From First Principles to Continuum--
the ORNL Materials Advantage

The ability to calculate realistic materials properties at the atomistic level not only provides great insight for understanding complex mechanical and electronic behavior, but it offers a whole new approach to materials design and fabrication.

Increasingly the wealth of information available from a well constructed first principles simulation is providing insights into basic principles not available from experiment.

Ultimately, materials simulation should contribute, not only to development of new and improved materials, but also to a reduction in the effort and cost of development programs.

Some of ORNL's Materials Research Highlights...


ORNL's Materials Research has been outstanding from the start, expecially when computational simulation is involved. In 1990 alone, materials project won a Gordon Bell award for price/performance, 1st prize for Scientific Excellence in the IBM Supercomputing competition, and the Cray GigaFLOP award for superconductivity calculations. And that was just the beginning!


Materials Grand Challenge Begins

with the development of a locally self-consistent multiple scattering code ("Parallel Real-space Multiple-scattering") to solve problems ranging from high strength alloys to high temperature superconductors.

Multiple Scattering Memory Requirements Cut in Half

in a new version of the "Parallel Real-Space Multiple Scattering" code developed by Bill Shelton of ORNL. The new version is also significantly faster, allowing more challenging materials problems to be solved.

New Message Passing Algorithm has Constant Memory Requirements

regardless of the size of problem being solved, unlike previous algorithms, where systems of a few hundred atoms exceeded the available memory (16 MB) on the Intel Paragon compute nodes.


Order N scaling for LSMS using QMRPACK
(first code to be run on 512 nodes of Intel Paragon XP/S 35)

Conventional methods for quantum mechanical calculations of the properties of condensed matter are referred to as N-cubed methods because the computational work required grows as the third power of the number of atoms (N) in the simulation cell. This limits the size of problem that can be run, and thereby what properties can be studied.

ORNL developed truly order-N methods (methods for which the computational work grows linearly with system size) by replacing the direct solver with an iterative solver, selected from the several available in QMRPACK. The new approach scales much better than the old one, as well as being significantly faster.


Simulation of Charge Transfer in Brass Validates LSMS

Electrons move from Zn atoms to Cu atoms when alloys are being formed. Careful study of the results shows that the amount of electrons gained or lost depends on the local environment of the atom. The more Zn atoms in its neighborhood, the more electrons a Cu atom gains. Conversely, the more Cu atoms in its neighborhood, the more electrons a Zn atom loses.

Microscopic theory of the vortex state in superconductors

solved using the computational power afforded by the Intel XP/S-5 and XP/S-35 Paragons within the Center for Computational Sciences at ORNL. These calculations allow the physical properties of all individual states and excitations of the system to be studied, rather than just the average or total effect as confirmed by experiment in the 1950s.

distant view, details

CETEP Calculation of Energy Surface for Ge Adatom

The CETEP parallel ab initio pseudopotential code developed at Cambridge University has been ported to the XP/S 5 and XP/S 35. The code has been used to identify the binding site and diffusion pathway of a Ge adatom on Si (of interest because Si/Ge thin multilayers are considered to be a promising direct-gap material for optical devices). details

Microscope Directly Images Atoms in Materials

Steve Pennycook, a scientist in ORNL's Solid State Division, uses a powerful electron microscope designed to apply his "Z-contrast imaging" technique to give the sharpest direct images yet of atoms in a solid. The 300-kilovolt scanning transmission electron microscope (STEM) can resolve features as small as 1.3 angstroms. An angstrom is one ten-billionth of a meter and about one millionth of the diameter of a human hair.


LSMS run up to 1024 atoms on XP/S 150


GMR research leads to higher density disks

GMR (giant magnetoresistance) is a large change in a magnetic material's electrical resistivity caused by an applied magnetic field. This effect allows GMR "read sensors" to read data crammed into high-density disks as tiny regions of magnetization (the extent to which the material is magnetized).

Bill Butler, Xiaoguang Zhang and Don Nicholson of ORNL, and Thomas Schulthess of LLNL received the DOE-BES Award for Outstanding Scientific Accomplishment in Metallurgy and Ceramics for 1995.

"We used the Kendall Square Research and IBM SP2 parallel computers to calculate conductivity, the inverse of resistivity," Butler says, "and we are using the Intel Paragon XP/S 150 to calculate the magnetic fields required to align fields in the magnetic layers. "

press release, details, contact: Bill Butler

In Nov 1997 IBM announced the world's highest capacity desktop PC disk drive with "new breakthrough technology called Giant Magnetoresistive (GMR) heads."

Evidence for the Hexatic Phase in Two-Dimensional Melting

The theory proposed by Kosterlitz, Thouless, Halperin, Nelson, and Young (KTHNY) in a series of papers in the 70's predicts that a third phase, called the hexatic, will exist between the solid and liquid.

Kun Chen, Ted Kaplan and Mark Mostoller of ORNL have used the Intel Paragon machines at the CCS to do very large-scale simulations of melting in 2D Lennard-Jones systems. Simulations with millions of time steps were performed on systems of 576, 4096, 16384, 36864, and 102400 atoms.

In runs of 4 to 8 million time steps on the two largest systems, the hexatic phase appears after a minimum of about 1 million steps and persists for another 1 to 3 million steps. Since extensive simulations on all smaller systems did not reveal any metastable states, we conclude it is the large system size which is key.
contact: Ted Kaplan

Electronic Properties of (a/2)[110] Edge Dislocations in Silicon

Charge density contours for the gap states. (The dislocation lines run perpendicular to the page.)

Calculations for undoped dislocations in Si have been done for systems of as many as 576 atoms (the largest such calculations to date for this material) using the Intel Paragon XP/S 35. The calculations for the largest systems required roughly 20,000 node hours. Results showed electronic states rather deep (~ 0.2 eV) in the band gap, indicating that the dislocations may be electrically active.
contact: Ted Kaplan


Magnetic moment inhomogeneities in disordered NiCu alloys

A sample of 256 atoms in an essentially random alloy of 80% Ni and 20% Cu -- the magnetic moments of the individual Ni atoms are shown by the arrows.

Copper/Nickel is an important and prototypical magnetic alloy. Magnetism has a profound affect on many of the properties of alloys: phase stability, thermal expansion, electrical transport properties to cite but a few. The LSMS method has been used to study, for the first time, the nature of the magnetic inhomogenieties that have long been thought to be a feature of the onset of magnetism in this system. details

Magnetic scattering cross sections for neutrons are in excellent agreement with experiment confirming the microscopic picture of the distribution of magnetic moments obtained in the theoretical snapshot.

The Linked Computing Project: Heterogenous Distributed Supercomputers

Supercomputers at Oak Ridge National Laboratory, Sandia National Laboratories, and the Pittsburgh Supercomputing Center have been linked via high speed networks using PVM software, so that scientific researchers could use two or more of these machines as a single resource. details

PVM (originally developed at ORNL and the de facto standard for intermachine computation) was used for intermachine communications. LSMS was used as the initial application because of its scalability.

Linked run of LSMS on ORNL's Paragons ('35 and '150)

Linked run of LSMS over ATM on 512 nodes of ORNL/SNL Paragons

Linked run of LSMS on ORNL paragon and PSC Cray T3D


Predicting Structure of Grain Boundaries in Si

Experimental Z-contrast image of a (510) symmetric tilt boundary in Si.
M. F. Chisholm and S. J. Pennycook, ORNL.

436 atom simulation of Si grain boundary
J. R. Morris, K.-M. Ho, D. M. Ring, Z.-Y. Lu, Ames Laboratory; C.-L. Fu, ORNL

Grain boundaries are important in semiconductors, due to their mechanical and electrical properties, but are difficult to study theoretically. Our recent advances in algorithms, combined with the power of the new Cray T3E computer at NERSC, allow accurate energy calculations for systems of up to 1000 atoms on a routine basis. These can be combined with more empirical search techniques to predict grain boundary structures and properties.

The illustrations on the left show a comparison of a 436 atom simulation of a Si grain boundary with a corresponding experimental image obtained using Z-contrast imaging.


Non-colinear Magnetism NiFe Invar Alloys

Perspective view of Fe (large arrows) and Ni (small arrows) moments in Fe65Ni35 Invar alloys.

View of non-colinear Fe and Ni moments projected onto a common Fe and Ni origin.

By combining first principles scalable methods with the power of the Intel Paragon XP/S-150 at ORNL, we are able investigate of a wide range of magnetic properties previously not accessible to ab initio theoretical study.

The results point to unusual magnetic behavior involving non-collinear arrangements of the local magnetic moments associated with Fe rich clusters within the otherwise disordered NiFe (Invar) alloy.

These magnetic structures have important implications for our understanding of magnetism in this material, implications that are currently being investigated experimentally using neutron diffraction. details

Work performed by: Yang Wang, D. M. Nicholson, W. A. Shelton and G. M. Stocks, ORNL; Vladimir Antropov and B. N. Harmon, Ames Laboratory

LSMS run 116 iterations with 1372 atoms/1372 nodes ORNL paragons linked


Gordon Bell submission--CLM version of LSMS achieves 657 Gflops

512-atom CLM state of prototypical paramagnetic bcc Fe. The top frame shows the SCF magnetic moments, the bottom frame shows the corresponding transverse constraining fields. Atom positions are denoted by spheres, magnetic moments by arrows and constraining fields by cones. The magnitudes are color coded as indicated. These calculations are for LIZ27 and were performed on the T3E900 LC512 at NERSC.

ORNL researchers combined CLM (constrained local moment--which maintains the orientational configuration) with the LSMS method and validated it on a bcc Fe system (where results at certain phases can be confirmed by experiment), achieving 657 Gflops in the process.

In initial testing, we found that the constraining fields converged rapidly (a few tens of iterations), while convergence of the charge and magnetization densities required 800 iterations. Multiple scattering processes outside a local interaction zone (LIZ) centered on each atom are ignored.

details, contact: Bill Shelton

The largest simulation performed was of 1024 atoms on 1024 processors of the USG site T3E1200 LC1024/512 for which we obtained a performance of 657 Gflops.


Lightweight, high energy density batteries

Problem: The effects of disorder induced by lithium insertion and extraction in the metal oxide cathodes during battery operation.

Objective: ORNL researchers will be applying the MKKR-CPA method to the systematic exploration of mixed transition metal ion compositions. The materials involved are highly disordered and thus require computational cells containing thousands of atoms. The most promising compositions will be tested experimentally. Improvements in battery capacity and power density are needed for electric cars.
contact: Bill Shelton

Improving performance/reliability of ceramic coatings

Problem: Cracking and debonding of thermal barrier coatings (TBCs) on turbine blades. Improvements in TBCs can lead to significantly higher efficiencies and safer operation of land-based power generators and aircraft.

Objective: ORNL is developing unique simulation capabilities to realistically treat failure in solids, including a predictive capability which can be used to optimize the design of the coatings process and minimize the potential for failure.
contact: Ted Kaplan

ORNL || CsM || ORNL SC98 exhibit