From his third-floor office in the South Mudd Building, Jeroen Tromp is carving up the globe, as he develops computational methods that provide a more accurate view of how earthquakes propagate.

SEISMIC SIMULATOR

By Michael Rogers

Jeroen Tromp, the new head of Caltech’s Seismological Laboratory, readily admits that he had never been shaken by a sizable earthquake until he found himself in a swaying high-rise several years ago in Japan. Afterward, he ran to find his Japanese colleagues and excitedly quizzed them about the quake. “Was that an earthquake? Did you feel it?” he demanded to know with all the enthusiasm of the neophyte.

Tromp says that he got a few strange looks in response. It turned out that the source of his excitement was a minor quake that had caused no damage—the kind of event that happens regularly in Japan. “It was only a 4 on the Richter scale,” says Tromp, sounding a bit disappointed.

At the time, Tromp was a professor at Harvard. He didn’t come to Caltech until 2000, and because of the recent lull in local seismic activity, he still hasn’t lived through a “big one.”

Tromp, who is also the Institute’s McMillan Professor of Geophysics, may be forgiven for his lack of actual earthquake experience. He grew up in the Netherlands and went to graduate school at Princeton, two places that aren’t exactly famous for their shake, rattle, and roll. And, perhaps not surprisingly for a guy who was raised in the lowlands of Holland, he is a theoretical seismologist who has spent a lot more time staring at equations on whiteboards and computer screens than he has looking at actual fault lines.

But the distinction between theoretical and field research in seismology may be eroding thanks in part to work by Tromp himself. In recent years, he has branched out from theoretical seismology to embrace a computational approach that’s more directly related to observation and analysis of actual earthquakes. It all started in 1998, when he and a colleague wrote a comprehensive textbook called Theoretical Global Seismology, surveying the history of seismological research and the major recent theoretical and observational advances in the field. Once that was published, Tromp says that he began to think about applying his knowledge more directly to actual events rather than to hypothetical ones.

“I started thinking, ‘Theoretical seismology is important, but ultimately seismology is about the earth,’” he says. “It doesn’t mean that you don’t need a strong understanding of theory, but at some point you want to be able to produce results related to actual events.”

Since coming to Caltech, Tromp has devoted much of his time to writing software and building an innovative computer network designed to create three-dimensional simulations of seismic events and provide a better understanding of what happens in the intervals between big earthquakes. (He defines a big quake as 7 or greater on the Richter scale.) To track temblors, seismologists have typically relied on data collected from the Global Seismographic Network of seismic recording stations and on computational methods based on a one-dimensional Earth model. But Tromp says that these methods ignore the diverse geological features within and below the crust—variations that can have a dramatic effect on how earthquakes propagate. “When geological variations become large, the classical methods break down,” he says.

With help from a former Caltech postdoc, Dimitri Komatitsch, Tromp has created a computer model in which the earth is divided into 2.6 million elements that are each 40 kilometers on a side. Each cube has different geological features, which affect the behavior of the seismic waves that pass through them.

In Tromp’s 3-D computer models, the earth is divided into cubes. Each cube represents a discrete region whose unique geophysical properties can affect the movement of temblors.

Tromp says that the model is like a CAT scan of the earth, in that it allows researchers to track the paths followed by seismic waves, much as CAT scans —with far greater precision—monitor the propagation of X-ray signals to build up a 3-D picture of the brain. “An earthquake is like an X-ray source,” he says. “But while a doctor doing a brain CAT scan knows where a signal originates and its intensity, we get a poor man’s version of a CAT scan using data from seismic stations since we don’t know exactly when and where the earthquake occurred and what happened along its path.” With his simulations, a more complete picture of a quake can be developed.

At Caltech, Tromp runs earthquake simulations on a parallel arrangement of 150 personal computers known as a Beowulf cluster, packed into a room on the second floor of the South Mudd geology building. Each simulation involves tens of millions of operations per second, as the progress of the quakes’ seismic waves is mapped from one cube to the next, gathering speed, slowing down, changing direction, and altering in other ways that depend upon the geological characteristics in that part of the earth. The models also account for the fact that seismic waves can travel at different speeds in different directions away from the quake’s epicenter. Although detailed information about the earth’s geology has existed for decades, the relatively recent development of the computer cluster plus advancements in 3-D modeling have allowed Tromp to put that information to use in his simulations.

Tromp has also collaborated with Seiji Tsuboi at the Japan Marine Science and Technology Center, which operates the Earth Simulator, a machine in Japan that is considered to be the world’s fastest supercomputer. For those simulations, they created a model of the earth with 200 million elements. For a simulation performed there in 2002, they were awarded the Gordon Bell Prize for peak performance at last November’s Supercomputing 2003 conference.

Tromp’s colleagues say that his approach to computational seismology has greatly advanced the field. “The computer code he developed for simulating seismic wave propagation is the most comprehensive in the world, enabling the first complete solution for wave propagation in a 3-D Earth model,” said Thorne Lay, PhD ’83, professor of Earth sciences and director of the Institute for Geophysics and Planetary Physics at UC Santa Cruz. “The initial implementation of Tromp’s spectral element method on the Beowulf cluster in the Seismological Laboratory was the dawn of a new age in seismology. Prior methods used approximations of Earth’s geology and approximations of elastic wave propagation equations, limiting our ability to resolve global Earth structure or details of earthquake ruptures. With the ability to reliably compute ground vibrations in a 3-D Earth model, iterative approaches to refining our global models will be revolutionized as we will no longer be limited by the accuracy of the simulations. As computer technology blossoms, complete numerical solutions like those provided by Tromp’s code will become the standard tool for all seismologists, sweeping aside approximate methods that the field has been constrained to use for the past century.”

In the basement of South Mudd, Tromp is now developing a new computer facility that can accommodate a Beowulf cluster of 1,200 personal computers. With the cluster’s additional memory and faster speed, he’ll be able to increase the amount of information seismologists can process about earthquakes, reduce the size and increase the number of elements in the 3-D simulations, and create more accurate models in less time. The cluster will also be used to study other geophysical phenomena, including volcanoes and glaciers. “This will enable us to start mapping the earth in greater detail and complexity,” he says. To fill the facility to capacity would cost approximately $4 million, and would enable simulations that are 10 times faster and have 10 times better resolution than the current Beowulf system.

Tromp’s colleagues say that they are eager to use the new computer cluster. When Donald Helmberger, the Smits Family Professor of Geophysics and Planetary Sciences, saw Tromp putting together his first batch of computers, he says he told him, “Do you think this is going to work?” But he says that he never doubted that it would. “If he could make it work for 150 computers, why not 1,200? I think it’s the future.”

Since moving to Pasadena, Tromp has naturally gotten more interested in the seismic activity in Southern California and has created a detailed 3-D model of the region, using an oil company’s geological maps. In this model, the grid consists of elements that are 300-meters long on each side. Because the cubes are smaller than in the global model, the resolution is better, making it possible to gather more detailed information about local earthquakes.

Besides increasing basic knowledge of earthquakes and the geophysics of the earth’s interior, Tromp’s simulations have other practical uses. Although they would not play a role in an earthquake early-warning system, he thinks that civil engineers could use the information when determining where to build high-rises and other structures, since the simulations model the varying intensity of shaking in different locations.

In fact, one of his goals as Seismological Lab director is to increase collaborations between the Seismo Lab and Caltech colleagues in civil engineering. “We’re starting to collaborate with engineers so they can use our simulations as input to shake their buildings and numerically assess what might happen,” he says.

Tromp says that he’d also like to increase the lab’s public outreach. One plan is to make animations that show how local earthquakes affect different neighborhoods and to make them available to the public. He also hopes to secure funding to expand and maintain the Southern California Seismic Network (SCSN) of 150 broadband sensors. SCSN, a collaborative project of Caltech, the U.S. Geological Survey, and the California Geological Survey, provides the public with information about where earthquakes occur, how big they are, and what type of faulting is involved, within minutes of an earthquake. “Just coming here and seeing the data from the network is a gold mine,” Tromp says. “How could you not be interested in working with it?”

Speaking as a local resident and the father of a young daughter, Tromp says that he’d be “perfectly happy if not a single big earthquake occurs during the time I’m in L.A.” At the same time, he is well aware of the invaluable seismological data, not to mention the enormous public relations boost, these regional temblors provide to the Seismo Lab. Anytime there’s a big quake in California, the media trucks converge on the Caltech campus, Seismo Lab staffers pop up everywhere as spokes-people, and Caltech earthquake science acquires a popular following Geraldo would envy. One of the key challenges facing Tromp as an administrator is how to maintain the public’s interest in and support for the lab’s work during this latest earthquake lull.

“After the Northridge quake in 1994, people were aware of the dangers of earthquakes, but if there hasn’t been a recent big quake, people forget,” he says. “A lot of the things we’re doing involve education and outreach. Hopefully, a big quake won’t happen here during our lifetimes, but people need to be prepared and understand what happens during an earthquake.”