The term “informatics” has long been used to describe the entire field of computer science from computer engineering to information systems and related fields. This conference will focus on “material informatics” – the application of computational methodologies to process and interpret materials science and engineering data. Modern materials science and engineering research produces a large amount of heterogeneous data. As a result computational methods that are used to organize, manage, interpret and analyze these data and are now becoming essential tools for materials science and engineering community.
Materials informatics can be broadly divided into two main parts – data management and knowledge discovery. The conference theme will be the knowledge discovery component. New experimental methods, such as combinatorial materials science, are generating large amounts of structural and property data – physical, chemical and engineering. More recently ‘computational’ combinatorial methods are being employed to calculate systematically the structure and properties of materials. The use of these complementary methods – experiment and theory – to generate data, enables the filling in of ‘holes’ in materials property space and provides routes to estimate properties – both physical and engineering, and improves property prediction or material selection capabilities.
For successful design of components for engineering applications, data and knowledge, relating to the selection of the material, the component geometry and the processing route need to be optimized to maximize performance and minimize cost and adverse environmental impact. Often, to improve selection, other descriptive information of a relational nature need to be extracted from text, figures, tables, graphs and captions.
Materials scientists’ interests cover a wide range of materials, together with a variety of techniques used in their studies, both experimental and theory. A review of the scientific literature over the past ten years has shown that very little use of materials informatics methods, and related computational techniques, have been employed to solve materials problems. While experiments continue to be costly, the cost of computing is shrinking rapidly, at the same time the accuracy of computational calculations and the results from data mining algorithms are improving steadily. Since computational power continues to follow Moore's law, doubling every 18 months or so, our ability to address complex materials issues is dramatically improving, e.g., it is possible to run thousands of potential material calculations with current computational power and generate notable “theoretical databases.” Databases with derived materials, with calculated physical and engineering properties, will no doubt play an increasingly important tool for researchers and engineers working in fields related to materials development.
During this same period electronic data sources containing materials data have increased, although not all as databases. Improved computing hardware speeds and storage capabilities, have enabled complex data dissemination capabilities. Improved methodologies in data visualization and data mining have enabled effective problem solving capabilities. These advancements, coupled with the applications of quantum mechanical methodologies, offer mature informatics systems for many applications in the materials sciences. The coupling of materials informatics with experimental methods then translates into increased productivity.
Some of the techniques of interest in materials informatics are standard, e.g., quantum methods for computing the stability of materials, or information techniques such as `data mining', or `data analysis' (in statistics). However, combining these techniques to exploit materials informatics is not “standard” and offers us a novel approach to materials design. In particular, with materials informatics one can address a new set of questions. For example: If I am given this database of known materials, what unknown material has desirable properties most like a known one? Or, can materials be clustered into sets so as to discover useful patterns? While techniques of this category have penetrated other scientific areas, they remain at an embryonic stage in materials science. We believe the time is ripe to explore the diverse techniques of data mining with quantum methods for designing and discovering new materials.
The challenge for the materials development community is to develop new materials technologies at a faster rate, and in a more cost effective way, than previously done. This challenge is to reduce development time both for the discovery of new materials and the prediction of their properties and process. This accelerated method provides closer alignment to the product development cycle while contributing to increased product performance.
A review of current state of the art approaches to materials informatics can be found in a recent issue of the Materials Research Society Bulletin . This issue gives an overview of various approaches to understanding and predicting the properties of materials. Approaches for data mining driven predictions of structures, design of ordered alloys, combinatorial experiments and the modeling of materials were discussed.
In much of this work, the role of quantum mechanical methods is stressed [1,2]. The advent of new computational algorithms has gone hand-in-hand with the development of hardware advances. Today, it is possible to calculate literally thousands of hypothetical structures and determine many of their properties accurately. This origin of such computational databases can augment experimental ones to explore and discover new materials and relationships between materials with data mining techniques or via other statistical means [2,3].
Alternatively the quantum mechanical methods can be used without experiment to help design materials with specified properties, e.g., new catalysts [4,5] and new thermoelectrics .