IBM and LLNL scientists show supercomputer advance in predicting materials strength

In the most extensive computer calculations of their type to date, the scientists used the ACSI White supercomputer, which was built last year by IBM for LLNL, to create and then deform simulated cubes of as many as 1 billion atoms. Creative computer visualisation techniques revealed the inner workings of the atoms' response to the stress: stunning images and videos showing cracks moving at surprising supersonic speeds as well as the expanding tangle of defects deep inside the cube that can harden a tough, flexible material to the point of brittle fracture.

Details and results of the computer simulation experiments are published in two technical papers. The first paper describes a 20-million-atom simulation that shows how brittle-fracture cracks can travel far faster than theory had previously predicted. This result is expected to be important in helping scientists understand a wide range of fractures, from shallow earthquakes to the sudden failure of fiber-reinforced composite materials, such as those used in airplanes.

The second paper recounts a 1-billion-atom simulation of "work hardening", the process by which deformation strengthens a material but can embrittle it if overdone. Bending a paper clip back and forth is an example of work hardening: the metal is initially rather flexible, but it soon stiffens and breaks where it was repeatedly stressed. Work hardening also strengthens materials during forging, an important manufacturing technique used to make products as diverse as critical auto parts and golf clubs.

In each of the simulations, which required up to 10 days of around-the-clock computations, the supercomputer calculated the forces between each of the atoms and its neighbours and their positions as the edges of a notched cube or atoms were pulled apart. In the brittle fracture simulation, a crack formed at the notch and travelled rapidly through the material as stress concentrated at the crack tip and ripped apart the chemical bonds that held nearby atoms together.

The IBM/LLNL scientists found that when the material is given the property of becoming stiffer, not weaker, as it is stressed, as occurs with certain polymers and rubbers, the crack tip can shoot through the material at supersonic speeds (that is, faster than the speed of sound in that material). Such behaviour was long thought to be impossible. But in recent years, supersonic crack speeds have been observed directly, or suspected, in both laboratory experiments and two devastating 1999 earthquakes in Turkey.

The IBM/LLNL simulation gives a sound theoretical footing to such claims and will result in improved tools to understand and predict the behaviours of earthquakes and to design new materials that can resist brittle fracture.

In the work hardening computation, the simulated material was made to be tough, not brittle. That meant that the atoms would initially respond to stress by sliding past each other rather than simply breaking apart. The offset atoms create lines of misalignment in the periodic structure of the material that are called dislocations.

In a soft metal under stress, such dislocations simply pass through the material as deformation occurs. But in a stronger or more complex material, various dislocations collide, which halts further atomic motion at each intersection. As deformation continues, these pinned dislocations accumulate, initially increasing the strength of the material because it can resist a greater force. But if the stress continues, the density of pinned dislocations can become so great that the material turns brittle and breaks.

In addition to Farid F. Abraham and Diaz de la Rubia, co-authors were Robert Walkup of IBM's T.J. Watson Research Center, Yorktown Height, New York; Huajian Gao, a visiting scientist at IBM-Almaden, now at the Max Planck Institute for Metals Research in Stuttgart, Germany; and Mark Duchaineau and Mark Seager of LLNL.

Over the past three decades, Farid Abraham has been a pioneer in using the most powerful computers available to calculate and predict materials properties. In 1985, he was able to model 200,000 atoms, a flat square having only 450 atoms on a side. In July 1994, he published a million-atom simulation.