The Materials Genome Initiative and several other global efforts were introduced to develop the infrastructure needed to quickly discover and deploy advanced materials. At the heart of such efforts are high-throughput data-driven studies that quickly find the most promising set of candidate materials without the need for extensive synthesis and/or experimentation, thus saving both time and money. Descriptors, or features as they are known to machine learning practitioners, play a very important role in data-driven studies in materials science. A good set of descriptors can lead to a better insight into the underlying physical phenomena and in some cases, even provide useful guidelines for the design as well as synthesis of new materials. Descriptors for inorganic materials tend to be motivated by the problem at hand which affects their transferability across different application areas. To highlight their importance and to discuss key issues related to them, we survey the recent literature for different descriptors for inorganic materials. In the process, we point out the most important descriptors (if reported) for specific materials science problems. Additionally, we highlight the crucial significance of a good set of descriptors via a case study of predicting formation energies. Further, we discuss issues related to the construction and transferability of new descriptors as well as the combination of different types of descriptors for the same material.
Virial coefficients are unique thermodynamic properties of a system owing to their link between interactions at the molecular level to macroscopic quantities such as the pressure. In my research, we take advantage of this feature and compute virial coefficients of a variety of systems by performing simulation studies. The nature and quality of the interaction potential used in such studies highly affects the quality of the resulting virial coefficients. Therefore, we have employed ab initio based interaction potentials that are state-of-the-art and have been developed using high quality quantum chemistry calculations. Naturally, the complexity of such simulations is a strong motivator for the development of algorithms that are highly efficient and yield precise results. In this regard, we have developed two efficient and novel algorithms for use in Path Integral Monte Carlo (PIMC), a method used to incorporate nuclear quantum effects in virial coefficient calculations for diatomic molecules. We have successfully applied these algorithms to compute virial coefficients including quantum effects or, in short, quantum virial coefficients, for H2 and N2 systems. In addition to applying these algorithms to study diatomic molecules, we have also investigated other algorithms like PIMC using semi-classical beads and compared them to conventional PIMC, for He as well as N2 systems.