DOESciDAC ReviewOffice of Science
NANOPHASE MATERIALS SCIENCES
The Computational Materials END STATION at the Center for Nanophase Materials Sciences
From compact cassettes circa 1980 to today's hard drives, the capabilities of magnetic storage have grown at an astonishing rate during the past few decades, with individual devices rising in capacity by six orders of magnitude, from megabytes to terabytes. Indeed, last year's Nobel Prize in physics recognized such efforts, with Albert Fert of France and Peter Grünberg of Germany sharing the honor for having discovered giant magnetoresistance nearly a decade before.

If the growth trend is to continue and accelerate, so must the research. Since 2005, Oak Ridge National Laboratory (ORNL) materials scientists have pursued new candidate nanoparticles for magnetic storage and then simulated the particles on the laboratory's Jaguar supercomputer (sidebar "Recycling Code," p16). Such efforts not only propel the research but also make the process itself more efficient. It optimized the popular Vienna Ab-initio Simulation Package (VASP) for use on ORNL's Cray XT Jaguar supercomputer, and it offered this tool to researchers across the community through ORNL's newly created Computational Materials End Station.
End Station Makes Life Easier for Researchers
The challenges of high-end computing in nanoscience, condensed matter science, and materials science are similar to those faced by large experimental facilities such as synchrotrons and neutron sources, an analogy pointed out by Malcolm Stocks and Bruce Harmon in late 2002 (figure 1). These highly complex facilities are run by specialists who focus on delivering stable operations, but users are not confronted with the complexity. Rather, they take their measurements at an end station that provides an environment similar to a home laboratory, assisted by instrument scientists who are familiar with every aspect of the measurements they are taking.
The end station provides researchers with tools and it gives them access to a staff of computational experts, allowing scientists to devote less attention to the simulations and more to the science being revealed by those simulations.
ORNL adopted this customer-focused approach for the Computational Materials End Station, created within the Center for Nanophase Materials Sciences (CNMS) Nanomaterials Theory Institute (NTI). The end station provides researchers with tools including VASP, and it gives them access to a staff of computational experts. This approach allows scientists to devote less attention to the simulations and more to the science being revealed by those simulations.
Organizationally, the end station consists of three equally important parts:
  • Pursuit of high-end computational science projects
  • Development of a software repository and scalable application libraries
  • Integration with the user program of the CNMS/NTI
Adaptation: A. Tovey; Source: B. Harmon and M. Stocks
Figure 1. The analogy between experimental and computational "instrumentation," adapted from a December 2002 presentation to the DOE Council on Materials Science and Engineering, led by Peter Flynn and David Mermin.
High-end computational science projects typically grow from priorities of the internal research team and partner users, with successful projects leading to new capabilities that will be incorporated into the Center's user program. Challenging science problems promote new computational approaches, algorithms, and software, which are designed with special attention to efficiency and scalability on the high-end computing platforms of the National Center for Computational Sciences (NCCS) at ORNL. Some projects under way include the following:
  • The first microscopic simulations of phase separation in the giant magnetoresistance materials used in hard disk read heads
  • Systematic solutions of the most popular models currently used to explore high-temperature superconducting materials
  • The first quantum mechanical studies of novel magnetic nanoparticles that are candidates for a new generation of higher-density magnetic recording media
  • Development of models focused on the growth of catalytic oxide nanoparticles for alternative energy technologies
  • Modeling of platinum catalysts for use in fuel cells
  • Calculations of the structure of advanced nanostructured thermoelectric materials for use in heat recovery
Scientists of the CNMS/NTI collaborate closely with colleagues in ORNL's Computing and Computational Sciences directorate to form well-focused and highly efficient teams similar to SciDAC teams.

Nanomagnetism
The impressive growth in magnetic storage capabilities has required the development of new materials and recording methods and even the exploitation of new phenomena. Further increases in storage densities, however, will require further advances in materials technology. While current technology involves recording individual bits of magnetic information on a thin film, one nanoscale option would move to recording bits of magnetic information on either individual magnetic nanoparticles or groups of such nanoparticles. This approach would open new options for magnetic recording by exploiting the fact that nanoparticles have very different properties from bulk materials.
One candidate for this approach is the metal alloy iron-platinum. Nanoparticles of this material are predicted by simple models to offer the correct combination of properties for magnetic recording, in particular stable storage of the magnetic information at typical operating temperatures. Questions about the detailed behavior of this material remain, providing the first area of application for the computational end station.
Scientists of the CNMS/NTI collaborate closely with colleagues in ORNL's Computing and Computational Sciences directorate to form well-focused and highly efficient teams similar to SciDAC teams.
Of particular interest when modeling nanoparticles is the way properties vary with the size of the particle. By extracting trends in this information, simplified models can guide the choice of materials. In this investigation, idealized nanoparticles up to 1,289 atoms were constructed, allowing the magnetic properties to be studied up to a system of size close to the 3 nm target considered by industry.
The calculations of iron-platinum nanoparticles were used as an initial basis for optimizing the VASP code on the Cray XT3 machine. Parallel scalabilities up to around 10,000 processors were found for the largest nanoparticles; this allowed for high-quality calculations in which atomic positions were routinely relaxed to the positions in which they would be found in an actual system. These calculations were particularly challenging due to the delicate nature of magnetic interactions and the large number of transition metal atoms in the calculation. Figure 2 illustrates the structure of some of the nanoparticles considered and the scalability of the code obtained.
Illustration: A. Tovey; Source: P. R. C. Kent
Figure 2. Calculations on ferromagnetic FePt nanoparticles, including (a) structure of an FePt nanoparticle (b), calculated magnetic moments, and (c) scaling of calculations on the Cray XT3. The time shown is for one electronic iteration. Larger nanoparticles can effectively use more processors, but the time to solution is also longer. The limit of parallel scalability with the algorithms used in the present calculations comes from parallel dense linear algebra.
Solving the Structure of Nanowires
The novel properties of nanoscale systems often depend on the position of individual atoms. While this atom-by-atom breakdown is beyond the capabilities of modern observation and experiment, it is well within the abilities of computer simulation. In the case of catalysts, the location of atoms in metal clusters containing only a dozen atoms can dramatically change the efficiency of the catalyst. In the case of molecular electronics, conductivities can vary by orders of magnitude with small changes.
Researchers at the University of Tennessee and CNMS recently demonstrated the role of computation in elucidating structure down to the atomic level by focusing on newly synthesized yttrium silicide nanowires. Their work is published in the July 2008 issue of Nature Materials. The wires were as small as a few atoms wide and showed new behaviors because they were nearly one-dimensional. These nanoscale effects may one day be used in very sensitive electronic circuit elements, or they may detect certain molecules, such as individual components of DNA.
Recording bits of magnetic information on magnetic nanoparticles would open new options for magnetic recording by exploiting the fact that nanoparticles have very different properties from bulk materials.
Microscopy images produced a range of possible structures for the wires, but microscopy can see only the topmost atoms of the wires. Information about the complete structure came from computer simulation. Large-scale density functional calculations with models containing up to 1,000 atoms identified the most likely candidate (sidebar "Optimized Density Functional Calculations," p18). Comparisons between electrical conductivity measurements and simulated data from the candidate structures provided final confirmation, with very close agreement between experiment and theory.
Research into the wires is ongoing. The aim is to improve understanding of the ways in which properties of the wires depend on small changes in their structure. Figure 3 (p19) illustrates the structure of the wires, as well as measured and simulated microscopy data.
Source: C. Zeng, P. R. C. Kent, T.-H. Kim, A.-P. Li, and H. H. Weitering
Figure 3. Simulated structure and wave functions of YSi2 nanowires on silicon (a and b). The simulated tunneling spectra (c) are in excellent agreement with measured room temperature data.
On the Edge: Titanium Oxide
Not all materials are alike, even if they are made of the same elements.
Depending on how these materials are put together, they can change color, conductivity, and a range of other properties. These reversals can lead to nanoparticles with valuable new qualities, or they can create problems for industry, introducing unwanted forms of a material during synthesis.
Researchers from Pennsylvania State University, the Smithsonian Institution, and CNMS were able to explain such a reversal for titanium oxide, a material with a range of industrial applications. Nanosized titanium oxide particles show enhanced electrical and optical properties and are widely used in pigments, sunscreens, and food additives. They are also critical to several alternative energy technologies, serving as catalysts for the creation of hydrogen fuel and components of solar cells. Recently, nanoparticles of titanium oxide were also used to create the fourth fundamental circuit element, known as the memristor ("Cortical Computing with Memristive Nanodevices," p58).
Not all materials are alike, even if they are made of the same elements. Depending on how these materials are put together, they can change color, conductivity, and a range of other properties.
Titanium oxide has two competing crystal structures, rutile and anatase. Although bulk calculations predict rutile to be stable at all relevant temperatures and pressures, it is anatase that dominates samples of nanoscale material, in the laboratory and even in natural settings.
Until now, it had been assumed that anatase's dominance at the nanoscale is a function of the high surface-to-volume ratios of small nanoparticles, with the energetics of the surface winning out over the bulk energetics at the nanometer scale. This theory could not be tested, however, until large-scale, high-quality calculations could be made on whole nanoparticles.
The team used the VASP code on NERSC's Franklin system, calculating nanoparticles up to 3 nm with various possible surfaces for both rutile and anatase. With these calculations the team was able to identify the different energetic contributions to the stability of each structure. The calculated energies can also be used to test existing models of nanoparticle growth, which are usually based on the bulk energetics and the surface energies of typical surfaces.
In order to get accurate calculations of the nanoparticle energies, the team had to minimize systematic errors by using very large supercells to contain the nanoparticles and using very high-quality convergence criteria. The atoms near the surface of the nanoparticles move concertedly away from their idealized bulk positions, requiring careful optimization of all atomic positions.
Time-resolved X-ray diffractions performed in conjunction with the calculations were able to follow the nucleation of nanoparticles with time. These experiments not only identified their structure but also measured the spacing between atoms. The measurements were used as a cross-check on the calculated nanoparticles and gave confidence that the methods used were accurate and that the structures of the model nanoparticles were representative. Some of the calculated structures are shown in figure 4.
Source: P. R. C. Kent, D. Hummer, and J. Kubicki
Figure 4. The structure and charge density isosurface of 3 nm TiO2 nanoparticles in the energetically competing anatase (left, 816 atoms) and rutile phases (right, 627 atoms).
The calculations revealed for the first time that anatase's greater stability cannot be explained by the simple picture of surface energies winning out over bulk energetics. Instead, the energetics of the edges, defects, and corners of the nanoparticles were also very important. Even for 3nm anatase nanoparticles, around half of the formally defined surface energy comes from the nanoparticles' edges.
Nanosized titanium oxide particles show enhanced electrical and optical properties and are widely used in pigments, sunscreens, and food additives. They are also critical to several alternative energy technologies, serving as catalysts for the creation of hydrogen fuel and components of solar cells.
These surprising results show the complexity that takes place at the nanoscale, revealing the necessity to revisit and modify existing models of nanoparticle growth. The researchers are now studying larger nanoparticles, and eventually, more difficult calculations that consider the actual dynamics of growth might reveal how the growth of these industrially important materials can be engineered to obtain desired materials.
Nanostructured Thermoelectrics
More than 60% of the energy created by your car's engine is lost as waste heat, but thermoelectric materials promise to take that heat and turn it into electricity. By converting a thermal gradient into electricity—or vice versa—these materials are able to harvest waste heat to run a vehicle's lights, radio, or air conditioner, for instance, or to provide active cooling.
These materials hold the potential for enormous energy savings, and with rising energy costs they are becoming more and more attractive. To be a viable and widespread technology, however, they must perform well enough to overcome their cost, weight, and complexity.
Because of the difficulties involved in creating high-performance thermoelectrics, they have traditionally been limited to only a few niche applications. Nevertheless, nanoscience has made several promising advances in this mature field, and several nanostructured thermoelectrics have recently shown very high performance in laboratory settings. To make continued progress with these and other materials, researchers must develop a complete understanding of why they perform better than traditional bulk materials.
The performance of a thermoelectric is a complex balance between its electrical and thermal properties. These properties must be optimized for the material's operating temperature, which will vary depending on how it is being used. The study of nanostructured materials must begin with a full understanding of the material—the location and types of atoms within it—only later coupling this structural knowledge with the material's electrical and thermal properties. If these links can be established, they may allow researchers to boost performance even higher by determining the ideal composition and configuration of the material.
Thermoelectric materials hold the potential for enormous energy savings, and with rising energy costs they are becoming more and more attractive.
Researchers from General Motors Corporation, the University of Nevada, and CNMS are using this approach on a promising nanostructured thermoelectric, a lead telluride material containing small nanoscale regions of silver, antimony, and tellurium. These nanoprecipitates are the likely cause of improved performance in this material, but a detailed picture of its structure has been lacking.
The biggest challenge in modeling the structure of nanoprecipitates in thermoelectric materials is the large length scales required: the nanoprecipitates are typically several nanometers in size, which equates to several thousand atoms. The team used the optimized VASP code to perform many high-accuracy calculations for precipitates of differing size and composition in a massive, 1,728-atom supercell. This effort, taken with the support of an INCITE grant, created the first simulations ever of whole nanoprecipitates and provided a realistic picture of these materials.
The team also used electron microscopy to image the nanoprecipitates at the atomic scale. Although these measurements cannot identify all the atoms, they are able to provide a valuable cross check on the models. The team found excellent agreement between the measured and simulated systems, adding strong support to the methods and models it used (figure 5).
Source: X. Ke, C. Chen, J. Yang, and P. R. C. Kent
Figure 5. Left: supercell (1,728 atoms) of thermoelectric PbAgSbTe. A matrix of PbTe (gold and gray atoms) holds nanoprecipitates containing additional silver (white) and antimony (purple) atoms. The supercell is large enough to hold a precipitate of size comparable to that found in experiment. Right: silver and antimony atoms only, indicating the elongated shape of the precipitate.
The researchers have used extensive calculations of different atomic configurations and compositions to untangle the complex energetics of nanoprecipitate growth. The very large supercells used in the calculations allowed them to accurately study the effect of small changes in the structure and uncover the role of strain and electric charges induced in the material. Together, these two effects govern the growth directions and eventual size and shape of nanoprecipitates in the materials, with quantitative measures providing the basis for tailoring the structure of the materials.
The next challenge is to connect the details of the known structure, revealed by these calculations, with the materials' properties, and the team is focusing on the thermal properties of the materials. These are significantly more expensive calculations: while determining the structure of the system requires a calculation scaling formally with the cube of the number of atoms, the response of the system requires a calculation scaling with the number of atoms to the fourth power. These calculations are able to efficiently use the entirety of the current 263 teraflop Jaguar system.
The overlap of length scales that can be achieved through experiment and simulation presents many opportunities for scientific discovery using computation in nanoscience.
Future Developments
The overlap of length scales that can be achieved through experiment and simulation presents many opportunities for scientific discovery using computation in nanoscience. Demands for increasing accuracy as well as longer length and time scales, however, will require a great deal of work as researchers develop and hone new methods for exploiting the power of new supercomputing resources (sidebar "Optimized Density Functional Calculations," p18). The computational end station will play an important role in this evolution by tuning codes and methods to work with real systems and real problems, rather than idealized systems and idealized problems (sidebar "Benchmarks").
Contributors: T. C. Schulthess, P. R. C. Kent, and P. T. Cummings

Acknowledgments: Nanowires—Changgan Zeng, Tae-Hwan Kim, An-Ping Li, and Hanno H. Weitering; Oxide Nanoparticles—Daniel Hummer and James Kubicki; Thermoelectrics—Xuezhi Ke, Changfeng Chen, and Jihui Yang
 
Published by IOP Publishing in association with Oak Ridge National Laboratory, for the US Department of Energy, Office of Science. Copyright © 2008 by IOP.