ACADEMIA
New Wave Supercomputers Catch Big Waves
ALBUQUERQUE, NM -- The new wave in computing - super-fast machines churning out three-dimensional models viewable in high-tech, immersive theaters - may teach us more about the big waves that sometimes threaten people who live near the seashore. Although earthquakes cause most of these giant waves, called tsunamis, researchers at the National Nuclear Security Administration's Los Alamos National Laboratory recently completed the largest and most accurate simulation of tsunamis caused by asteroids. They presented the first data from that model today to the American Astronomical Society meeting in Albuquerque, The scientists aren't working on a sequel to the Hollywood blockbusters Deep Impact or Armageddon. They reason that since a large percentage of the world's population lives on islands, bays or coastlines, a better model could help predict how tsunamis behave, aiding emergency responders. Most tsunamis often result when earthquakes send huge landslides tumbling into bays or oceans. Recent studies of a 30-foot-high tsunami that killed more than 2,100 people on Papua New Guinea in July 1998 showed the cause was an underwater landslide more than 2,000 miles away. A landslide in Lituya Bay, Alaska, in July 1958 inundated the shore of Gilbert Inlet nearly a third of a mile above the high tide line, and its monster wave is the largest ever documented. Computer scientists Galen Gisler and Bob Weaver from the Los Alamos' Thermonuclear Applications Group, and Michael Gittings of Science Applications International Corp., created simulations of six different asteroid scenarios, varying the size and composition of a space visitor hitting a three-mile-deep patch of ocean at a speed of 45,000 miles an hour. The Big Kahuna in their model was an iron asteroid one kilometer in diameter; they also looked at half-sized, or 500-meter, and quarter-sized variants, and at asteroids made of stone, roughly 40 percent less dense than iron. "We found that the one-kilometer iron asteroid struck with an impact equal to about 1.5 trillion tons of TNT, and produced a jet of water more than 12 miles high," Gisler said. The team's effort builds on the pioneering research of Los Alamos' Chuck Mader and Dave Crawford of Sandia National Laboratories. More accurate models of tsunami behavior are now possible, thanks to recent improvements in high-performance computers and the codes that run on them funded by the NNSA's Advanced Simulation and Computing program. "Although this is important science and has potential value in predicting and planning emergency response, it's an great way to test and improve the code," Gisler said. "We can do the problem better now by simulating an entire tsunami event from beginning to end and bringing more computing power to bear on some of the key variables." The code, called SAGE for SAIC's Adaptive Grid Eulerian, was developed by Los Alamos and SAIC. A majority of large simulations come in one of two flavors: Lagrange, in which a grid or mesh of mathematical points matches with and follows molecules or other physical variables through space; or Eulerian, in which the mesh is fixed in space, thereby permitting researchers to follow fluids as they move from point to point. SAGE's power lies in its flexibility. Scientists can continuously refine the mesh and increase the level of detail the code provides about specific physical elements in the mesh. The new Los Alamos simulation uses realistic equations to represent the atmosphere, seawater and ocean crust. To follow a tsunami from the point of splashdown to a city like Honolulu or Long Beach, Gisler and his colleagues needed to model in great detail the interactions between air and water and between water and the surface of an asteroid. Then they followed how the shock waves moved through the ocean and the seabed below and how water waves propagated through the water. "We looked in some detail at a couple of the key variables, especially the heights of tsunamis as a function of their distance from the point of impact; we modeled the heights of individual waves and studied how densely spaced they would be at various distances," Gisler explained. When the enormous simulation was done - more than a million hours of individual processor time, or three weeks on Los Alamos' Blue Mountain supercomputer and the ASCI White machine at Lawrence Livermore National Laboratory - the team found they had some good news and some bad news for coastal dwellers. "The waves are nearly double the height predicted in the earlier simulation, that's the bad news, but they take about 25 percent longer to get to you, which could help more people get to higher ground if they had some warning," Gisler said. The model predicts that wave velocities for the largest asteroid will be roughly 380 miles an hour, while the older model calculated their speed at close to 500 miles an hour. However, the initial tsunami waves are more than half a mile high, abating to about two-thirds of that height 40 miles in all directions from the point of impact. The earlier model of asteroid-caused tsunamis actually was a patchwork of three different computer codes, Gisler said. The first code simulated the big splash and formation of the cavity, the second depicted how the water collapsed to create the tsunami and a final code followed the tsunami wave through the ocean. "With the SAGE code, we were able to avoid a series of potential mistakes that happen when the model doesn't understand the conditions that you're passing on from each separate code," Gisler said. In addition to learning more about how wave height and density vary with distance from the asteroid impact, the Los Alamos team also improved the way the computer model represents the strength of materials, which can be applied to other codes with industrial, defense and scientific applications. As the asteroid strikes the water, its overall density decreases rapidly. One challenge for the team was to model accurately how acoustic waves propagate through the asteroid as it vaporizes. Initially, that problem appeared insurmountable because both the earlier codes and SAGE showed the acoustic waves -moving at physically impossible speeds through the highly mixed materials. By adjusting how the cells in the mesh represent those rapidly changing materials, the team was able to model the acoustic waves accurately. Gisler said the team produced both two-dimensional and three-dimensional versions of the SAGE tsunami code. The 3-D code required more than 200 million separate cells and ran for three weeks on one-eighth of ASCI White. Clever code writing and the enormous computational power in the 3.1 teraOPS Blue Mountain and 12.1 teraOPS ASCI White weren't the only crucial factors in building the model. "It's not all about better and better resolution," Gisler said. "You must have good visualization techniques, such as the three-dimensional power walls we use at Los Alamos, if you're going to make sense of the data from these huge calculations." The modeling continues. Gisler, Weaver and Gittings next plan to study in three dimensions how an asteroid-induced tsunami will behave if the space rock strikes a glancing blow, 30 degrees from the horizontal, instead of the 45- and 90-degree angles they've already calculated.