no one in his division could do the stress analysis on it. No analytical methods could do the job.
“We ended up asking the ANSYS guy down the hall, and he was able to simulate it for us,” says Gunney. “And the design was accepted. When I worked in the defense industry, I saw that computational fluid dynamics could do the same thing for aerospace engineering problems, that’s when I realized what computing meant for ‘unsolvable’ problems.”
As a PhD student supported by the DOE Computational Science Graduate Fellowship, Gunney was able to get more involved with algorithms and high performance computing. He did his postdoc at the Institute for Mathematics and its Applications at the University of Minnesota and worked for 2.5 years in the medical device industry, developing numerical methods for analyzing cardiac leads. In 2000, looking to do more leading-edge scientific computation at a premier research center, he joined LLNL.
In more than two decades here, Gunney has worked with several generations of LLNL supercomputers (including Sequoia and Sierra) and is currently working on computational geometry capabilities in the Axom and Ares codes in preparation for the arrival of the Lab’s first Exascale computer, El Capitan. Part of the Exascale Computing Project, Axom is a tool set for the simulation codes used by many researchers across the Lab who require robust, high performance, mid-level functionalities. Gunney works on a code that finds which parts of the computational mesh a particle in the flow is closest to, and one code that reconstructs material interfaces as they evolve during the simulation.
Gunney’s work on these codes follows many years spent working on the SAMRAI mesh refinement project, which began in 1996 as part of the Advanced Strategic Computing Initiative. SAMRAI, which stands for Structured Adaptive Mesh Refinement Application Infrastructure, is a framework for large, dynamically adaptive, multiscale, and multiphysics simulations run by the Lab’s massively parallel computers.
“The physics problems we’re interested in don’t have easy solutions, so we use meshes to divide space into small pieces where we can make solvable approximations. The smaller the pieces, the more accurate the approximations. So, we have billions of small problems to solve. That’s where big computers come in. The adaptive meshes that help us put that computing power where we need it most are so large and complex that even their descriptions were too big to process,” explains Gunney. “We distributed that, which means no processor ever has a full picture of what the entire mesh looks like. We came up with new algorithms that did the job efficiently even with the incomplete picture.”
Gunney knows the codes he works on have to be big and complicated because so are the problems they’re designed to solve. The challenge of this work still excites him most about working at the Lab. Each generation of supercomputers has given Gunney and his fellow computer scientists new potential and forced them to come up with creative solutions and new algorithms to realize that potential.
“Ten years ago, the simulations we did to support the NIF ignition shot would have been hard to even imagine,” states Gunney. “The increasing power of the computers lets us tackle bigger and bigger problems, so it’s fun to think about which problems that are ‘unsolvable’ today will be solved in another ten years.”
Having worked in a variety of fields with researchers who took different approaches and had schools of thought for solving problems, Gunney is grateful to be here at the Lab working among his fellow computational scientists and computational physicists. “It makes me better at what I do, and it’s nice to have management that not only understands the technical problems but also the people working on solving them,” says Gunney.
In his spare time, Gunney is a published author whose fantasy book about dragons qualified him as a finalist in the American Fiction Awards. “It started as a story I made up for my daughter when she was very young, but I’m an aerospace engineer,” Gunney notes. “I’ve always wanted to see a good fantasy story that doesn’t offend my scientific sensibilities. So, I calculated what it would take for a dragon to fly: its weight, wing size, speed, how much it could carry, and how tired it would feel.”
When Gunney is not presenting his conclusions on the feasibility of the dragons from Game of Thrones at local public libraries or navigating the invisible topography created by the advanced simulation codes on the Lab’s supercomputers, you may find him traversing the Sierra Nevada backcountry passes and canyons as a backpacking guide for the Sierra Club of San Francisco.