Skip to main content

Big-Data Analytics for the Thermal and Electrical Conductivity of Materials from First Principles

Periodic Reporting for period 2 - TEC1p (Big-Data Analytics for the Thermal and Electrical Conductivity of Materials from First Principles)

Reporting period: 2019-04-01 to 2020-09-30

Thermal conductivity (TC) and electric conductivity (EC) are key characteristic of many materials, particularly those used in the energy and environment sectors. Finding novel materials with particular behavior of TC and EC is highly relevant for developing new technologies for a more sustainable energy use, e.g. materials for thermal-barrier-coatings for more fuel-efficient turbines (for aircrafts), and thermoelectric devices for the conversion of waste heat to electrical current.

However, TC is largely unknown − of the 225,000 identified inorganic semiconductor and insulator crystals, only about 100 have any TC data available. To address this, we will use ab initio molecular dynamics, a method in which computer simulations are used to understand the movement of atoms and molecules from first principles, and combine it with big-data analytics (e.g. artificial intelligence). This way, we plan to generate quantitative values and understanding of TC (and electrical conductivity (EC)) for most of these 225,000 materials, as well as for materials not yet discovered. The final goal is, in analogy to Mendeleev’s table of the elements, to build maps that arrange existing and predicted materials according to their TC and EC properties.
Achieving the main objectives of TEC1p involves three tasks, i.e.
(a) improving and developing methodologies which start from first principles for assessing heat and charge transport in materials with unprecedented accuracy,
(b) embedding these techniques in high-throughput frameworks so one can scan over the material space with high efficiency, and
(c) developing and improving artificial-intelligence approaches to find materials of interest.
In the first period of the project, there has been impressive progress in all three tasks. With respect to task (a), we are building on our earlier achievements in accurately computing TCs via ab initio molecular dynamics. The method has been advanced and presently we are extending key aspects to also cover electronic transport and thus to calculate EC.
Concerning (b), we have concentrated our efforts into the development of a high-throughput framework that is able to perform the complex workflows necessary to compute TC and EC. The work has progressed significantly and we are presently preparing a publication describing the science behind it and the practicability of this approach .
Regarding task (c), we focused on further developing artificial-intelligence methods specifically adapted for materials science for the prediction of (materials) properties of interest and to recognize trends and anomalies in materials big data. Furthermore, we applied the improved methodologies to materials-science relevant challenges.
For the first period of this project, we would like to highlight the above-mentioned development of a method to predict electronic properties of materials. We can now make predictions up to the melting point, thus covering also those thermodynamic regimes for which previous approaches struggled. Another break-through was the development of artificial-intelligence methods contextualized and tailored to materials science, the latter being a major open challenge in the field of big-data science.

The methods that are developed in TEC1p and the extensive calculations that are executed are both innovative and timely. They will greatly progress scientific knowledge of the physical properties of materials. The impact of the concepts, methodology, and results will be far reaching for materials science, novel materials discovery, engineering, and industry. The data will be presented in a FAIR manner in the NOMAD Laboratory (