Texas A&M University
After a rough start that included dropping out of high school at 15, Richard Vega was doing well in the mid-2000s. He’d learned carpentry and had his own business in the Miami area.
Then the Great Recession hit. “People weren’t eager to spend a lot of money on custom furniture,” Vega says. The business “fell apart and I figured it would be a good time to go back to school.”
It was the start of an academic career that led Vega to his high school diploma, top grades at a community college and bachelor’s degrees in physics and nuclear engineering at Texas A&M University. “I picked the hardest-sounding thing on the menu, which at the time, to me, was nuclear engineering” – even though he knew nothing about the subject.
Vega’s quest for challenges has been rewarded. He earned a year-round internship at Sandia National Laboratories in New Mexico as an undergraduate and later won a Department of Energy National Nuclear Security Administration Stewardship Science Graduate Fellowship (DOE NNSA SSGF). Vega is still at Sandia, pursuing his doctorate (also from Texas A&M) remotely and improving complex computer simulations of subatomic particle movement.
Vega’s research, under advisor Marvin Adams, seeks new, innovative ways to solve the Boltzmann transport equation, nuclear engineering’s cornerstone. It determines the distribution of neutrons or other subatomic particles in a radiation environment. The transport equation is key to calculating nuclear reaction rates and other factors governing nuclear power and weaponry.
In such simulations, algorithms typically divide the physical space being modeled into a mesh of cells. A different computer processor then calculates the physical processes in each cell. This massively parallel approach cuts the time needed to reach a solution.
Computational scientists have different ways to do this division, called discretization. Vega focuses on the slice balance approach (SBA), which extends one-dimensional discretization to a more complex three-dimensional configuration. The technique has a 20-year history, but it’s impractical for the linear discontinuous finite element discretization scheme Vega wanted to implement. As the algorithm continuously sweeps through the domain it models, calculating particle flux and other properties, the resulting data exceed storage capacity. That means every property must be recalculated with each of thousands of sweeps as the simulation progresses. “Every time you did a sweep, 90 percent of your time was spent recalculating stuff you just calculated a minute ago.”
Vega recognized that graphics processing units (GPUs) were suited to accelerate the calculations. GPUs were first developed to repeatedly redraw pixels in video games. They have thousands of computing cores, each slower than the typical central processing unit, but are capable of quickly performing identical operations on different data. The latest high-performance computing systems often pair general-purpose GPUs with standard CPUs to accelerate calculations while limiting power consumption. The downside: GPUs require extra programming skill.
Vega implemented the SBA with GPUs, reducing time spent on recalculation in each sweep from the previous 90 percent to around 5 percent. He also recognized that the SBA often dealt poorly with discontinuities – domain irregularities that generate large changes in conditions and require added computation. Vega created sub-slices that allow discontinuities to propagate into downstream cells, improving the simulation’s accuracy.
Since the extended SBA, as Vega named his technique, is designed to quickly and accurately solve the Boltzmann equation, it could be used to calculate particle transport in nuclear reactors, plasma physics and more. For his Sandia work, the technique could improve calculations in NuGET (neutron gamma energy transport), which models how materials respond to high-radiation environments.
In his 2016 Lawrence Livermore National Laboratory practicum, Vega worked with computation specialist Thomas Brunner on another Boltzmann equation approach: the implicit Monte Carlo method. This time the target was calculating heat transfer through radiation in high energy density plasmas. “Monte Carlo” means the simulated particles move and interact randomly.
Vega tweaked the algorithm to cope with higher-order meshes – ones with irregular, twisted cells – by making particles appear to move through simple, regularly shaped cells. He’s continuing to work with the Livermore group on making the code truly implicit – capable of solving an equation involving both the system’s present state and its later state.
Vega remains at Sandia, an arrangement he says wouldn’t have been possible without flexibility from the DOE NNSA SSGF steering committee. Its accommodation “has improved my life dramatically,” he says. Vega’s wife now works at Sandia and they’ve bought a house near the lab, where he hopes they remain after graduation in spring 2018.
Image caption: To illustrate the extended slice balance approach’s ability to handle complex geometries, Richard Vega used a photo of his dog to create a meshed domain with a radiation volumetric source in the animal’s shape. Credit: Richard Vega.