By Jarrett CohenPhotos by Debora McCallum

Richard DeVore of Naval Research Laboratory (NRL) describes how radio telescopes observe radiation from the sun. While NRL's radio telescope is no longer active, the latest data from the Solar and Heliospheric Observatory satellite guide DeVore and colleagues in simulating solar phenomena.

Visualization by NASA/Goddard Space Flight Center, Scientific Visualization Studio

NASA's Polar satellite observed the full extent of the aurora borealis, popularly known as the northern lights. One cause of the phenomenon is a coronal mass ejection (CME) from the sun that passes through the Earth's atmosphere. Under the right conditions, a CME also can knock out satellite communications or power grids. NASA HPCC software is helping Naval Research Laboratory scientists work toward the goal of predicting CMEs before they leave the sun.

In the simulation, Naval Research Laboratory (NRL) scientists were testing their theories about coronal mass ejections (CMEs). Their goal is to predict when and where these chunks of the star's corona, or outer atmosphere, will be thrown out. To predict a CME, one has to consider everything from the 30-million-mile expanse of the sun and its atmosphere down to the 5,000-mile-wide magnetic maneuvers that likely push the mass out. With a 6,000-fold difference in scale, "you need a telescope and a microscope to see what's happening," says Spiro Antiochos, solar physicist at NRL.

The Solar and Heliospheric Observatory captured this gigantic coronal mass ejection on November 6, 1997. The inner white ring at the center outlines the sun.

Traditional computational methods cannot focus on all those sizes at once. NRL requires a tool that concentrates limited resources on where things are changing, and NASA's High Performance Computing and Communications (HPCC) Program is supplying such a capability in adaptive mesh refinement (AMR). The NRL-NASA collaboration will ultimately help government agencies tasked with forecasting CMEs and other forms of potentially destructive solar weather.

NRL's pre-AMR simulations treated the sun like a giant Rubik's Cube, solving physics equations inside millions of identically sized blocks -- a concept known as a structured mesh. The mesh blocks were too big to reproduce the small-scale behavior triggering a CME, so the software crashed. Substituting a larger number of smaller blocks to get higher resolution is not the answer because they would overwhelm a computer's memory.

This coronal mass ejection (represented by thick black lines) almost escapes the sun in a simulation that crashed because it could not focus on the small-scale magnetic features that likely push the mass out.

The secret is to substitute smaller blocks only where they are necessary, and that is what AMR is all about. Marsha Berger pioneered structured mesh AMR during the early 1980s while pursuing her doctorate at Stanford University. She developed the technique to address aerospace and Earth and space science problems in which extremely small features move through a mesh.

NASA HPCC's PARAMESH software tool makes the NRL code AMR-capable. PARAMESH builds an initial mesh of large blocks that automatically divide into smaller blocks to follow interesting physics. Smaller blocks then recombine into larger blocks when no longer needed. "An AMR code gives you both a large-scale view and small-scale analysis," according to Antiochos, who foresees the possibility of CME prediction within five years by harnessing PARAMESH's ability to encompass many scales.

Snapping to erupt

AMR is helping NRL solve the mystery of how the sun's gas and magnetic fields interact to produce CMEs. "When I began in solar physics 20 years ago, I worked with pencil and paper," Antiochos reflects. "Since you could only calculate very idealized physical systems, all you could say was 'this idea may or may not work.' Now, you can do a calculation, then compare it with your observations. For me, it is an exciting time."

The latest observations lie at Antiochos' fingertips. One floor above his office is the real-time data room for the Solar and Heliospheric Observatory (SOHO), a satellite built in part by NRL. SOHO shows that most CMEs start out as prominences -- arches of relatively cold material that become suspended above the corona. Antiochos had a theory about what makes the gas condense and cool to form prominences, but his attempts to simulate it crashed codes for seven years, until PARAMESH entered the equation.

NRL's Spiro Antiochos is excited about using NASA HPCC's PARAMESH adaptive mesh refinement software to capture phenomena ranging in size from 30 million miles down to 5,000 miles.

Peter MacNeice and Kevin Olson of Goddard Space Flight Center (GSFC) developed PARAMESH originally for the NRL team, whose members are investigators with NASA HPCC's Earth and Space Sciences Project. The software has matured and is now flexible enough to work with most structured mesh codes. Ninety researchers worldwide have acquired it.

Antiochos' condensation theory states that heating the bottom layer of the sun's corona drives gas to accumulate throughout the corona. As the mass increases, runaway radiation in the upper layers sucks heat out of the gas, dropping its temperature catastrophically. On a supercomputer, "it had been a problem to produce anything with high density," MacNeice says. "The key was the simulation didn't have a fully adaptive mesh." Coupled with a gas dynamics code written by GSFC's Dan Spicer, PARAMESH subdivided simulation blocks up to seven times. Much like soldiers gathering on the most active sites along the battlefront, a hierarchy of blocks constantly shifted to capture condensation's intricacies.

"The condensation region is very small," Antiochos says. "The AMR calculation shows how cold material is held up and how it moves along. There are shock waves and a sequence of condensations. We could never have calculated it without PARAMESH putting resolution only where we need it."

Before using PARAMESH, NRL also faced challenges simulating the magnetic fields that keep the tightly packed gas aloft -- and eventually force it out from the corona. "We have some observations, but it is hard to measure magnetic fields above the surface of the sun," says NRL physicist Richard DeVore. "We infer the magnetic fields from looking at gas density and temperature." When DeVore simulates highly stretched magnetic field lines tied down at their midsections, they suspend the cold gas above the corona.

Stretch the magnetic field lines out even further, DeVore points out, and "strong electrical currents form in very small regions of space. The field lines snap there and reconnect to form new lines." The result is a tangled structure of helical fields resembling observed prominences.

Antiochos and colleagues believe that an uproar of magnetic reconnection then triggers a CME. "A prominence is like a helium balloon tied down with bungee cords," Antiochos explains. "The field lines are perfect bungee cords -- they can stretch forever but you can't cut them. They keep the balloon from expanding the way it wants to, so it builds up energy. If you have the right magnetic field pattern, it is possible to get rid of those bungee cords very quickly."

Observations show that most CMEs start out as prominences, or arches of gas that become suspended over the sun. In snapshots from an NRL simulation (upper left), magnetic field lines reconnect on the way to forming a prominence. The NRL scientists believe that a subsequent uproar of magnetic reconnection triggers a CME. Goddard Space Flight Center's Peter MacNeice is running new simulations (lower) to confirm that theory. With PARAMESH software adaptive mesh refinement, blocks in the computational mesh automatically divide as CME physics occur on smaller and smaller scales.

MacNeice has begun using the combined PARAMESH-NRL code to resolve that reconnection pattern. "Details in tiny regions can have a huge effect," MacNeice adds. "Throwing a switch in a telephone system can connect places separated by many miles. On the sun, magnetic reconnection can redefine the pathways along which energy travels. Over time, areas of reconnection move outwards, and the mesh follows them. Otherwise, you get the wrong answer."

The original prominence-to-CME simulation attempt took 300 hours on a single CRAY C90 computer processor. A PARAMESH simulation will cut the run-time to 50 hours using 64 parallel processors on a supercomputer such as the CRAY T3E. Moreover, an adaptive mesh achieves peak resolution with less memory than a uniform mesh needs. "You can fit a larger problem on the same size computer and get a faster solution," DeVore says.

Routine prediction

Above: The PARAMESH software applies adaptive mesh refinement to explore the details of a supersonic astrophysical jet. Below: Taking a closer view, mesh blocks refine four times in one section.

A CME prediction code will have to run fast. Anything less than real time is "predicting history," Antiochos says. "It does you no good if it takes 10 days to calculate three days of reality, which is where we are now." This shortcoming is in part why operational agencies' current CME forecasting techniques largely rely on tracking and statistical prediction.

For example, the National Oceanic and Atmospheric Administration (NOAA) uses ground and satellite observations to watch for CMEs leaving the sun. They then simulate the paths CMEs are likely to take. If CMEs come straight at the Earth, resulting magnetic storms can interrupt satellite communications or electrical power distribution and pose hazards to astronauts and high-altitude pilots. Shutting down systems or rescheduling flights minimizes damage from solar weather.

By 2005 NRL aims for their code to pinpoint the location and timing of CMEs before they erupt. The armed forces, NASA and industry will have more time to prevent disasters. Likewise, solar weather simulations will be more accurate with solid physics starting them off. Beyond speed, NRL's code must be reliable for NOAA to use it routinely. The code needs to predict CMEs and ingest observations every day. Reliability also depends on a supercomputer that is up all the time and on satellites that provide observations daily.

Daily forecasts are necessary because CMEs are rather frequent, occurring once per day on average. The daily count doubles when solar activity reaches its maximum every 11 years. The next solar maximum begins this year, and a strong effect on the Earth is likely once per month.

While important now, accurate CME warnings will be critical by the next solar maximum, as Antiochos explains: "Our reliance on space technology is increasing with time. If a CME hit 100 years ago, so what? You got the northern lights. In the future, when everybody has a cell phone, you can't tolerate interference. You become more and more sensitive to your space environment."

For more information, visit the following: http://www.physics.drexel.edu/%7eolson/paramesh-doc/Users_manual/amr.html and http://www.lcp.nrl.navy.mil/hpcc-ess/.

Credits for Insights Magazine go to the following people along with the writers and photographers who contribute to each issue and the researchers and specialists whose material is highlighted:

Program manager: Dr. Eugene L. Tu
Deputy program manager: Dr. William R. Van Dalsem
Editor, writer, photographer and designer: Judy Conlon
Web designer and programmer: Sue Cox

Insights was published by the HPCC Program office and produced by Raytheon contractor staff at NASA Ames Research Center.