Opening a new window on the universe — that’s the promise of gravitational wave astronomy, and its fulfillment presents a scientific computing challenge that might almost be akin to pulling light out of a black hole, if that were possible. Or maybe the more appropriate analogy is water in a desert, where sometimes the solution is cactus.
Make that a capital “C” — as in Cactus, an open, collaborative software framework for numerical relativity that since 1997 has enabled research that underlies more than 200 scientific papers and 30 student theses. That and more than that, in a fast-moving, information-packed presentation, was the topic of Gabrielle Allen’s keynote talk, Tuesday, August 3 at TeraGrid ’10, the fourth annual conference of the TeraGrid, in Pittsburgh, Pa.

Allen is associate professor in Computer Science at Louisiana State University, and a faculty member at LSU’s Center for Computation & Technology. Before moving there in 2003, she led the computer science area of the Max Planck Institute for Gravitational Physics (Albert Einstein Institute) in Potsdam, Germany. At AEI, she was a PI for the European GridLab project, and led the initial development of Cactus.

She began her talk, “Cyberinfrastructure for Numerical Relativity,” by noting that she’s been a TeraGrid user since 2001. “Accurately modeling astrophysical systems that are governed by Einstein’s Equations of General Relativity, such as black holes, stellar core collapse or gamma ray bursts,” she added, “requires the use of cutting-edge computational resources and software.”

Solving the problems of this field of science, Gravitational Wave Physics, depends on interactions between modern theory, observation and computation, and all three aspects, says Allen, are leading to new discoveries. Gravitational waves are one of the startling aspects of Einstein’s predictions from general relativity. Measurements of the decaying orbits of binary pulsars agree with Einstein’s prediction of gravity waves, yet even now these waves haven’t been directly observed.

Two large projects have mounted gravity wave detectors — the U.S. LIGO (Laser Interferometer Gravitational Wave Observatory) project and GEO 600 in Germany — to test Einstein’s theory, but these extremely sensitive instruments need to be precisely tuned and use complex data analysis to recognize the delicate signatures of gravity waves from events in deep space. For this the physicists need numerical simulations.

The numerical problem is finding ways to solve the Einstein equations that govern gravity-wave phenomena. “There are thousands of terms on the right-hand side,” says Allen, “and these equations are very difficult to work with.” The initial challenge has been modeling binary black holes — two black holes in orbit around each other — a relatively “simple” system with relatively few parameters, as a test case for LIGO.

Recent work using TeraGrid resources at multiple sites, a project of an international team that included Allen’s LSU colleague Erik Schnetter, a research professor in the Department of Physics and Astronomy, modeled the binary black hole problem with unprecedented detail. Allen describes this work, featured on the cover of the 2009 TeraGrid Science Highlights publication, as an outcome of what has so far been a 40-year plus effort to model gravity waves from binary black holes, which has only now arrived at numerically generated waveforms. “We still can’t do extreme mass ratios or very fast spins,” she says, “but this has opened the door to modeling more complex scenarios, such as general relativistic hydrodynamics.”

Allen went on to describe the essential elements of cyberinfrastructure needed to move this work forward, and elaborated on the Cactus framework — so-called from its design of a central core (“flesh”) which connects to application modules (“thorns”) through an extensible interface. It’s a modular system, with thorns that are defined by parameters, variables and methods, and the flesh binds it together.

Cactus derived originally, Allen explained, from a mid-90s Black Hole Grand Challenge project, with multiple groups collaborating. “This came out of the vision of Ed Seidel,” she said. Seidel, who recently ended a term as director of NSF’s Office of Cyberinfrastructure, worked during this period at AEI in Germany, and recognized needs — that have been implemented through Cactus — for modularity, for easy code reuse, community sharing and development.

A recent set of Cactus thorns, Allen pointed out, has implemented adaptive mesh-refinement (AMR). Developed by Schnetter, this has allowed many groups to have access to AMR with little code change. “We can scale the AMR up to around 16,000 processors,” noted Allen. Cactus also implements automatic code generation through “Kranc” — a Mathematica tool to generate Cactus thorns from PDEs. “Your turn the Kranc and it spits out complete thorns of Cactus.”

Cactus interfaces with the Einstein Toolkit, a consortium that develops and supports open software for relativistic astrophysics. “Our aim,” said Allen, “is to provide the core computational tools than can enable new science, broaden our community, facilitate interdisciplinary research and take advantage of emerging petascale computers and advanced cyberinfrastructure.” The consortium includes 55 members at 17 sites in nine countries.

Among many challenges to be faced, Allen observed that changes in academic culture are needed to support the model of open collaboration versus competition among research teams. “We need incentives for faculty to encourage postdocs and students to use and contribute to community software.”

“Everything is a challenge,” she added, “in this kind of work. Nothing works as well as you’d like. The TeraGrid has been a big friend of numerical relativity, and has helped us to develop the kind of community we need — especially for students, it has been amazingly helpful. It provides access for students to the hardware we use, and the software and best practices. All these things are crucial.”

The biggest challenge ahead, she added, is how to handle tremendous amounts of data. “Everything is going to be about data very soon. We need to be ready for that. It is changing the world of science. There is a whole sociology of how data is going to be used in academia. We have a big chance to do this properly.”