DEVELOPING AND DESIGNING NEW CHEMICALS AND MATERIALS - Chemistry: Notable Research and Discoveries - page 24

choice for computer storage applications. The researchers discovered
that a lot of the fluctuations occur on the boundary of the 14,400-atom
block, leaving the interior unaffected. This result suggests than even this
small piece of matter, which is on the scale of nanometers, can reliably
store information.
PrEdICTIngTHEProPErTIES
Another design technique being actively explored involves searching
for patterns. Whereas the method described in the previous section at-
tempts to understand the behavior of atoms and molecules, this method
takes its cue from combinations of these particles. By means of X-ray
crystallography and other methods, scientists have already determined
the structures of thousands of crystals, composed from a wide variety
of elements and compounds. The set of these structures represents an
enormous amount of data. Some researchers have begun to use com-
puters to sift through this data, looking for clues as to what elements
and compounds produce which structures. With these clues, prediction
of the properties of new materials may be possible.
Considering the huge number of compounds and the complexity of
materials, this procedure is unlikely to be able to specify the exact struc-
ture of a new material in advance. But the clues gained from previously
acquired data may narrow the possibilities to a much smaller number,
giving material designers a much better handle on what sort of material
they require. Since the amount of data is so large, fast computers must
be used, and algorithms—computer procedures—must be written to
make an appropriate search. Sifting through large data sets for informa-
tion that is not readily apparent is a process known as data mining.
The Massachusetts Institute of Technology scientists Christopher
C. Fischer, Gerbrand Ceder, Kevin J. Tibbetts, and their University of
Wisconsin colleague Dane Morgan developed a data-mining algorithm
in 2006 to look at crystal structures. The researchers examined a da-
tabase containing 28,457 structures of binary alloys (which are com-
posed of two metals). With their algorithm, the researchers gathered
clues about the possible structures of a given new material. To test the
algorithm's accuracy, its predictions could be evaluated in the case of
materials whose structure is known but was not in the database.