• Epa, Vidana Chandana
• Winkler, Dave
• Le, Tu

Abstract

Materials design and synthesis is undergoing a rapid paradigm shift, moving from single, relatively simple materials to libraries of complex substances with properties increasingly designed to be ‘fit for function’. Although good, predictive computational modelling methods exist for the current “one-material-at-a–time” scenario, few methods exist that are able to make robust predictions of properties of large numbers of new materials generated by high throughput synthesis and characterization methods, now increasingly being developed and deployed. These paradigm shifts in materials design, synthesis, and characterization are similar to that which took place in the pharmaceutical industry two decades ago, when combinatorial chemistry and high throughput synthesis methods were developed.Very efficient and effective methods for developing predictive models from these data sets were developed to meet the challenges posed by these new methods. One of the most useful techniques employs statistical and machine learning methods, coupled to novel mathematical descriptions of molecular properties, to generate quantitative structure-property relationships. The depth of experience in the pharmaceutical industry, and recent refinements of these methods using new discoveries in mathematics and statistics, is potentially transferable to high throughput materials science.Surprisingly, compared to the pharmaceutical science, relatively little published work exists on how these methods can be used to model materials. This review summarizes the state of the art in quantitative structure-property modelling methods applicable to materials and summarizes published studies of materials structure-property relationships.We provide a concise and timely overview of the computational techniques that are capable of robust, versatile, and predictive modelling of large materials data sets. We also illustrate by examples, how these methods are applied to moderate to large materials property data sets, and provide a perspective on the value of methods to the development of the novel materials.

Topics

• impedance spectroscopy
• machine learning