| Summary: | A third-generation photovoltaic technology, PbS quantum dot solar cells boast tunable, infrared-compatible bandgaps, air stability and scalable production. Their unfortunately limited photovoltaic efficiency, a result of low carrier lifetime, has been tentatively ascribed to polydispersity, fusing, or mid-bandgap trap states, the exact nanostructural origin of which remains a subject of discussion. We seek to use DFT simulations to understand the structure-property relationship of PbS QDs. To aid in the computational efforts, we first developed a Python library to automate DFT calculations and facilitate analysis, a framework based on nested directed graphs to manage and schedule concurrent and consecutive tasks, and a sigma.js- and Flask-based web frontend. In an effort to reduce the high computational cost of geometrically relaxing a PbS quantum dot, we explored the use of the Behler-Parrinello approach, recurrent neural networks, as well as a discretized and regularized manybody cluster expansion formulation alongside multilayer perceptrons, in training neural network potentials, as well as in directly accelerating geometry relaxations for PbS quantum dots. Noticing some unexpected nontrivial behavior during the geometry relaxation runs, we sought to quantify and clarify the observed intuitively pathological behavior, both phenomenologically with outlier detection, and physically by inspecting bonds on the quantum dots’ surfaces, touching on their electronic structure in the process.
|
|---|