Database Reference
In-Depth Information
PLS Prediction with Lattice Constants, Space Groups,
and Secondary Descriptors
2
26
1.95
27
20
25
31
17
32
21
1.9
22
30
24
34
23
11
28
1.85
16
8
9
7
6
13
19
10
18
3
2
1.8
12
15
14
5
33
1.75
29
1.7
1
1.65
35
1.6
1.6
1.65
1.7
1.8
Measured N-O Bond Length
1.75
1.85
1.9
1.95
2
Figure 8.7
The PLS prediction with 12 space group descriptors, 4 lat-
tice constants, and 4 secondary descriptors. M-O bond lengths are given in
Angstroms.
that describe the local topology of the frameworks and have used these statis-
tically derived variables to enhance the prediction of structural properties. In
our study of zeolites, we first use PLS as a predictive tool based on only the
primary descriptors from the database.
86
,
87
We then added secondary topo-
logical descriptors such as c/a ratio (chosen using principal components) and
compared the results (Figure 8.7). This illustrates how we can enhance a mate-
rials database by adding statistical secondary descriptors tailored for specific
applications.
8.5.4 Future Challenges
With crystallographic and thermodynamic databases as an exception, in ma-
terials science, the building of databases is still largely an ad hoc process.
What data is collected, compiled, and managed is primarily defined by the
community that may use that data. As such databases in materials science
are not set up to easily integrate, correlate, and process the diversity in data
that needs to be considered for solving many complex problems. Crystal-
lographic and thermodynamic databases have a well-established formalism
and scientific rationale for organization of data that lend themselves well to
be used and applied to experiments and models. For instance, group theory
