PARALLEL PROCESSING ENABLES RAPID COMPUTATION OF X-RAY ABSORPTION.

A popular computer code for x-ray absorption spectroscopy (XAS) now runs 20-30 times faster, thanks to NIST. XAS is widely used to study the atomic-scale structure of materials and is currently employed by hundreds of research groups in a variety of fields, including ceramics, superconductors, semiconductors, catalysis, metallurgy, geophysics, and structural biology. Analysis of XAS relies heavily on ab initio computer calculations to model x-ray absorption in new materials. These calculations are computationally intensive, taking days or weeks to complete in many cases. As XAS becomes more widely used in the study of new materials, particularly in combinatorial materials processing, it is crucial to speed up these calculations.

One of the most commonly used codes for such analyses is FEFF. Developed at the University of Washington, FEFF is an automated program for ab initio multiple scattering calculations of X-ray Absorption Fine Structure (XAFS) and X-ray Absorption Near-Edge Structure (XANES) spectra for clusters of atoms. The code yields scattering amplitudes and phases used in many modern XAFS analysis codes. FEFF has a user base of over 400 research groups, including a number of industrial users.

To achieve faster speeds in FEFF, NIST researchers developed a parallel version. In modifying the code to run on the NIST parallel processing clusters using a message-passing approach, they gained a factor of 20-30 improvement in speed over the single processor code. Combining parallelization with improved matrix algorithms may allow the software to run 100 times or more faster than current single processor codes. The latter work is in process.

The parallel version of the XAS code is portable and is now also operating on parallel processing clusters at the University of Washington and at DoE's National Energy Research Scientific Computing Center (NERS C). A speedup of 30 makes it possible for researchers to do calculations they only dreamed about before. One NERSC researcher has reported doing a calculation in 18 minutes using FeffMPI on the NERSC IBM SP2 cluster that would have taken 10 hours before. In 10 hours, this researcher can now do a run that would have taken months before and hence would not have been even attempted.