Atomistic and ab initio prediction and optimization of thermoelectric and photovoltaic properties

Abstract

The accurate prediction of physical properties in the vast spaces of nanoscale structures and chemical compounds is made increasingly possible through the use of atomistic and ab initio computation. In this thesis we investigate lattice thermal conductivities KL and electronic band gaps E,, which are relevant to thermoelectric and photovoltaic applications, respectively, and develop or modify computational tools for predicting and optimizing these properties. For lattice thermal conductivity, we study SiGe nanostructures, which are technologically important for thermoelectric applications. From computing aL for various SiGe nanostructures, we establish that the Kubo-Green approach using classical molecular dynamics (MD) gives additional quantitative predictions not available from phenomenological models, such as the existences of a minimum value of KL as the nanostructure size is varied and of configurational dependence of KL. We carry out the minimizatin of KL in the space of atomic configurations in SiGe alloy nanowires and demonstrate the feasibility of using the cluster expansion technique to parameterize KL. We find that the use of coarse graining and a meta cluster expansion approach is effective, in conjunction with a genetic algorithm, to find configurations which drastically lower KL. The low values of KL obtained, close to the bulk amorphous limit, are due to the absence of long-range order, and such absence allows a local cluster expansion approach to optimize KL. We examine ab initio bandgap prediction for semiconductor compounds, and address the large errors of Kohn-Sham band gaps in density functional theory (DFT).

(cont.) We apply corrections using the self-energy approach in the GW approximation, which includes non-local screened exchange and correlation, and find that the G₀W₀ approximation significantly reduces prediction errors compared to Kohn-Sham band gaps, though at much higher computational cost. We propose a new method involving total energies in DFT to predict the fundamental gap, by use of the properties of the screening or exchange-correlation hole in an electron gas. With this method, we are able to efficiently predict band gaps that are in agreement with experimental values.