Simulating supernovae on supercomputers

Tony Mezzacappa sits before a computer visualization of an exploding star. The visualization illustrates convection in a core-collapse supernova explosion. The red plumes, composed of material that has been heated by neutrinos, are rising, and the yellow fingers, composed of cooler material, are moving inward. (Photo by Curtis Boles) Click here for more photos.

Billions of years ago, the universe became the stage for spectacular stellar explosions. These momentous celestial events marked the end of a star’s life but planted the seeds that would ultimately result in the explosion of life on the earth. After millions of years of evolution, massive stars were disrupted in a creative process. These core-collapse supernovae spewed forth lightweight, life-giving elements such as carbon and oxygen that eventually reached the regions out of which our solar system formed. They also synthesized heavy elements and disseminated them to interstellar space. In addition to the elements they created, supernovae also left their mark in the form of neutron stars and black holes.

“Life as we know it would not exist if not for these incredible explosions of stars,” says Tony Mezzacappa, task leader for astrophysics theory in ORNL’s Physics Division. “When stars die in these explosions that generate energy at the rate of billions upon billions of watts, elements necessary for life are strewn throughout our galaxy and become part of the ‘soup’ from which our solar system formed.”

A major challenge for computational astrophysicists is to solve the “supernova problem,” using massively parallel supercomputers to model stars greater than 10 times the mass of the sun. One goal is to predict whether these stars will explode like Supernova 1987A (a much observed supernova in a nearby galaxy). The other is to predict all of the observed phenomena associated with such stellar explosions. Three-dimensional (3D) simulations that can be run only on terascale supercomputers (such as the IBM supercomput-ers at ORNL) are being developed to do just that.

A core-collapse supernova explosion occurs in only a few hours in a star that has evolved over millions of years. This event is thought to be caused by a shock wave that arises when the star’s iron core collapses on itself, compressing its subatomic particles to the point where they repel each other and force the core to rebound. Astrophysicists believe the shock wave stalls while trying to propagate through the stellar core and is reenergized by neutrino heating. Neutrinos are particles with no charge and small mass that interact very weakly with matter. At the core of a supernova is a neutrino “bulb” that radiates heat and energy at the staggering rate of 1045 watts.

ORNL leads the field in simulations of neutrino transport and must now apply this expertise to developing 3D simulations that will explore the role played by convection—transfer of heat by the circulation of the core’s proton-neutron fluid—in aiding this shock revival process, as well as the role played by the star’s rotation and magnetic field. ORNL researchers are also interested in using these simulations to predict the emitted gamma rays and gravitational waves (ripples in space-time) from supernovae.

Supported by $9 million in funding over five years from the Department of Energy’s Office of Science initiative called Scientific Discovery through Advanced Computing (SciDAC), ORNL, the National Center for Supercomputer Applications (NCSA), and eight universities plan to obtain a detailed understanding of how a star explodes. Their approach is to perform 3D simulations of the radiation of the enormous amounts of neutrino energy and the resulting turbulent fluid flow (hydrodynamics) that together may propel material into outer space. Computational predictions will be made consistent with data obtained from astronomical observations.

“Thanks to the growing wealth of observational data from ground- and space-based facilities and the growing computing power afforded by massively parallel supercomputers at ORNL and elsewhere, we are presented with a unique opportunity to finally solve one of nature’s most important mysteries,” says Mezzacappa.

The simulations will uncover how supernovae synthesize elements and disseminate them into interstellar space for processing by other astrophysical systems. Additionally, they hope to determine how the “neutrino-driven wind,” which arises from the proto-neutron star left behind after the explosion, synthesizes elements heavier than iron in a process of rapid neutron capture.

The simulations will also be important in ultimately understanding how the cooled-down remnants of supernovae give rise to neutron stars and black holes, creating the basic building blocks of rotating neutron star (pulsar) and X-ray binary systems.

“We will try to predict whether a black hole or a neutron star will be left behind in the next supernova explosion in our galaxy,” says Mezzacappa. “These events occur in our galaxy two or three times each century.”

One of the collaborating institutions is the University of Tennessee (UT), where Jack Dongarra and research faculty member Victor Eijkhout will be working on mathematical solutions (algorithms) to help solve the equations that govern the motion of neutrinos through the stellar material.

In addition to UT, ORNL’s collaborators for the project are the State University of New York at Stony Brook, the University of Illinois at Urbana-Champaign, the University of California at San Diego, the University of Washington, Florida Atlantic University, North Carolina State University, and Clemson University. Mezzacappa’s co-investigators at ORNL are David Dean and Michael Strayer, both of the Physics Division, and Ross Toedte of the Computer Science and Mathematics Division (CSMD).

The ORNL-centered SciDAC team led by Mezzacappa is also working with five other SciDAC teams (“ISICs”) on issues that include scalable solution of large sparse linear systems of equations, code architecture and design, management and analysis of terascale datasets, code performance, and adaptive meshes, as well as with a supporting “base project” to address networking issues for this distributed collaboration. These collaborations involve a number of ORNL staff in CSMD.

This “mother of all applications,” as Mezzacappa calls the core-collapse supernova problem, should provide new insights into radiation transport and fluid flow relevant to many phenomena. “Our work addresses very broad themes important to DOE’s national mission,” Mezzacappa says. “Our ability to model the movement of radiation through matter may help ad-vance DOE’s energy and basic research missions.”

Examples of interest to DOE are combustion processes in internal combustion engines, effects of increased atmospheric greenhouse gas concentrations on future climate, simulated nuclear weapon explosions (stockpile stewardship), production of fusion energy reactions in hot plasmas, and the effects of radiation therapy on tumors and normal tissue. Supernova simulations on supercomputers will likely be a shining star in the astrophysics and other scientific communities.