Lunch & Learn: Faculty Use of High Performance Computing with Jeroen Tromp

Faculty are taking full advantage of Princeton’s TIGRESS High-Performance Computing Center.
Professor Jeroen Tromp, the Blair Professor of Geology and Professor of Applied & Computational Mathematics came to Princeton in July from Caltech. Among his many awards, he received the John von Neumann Prize in Supercomputing in 1988, the Gordon Bell Award in Supercomputing in 2003) and the Medal of the Vening Meinesz Research School of Geophysics in 2004.

In “Simulating the Big One” at the October 22 Lunch ‘n Learn, Tromp illustrated what we can do today to simulate waves that are generated from earthquakes. He showed off large-scale 3D numerical simulations of seismic waves generated by real and hypothetical earthquakes and the resulting response of differently engineered structures.
Tromp explained that earthquakes generate compressional (acoustic) waves and shear waves that move through solids albeit at slower speeds, usually on the order of 5-6 kilometers per second. But when such waves hit sedimentary areas like the Los Angeles basin, the waves slow down considerably and the motions get amplified. The result, for those on the ground, is a longer duration event and far greater shaking.
He showed an animation of an earthquake that occurred on June 12, 2005, a 5.1 magnitude event in Big Bear. The animation showed the waves radiating away from the epicenter, and illustrated the up and down motion of the ground using red for up and blue for downward motion.
A Southern California Seismic Network (SCSN) now records such ground motion throughout the year. The data is available to the public.
JeroenTromp.jpgHe then illustrated how the shockwaves would affect a typical 17 story steel office structure in a theoretical 7.9 earthquake, the estimated size of the 1857 San Andreas rupture. The shaking from such an event would last about a minute and a half. The simulations now incorporate the kinds of ground motion that you might experience in the LA basin owing to an earth quake.
He also produced a three dimensional simulation of an 18 story steel frame structure, a building identical to one that was severely damaged during the Northridge quake, as well as comparable building redesigned according to today’s standards.
He explained that the danger for such a structure is that if you move one story too much relative to its neighboring story, the entire building can pancake. The redesigned build stays up, but it nonetheless can develop a permanent kink, what actually happened to the building during the Northridge quake. After the quake, the elevators would no longer run because the shafts were out of alignment. He hopes that more buildings in Southern California will be instrumented to aid in our understanding of the impact of the waves on different types of structures.
Such simulations can help to guide future design and budgetary considerations.
Caltech shares shakemovies. After each event of magnitude 3.5 or greater, they generated an animation of the earthquake at There, you can even sign up for e-mail alerts. He noted that efforts are under way to mirror the shakemovie site here at Princeton and to begin a new site that will provide comparable data for the entire globe for events of magnitude 5.5 or greater. Such efforts are useful, he emphasizes, for understanding events like the earthquake off Sumatra, which generated not only a huge tsunami but also waves that repeatedly circled the entire globe.
Anatoly Spitkovsky received his PhD from the University of California (Berkeley), worked as a post-doctoral fellow at Stanford, and received the Sloan Fellowship in 2007. Since 2006 an Assistant Professor in Princeton’s Astrophysical Sciences department, Spitkovsky presented “Simulations of Astrophysical Shocks,” first-principles plasma simulations of shock waves in astrophysics. His work seeks to address a very simple-sounding problem — what happens when two clouds of ionized gas collide in space?
On earth, most shocks are mediated by physical collisions that involve density, temperature, and pressure changes. In space, although the shocks are fundamentally collisionless (a much, much longer path between particle contact than on earth), shocks still form and result in deceleration of supersonic flows on scales much smaller than mean free path. Spitkovsky used the Princeton’s high-performance computing facilities to simulate collisions between two clouds of plasma in order to understand the physics that drives collisionless shocks.
In astrophysics, shock waves span a wide range of velocities. There are relatively slow solar winds at a few hundred kilometers per second colliding with the surrounding interstellar medium, and exploding supernovae whose remnants travel at a few thousands of kilometers per second. By contrast, jets from a galactic nucleus would travel at relativistic speeds, close to the speed of light.
Of interest, says Spitkovsky, all of these shock waves appear to be related. They all decelerate supersonic flows, without colliding, and manage to accelerate particles out of thermal equilibrium. They also seem to be able to amplify magnetic fields. Where there are no magnetic fields, the shocks create them from scratch.
His simulations involved hypothetical, relativistic waves passing through each other. Without collisions, one might have expected no effect. But indeed, shock waves are created and magnetic fields appear due to plasma instabilities. An important result also emerged: some particles were self-consistently accelerated in simulations, and gain energy by repeatedly crossing the shock front. These particles are responsible for the high-energy emission observed in astrophysical shocks, like supernova remnants and jets from active galaxies.

Spitkovsky’s use of Princeton’s high-performance computing facilities involves code parallelization. The longest run involved two months on Woodhen (256 cpus). His conclusion? The availability of such resources immediately has an impact on science, especially in problems where the exploration of different regimes requires fast turnaround, very difficult to obtain at much larger national centers. University researchers interested in using these facilities should contact Curt Hillegas, Director of the TIGRESS High-Performance Computing Center and Computational Science and Engineering within OIT’s Academic Services.
A podcast is available.