Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Material Informatics: Data, Methodologies, and Applications
Informatics for the Materials Genome: a Minimalist Perspective: Krishna Rajan Iowa State University http://cosmic.mse.iastate.edu
July 13th 2011
Acknowledgements: • National Science Foundation • AFOSR
Dept. of Homeland Security • Army Research Office • Office of Naval Research • Dept. of Energy • DARPA
Do we really need more data?
Data vs Knowledge
http://www.genengnews.com/
Krishna Rajan
Functionality = F ( x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 ……)
Issues: • how many variables? • which variables are important? • classify behavior among variables • making quantitative predictions …relate functionality to variables …
• traditionally we describe them by empirical equations: •Quantitative Structure Activity Relationships (QSARs) are derived from data mining techniques not assuming a priori which physics is the most important
Need to build database with these variables
Krishna Rajan Krishna Rajan
High Dimensional Data
Multidimensionality of data
Krishna Rajan
Hume-Rothery 1926, 1934
Laves 1956, 1967
Engel-Brewer 1964, 1967
Pearson 1972
Villars 1995 ……………………….
http://www.chem.ox.ac.uk/icl/heyes/structure_of_solids/Lecture3/Lec3.html#anchor5
Multidimensionality of data
Krishna Rajan
n
i
ii ppH1
log
Classification (partitioning of feature space)
= Minimization of information entropy
= Maximization of information gain
: Probability distribution of AB2 structure types occurred in Linus Pauling File (LPF)
Information Entropy
Krishna Rajan: ICME Symposium- MS&T
Krishna Rajan
Kong & Rajan-2012
Ranking descriptors
Krishna Rajan
Information entropy based “phase diagrams”
Krishna Rajan
840 compounds
(34 structure types)
140 compounds
(14 structure types)
22 compounds
(2 structure types)
Recursive partioning to track Evolution of design rules
Tracking Structural Correlations
Krishna Rajan
Developing a design rules for intermetallics
Krishna Rajan
Size factor
Electrochemical factor
Valence-electron factor
GeX2
Crystal-structure design rules
Tracking design rules
Entropy scaled Structure map
Krishna Rajan
Kong & Rajan (2009/ 2012)
The possible crystal structures of a hypothetical compound AuBe2 are suggested from the classification tree constructed by using known data.
A compound the structure is unknown
Potential structure types
Data-driven crystal chemistry (if-then) rules
Guiding Structure Prediction
Krishna Rajan
“Minimalism”: Linking information entropy to irreducible representations
Krishna Rajan
Linking Crystal Chemistry with Crystal Symmetry
Krishna Rajan
Limited Data Problem—no data deluge!
Data Diversity
Modeling with Data Mining
“omics” materials design Non-“omics” materials design
Accelerated Design- value of “omics” design
Krishna Rajan
2010/11- Rajan
Data Driven Modeling
Closing the gap : developing a QSAR- a new figure of merit
Knowledge: a unified & accelerated model of materials behavior
Information: the ‘tolerance factor’ Data: developing a descriptor database
Krishna Rajan
Discovering Classifiers
Krishna Rajan
Ranking Descriptors
Krishna Rajan
Structure Maps from Data Mining
Krishna Rajan
The 4 V’s of Materials Data
VARIETY: data in many forms
Engineering design / materials insertion
Multiscale data
VOLUME: data at rest
Materials reference data Thermodynamic Crystallography Property
VELOCITY: data in motion Materials Characterization
in-situ materials dynamics ( x-ray, ..)
Time-of-flight data
VERACITY: data in doubt
Incomplete data, ambiguities, missing
data
Phase diagrams, Property maps
Modeling
Krishna Rajan
veracity variety velocity volume :Discovery Materials
Summary: “Closing the Gap”’
Experiments and physical models
Informatics, statistical learning
To transform the “Materials Genome” from a concept to reality we need an information system that can enable and accelerate the Data to Knowledge transformation (the new paradigm for Materials-
by-Design)
Krishna Rajan