Computational Materials Sciences Workshop

Materials are critical to each of DOE's missions--Energy, Environment, Science and Technology, and National Security, as well as to economic growth and to the prosperity of the nation. Every technology has been and continues to be limited by the performance of materials used for technological applications. Great improvements in energy efficiency and alternative energy resources and the development of a new generation of efficient vehicles for transportation demand the right new materials with much improved properties. In addition, environmentally benign processes for materials manufacturing, for recycling, and for disposal must be developed.

The economic and environmental impacts of new and improved materials are enormous. For example,

Major global issues demand fundamental changes in our generation and consumption of energy. The rate of CO₂ production must be reduced through development of more efficient means of transportation, heating, cooling, and lighting. Alternative ways for energy generation must be developed. To help accomplish these, we target a few areas that are central to the most urgent missions of the DOE and which the materials science issues are amenable to large-scale computer simulation. These are listed below:

To better meet future energy needs in the U. S., materials research and development requires a radically different approach than used previously. High-performance computing and materials modeling offers a unique and timely opportunity to accomplish this. For materials science, the results of a simulation initiative will be to:

As history has proven, advances in technology depend on advances in materials performance and on developing new materials with superior properties. These advances were made in the past largely by an experimental approach, using extensive laboratory tests on many variants of existing materials, and gathering and sorting data through practical experience. Although it has worked well, this Edisonian approach is very time consuming and slows the pace of technological innovation. It still requires, for example, about 10 years to explore, measure, test, and certify for design a new alloy for a jet engine.

Not only is this experimental process time consuming, it also severely restricts the search for unknown materials. Instead, well-tested and proven materials have constantly been fine-tuned and tailored for a specific application, rather than revolutionary ones developed. The austenitic steels with their many variants represent a good example of such a well-proven class of materials.

The realm of new materials is, however, huge and hardly explored in a systematic and exhaustive manner. There remain vast numbers of materials to be discovered with potentially useful and exceptional properties. However, employing only the traditional experimental approach, there simply are not enough materials scientists and laboratories available to carry out the necessary exploration and discovery of needed new materials.

We propose a revolutionary change in this process, driven by the vastly increased computer power and communications that will become available shortly in sufficient quantities for the first time. Computational search and materials simulations provide a revolutionary opportunity to change the pace of discovery of new materials. Computer simulations and modeling enable prediction of the relevant properties of complex materials composed of any number of elements from the periodic table with sufficient precision to target promising candidates for subsequent experimental testing and further development. The computer models encompass not only the calculation of chemical, physical, and thermodynamic properties, but also the processes by which these materials are manufactured and by which these materials degrade in a given technological application. Materials will still need to be synthesized and tested, which is a time consuming and laborious experimental procedure. However, a computational search and simulation provides a rational guide to the materials for such testing.

The aging and failure of materials in our present society is all around us: our cars are not as crashworthy as they could be given the need for lightweight vehicles; metal fatigue in airplane structures causes expensive catastrophic failures and retrofitting; bridges collapse. The list goes on. A major reason for these surprises is that the aging and deterioration of materials has been too difficult and too complex to quantify in order to be factored into the design of the components of our technological society. Nevertheless, this must be done in the future, and it must be done before manufacturing the components or rebuilding the energy and transportation infrastructure. Moreover, it must be done without the benefits of having learned it by trial and error, i.e., by the empirical approach. Computational simulation of the fundamental aging processes that change materials properties and lead to degradation phenomena is essential for safer designs, for developing degradation sensors, and for predicting the reliability of our implements used in everyday life.

To develop a new and time-efficient simulation approach to materials discovery, various computational methods already used in specialized research projects must be further advanced, extended, and above all integrated into highly interactive tools. The new paradigm of this approach is now widely discussed in the materials science community, and it is known as multi-scale modeling. The name captures the following challenge:

Properties and performance of materials can be ultimately explained in terms of the properties of atoms and their mutual interactions, as well as their vibrational and diffusive motions at finite temperatures. While this fundamental statement is true in principle, it is impossible to use only atomistic theories to simulate real materials on a computer by including larger and larger numbers of atoms. The astronomically large number of atoms and the equally large number of vibrations can never be simulated completely on a computer to predict macroscopic properties. Nature creates new phenomena when assembling this vast number of atoms, and these are referred to as collective phenomena. These collective and complex phenomena need to be described by additional mathematical models that do not simply emerge from the atomic scale theories and models. Nevertheless, they require information and input data from the theories and models at ever finer length and time scales, all the way down to the smallest and shortest, the atomic scale, the realm of electrons and atomic nuclei.

The multi-scale approach is actually very old and time-honored, and chemists, physicists, materials scientists, and engineers have all practiced it rather imperfectly throughout their careers. What is different now with the advent of supercomputers is the ability to execute this approach in terms of quantitative accuracy, of complexity, and of seamless integration over all length and time scales. The old approach was imperfect because the mathematical models had to be simplified so drastically for computational convenience that their predictions were often no better than the scientist's intuition or experience. Previously, including issues of complexity in a model meant leaving out the details.

Today, atomistic calculations based on local density functional methods have reached the point of predicting, from first principles, the energy, crystal structure, and optical properties of most materials. These calculations incorporate aspects of chemical bonding and alloy formation in a parameter-free way leading to accurate predictions of the properties of materials that have not yet been synthesized in the laboratory. While these predictions are presently limited to small aggregates of atoms, say 100's to 1000's; new algorithms are being developed which will extend these predictions to larger length scales. First, techniques for finding effective interatomic potentials have been developed and applied to systems with 10,000 - 100,000 atoms. Finally, these calculations can provide critical input to continuum (elasticity and plasticity) theories needed for the actual design of new materials. However, microstructural scale models must be developed before this information can benefit continuum models.

Materials scientists agree that microstructure plays a critically important role. Most materials are complex mixtures of small grains or crystallites with different orientations separated by grain boundaries. Most engineering materials are complex mixtures of different atoms, some purposely added, some present as impurities. Deviations from perfect crystallinity, such as the grain boundaries, inclusions, dislocations, and impurities, determine the strength, magnetic properties, or ductility of an alloy. Furthermore, the microstructure depends on the thermal history (melting, annealing, etc.) and processing. The microstructural scale is the scale where materials scientists optimize and design properties using empirical rules and intuition.

Understanding the microstructural scale requires a very large range of length scales, from atomistic (where impurity and grain boundary effects are important) to macroscopic (where the interactions among dislocations and long range strain or magnetic fields are important). The question then becomes how can we tailor the microstructure given the very large number of variables in the problem? Experimentally, the length scale of 0.1 to 10 microns is especially critical because it corresponds with the length scale of microstructural features (such as grain or crystallite size) in functional materials that may be systematically varied and controlled via conventional materials processing so as to optimize the performance of materials. Large-scale simulations will be able to predict the behavior of materials in this critical length scale.

Therefore, much of the initiative in materials science will involve multiscale modeling because many crucial materials discovery issues involve properties at a variety of length scales. As one illustration, consider the case of high-temperature materials, such as metallic alloys and ceramics with melting temperatures above 2000° C, which are important because the efficiency of any combustion driven process increases with temperature. Often these materials are brittle or difficult to fabricate and finding ways to ductilize and form them has a large potential benefit both to the economy and to the environment. A second, similar illustration is provided by lightweight alloys and polymers, which have great potential to reduce the weight of cars and light trucks without compromising safety. Multiscale modeling plays a critical role because both the chemical nature of small alloying additions (atomistic scale) and the microstructure (e.g., grain size and orientation) are important. First principles calculations can predict the bonding, crystal structure, and impurity effects which will determine the ductility as well as the most likely precipitates that act to pin dislocations and determine mechanical strength. Calculations at the atomic scale using effective potentials to treat 10⁶ to 10⁹ atoms can yield insights into the likely microstructures that result from various processing steps. Microstructural scale models provide understanding of deformation and grain size effects. Adding in the range of possible alloying additions, concentrations, and processing history produces a problem so large that it can only be explored with large-scale simulations. These calculations, in conjunction with synchrotron x-ray and other experiments can also predict the corrosion behavior and the need for additional surface coatings to prevent corrosion. Finally, continuum modeling using the finite element method can provide design engineers with the likely behavior of different microstructures and alloying effects in such important areas as crash testing.

A third illustration is provided by magnetic materials, which are widely used in energy conversion (e.g., electric motors consume 64% of the electricity generated in the U.S.), transportation (e.g., a typical car has more than 40 magnets) and in data storage. These materials are not achieving their theoretical potential because the microstructure is not ideal. For example, a single crystal of Nd₂Fe₁₄B is a poor permanent magnet, but with controlled introduction of defects has become one of the most widely used of magnetic materials. A modeling effort which coupled a microscopic understanding of the fundamental magnetic properties with realistic models of the microstructure including the effects of grain boundaries and the pinning of magnetic domains could lead to the discovery of new even better magnetic materials.

As an excellent illustration of the need for a computational materials search process consider the case of solar cells, which have great promise to provide electricity, generate fuels (such as hydrogen), and impact chemical processing. Despite 50 years of research, current cells are manufactured from a handful of materials, achieve efficiencies of 30% at best, and are too expensive to displace conventional sources in most applications. Expanding the number of materials is an astronomically large task given the impossibly large number of combinations of elements, crystal structures, and processing steps available - especially with new molecular beam epitaxy techniques. A concerted computational combinatorial materials discovery effort would accelerate the discovery and testing of new photovoltaic materials.

Other materials do not require multiscale modeling but instead need electronic structure calculations to understand the effect of various substituents on the electronic properties. For example, batteries and fuel cells have the potential to provide the power, energy storage requirements, and cycle life for the next generation electric cars and small electric turbine engines for use in the power generation industries. However, several essential problems must be resolved before these materials can be used in industry, including the effects of chemical substitution on the disorder and crystal structure. These require treating many electrons and can only be addressed by large-scale first principles electronic structure methods, as was recently demonstrated by the discovery of a new class of electrode materials based upon the predictions of such calculations.

The present initiative would supplement and extend those of the DOE 2000 materials co-laboratory and would overlap strongly and benefit from the ASCI efforts in multiscale modeling of materials in weapons systems. Indeed, an ER computational science initiative would accelerate progress to coincide more with the time at which the ASCI and other stockpile stewardship programs of DOE will need results. Computational materials sciences will also impact the productivity of major facilities supported by DOE and other agencies such as the four advanced synchrotron light sources, atomic and electron microscopes, and neutron sources. These increasing detailed experimental measurements of materials properties often require the results of equally complex and detailed calculations for effective interpretation.

Accelerating progress in computational materials science will require multiple institutions to make advances across multiple disciplines with a common focus. Significant scientific and modeling challenges will require universities to interact with enhanced scientific efforts in the DOE laboratories. As such a new management model is proposed that involves large-scale inter-laboratory cooperation with the computational materials modeling community acting organically to solve major problems. Specifically, we propose a "Computational Materials Science Network" as an integrated management approach. This model preserves the traditional strengths of the materials community while recognizing that making optimal use of a multi-teraflop computer will require some fundamental changes in the way that this type of research is performed. It involves large multidisciplinary teams, the development of standardized and optimized codes, central human, computational, visualization, data storage resources. It involves a community-wide management approach that can dynamically set focus areas and assign appropriate financial resources to accomplishing goals. Finally, the network provides organizational structure through which the materials community goals for the development and usage of high-end computational resources can be articulated.

Success in this effort will require multidisciplinary teams that include computational materials scientists, computational molecular scientists, computer scientists, and applied mathematicians. A few examples of the needed contributions include:

Such a program would have tremendous scientific impact. Currently, there is no full understanding how the mechanical or magnetic properties of engineering materials are determined by the underlying structure of individual grains. In addition, we do not fully understand how the chemical composition and thermal history determine the structure of the grains themselves. Macroscopic properties such as stress-strain relations, plasticity, strain rate, or magnetization are complex averages over such grains. These properties are currently represented by phenomenological parameters fitted to experiments. However, to advance the field a deeper understanding is needed. Teraflop (Tflop) computing provides for the first time sufficiently realistic simulations to contribute to this advance.

At the smallest length scales, we do not fully understand how individual electrons interact to produce such complex phenomena as magnetism or superconductivity especially in oxides, polymers, and so-called bad metals. There is currently a gap between the theoretical description of such systems and that of ordinary metals and semiconductors where the local density functional method is appropriate. The only known technique to solve this quantum many body problem is statistical Monte Carlo sampling. Because Monte Carlo methods involve averaging over independent samples, they are intrinsically suited to parallel architectures. Thus with massively parallel multi-teraflop machines, we can expect significant new insights into the many body problem – the fundamental problem of condensed matter physics.

In education, we note that high performance computing and simulation form the third aspect of a comprehensive scientific research program along with theory and experiment. Thus the next generation of materials scientists must be well trained in large-scale simulation in order to discover produce the materials of the 21st century.

We emphasize that as a discipline, computational materials science is ready for such a simulation initiative. Researchers have demonstrated their ability to utilize parallel architectures similar to those envisioned for a multi-teraflop machine. In addition, they have already developed algorithms that scale nearly linearly with system size for both electronic structure and molecular dynamics. Achieving the goals outlined here is most dependent on increases of 10-1000 in computational power.

To summarize the goals of this computational initiative, we point out that three recent trends in science and technology are converging which make it timely and urgent to embark on this new branch of materials science:

To facilitate the development and integration of the computational methods, we propose a "Computational Materials Science Network." This revolutionary capability will enable us to build a third pillar in materials research in addition to the existing ones in experimental and theoretical materials science.

All modern solar cells, other than Si-based, require as the active material one or more semiconductor alloys. Consider for example, the structure of an AlGaAs/GaAs solar cell. Surprisingly, while we know what properties are required of the active alloy material for ideal performance (see below), laboratory search for this elusive material continues. We propose to combine modern advances in electronic-structure simulation theory with high-performance computers to (i) search for and discover the ideal photovoltaic semiconductor alloy(s), and (ii) predict its relevant properties, so as to optimize its performance, thereby reaching high (30%) and ultra-high (~50%) solar energy conversion efficiency.

Perhaps the most intriguing aspect of the spectacular success that semiconductor-based high technology has had in the past 50 years is the tiny number of species (core materials) on which these technologies are based. Even considering a broad range of semiconductor devices (e.g., transistors, computer chips, solid state lasers, detectors, solar cells, light-emitting diodes, etc.), one finds only on order of ten basic semiconductors that enable these technologies. Moreover, all belong to the same crystal type! For photovoltaics, the number of known active materials is even smaller (e.g., Si, GaInP, GaAlAs, CuInSe₂). This is a strikingly narrow material base, considering the number of materials used in other technologies; e.g., the 10³-10⁵ species used in metallurgy, polymer technologies, biotech, and the pharmaceutical industry. The currently used "high tech" semiconductors also provide but a limited set of relevant materials properties, such as band gaps, lattice constants, effective masses, and mobilities.

Of course, there are good historical reasons for this narrowness of material base, ranging from the stringent criteria that electronic devices place on material perfection and purity, to the natural human inertia associated with the large investments that have been made in the first semiconductor to work in a big way (silicon). Thus, it is entirely possible that we are currently missing the crucial breakthrough material for efficient solar energy conversion! Given, however, the remarkable progress in our ability to grow high-purity artificial structures, (even in defiance of conventional equilibrium thermodynamics), and the increasing need to diversify materials properties in new device architectures, one wonders whether time has come to take a bold, systematic look at enlarging the database of potentially useful photovoltaic materials and structures.

The obvious approach to this need is to use educated, phenomenological trial-and-error techniques that have brought us, among others, new superconductors, ferroelectrics, and quasicrystals. It is almost certain that a combination of such Edisonian approaches with a considerable amount of "guided luck" will continue to provide us with exciting new materials. There is, however, a possible complementary approach: use of solid state theory and high-performance computer simulations as a guide to selecting promising new materials. Indeed, while our theoretical understanding of the microscopic makings of ferroelectricity, quasicrystallinity, and unconventional superconductivity is yet to reach the state of a-priori material-predictive ability, our theoretical understanding of the structure versus function relationships underlying semiconductivity is considerably more advanced. We thus propose to use such theoretical approaches to identify and characterize new breakthrough semiconductors for efficient solar energy conversion.

A number of properties are required of candidate PV materials and device structures. Among the most essential are:

Of course, the proposed materials should meet as many of these desired properties as possible. For more details, see Special Issue of the Journal of Electronic Materials, Vol. 27 No. 1, pages 1-65, containing the conclusion of DOE's - "Research Assistance Task Force" on "Research Opportunities in Photovoltaic Semiconductors."

Observe that after 40 years of laboratory empirical search for new materials, we are still working with less than 10 candidates!

Electronic structure theory of crystalline semiconductors has reached a stage where one can predict reasonably accurately the electronic, elastic and optical properties, once the structure of the material (i.e., crystal type) is given. However, we do not know how to select, from the many possible crystal-types of a given alloy system, the one that is the most stable. This can be recognized by noting that even in a simple binary alloy A(x) B(1-x) [where, for example, A = GaAs and B = AlAs] there are 2^N possible atomic configurations if there are N possible lattice sites, and that even for modest N values, this is an astronomic number of possibilities. Yet, the structure matters a lot! For example, the band gap of the GaAlAs alloy at 50%-50% composition can change by 0.5 eV if the alloy is random, or if it has the chalcopyrite, or the CuAu, of the CuPt crystal structure. In fact, in the latter structure, the band gap is "direct" (i.e., strong absorption), while in the former structures it is "indirect" (i.e., weak absorptions). Thus:

Challenge No. 1: Find a systematic way to explore the astronomic space of structural possibilities of semiconductor alloys, in search of the most stable configurations. This entails quantum-mechanical total energy calculations on a large number of crystal types.

Challenge No. 2: Given a stable configuration and composition, predict its photovoltaic properties (see section 3 above). This entails prediction of optical and transport properties of crystalline, but possibly configurationally random (or partially ordered) alloys of a given structure and composition.

Challenge No. 3: For polycrystalline materials (see item b of currently available materials above), predict the effects of the microstructure of a given alloy system on its photovoltaic properties. This will include phase-separated domains, local clustering and composition fluctuations. Of particular interest here is the understanding of nanostructures such as semiconductor dots that are present within the active layer. This entails performing electronic structure and atomic structure calculations on 100 Å to sub-micron length scales.

The best approaches to the three scientific challenges are yet to be selected. Here we outline our current understanding of the main strategic possibilities:

Possibilities include (i) genetic algorithms (e.g., K. M. Ho, et. al.), or (ii) "linear expansion in geometric objects" (LEGO), coupled with Monte-Carlo and simulated-annealing ground state search of this type of Ising-like expansion.

Challenge No. 2: Electronic Structure of Configurationally-Complex Crystalline Alloys.

Choice of "Hamiltonians" includes density-functional and beyond, tight-binding. Choice of approaches to the Hamiltonian solutions include coherent potential approximation, large supercells, Car-Parrinelo, etc.

Challenge No. 3: Alloy microstructure and nanostructure possibilities include classic or quantum molecular dynamics, continuum elasticity, etc.

A concerted effort to identify and characterize new photovoltaic semiconductor alloy(s) could lead to an immediate breakthrough in conversion-efficiency and conversion cost. Current experimental R&D in the areas of semiconductor growth and device design are so advanced that a new proposed photovoltaic material could be made and tested almost immediately. The impact on photovoltaics and energy conversion will be enormous. Furthermore, such an advance could impact other energy-related technology areas that rely on photon-absorption and carrier-collection, such as light-emitting diodes. The DOE office of Energy Efficiency and Renewable Energy (EERE) is the main government arm leading with these technologies. They have spent, on average, $100,000,000 per year in the past 20 years in support of photovoltaics.

Two major problems confronting the U.S. are the need to reduce CO₂ emissions and our dependency on fossil fuels. Roughly, one third of the CO₂ emissions in the U.S. can be attributed to vehicles. A major focus in the automotive industry, with possible near-term impact, is the development of new lightweight, high-strength materials (LWM) to significantly improve fuel efficiency and reduce CO₂ emissions. In fact, for every 10 percent decrease in vehicle weight, there is a corresponding 5 percent increase in fuel economy. A major milestone is to reduce the current weight of vehicles by 40 percent. This alone would result in a large reduction in CO₂ emissions equaling a 6.7% decrease in emissions. A major challenge is to achieve such weight reduction without compromising passenger safety, comfort, and cost. This will require an accelerated research and development effort by both industry and government in the area of materials science.

Currently, vehicles are made primarily of mild steel. The materials that offer the best potential to replace it in significant proportions include high-strength steel, aluminum, magnesium, and polymer-based and metal matrix composites. Although these materials offer significant benefits for weight reduction, each also has substantial barriers to overcome in the areas of cost, performance, design, and manufacturing.

To reduce the weight of a vehicle, almost every vehicle system and structural component needs to be reevaluated and redesigned. The general LWM vehicle tendencies are toward thinner sections, more slender components, high-strength materials, and the reduction of structural redundancy in the design. The push for a full utilization of the material capabilities places more emphasis on the behavior of the material during fabrication and service. The microstructure of a material determines its materials properties and the materials processing conditions defines the final microstructure. Therefore, a fundamental understanding of the science of LWM synthesis, processing, and performance in a wide variety of in service conditions is needed and is a necessary requirement for increasing the national energy efficiency and reducing CO2 emissions.

These needs are time critical requiring a multiscale high performance computing approach to aid in the development of LWM and thereby, accelerate the materials science research to meet the demands for increased energy efficiency within the next 10 years. In the following sections, some of the most pressing issues in the use of LWM for transportation applications are examined. The identified topics make a strong case for a computational materials science initiative in order to gain fundamental understanding of the relationship between processing and material behavior of LWM. Advances in these areas are crucial for the reliable use of LWM for transportation needs. The use of modeling at different length scales is an effective method to address and accelerate the successful utilization of LWM technology. To accurately simulate the evolution of the microstructure requires a multiscale approach since a number of mesoscale phenomena depend on microscopic properties, such as alloying effects, stacking fault energies, Warren Cowley parameters, etc. The microscopic properties would be calculated using first principles methods and would be used either directly as input or as fitting parameters in molecular dynamics and continuum simulations.

A variety of issues related to the processing of lightweight metals such as aluminum and magnesium, and the fabrication of structural components from them, needs to be addressed for their successful utilization. Increased raw material costs must be offset by the use of innovative processing and production methods. The replacement of steel by aluminum alloys in autobody applications requires the development of new processes for the production of cheap aluminum sheets. The current trend is to cast sections that are closer in thickness to the final sheet so that expensive intermediate thermomechanical processing steps can be eliminated to offset the higher raw material cost. This requires a more fundamental understanding of the solidification of complex aluminum alloys and the constitutive behavior of the evolving solidification microstructures during thermomechanical processing so that optimum process parameters that minimize variability in the sheet properties can be determined and implemented in processing.

The formability of aluminum alloys is an important area where the type of process used to form aluminum sheets is strictly determined by cost. Currently, aluminum sheets are formed by inexpensive high volume rate stamping processes. However, the formability of aluminum alloys by stamping is vastly inferior to that of steels. Significant advances are needed in understanding and enhancing the formability through novel changes in alloy composition and processing design. A phenomenon that continues to remain poorly understood in both steels and aluminum alloys is the springback of body panels after stamping. This affect results in significant losses in revenue through reworking of stamping dies. Springback is due to residual stresses in the material, which cause it to relax back into its original shape after removal of load during stamping. This problem is more severe for aluminum alloys than for steels. Furthermore, variations in the mechanical properties of the sheet from coil to coil often compound the problems making the springback issue more difficult to resolve.

The weldability of aluminum alloys is a major concern since the weldability of these alloys are, in general, inferior to that of steels. The utilization of new automotive sheet alloys requires a fundamental understanding of the weldability of these alloys. Significant advances in the understanding of the interaction of the welding heat source with the material, the rapid melting, mixing and rapid solidification of the alloys and the evolution of the weld and heat affected zone microstructures are required to increase the quality, reliability and in service use of welds.

There is tremendous potential for weight saving using lightweight materials in the engine block. Both aluminum and magnesium alloys are candidate materials. However, the high raw material cost has to be offset by increased life, reliability and performance of the engine. Surface modification of the aluminum engine bores through flame spraying of wear resistant materials is under development. A fundamental understanding of the deposition process and the influence of process variables on the microstructures of the deposit and the substrate are required to utilize the full potential of this technology.

One design concept in the powertrain of the new generation vehicle is the use of turbines for power generation because of their higher power density, high reliability, low emissions and ability to run on a variety of fuels. However, their efficiencies for low power applications (such as those required for PNGV) are inadequate. To be successful, fundamental advances in high-temperature structural ceramics, coatings, elastomers, and insulators are required. A significant problem in ceramics is the potential for catastrophic failure. To overcome this problem requires an understanding of mesoscale crack dynamics in these materials and thereby, gain important insight into the influence of microstructural variables on the fracture toughness.

High strength steels (HSS) offer an increased strength to weight ratio over steels currently used and therefore are a good candidate for reducing vehicle weight without comprising occupant safety. However, the use of high strength steel has raised several material and design issues.

The use of thinner gauge steel sheets requires the development of new autobody designs that can accommodate thin sections as well as providing for the inherent lower formability of high strength steels. Additional constraints are required in design since HSS components have to be used in critical locations where crash performance outweighs other considerations.

The time-critical needs of increasing energy efficiency and reducing CO₂ emissions place extreme urgency in the development of such designs for the future car. High performance computing is the only way to quickly test such designs for performance and thereby, accelerate the development process. In addition to the above design issues, many processing and fabrication issues have to be overcome in the near future.

One of the major issues in processing is to control the variability in mechanical properties. This requires a fundamental understanding of the evolution of the material microstructure during processing, and the influence of processing variables on the product microstructure. The forming of thin-walled tubes by hydroforming requires the development of steel tubes with adequate ductilities both in the longitudinal and transverse directions. This places a stringent limit on the non-metallic inclusion content in the steels. It is also important to understand the evolution of texture during the tube forming process. Additional fabrication issues include the weldability of HSS especially with other steel compositions as required by the design. Significant advances are needed in the understanding of weld microstructures and their influence on the performance of the body structure. Another important issue is the corrosion of thin-section members during service. Scientific advances are required in the development of new corrosion resistant coatings.

The LW composites are currently considered the best hope for providing strong, lightweight, chemically inert, and temperature resistant materials that can be 'tailor-made' for particular applications. There are a number of materials properties and processing issues to overcome in order for these materials to be technologically useful. Furthermore, a true fundamental understanding of the materials processing questions is desperately lacking. Several of these issues will be described below.

Inelastic deformation in polymers and LW composites is not well defined compared to materials such as metals that have microstructure that is chemically and geometrically ordered on a much finer scale. No recurring kinetic strain produced forms such as dislocations, disclinations, etc. can be recognized inside the transformation volumes during inelastic deformations. For example, in different polymers, shear transformations can either be dilatational or compacting which can have a dominant effect on fiber-matrix adhesion mechanisms.

Brittle and ductile-brittle transition theories for lightweight polymer composite materials and their constituents are not developed yet to the extent that they can relate the failure process to material (intrinsic) and external (extrinsic) properties. For example, the transition from stable nucleation of microcracks into unstable damage evolution can occur through development of a crack with a critical length or through development of cooperative mechanisms that lead to instability and failure. Theories are needed that can relate the inelastic deformation, localization and failure processes in composites and polymers with particular material microstructural and chemical features, their evolution and interaction with external influences for complex loading conditions. Strength of brittle LW composites also depends on the volume of stressed material and the nature of stress distribution. The brittle materials are flaw sensitive and lack plasticity effects that would reduce stress concentrations arising from defects and local fluctuations of the microstructure. There is a need for theoretical and experimental developments that would explain and develop methodologies for modeling the size effect in composites.

The fiber/matrix interaction mechanisms in LW composites are not fully understood. The interphase/interface region has chemical and morphological structure different from the bulk resin and fiber. The interaction mechanisms are difficult to quantify and may be influenced by a number of processes, including: selective adsorption of matrix components; conformational effects; penetration of polymer components into the fiber surface; diffusion of low molecular weight components from the fiber; catalytic effects of the fiber surface on polymers; surface-induced or surface-modified crystallization. The fiber/matrix adhesion mechanism is the critical characteristic of the composite. The quantitative relationships between the interface processes and mechanical properties of the fiber/matrix bond have yet to be established.

The LW composites are increasingly used for safety-critical structural applications in dynamic loading environments. It has been observed that different deformation and energy dissipation processes are active at different strain rates. The process of activation, transition and interaction between different deformation processes is not understood enough to allow for a wider structural application of LW composites.

The successful development and application of LWM depends on a fundamental understanding of the influence of the material microstructure on the macroscopic behavior and on the effect of processing conditions on the resulting microstructure. It is insufficient to just model the behavior at the continuum or macroscale although the final interest is in the material behavior at the size scale of the structural component. To accurately simulate the evolution of the microstructure requires microscopic inputs since a number of mesoscale phenomena depend on interactions occurring at the atomistic level. Therefore, a multiscale modeling approach is required to study the research and technological issues that occur over varying length and time scales. A more detailed understanding of the relationship between material composition, processing, microstructure, and properties is needed. Currently, our ability to uncover the important physical mechanisms that will enhance our scientific understanding of these complex physical process is limited by the computational resources required to simulate the interplay between these variables at the appropriate length and time scales.

The lack of computational resources has resulted in the development of theories and methodologies that incorporate a number of approximations. Furthermore, this has resulted in a number of constitutive relations (i.e., crack propagation) that are based on heuristics rather than on a sound theoretical underpinning. However, the next generation massively parallel computing architectures would provide the necessary computational tools to develop models and constitutive relations that are more realistic, and equally important, incorporating new advanced applied mathematical techniques to render a numerical solution.

To obtain realistic properties from first principles methods that can be used either directly as input or as fitting parameters for larger scale simulations requires simulations of a large aggregate of atoms ranging from hundreds to possibly tens of thousands of atoms. New first principles full potential electronic structure and molecular dynamics algorithms that scale linearly with system size and floating point operations need to be developed in order to realistically perform simulations of such large collections of atoms. A new method based on a wavelet expansion that results in a sparse formulation theoretically has the desired linear scaling.

New sophisticated parallel computational methods to treat large-scale interacting defects such as cracks and voids in three dimensions in the continuum/mesoscale level are needed. This will require the combination of several advanced mathematical techniques. First, finite elements are needed to treat non-linear problems such as plasticity, crack tip analysis, large deformations, etc. Second, boundary integral methods are an effective method for treating linear problems where a major benefit being the reduction of problem dimension by one. This is especially advantageous for situations where the domain evolves in time (e.g., crack propagation and void evolution), as rediscretizing the new problem becomes considerably easier. Due to the limited applicability of the integral equation approach, effective hybrid code containing the finite and boundary integral method will be essential. Third, the level set method is a powerful alternative to traditional front tracking techniques because it is capable of following discontinuous changes in topology, splitting or merging of surfaces, and singularities (cusp formation). Boundary integral methods have the capability of handling complicated geometries, but the calculations are expensive. The fast multipole method provides a highly effective technique for significantly reducing the computational cost for large-scale problems. The computational resource requirements for typical continuum/mesoscale methods are provided below.

For example, a fundamental understanding of the dislocation dynamics during thermomechanical processing requires multiscale modeling of the alloy system. Current developments on warm forming of automotive aluminum alloys rely on a judicious choice of alloying elements. The effects of alloying must be simulated from first principles. Once an appropriate alloy candidate is identified, parameters for larger scale modeling can be obtained from the first principles calculation. The process temperature and strain rate interact with the evolving dislocation structures to produce high strain rate sensitivity. Parameters for the constitutive relations used in the continuum model for dislocation structures and strain rate can be obtained by a combination of first principles and molecular dynamics simulations.

Accurate quantification of the solidification microstructure requires modeling at both the atomistic scale and the mesoscale, and a coupling of the phenomena occurring at these length scales. For example, the relationship between velocity and undercooling during rapid solidification of a multicomponent alloy system cannot be readily obtained using existing computational capabilities. Atomistic models that capture the fundamental phenomena of atom transport across a moving interface are required. These relationships can then be coupled to mesoscale models that describe the evolution of the mesoscale in terms of the dendrite size, shape, interdendritic spacing, volume fraction and spacing of second phase, etc. These mesoscale variables determine the mechanical response of the cast material to thermomechanical processing.

The various issues discussed above point to the need for a computational materials science initiative to take the lead in developing innovative simulation methodologies applicable at the appropriate length scales to study the processing and fabrication issues. A combination of continuum and atomistic methods would provide a powerful method for simulating deformation and recrystallization. The finite element simulations of the deformation process at the mesoscale provide the quantitative data on the orientation and slip resistance of the grains that can be used as inputs to a Monte Carlo (MC) simulation of recrystallization. Recrystallization is a phenomenon that occurs during heat treatment following the deformation process and that is driven by the reduction in the stored energy of deformation. The simulation requires initial values of the orientation and stored energy of the MC sites used to represent the grain structure. The novel feature of a coupled method is that in addition to predicting the kinetics of the recrystallization process, it is able to predict the resultant texture based on the final orientations of the sites, as well as the final microstructure.

The computational requirements for these methods are very similar to the previously discussed hybrid model. For the microstructure to be truly representative of bulk material behavior, it is necessary to consider a statistically significant number of grains that would be of the order of a million. Discretization of each grain having ~1000 elements translates to a mesh size of 10¹² elements. If one then adds the complexity of the crystal plasticity model, the problem quickly becomes intractable without significant increases in both computational resources and algorithmic developments.

Assume a 42 Tflop machine consisting of 42,000 one-Gflop processors each with 128 megabytes of memory. The machine memory is 4.2x10⁴ x 1.28x10⁸ = 5.4 x 10¹² bytes or 5.4 terabytes

For complex geometry in 2D solving LAPACK equation requires a dense matrix of dimension N=6600 (348 Mbytes). The time to solve for one time step on a single IBM RS6000/590 with 512 Mbytes of main memory is 373 seconds. LAPACK routines were used to perform the linear solve running at 150Mflop/s (LAPACK peak 168 Mflops). In addition, due to the complex geometry it takes nearly an equal amount of time to set up the matrix.

To perform a 3D calculation for fracture in typical ceramic material (i.e., Si-N) consisting of multiple cracks in a complex microstructure requires the construction of a matrix of dimension 500,000<N<750,000. Extrapolation of the above values for the storage of the N=660 matrix yields a total required storage of 2-4.5 terabytes. Hence, the problem is well suited for the total memory on the machine (5.4 Terabytes).

Machine	Peak Performance/Node	Time per time step per node
IBM-590	330 Mflop/s	373s
MPP on each node	1Gflop/s	123s

2/ the problem is well suited to iterative methods with preconditioning resulting in an algorithm that scales as N²;

Time for 1000 steps

on 42,000 Nodes

each one is 1GFlop/s

750,000

10.3 hrs

500,000

4.7 hrs

It is apparent that this approach would provide fundamental insights into microstructural evolution during processing that, in turn, could be used to tailor processing conditions to obtain the desired final microstructures. The availability of MPP Tflop computing power will make this possible.

Progress in developing increased energy efficiency in such strategic applications as jet engines, turbine power generators, rockets, hydrogen production reactors, etc., is controlled in large part by the development of structural materials with greatly improved properties at increasingly higher temperatures. Materials for high-temperature applications require a mixture of strength, toughness, creep resistance, oxidation/corrosion resistance, ability to handle thermal shocks and fatigue, etc., combined with adequate response at low temperatures. Devising materials with the needed properties balanced with the need for low weight for energy efficiency is a challenging task involving physical phenomena that span from the atomistic control of interfaces and impurity properties to the linkage between continuum response and the materials microstructure. The traditional Edisonian approach of trial-and-error materials development will lead to evolutionary advances, not the revolutionary changes needed for rapid improvements in required materials properties - modeling and simulations that couple details on an atomistic level with overall materials response are essential. To span the needed length and time scales will require huge increases in computational capabilities, as outlined below.

Candidate materials depend on the temperature regimes for the application. Typically, for temperatures in the 800-1000 °C range, nickel-based superalloys can be used. These alloys have excellent oxidation and mechanical (strength, toughness, and creep) properties, in addition to being cost effective. They suffer from rather high density and that for some applications, for example turbine blades in aircraft engines; they must be cooled, which can require complicated and expensive machining. One way to extend these materials to higher temperatures, in addition to cooling, is to put on thermal barrier coatings, which are typically ceramics. Issues include the adhesion of the coatings as well as their strength and toughness. Recently, certain aluminides have been suggested as possible replacements for the superalloys, largely because of their lower density. However, they suffer from poor oxidation resistance and must be coated. Certain fiber-reinforced composites have also been proposed for such temperature ranges.

For temperatures above about 1000 °C, there are severe requirements for oxidation resistance. Given the desire for low densities combined with the properties listed above, the best candidate materials seem to be silicon-based ceramics, aluminides, and silicides. None of these materials is ideal, and all suffer from brittleness over at least some part of their temperature range. For example, the aluminides are brittle at room temperature, have low strength and creep resistance at high temperatures, and lack a long-time oxidation/corrosion resistance at temperatures greater than about 1200 °C. Silicides have better high-temperature properties (while maintaining low densities) yet are brittle at low temperatures. The goal is to have lower-density materials that can be used at higher temperatures to achieve the consequent energy savings.

The key issues for modeling and simulation are how to design materials with a balance between the various requirements for materials properties. Exactly the approach depends on the materials classes of interest. For example, the use of superalloys is restricted because of their melting points - use over about 1000 °C is unlikely. Thus, we are must cool and/or insulate them. Insulation has been done with thermal barrier coatings, which are made through atomistic deposition processes yet develop specific micron- sized microstructure. How atomistic processes on the scale of Å and ps lead to micron-sized features is unknown.

Perhaps the most promising of materials are the ceramics, silicides, etc., all of which suffer from a lack of fracture toughness at low temperatures. Fracture toughness is controlled by a number of physical phenomena, including relative amounts of plasticity, improvements in interfacial properties, impurity levels, etc. How to control fracture toughness is not well understood. Modeling and simulation offer the possibility of connecting the macroscopic response to the underlying atomistic processes. Again, these simulations will require enormous computational resources to link the various scales.

Typical engineered systems are cm to m in size, yet their properties depend critically on their microstructure, which is generally on the order of µm. Controlling those structures, which in turn depend on the underlying atomistic behavior is a critical part of the manufacturing process. Simulating over such scales presents a formidable challenge. As an example, consider a casting process that is often used in the creation of high-temperature parts made of superalloys.

The computational challenge arises from a number of issues. First, the geometries are inherently three-dimensional. The prediction of microstructure requires a scale ratio of about 1000 over current simulations, i.e., approximately10⁹ computational cells or elements. Based on current codes, a full simulation of a casting including flow, solidification, microstructure, etc. would require approximately:

Physical requirement	Computational requirement
N=10⁹ cells with 500 words per cell	4 terabytes of memory
1000 s of "real time"	10⁴ time steps
full physics of fluid flow, heat transfer and solidification	10 conjugate gradient solves per time step

Putting these together yields 10 CG solves times (10⁹)^1/3 iterations time {60 10⁹ flop per iteration} times 10⁴ time steps for a grand total of 6x10¹⁸ flop per calculation. On a 40 Tflop machine, this would take 42 hours. These figures are undoubtedly an overestimate of the required resources. Advances in algorithms are expected to shrink the computational requirements. However, the scope is reasonable.

The scientific umbrella covering magnetic materials is enormous, ranging from basic research on atomic scale memory for quantum computing to disk drives, motors, transformers, sensors, and new media for refrigeration. The annual market for magnetic systems is $150 billion dollars, with $50 billion just for magnetic recording. Many of the other applications relate to energy efficiency. For example, 35% of the energy consumed in the U.S. is used to produce electricity, and of this, electric motors currently use 64% of the electric power generated. Improvements in efficiency can be achieved through size and weight reductions and by the use of new permanent magnetic materials with higher energy densities. The key to the development of improved magnetic materials for motors and for high technology devices is the understanding and control of the microstructure at small length scales. For permanent magnets the microstructure involves the size, shape, and orientation of crystallites of the compound and also the nature and distribution of secondary phases and impurities, which control domain-wall formation and motion, and hence determine the magnetization and demagnetization behavior. For example, a single crystal of Nd₂Fe₁₄B is a rather poor permanent magnet, but with pinning sites introduced by rapid thermal quenching, this material becomes one of the new 'miracle' permanent magnets that is revolutionizing our society. The high technology applications are dominated by disk drives where the area assigned to a single bit of information has decreased in size to the point where fundamental magnetic interactions at mesoscale lengths are being encountered.

This is a particularly opportune time to initiate a major project involving magnetic materials modeling. There have been major discoveries in the last ten years that have led to a renaissance in the field of magnetism (i.e., witness the overflow of abstracts to magnetism conferences). Spin valves, giant magneto-resistance, memory devices, new permanent magnetic materials, new magneto-elastic materials, new magneto- thermal materials, and new magnetic molecules have all played to packed conference rooms and led to numerous publications. Theoretically, there have breakthroughs both at the microscopic and microstructural length scales that if properly exploited can lead to a new modus operandi in magnetic materials development.

As in many of the materials categories discussed in this initiative, relevant magnetic properties for bulk size applications are determined by phenomena occurring at mesoscopic or even microscopic levels; and in order to realistically and accurately simulate these magnetic properties it is necessary to address a range of length scales. We propose the integration of first principles methods for the accurate evaluation of basic magnetic interactions (1000's of atoms level) with micro-magnetics techniques for the complete simulation of technologically important materials and devices. Such integration would obviate the need for much of the expensive trial and error approaches currently necessary to optimize the properties of magnets used in commercial applications. At the quantum end, methods that are currently capable of accurately handling a few tens to a few hundreds of atoms must be extended to thousands in order to describe the exchange interactions near grain boundaries, dislocations, interfaces, impurity concentrations, surfaces, etc. First principles spin dynamics methods must be implemented to treat time dependent phenomena and particularly the effects of finite temperature, an important consideration in all applications. From the other end, mesh sizes used in micro-magnetics calculations must be reduced such that three-dimensional simulations of complex microstructures and device geometries can be performed at near atomistic resolution. Techniques must be developed for extracting from the fine grained, first principles calculations the necessary parameters, and the relevant terms in the model Hamiltonians required for the micro-magnetics simulations. Furthermore, the general multi-scale modeling of microstructure envisioned elsewhere in this materials initiative involves the treatment of mesoscale materials features such as dislocations; and the results from these modeling efforts feed directly into the needs of the magnetic modeling. That is, magnetic properties can be optimized by understanding and developing processing techniques that can control the microstructure. Thus, the magnetic part of this initiative is intimately coupled with the whole research effort on materials.

Approaching the material from the atomistic or electronic level where the basic magnetic interactions originate, recent progress in performing first principles spin polarized calculations on massively parallel computers has pushed the viable system sizes to 100's of atoms (1000 maximum). Among the major accomplishments is the development of first principles methods that scale linearly with the number of atoms (order N) and allow non-collinear magnetic order to be treated. In addition, there are related new methods to allow temperature to be incorporated into these calculations, e.g., moment directions and magnitudes vary. Hence, spin dynamics is being developed in a manner similar to the very successful molecular dynamics approaches that make up an important part of this materials initiative. These new methods are indicative of the progress made by the electronic structure community in exploiting the power of succeeding generations of vector and massively parallel computers. On the continuum side, the micro-magnetics (m-m) community has extended their techniques to smaller distances in order to model nanoscale particles and structures. Thus, the systems that are both interesting and viable for simulations by the two communities are beginning to overlap. With multi-teraflop computing, the first principles calculations can be extended into a qualitatively new regime to give exchange constants and anisotropy parameters near extended so that coarse graining of the atomistic level calculations can lead directly to the parameters used in the m-m simulations. Currently the parameters used for the m-m calculations are empirical and typically obtained from experiments on bulk materials when feasible. The coupling of the two communities offers a tremendous opportunity for bringing together the basic and applied sides of the whole magnetic research effort for truly synergistic interactions.

Traditionally the first principles electronic structure and micro-magnetics communities have worked largely independently of each other. Thus, a fundamental cultural change is required if a fully integrated approach is to succeed. The formation of an integrated magnetic materials community naturally fits the general strategy of a 'Materials Simulation Network' with an 'Interest Group' focusing on magnetic materials envisioned within the materials simulation effort.

Magnetism at the fundamental atomic level is notoriously difficult to probe with small laboratory experiments, but major neutron scattering facilities, pioneered by the DOE, have been invaluable in providing the kind of detailed microscopic information needed for the validation and motivation of breakthrough theoretical efforts. More recently, the advent of high intensity synchrotron sources has allowed the magnetic x-ray scattering to come into its own as a technique which can provide information complementing that obtained with neutron scattering. The theoretical work proposed in this initiative is critical for the interpretation of magnetic scattering results on complex structures (e.g., magnetic multilayers used in spin valves). There are already strong interactions among the scientific groups which would participate in this magnetics simulation initiative and groups located at the major DOE synchrotron and neutron scattering facilities. The facility's groups need sophisticated modeling for interpretation of their data taken on the complex materials systems that are relevant for modern science and technology.

A major focus in the power generation and automotive industries is the development of new lightweight rechargeable high energy density batteries. CO₂ emissions in the U.S. come largely from vehicles (1/3) and the power generation industries (1/3). This technology would lead to a significant reduction of CO₂ emissions as well as drastically reducing this country's dependence on fossil fuels. For example, lightweight rechargeable batteries are crucial to the advancement of electric cars, photovoltaics, and fuel and solar cells since each of these devices requires a power source and/or an energy storage device.

The important properties of a battery are voltage, power, capacity, cycle life, and cost. The two main materials in the development of new batteries are lithium and nickel-metal hydride based compounds. The nickel-metal hydride based materials are considered a mid-term technology that potentially could double the range and performance of electric vehicles compared to other battery technologies used today. On the other hand, the lithium-based batteries are a long-term technology that offers the potential for higher energy and power densities at a lower cost. Furthermore, it is believed that they have the potential to provide enough power to make the range and performance of electric vehicles comparable to that of today's gasoline-powered engines.

These properties are determined largely by the material used to make the cathode. Although the cathodes determine the important technological properties of rechargeable batteries, there have been no satisfactory theoretical calculations that provide adequate guidelines for improving the properties of these materials or for developing new materials. Recently, it has been shown experimentally that partial substitution of a second transition metal in the lithium based compounds LiMn₂O₄ and LiCoO₂ can increase cell voltage, capacity, and cycle life. These results, however, have been achieved largely through trial and error. Considering the number of transition metal ions that can be partially substituted in these compounds, investigating all possible compositions experimentally is impractical.

The lithium based systems have the potential to provide the power, energy storage requirements, and cycle life for the next generation electric cars and small electric turbine engines for use in the power generation industries. However, several important problems must be resolved before these materials can be used in industry. These problems include the effects of substitutional disorder on the open circuit voltage and equilibrium crystal structure. The openness of the equilibrium crystal structure determines the power through mass transport, and the cycle life determines the battery lifetime for in service use.

To address the main research issues of lithium-based cathode materials a multiscale approach is needed since the physical phenomena important to batteries occurs on various length and time scales. Simulations must be performed on the microscopic scale for the electronic structure, on the mesoscopic scale for mass transport and continuum level calculations for the cycle life that occurs on a time scale of thousands of hours. The important physical properties needed to develop new batteries depend on the underlying electronic structure of the materials. In addition, both mesoscopic and continuum level simulations need electronic structure calculations to generate reliable inputs for the parameters making up the constitutive equations.

The cathode materials all contain some form of substitutional disorder such as, alloying, antisite defects, and vacancies. The disorder affects the open circuit voltage, an important technological property for batteries as well as determining the equilibrium crystal structure. In fact, certain cathode materials, such as LiNiO₂ based compounds, have some form of disorder on all the sublattices comprising the material. There can simultaneously be both vacancies and antisite defects on the lithium sublattice: vacancies on the oxygen sublattice and alloying on the metal sublattice. Furthermore, relaxation is very important in these systems because it also affects the open circuit voltage. To address the effects of disorder and relaxation on the open circuit voltage requires the calculation of the electronic structure of the cathode material. Due to the large amount of disorder and relaxation effects occurring simultaneously, large unit cells comprised of at least hundreds of atoms must be used in the simulations. A new wavelet based electronic structure method that scales linearly with system size and floating point operation is one potential method that may be employed to efficiently and accurately address these problems.

Disorder affects mass transport, which determines the power through the openness of the equilibrium crystal structure and influences the cycle life of a battery. To simulate the mass transport requires first principles quantum molecular dynamics and/or classical molecular dynamics. A new wavelet based quantum molecular dynamics simulation is a potential method for addressing this problem and generating input parameters for classical molecular dynamic simulation of mass transport. Continuum level calculations are necessary to simulate the cycle life of a battery to determine whether it is suitable for production usage. Again first principles methods combined with molecular dynamics simulations would be used to generate input parameters for continuum level simulations.

Catalysts are materials that increase the rate of conversion of reactants to products during a chemical reaction without being permanently altered. The number of inorganic materials know to be catalytically active is enormous; touching nearly every aspect of our material lives. Today, the discovery of new materials capable of catalyzing important reactions is largely advanced through a trial-and-error procedure that is occasionally guided by the current generation of theoretical methods. The best way to escape the combinatorial morass associated with today’s experimental Edisonian approach to catalyst discovery is to integrate the next generation modeling and simulation tools into a rational catalyst design process. Of course, this is not a new idea; the hope of multi-teraflop computing platforms, however, places this vision within the realm of reality. Such computer power will enable more realistic, atomic-level models of catalysts to be created and simulated in virtual environments. This, in and of itself, will lead to breakthroughs. Unfortunately however, a truly integrated rational catalyst design program can not be realized by simply building computers that are more powerful. Although progress can be made over the next five years with evolutionary changes in today’s theoretical methods, science must create a new generation of molecular and solid state quantum mechanical methods capable of more accurately describing many body physics at finite temperatures to make the next step. This new generation of methods must accurately evaluate the energy differences between the various moieties in a chemical reaction in order to predict the kinetics of the reactions under investigation. Errors present in today’s methods are so significant that predictions of the kinetics at or above room temperature are often meaningless or semi-quantitative. The ability to simulate the function of a catalyst under realistic temperature and pressure conditions using accurate models would change both the rate at which catalysts are discovered and lead to the design of more efficient and selective materials. Although, there are many candidate problems, the following examples were selected based upon their potential impact on the nations long range energy strategy and the degree to which simulation and modeling strategies could revolutionize the discovery process. Each of the examples below would profit from more accurate theoretical methods capable of predicting the kinetics of the various molecular scale processes associated with catalysts and device operation.

Electrocatalytic Materials are capable of both conducting charged species and catalyzing specific chemical reactions. Fuel cells are one of the most important energy related devices that require this technology. There is a need for improved electrocatalysts to dissociate and recombine the molecular fuels used in the fuel cell for either polymer electrode membrane (PEM) or solid oxide fuel cells (SOFC’s). In fact, one of the principle technical barriers associated with the widespread use of PEM fuel cells is the high cost and undesirable de-activation characteristics of Pt metal electrocatalysts in use today. For both types of fuel cells, catalysts that are stable and active at elevated temperatures and under electrochemical operating conditions must be identified and deployed. Simulation and modeling would be used to identify compositions and structures of candidate catalysts. The most promising of these would then be fabricated and tested under anticipated operating conditions. Theoretical investigations would elucidate gas/catalyst interactions such that reactant bond activation and scission could be controlled and would focus on understanding the transport charged atomic species across the catalyst/electrolyte interface and through the solid electrolytes. The engineering goal of these studies would be to selectively optimize the dissociation and recombination kinetics and transport rates of the charge carriers. An equally important aspect of this work involves designing electrocatalysts that neither promote undesired reactions nor are deactivated by the presence of trace contaminants in the fuel. Today, only a small fraction of these important subjects can be investigated due to a lack of computing power and suitably accurate theoretical methods. The development of portable fuel cells (for use in transportation) and large-scale electrical generators through discovery of new electrocatalysts would revolutionize the generation of clean energy.

Photoelectrocatalytic Materials are perhaps best understood by example. One important application is a device capable of utilizing solar energy from the visible portion of the electromagnetic spectrum to directly catalyze the decomposition of water into molecular hydrogen and oxygen. (Contrast this with generating electricity using a photovoltaic and subsequently using electrolysis to decompose H₂O into hydrogen and oxygen.) In essence, such catalysts would directly generate a useful fuel (H₂) from an abundant source (H₂O) using energy provided the sun. The hydrogen resulting from this process could be used to power portable and large-scale land based fuel cells or to enrich olefins derived from fossil sources for more efficient and clean consumption in combustion devices. Such materials are known to science today, however their current efficiencies (2-5%) would require a prohibitively large collection and processing devices. It is a challenging proposition to design a material that must overcome the difficulties currently being addressed in photovoltaic research while simultaneously understanding how to catalyze the decomposition of water. The technical challenges faced by this field of research are similar to those outlined above with the additional complexity associated with using solar energy to drive the water decomposition. The role of simulation and modeling in the design process would be to increase the rate at which candidate compositions and structures were explored and to provide the general rules and principles associated with the operation of these devices. Today’s electronic structure methods are well positioned to begin this exploration while the methods and computational tools to be developed in the next two decades may be able to solve it.

Heterogeneous Catalysts: One of the principle advantages of heterogeneous catalysts is that the materials are frequently of low cost and the processes that utilize them often produce less toxic liquid waste than homogeneous analogs. Since by definition, these catalysts function in multi-phase environments, understanding the complex interfacial physics associated with gas/solid, liquid/solid ad solid/solid interfaces is essential to making progress in any rational material design process. Simulation and modeling techniques provide a uniquely controlled environment were the details of the molecular steps involved in the catalytic process can be investigated and the general principles of how these systems function subsequently extracted. Although, there are many heterogeneous catalysts currently in use world wide today it is critical that we design catalysts that are more selective, efficient and tolerant to deactivation that can be used in olefin conversion and exhaust emission control. To increase the efficiency of today’s combustion based engines and thus reduce CO₂ production, it is necessary to efficiently produce hydrogen enriched olefins and to convert such compounds to alcohols and oxygenates which can be used in next generation heat engines. Once combustion takes place, it is necessary to scrub the exhaust stream to prevent undesirable combustion products such as NO_x and SO_x from being released into the atmosphere. The technology to build high efficiency (45%) diesel engines exists today. The widespread deployment of these vehicles is largely being delayed by our inability to produce reliable catalysts capable of converting nitrogen and sulfur oxides into harmless molecular species. During the next several years, simulation and modeling technologies could be directly applied to understanding nitrogen recombination in microporous media and to uncover the principles of how non-thermal plasma activated catalysts function. The resulting knowledge is the crucial step needed to begin designing materials with optimal characteristics. Although interfacial physics has been studied at the molecular level for several decades additional emphasis should be placed on this general field of research in combination with the previously mentioned need for more accurate modeling and simulation technologies.

A primary goal of this materials simulation initiative is to develop a capability to reliably predict the properties of real materials. Such a capability would have broad impact on virtually every aspect of DOE technologies including energy generation, storage, and utilization. To achieve this far-reaching goal one must be able to realistically simulate physical phenomena on a vast range of time and length scales. The illustrative examples in this report exhibit physical phenomena that range from electronic properties of semiconductors and magnetic materials at the quantum scale to mechanical properties of microstructural units at the macroscopic scale. Applications vary from photovoltaic materials for solar cells and magnetic materials for permanent magnets to high temperature ceramics for turbine blades and lightweight materials for automobiles. High Performance Computing provides the genuine possibility for revolutionary advances in our quantitative as well qualitative understanding of materials phenomena. High Performance Computing will also substantially enhance our ability to predict new materials that will enable the technological advances that will fuel the economy in the 21^st century. Challenges exist at each time and length scale. Major conceptual and capability advances will require the development of powerful, innovative methodologies at each of the characteristic time and length scales to enable reliable and accurate descriptions of the properties of real materials.

Perhaps the greatest challenge is the simulation of essential properties that depend critically upon phenomena or processes at very different scales of length and time [1-2]. The fundamental ideas of "multiscale modeling" have been central to materials science for many decades. Figure 1 illustrates characteristic scales and Table 1 shows corresponding typical ranges of time and length. The advent of high-performance computing presents the opportunity to move beyond qualitative, phenomenological connections between the scales to the creation of unified multiscale simulation tools. Teraflop computing and energized innovative computational materials scientists focused on high-impact outstanding materials research issues will provide the needed resources to uncover the elusive connections in the hierarchy of time and length scales.

	Quantum	Atomic	Mesoscopic	Macroscopic
LENGTH (meters)	10^-11-10^-8	10^-9-10^-6	10^-6-10^-3	> 10^-3
TIME (seconds)	10^-16-10^-12	10^-13-10^-10	10^-10-10^-6	> 10^-6

Viewed at the smallest scale, the simple building blocks of materials are electrons and nuclei, which obey the quantum mechanical Schrodinger equation and interact via Coulomb forces. The complexity arises from the fermion nature of electrons and the very difficult problems of describing many interacting particles. In principle, we could calculate all the properties and structures of any material from these known physical laws. However, it is impossible to simulate directly the huge number (~10²³) of electrons and nuclei with an objective of predicting macroscopic properties. Moreover, even if the computer were large enough, such an approach would be unlikely to provide much insight into the essential physics governing fundamental properties and structural features of materials. However, as is typically the case with large systems, after suitable averaging, new collective phenomena emerge. It is possible to describe such collective phenomena by new mathematical models obeying different equations. We refer to this as "coarse graining," because it arises from appropriate averaging over many degrees of freedom at a finer scale. The computational expense of a simulation will scale with the number of degrees of freedom, the complexity of the interactions and the number of simulation steps. In the process of coarse graining, one moves up the length scales, so that one can simulate ever larger systems (while keeping the overall computational expense approximately constant) until one is simulating the macroscopic material.

Simulations can work in various modes to lead to new understanding and accurate predictions in multiscale problems. The term for one of the most far-reaching purpose is the "discovery" mode, which is the mode of uncovering new paradigms for phenomena in materials. It has always taken a great deal of creativity and innovation to uncover the appropriate collective degrees of freedom and the laws for effective interactions at the coarser level. Simulation can provide information that is difficult or even impossible to obtain by experiment and yields new insight for this creative endeavor. Simulations can serve as a laboratory through which one makes discoveries until a fundamental understanding of the elusive phenomena emerges.

A second role for simulations is to perform "benchmark" calculations that provide critical tests of the faithfulness of models. The physics of the collective modes and their interactions at a coarser scale must be compatible with simulations from theories and models at the finer scale. Ultimately the model must be consistent with the shortest length- and time-scales, the realm of electrons and nuclei. At each scale, benchmarks using the most advanced computation possible are crucial for establishing the ultimate reliability of the multiscale simulations.

With the establishment of the validity of models coupled with increasing computational power, the third role of simulations emerges. In the "calculation" mode simulations at a finer scale directly provide the needed input to models at a coarser scale. Many of the calculations described in this report have identified examples where such input is critical to the goal of predicting the properties of materials. For example, electronic structure calculations often determine the interatomic force field that is the input for large-scale molecular dynamics involving many millions of atoms. In turn, these molecular dynamics calculations can provide information on defect structures and interactions that provide the input to next coarser level; this involves simulations of grain boundary and dislocation dynamics.

The "validation" mode uses computation to simulate complex phenomena with such realism that comparison with controlled experiments can provide definitive validation of the underlying assumptions and algorithms. With new methods and computational power, one can iterate between laboratory experiments and computer experiments (with exquisite control on the variables) and theoretical model development.

Combining all these modes, the high-performance computer becomes a powerful tool to accelerate conceptual advances that can arise from an iterative process between experiment (computer or laboratory) and theoretical interpretation. The interplay between simulation and experiment the potential to accelerate dramatically the solution of materials problems. In these roles, high-performance computers will become indispensable tools that can greatly enhance the productivity of major DOE supported facilities such as synchrotron light sources, neutron sources and electron microscopy.

The most advanced level of computational power is required to solve materials problems at the cutting edge of research at every time and length scale, as outlined in this report. At the most fundamental quantum scale, the great challenge is to solve the many-body problem of interacting electrons and nuclei. For example, quantum Monte Carlo (QMC) calculations on the electron gas have provided the benchmarks for all other methods. The input from QMC calculations on the homogeneous electron gas is the basis for the density functional effective Hamiltonians and can form the basis for the development of more accurate functionals. Density functional theory (DFT) has become the most widely used method for describing electrons in real materials problems. Compared to QMC, DFT requires less computational power by a factor of ~100-1000, and hence is applicable to much wider classes of problems. Because of the development and validation of the density functional methods, there has been a qualitative change in our ability to predict properties of complex materials. It has provided the capability to simultaneously solve the electronic equations while moving the nuclei thermally or exploring to find energetically favorable and stable structures. Present algorithms scale with the cube of the size of the system, thereby limiting calculations to no more than ~1000 atoms on the largest computers in the world. A typical large-scale simulation today might be a structural optimization that involves 100 atoms. It takes on the order of 10⁵-10⁶ Fourier components (or other basis functions) to describe each electron in the system. This requires ~10 gigabytes of memory. For a complex material, it can take thousands of simulation steps to find an energetically favorable and stable configuration. This relaxation simulation can therefore take a week on a 30 Gigaflop computer. The methodology is relatively easily adaptable to parallel computation. However, the largest scale multi-teraflop computers will be essential for reliable thermal simulations of complex materials and searches for new materials. The forefront areas of research at the quantum scale are continued development of linear scaling algorithms; theoretical and algorithmic development in excited state properties of materials; and, development of methodologies for linear and non-linear materials response.

In some cases, there is a direct relationship between the macroscopic properties of interest and quantum scale phenomena. For example, one can directly compute electronic properties, such as band gaps. The band gap is a materials property that critically affects the performance of solar cells and solid state lighting. Certain fundamental magnetic properties of materials are direct consequences of the quantum electronic system. It is also possible to compute bulk phase diagrams from a combination of statistical mechanics modeling coupled with energies calculated at the quantum or atomic scale. However, even in the case that the connection between the electronic and macroscopic properties is direct, the calculation of the electronic behavior still poses a formidable computational challenge. It will require developments in computer science, applied math, and algorithms as well as advances in fundamental theory.

In other cases, simulation of the essential properties requires successive coarse graining. Atomic level simulations can span the range from the quantum description of 10²-10⁵ (with high-performance computing) atoms to the classical molecular dynamics simulations of 10⁶ - 10¹⁰ atoms. This is the number of atoms necessary to begin to simulate the next coarser scale, the atomic. The atomic scale describes such phenomena as individual grain boundaries, crack formation, etc. The atomic scale is ripe for developments because it is beginning to be possible to accurately calculate the forces between atoms, which is the necessary input to determine the atomic scale behavior. Innovative methodologies also provide examples of "embedding" in which one embeds the quantum electronic calculation in a classical atomic scale simulation to describe critical aspects of the problem. For example, the tip of a moving crack is difficult to describe purely classically because the atoms are undergoing massive breaking and remaking of bonds as the crack propagates.

One of the greatest challenges is to develop methods to determine the long time behavior of systems. Direct applications of atomic-scale molecular dynamics are limited to times in the nanosecond range even on a multi-teraflop computer. Innovative work on novel ideas should be able to extend this time-scale by a couple orders of magnitude for systems dominated by infrequent events. For simulation of processes at longer times, one must use different approaches. One is to develop a continuum level model based on a thermodynamic description of the driving forces. This requires the determination of the thermodynamic properties of defects and boundaries and the mobility of these defects. For multi-component systems, quantitative thermodynamic calculations will require large atomic-scale Monte Carlo simulations based on accurate total energy electronic structure methods. Statistical accuracy can then require the computation of the energy of millions of configurations. Another approach is to first identify all the individual processes, and then construct and calculate the input for a kinetic Monte Carlo simulation. This transition state methodology works well in cases where one can identify the rate-limiting transitions. However, in many important cases, one does not know and cannot guess the individual transition states, and it is simply not possible to catalog the entire set of relevant transitions. High-performance computing combined with intelligent analysis can assist in the determination of the various unit processes possible for complex systems and so make such an approach feasible for a greater array of problems. In general, though, the extension of electronic/atomic scale information to the times of practical interest remains the greatest methodological hurdle for the field.

The connection to macroscopic properties often requires the extension of electronic or atomic level methodologies to a length scale that overlaps with existing phenomenological models. One can then use the phenomenological models to establish the desired mesoscopic or macroscopic behavior. There are many examples of such mesoscale studies in the applications described above. One is the prediction of the dynamics of magnetic domain boundaries and their interactions with defects. Batteries are an example that involves all the scales from molecular chemistry, to surface reactions, and large-scale transport of matter. Another example is the use of atomic scale calculations to determine the properties of dislocations such as core structures. The dislocation core structures can be used as input to larger scale simulations of individual dislocation motion, and networks of dislocations, and grain boundary motions.

At the next largest scale, long-ranged stresses and short-ranged "contact" interactions between dislocations as well as the atomic scale properties of the dislocation core, kinks, jogs, etc. govern the behavior of dislocations. Thus, realistic simulations must capture complex structures with dissipative dynamics. The nonlinear, stochastic, dissipative dynamics leads to pattern formation, and the patterns are best characterized using microns. Current dynamic simulations allow modeling of approximately 2000 dislocation segments, each 10^-8 meters in length using approximately a week of CPU time on a 50 Megaflop machine. At dislocation densities of 10¹² m^-2, one deduces that one may examine a cube 10^-5 m on a side using approximately one week of CPU time on a two Tflop machine. A 40 Tflop machine would begin to allow the necessary statistical analysis of this scale of simulation, and thus, one can begin to approach the mesoscale.

At the larger macroscopic length and time scale, there are also open challenges for the largest scale simulations of, for example, mechanical properties of structural alloys and composites, and simulation of entire complex structures. As described in the sections on high-temperature alloys and lightweight materials for transportation, macroscopic properties, plasticity for example, depend crucially on the microstructure of the grains that make up the material. Finite element methods can describe aggregates of grains under stress. Then can provide a continuum level description of the highly inhomogeneous deformation inside a grain and motion at the boundaries. They can handle nonlinear constitutive materials relations such as hardening and strain rate sensitivities. Significant advances in simulating critical mechanical properties will require computational resources to described 1000's of grains, with meshes of ~10¹² elements. In order to consider reorientation and regrowth of the crystallite grains, simulations based upon Monte Carlo methods can extend the time scale of the finite element calculations. This could allow the simulations to take into account larger statistical ensembles. The finite element mesh would need to map onto a Monte Carlo mesh also of order 10¹² elements. Coupled with the difficult non-linear nature of the equations at this macroscopic continuum level, simulations of this size are among the most significant challenges for multi-teraflop computing.

The constitutive relations, which are the input for simulations at the macroscopic scale, are one of the clearest examples of a critical need for inputs from the finer mesoscale in order to make a revolutionary advance in predictive simulation capabilities. Yet they are also one of the best examples of the need for discovery. This requires new paradigms for describing the complex phenomena at the mesoscale. General methodologies do not exist to treat the complex problems of interacting, dynamic dislocations and grain boundaries directly from the known relations at the quantum and atomic scales.

In summary, there are four unique roles for large-scale materials simulations: (1) to provide benchmarks to test the forms of the equations, (2) to directly calculate input for models where possible, (3) to discover new relations and paradigms for complex behavior, and (4) to ultimately validate models by careful comparison with experiments. Combining all these modes, simulation can become a powerful, revolutionary tool to accelerate conceptual advances, and to play a critical role in developing new materials and technologies for energy generation, storage, and utilization. As such, it will become an indispensable tool greatly enhancing the productivity of major DOE supported facilities and programs.

[1] See, for example, the report "Multi-scale Modeling of Polycrystal Plasticity: A Workshop Report," by G. H. Campbell, H. Huang, W. E. King, D. H. Lassila, D. J. Nikkel, T. Diaz de la Rubia, J. Y. Shu, S. M. Foiles, D. A. Hughes, and V. P. Smyshyaev, Materials Science and Engineering A, in press. Report number 97-12 of the Institute for Mechanics and Materials, University of California at San Diego, 9500 Gilman Dr., Dept. 0404, LaJolla, CA 92093.

[2] W. E. King, G. Campbell, T. Gonis, G. Henshall, D. Lesuer, E. Zwyicz, and S. M. Foiles, "Simulation and Modeling of Interfaces in Materials – Bridging the Length-Scale Gap", Materials Science and Engineering A, 191 (1995).

The traditional modus operandi of the materials simulation community is one of small groups (typically one or two senior researchers/professors plus a small number of post doctoral researchers and graduate students) acting independently. For the most part, this model has served the community well by accentuating the creativity and innovation of individual researchers and allowing for the rapid evolution of methods and computer codes. The computational tools of materials science have historically been used effectively on the largest computers available. Indeed electronic structure, molecular dynamics, and finite element codes used by materials scientists dominate usage of the current generation of high performance computers owned by DOE (e.g., Cray T3E at NERSC, Intel Paragon at ORNL-CCS, ASCI machines at LLNL and SNL).

Successful as this model has been it is clearly not well suited to the integrated approach to materials modeling proposed in this initiative. For example, it has no mechanism to foster direct collaboration and cross-fertilization between materials disciplines such as first principles, atomistic molecular dynamics, microstructure modeling, continuum theory, engineering design, etc. Even within sub-fields, the model leads to small groups developing their specific approaches. For example, members of the electronic structure community using the plane wave pseudo-potential method have typically use codes developed locally, which resulted in considerable duplication of effort when several of these codes were recently ported to the NERSC Cray T3E.

For this initiative to be successful will require a new model that involves large-scale inter-laboratory cooperation with the computational materials community acting organically to solve major problems. Specifically, we propose a ‘Computational Materials Science Network’ as an integrated management approach. This model preserves the traditional strengths of the materials community while recognizing that making optimal use of a multi-teraflop computer requires some fundamental changes in how this type of research is performed. These include the use of large multidisciplinary teams, the development of standardized and optimized codes, and the centralization of human, computational, visualization, and data storage resources. A community wide management approach is required to dynamically set focus areas and assign appropriate financial resources to accomplishing goals. Finally, the network provides organizational structure that can articulate the goals of the materials community for the development and usage of high-end computational resources.

1. To provide an effective management model to assure readiness to enact the goals of the SSI Materials Simulation Initiative.

2. To provide a forum for developing cross-disciplinary, cross-institutional sustained efforts

3. To provide a mechanism to rally a critical mass behind a challenging problem to secure adequate resources (financial and human) to ensure rapid and major advances in a focused area.

1. Nodes are the basic research unit since this is where the strength and vitality of the materials research community currently resides. Each node will have the following characteristics:

2. Interest groups are responsible for carrying out the major research goals of this initiative.

3. Technical board with broad representation from the community of nodes, which sets network priorities and has the ultimate responsible for ensuring that the goals of the initiative are met.

4. Central resources are the physical focal point of the initiative, probably situated at the computer site(s) with limited staff accountable to the technical board.

5. Network funding holds the initiative together by fostering the multi-length scale and multi-disciplinary approach that underpins the research.

In response to the unprecedented national security challenges emerging from the end of nuclear testing, the Defense Programs of the Department of Energy has developed a long-term strategic plan based on a vigorous Science-Based Stockpile Stewardship (SBSS) program. The main objective of the SBSS program is to ensure confidence in the performance, safety, and reliability of the stockpile because of a fundamental science-based approach. A central element of this approach is the development of predictive, ``full-physics'', full-scale computer simulation tools. As a critical component of the SBSS program, the Accelerated Strategic Computing Initiative (ASCI) was established to provide the required advances in computer platforms and to enable predictive, physics-based simulation capabilities. Foremost among the essential elements needed to develop predictive simulation capabilities, the development of improved physics-based materials models is a cornerstone. Consequently, the development and application of advanced materials simulation tools of broad applicability is a critical element of the ASCI program. These advances are of great significance to the entire materials research community and not restricted to the confines of the national security missions of the DOE. In fact, we anticipate that the materials modeling and simulation capabilities supported by the ASCI program will provide direct leveraging opportunities for the proposed DOE Strategic Materials Simulation Initiative.

Foremost among the objectives of the ASCI program is its vision of promptly shifting from a nuclear-test-based methodology to a simulation-based methodology for issues of nuclear safety, reliability, performance, aging, and manufacturing. Consequently, the ASCI goals are mostly focused on the development of next-generation simulation codes embodying three-dimensional geometry, high-fidelity modeling, improved sub-grid physics models, and full-scale integrated systems.

At the heart of the development of predictive simulation capabilities supporting the SBSS program is the need to improve our ability to predict the effects of materials properties on stockpile performance, safety, and reliability, especially as these properties change as a result of aging and/or re-manufacturing. Of particular importance to the prediction of stockpile performance is an accurate determination of the thermodynamic properties of materials, i.e., equation-of-state under a broad range of conditions of pressure, temperature, and density. In addition, since materials are subject to extreme strain and strain rates, the development of physics-based models to predict the high-pressure mechanical response (e.g., deformation and plastic flow, strength, spalling, etc.) of materials is necessary. In terms of methodologies, thermodynamic properties are determined at the electronic/atomic level. On the other hand, mechanical properties are determined by the collective motion of defects at different length scales and their prediction requires a multi-length-scale approach. As we indicate below, however, the materials simulations capabilities supported by the ASCI program are applicable to a wide class of materials and are not restricted to stockpile materials.

As a strategy to address the unprecedented materials simulation challenges in support of the SBSS program, the ASCI program encompasses several materials modeling and simulation efforts based on fundamental, ``first-principles'' approaches. Because these first-principles approaches lead to predictive materials simulation capabilities of broad applicability to a wide class of materials systems, and because materials science cuts across all DOE missions, we anticipate that the proposed DOE Strategic Materials Simulation Initiative will leverage the current ASCI-supported materials simulation efforts ongoing within the DOE.

Examples of such scientific leverages in the area of materials simulation include:

In addition to these obvious scientific leverages, the ASCI program is supporting the implementation of a broad class of advanced materials simulation tools (from the quantum to the continuum scales) on high-end scalable-parallel computer platforms. By leveraging these efforts, the entire computational materials science community will enhance its ``readiness'' to take full advantage of the surge in computing power afforded by these emerging parallel computer architectures.

In addition to the materials science applications developments indicated above, a key element of the ASCI program is an aggressive plan to acquire high-p