Computational Materials
Commentary
Computational Materials Science: A Powerful and Predictive Tool Jeffrey Hoyt
During the last 30 years, computer speed and memory has grown exponentially while the cost of performing computations has plummeted. Over the same time period, however, the Jeffrey Hoyt cost of performing laboratory experiments has risen. This divergent cost structure has lead to a fundamental rethinking of R&D methods. Industry, national laboratories, and government funding agencies have recognized that computational science can play a vital role in the research process. Because an accurate computation can significantly reduce the number of timeconsuming laboratory experiments required to test a new component design or process, the result is a faster launch of new products. Since, in today’s competitive business climate, just a few months delay in the time to market can mean a significant loss of market share, it is perhaps not surprising that, in many industrial research labs, computer simulation has become a major player. The next three articles provide examples of success stories in computational materials science. The papers are divided, loosely, in terms of length scales. Atomistic simulations model materials behavior at the level of individual atoms; mesoscopic or microstructural simulations provide information at the level of approximately one micrometer; and continuum-level simulations describe material properties at the centimeter scale. The division by length scale is convenient, but the reader should not conclude that the simulations are wholly independent. Many successful computations at one length scale use input from other, typically lower, length-scale simulations. In fact, there is currently considerable research that attempts to bridge length scales within a single simulation. Of course, every example of successful computational simulation cannot be provided in this short overview, but it is hoped that the examples discussed by each of the authors will provide sufficient insights into the capa14
bilities of state-of-the-art computersimulation techniques. The most attractive feature of atomistic simulations is that very few adjustable parameters are required to describe the energy of interaction between atoms. In fact, with so-called ab-initio or first-principles calculations, no adjustable parameters are needed whatsoever. Since the interatomic potential is specified and because the simulations can be performed at precisely controlled conditions including temperature and volume, atomistic studies can provide materials parameters that are difficult, if not impossible, to measure experimentally.
Industry, national laboratories, and government funding agencies have recognized that computational science can play a vital role in the research process. In the first paper of this series, M. Asta, V. Ozolins, and C. Woodward provide examples of first-principles calculations in their study of the Al-Sc alloy system. Using only the atomic numbers of aluminum and scandium as input, the authors compute the equilibrium solubility of scandium in aluminum as a function of temperature, the heats of mixing of various intermetallic compounds, and the interfacial energy between the aluminum matrix and the Al3Sc precipitates. The interfacial energy, in particular, is notoriously difficult to measure in the lab. A number of significant advances in the simulation of microstructural evolution have occurred within the last decade, such as the development of the phase-field model and its application to dendritic solidification and precipitate coarsening. In the second paper, E. Holm and C. Battaile of the Sandia National Laboratories, New Mexico
describe a further example—namely, modeling grain growth in metals and alloys. The authors employ the socalled Potts model to describe the energy of a polycrystalline material and utilize a statistical scheme, known as the Monte Carlo, to transfer material from one grain to a neighboring grain. The examples outlined in the paper demonstrate not only the tremendous advances in the study of grain growth, but the power of the Monte-Carlo technique. Monte-Carlo methods are used in a wide variety of simulation studies and are quite effective in that they correctly capture the random nature of thermodynamic processes. The final paper of this compilation describes continuum-level modeling, specifically finite-element modeling (FEM). For many years, FEM has been used with great success to examine the stress and strain behavior of plastically deformed materials, however, as M. Horstemeyer of the Sandia National Labs, California demonstrates, the technique is now capable of predicting the failure properties. By combining the evolution of stress as modeled by a constitutive relationship for the material with the evolution of damage (e.g., porosity), the FEM can reproduce the location of and the stress to failure. Horstemeyer illustrates the predictive capabilities of the FEM with examples of components made from cast 356Al. With the staggering advances in computer speed and the continued improvements in numerical techniques, computational materials science has become a powerful and, more importantly, a predictive tool. A common theme in the three articles to follow is the fact that simulations can now generate reliable estimates of materials properties or behavior. As the field of computational materials science continues to progress, the goal of replacing most laboratory experiments with computer simulations is fast approaching reality. Jeffrey Hoyt is a senior member of the technical staff at Sandia National Laboartories and is the advisor to JOM from the Chemistry & Physics of Materials Committee of the Electronic, Magnetic & Photonic Materials, and the Structural Materials Divisions of TMS.
JOM • September 2001
