WHAT happens to water under high pressures and temperatures? Why does silicon carbide break where it does when strained? How does DNA interact with water?
Complete, accurate answers to these deceptively simple questions-and others like them-require knowing what is going on at the atomic level. This means one must account for the behavior of individual atoms and electrons in an entire system: how they move, how they chemically bind, and how those bonds form and break. Not an easy proposition, once one moves beyond a small number of simple atoms.
Computers have long been used to model material behavior, both on the large and small scales. Macroscopic-scale modeling applies statistical mechanics methods to the system as a whole, ignoring details about how each atom responds. Atomic-scale modeling applies the laws of quantum mechanics-fundamental physics equations that describe electrons-but because of the complex nature of the equations, these models can handle only a few atoms at a time.
Lawrence Livermore physicists Francois Gygi and Giulia Galli and their collaborators are using the computer power of the Department of Energy's (DOE's) Accelerated Strategic Computing Initiative (ASCI) and Gygi's JEEP code to push the limits of atomic-scale modeling of complex systems. ASCI, guided by the Office of Strategic Computing and Simulation under the DOE Assistant Secretary for Defense Programs, is developing capabilities to simulate nuclear weapons performance in lieu of nuclear testing. To do so requires computers of unprecedented computational power and speed, as well as simulation codes such as JEEP.
JEEP, which Gygi began developing at the Swiss Federal Institute of Technology about five years ago, uses quantum molecular dynamics (QMD) methods to simulate the behavior of materials at the microscopic level. "Unlike the macroscopic-scale codes, we make no assumptions in using JEEP," explains Gygi. "Using QMD, we input only absolutely known quantities into the code-that is, the identities of the atoms and the laws of quantum mechanics. Combining this approach with ASCI's computational power, we can examine material systems of hundreds of atoms and thousands of electrons extremely accurately."
With JEEP, the ASCI computer becomes a virtual laboratory, where scientists can follow the trajectories of atoms and study the forming and breaking of chemical bonds. This powerful combination can be used to predict physical properties of various materials, investigate properties not directly accessible through physical experiments, and interpret and complement physical experiments.
QMD simulations have a number of applications, from deepening our understanding of materials under extreme conditions-a vital issue for DOE's Stockpile Stewardship Program-to forming a better understanding of complex biological systems.

Modeling the Impossible Experiment
The detonation of some high explosives produces hydrogen fluoride and water, both of which are hydrogen-bonded systems. Little is known about this mixture because hydrogen fluoride is toxic and corrosive, making experiments difficult.
Using the JEEP code on the ASCI computer, Galli, Gygi, and physicist Francis Ree conducted quantum-level simulations of hydrogen fluoride-water mixtures at high temperatures and pressures (Figure 1). These simulations were of unprecedented scale: a picosecond-long "peek" at the interactions of 600 atoms with 1,920 electrons required updating and computing 200 million unknowns. It took 15 days and the entire resources of ASCI's Sustained Stewardship Blue Pacific machine, with teraops (trillions of operations per second) calculational power, to simulate this one picosecond's worth of interactions.
The results provided a better understanding of high-explosive detonation products, revealing their molecular interactions and chemical reactions. The results also helped scientists better understand the equation of state of the mixture. The simulations indicated that hydronium fluoride and hydrogen difluoride anions are produced at high pressures, something that has been hypothesized but not yet observed.

Stretch and Break
In another numerical experiment, Gygi and Galli simulated what would happen when a microscopic chunk of amorphous semiconductor (silicon carbide) is stretched past its breaking point (Figure 2).
"One of the advantages of computational simulations over experiment, in general, is that we can define chemical purity at 100 percent," explains Galli. "We can create the configuration that we want, apply a strain to the system, and let the code run. We watch what happens, what bonds break and where, and how the resulting microfracture relates to the chemical properties of the material."
These quantum simulations provided numbers relating to the elasticity and hardness of the material that could then be compared to results gathered from physical experiments. Galli notes that this was the first time hardness had been computed from first principles for a disordered alloy. "In the past," says Gygi, "it's only been computed for crystal or ordered structures, because disordered systems are much more complex."
They also discovered that the simulated semiconductor material broke at a silicon-rich "island" and were able to define the surface where the material cleaved. "Most of the surface atoms were silicon," says Galli. "In laboratory experiments with this material, physicists find precisely that, so we were pleasantly surprised when we saw it in our numerical experiment as well."
Galli and Gygi plan to continue these studies by simulating atomic clusters residing on other types of semiconducting surfaces as well.

Delving into DNA
Their most recent work involves examining components of the familiar two-strand, double-helix DNA structure. Each strand of DNA consists of a "backbone" on which chemical bases attach. When the two strands are wound around each other in the familiar configuration, a base from one strand attaches to a base on the partner strand, forming base pairs that step up like ladder rungs (Figure 3).
Working in conjunction with Michael Levitt from Stanford University and Livermore's Eric Schwegler, Gygi and Galli are examining what happens to the DNA backbone in water, its natural medium. They plan to isolate a fragment of the DNA backbone and simulate how a molecule of dimethyl phosphate from that fragment interacts with water molecules. While the Livermore scientists use JEEP for the simulation, Levitt will be running a simulation code with his widely used model of this interaction. The numerical experiments at Livermore are expected to validate or invalidate some of the assumptions made in standard models and serve to improve them. The results will also deepen understanding of this complex biological system. The work is being carried out within a Laboratory Directed Research and Development strategic initiative on computational biology led by Mike Colvin.
In another DNA experiment, Gygi, Colvin, Raymond Fellers, and Daniel Barsky are extracting a single DNA base pair to see how the complementary bases interact. They expect to better understand what causes DNA to bind and which molecular interactions are key to binding. On a larger scale, they hope to find out how binding mechanisms affect the replication of DNA.
At stake is understanding the fidelity of DNA replication. Sometimes, because of damage or mutation, a different molecule replaces one of the bases. The question then is, will the DNA successfully replicate itself? "Understanding this issue is a very long-range goal," Gygi notes. "For the short term, we hope to calculate the energy needed to separate a base pair and, through the results, discount certain scenarios of DNA binding."
"One key thing to keep in mind about all this work," adds Gygi, "is that the JEEP simulations do not provide the full picture. But they do provide key pieces to a puzzle, which allow us to look at yet other pieces and say whether those are a part of the same puzzle."

Laboratory Is a Unique Environment
The Laboratory provides the requisite elements for performing QMD: top-notch physics, the biggest computational machines available, and state-of-the-art visualization tools to help view and interpret enormous amounts of data. "There are few places in the world where you can pull all this together," says Gygi. "We've been able to apply our methods to different problems and learn things that relate to other areas of research. The big machines with terascale computing power have made it possible to study more and more complex systems, more reliably and accurately. Finally, these big machines have made it possible to build a bridge between two worlds of simulation-the macroscopic-scale simulations using statistical mechanics and the atomic-scale simulations using quantum mechanics."
-Ann Parker

Key Words: Accelerated Strategic Computing Initiative (ASCI), computational biology, DNA, high explosives, JEEP, material behavior, quantum molecular dynamics (QMD), semiconductor.

For further information contact Francois Gygi (925) 422-6332 (gygi1@llnl.gov) or Guilia Galli (925) 423-4223 (galli@llnl.gov).

Back to July/August 1999 // Science & Technology Review 1999 // Science & Technology Review // LLNL Homepage