Modeling better catalysts
Argonne's petascale computing capability will accelerate nanoscale
experimental research. Argonne's advanced computing facilities have
already provided materials scientists with a map to more efficient
catalysts. These catalysts support the reactions involved in manufacturing,
petrochemical processing and pollution abatement.
In an office at Argonne's one-year-old Center
for Nanoscale Materials, materials
scientist Jeff Greeley is working to find a new chemical catalyst that
can transform propylene into propylene oxide, an organic molecule needed
for the production of many plastics. Greeley, along with materials
scientists Larry Curtiss and Peter Zapol of Argonne's Materials
Science and Chemical Sciences and Engineering divisions, compiles a list of
suspects and enough physical evidence to run through some of the fastest
computers in the country. The way that Greeley describes the investigation
sounds almost as much like forensics as pure chemistry.
“We use a few chemical parameters as a kind of ‘fingerprint' that
gives clues as to how catalysts actually work,” Greeley said. “Our
goal is to use advanced computing to run these ‘fingerprints,' so that
we can help experimenters to more closely focus their synthesis and
testing efforts on the most likely candidates.”
Greeley's search for a new catalyst is a prime example of how basic
scientific research plays a critical role in ensuring long-term economic
growth. From plastics manufacturing to petrochemical processing to
environmental cleanup and pollution abatement, catalysts play an essential
role in many aspects of our everyday lives. According to R&D magazine,
approximately 90 percent of commercially produced chemical products
involve the use of catalysts at some point during their manufacture.
Catalysts also form the core of many of the most promising future technologies,
such as hydrogen fuel cells. Greeley's research is funded by the Basic
Energy Sciences program in the U.S. Department
of Energy's Office
of Science to expand the scientific foundations for new and improved energy
technologies.
Most commercial catalysts consist of a transition metal bonded to
a non-metal base. They work by reducing the amount of energy needed
to initiate a chemical reaction. The best catalysts lower this quantity—known
as a reaction's activation energy—dramatically, providing a shortcut
from reactants to products that saves money and conserves energy. A
catalyst's function also depends on its binding energy, a quantity
that describes how strongly reactants attach to it.
By harnessing the combined power of Argonne's high-performance computing
resources, including the Jazz cluster
at the Laboratory Computing Resource Center (LCRC), the IBM Blue Gene ® /P
supercomputer at the Argonne
Leadership Computing Facility and the new Nanoscience Computing
Facility (NCF), Greeley and his colleagues attempt to predict which
catalysts will have the lowest activation energies and the optimal
binding energies.
These two principal quantities are relatively easy to estimate, but
ensuring that they produce accurate representations of actual catalysts
requires a significant amount of computing power, Greeley said. “We
need our models to have predictive power; that is, we want to be able
to evaluate one set of parameters for each catalyst candidate and make
meaningful predictions about how fast a particular reaction will proceed,
an environmental contaminant will be cleaned up or a particular fuel
cell reaction will be accelerated. Then we'd like to go through and
evaluate that parameter on hundreds or thousands of different alloys.”
To investigate the catalytic properties of a variety of materials,
Curtiss and Zapol have used the Jazz cluster to more closely examine
catalysts' fundamental atomic structures as well as the mechanisms
of the reactions they facilitate. Many catalysts, including those used
in fuel cells, consist of a metal nanoparticle that sits on a larger
non-metal support—fuel cell electrocatalysts typically use platinum
active sites on carbon-based structures.
The particular geometric and electronic configuration of the catalyst
on a support greatly influences its efficiency. As a result, finding
the optimal atomic arrangement is as important as determining the correct
component materials. In the case of propylene, for example, Argonne
researchers found that small clusters of silver atoms on an aluminum
oxide support helped to add oxygen to propylene's carbon-carbon double
bond, converting it to propylene oxide. Later hands-on experimental
research has validated that result.
“For a computational materials scientist,” Greeley said, “that's the
dream: to be able to make a prediction about a material that has improved
properties, and then to actually test it in the lab and to find out
that it works. But it takes a lot of computer power to go through all
the different candidates.”
The opening of the 10-teraflop NCF earlier this year has provided
Argonne's scientists with enough computing power to do many more of
these atomic-scale calculations than they could previously. “We used
to run these calculations on smaller systems, where we'd have to represent
catalysts with a smaller number of atoms,” Curtiss said. “But now we
can include many more. In addition, in the past many of the reactions
in which we were interested were just too complex, but now we have
an opportunity to study many more reactions in more detail.”
Argonne's materials scientists refer to these combined approaches
as “computational screening,” a process somewhat like panning for gold.
The computer models act as a sieve, allowing scientists to eliminate
the vast majority of possible alloys or configurations from consideration
without having to perform expensive and time-consuming tests on them
in the laboratory.
“Our goal,” Greeley said, “is to shorten the list of likely candidates
so that we can give guidance to our experimentalist colleagues to help
them focus their synthesis and testing efforts more closely on materials
that we think have a good chance of success. Just in terms of manpower
and computational expense, it would be quite costly to go through all
those 750 materials and test them in a lab from the start.”
Zapol agreed: “You can try different combinations much more rapidly
on the computer than in the lab.”
Typically, however, catalyst “gold” isn't gold at all, but rather
a metal cluster, metal oxide or other composite. While precious metals
like gold or silver do compose part of many of the most active catalysts,
the costs of using them on a large scale provide another incentive
for researchers to use computational screening to identify cheaper,
but equally effective, alternatives.
In the course of his more than 30 years at Argonne, Curtiss has become
widely known for his development of new computational techniques, including
the formulation of highly accurate quantum chemical methods for the
calculation of binding and activation energies, the two most significant
parameters that scientists must consider when assessing and modeling
the catalytic properties of different materials. “The availability
of faster computers and new methodologies provides materials scientists
with an unprecedented opportunity for designing materials with improved
properties,” Curtiss said.
Although Argonne's scientists do most of their computational screening
of catalysts on the Argonne NCF and LCRC computer facilities, Argonne's
recent acquisition of the IBM Blue Gene/P supercomputer promises to
facilitate additional breakthroughs in computational catalyst modeling.
For more information, please contact Dave Baurac (630/252-5584
or media@anl.gov) at Argonne.
|