hide
Free keywords:
-
Abstract:
Growths in the economy and population, a modern lifestyle, and a fully digitalized and connected world increase the global energetic demands each year. Currently, fossil fuels make up 80% of the global energy consumption, the combustion of which being the main driving force for the disastrous effects of climate change. Controlling and reducing CO2 emissions are therefore key challenges of modern society. Renewable energy sources such as wind and solar panel would be ideal solutions to this problem, however both are a.) not necessarily predictable and b.) not evenly distributed geographicaly. To enable an even energy distribution, we require efficient energy storage. In the past, the combustion of coal and oil has been so successful as that is what carbon-based chemicals are: extraordinarily efficient forms of energy storage.
Many applications, such as laptops, mobile phones, and electric vehicles, utilize lithium-ion batteries as their primary energy storage. While lithium-ion batteries using liquid electrolytes entered the market in 1991, all-solid-state lithium-ion batteries (ASS-LIB), although investigated for decades, are still not widely applied. They promise several advantages in comparison to liquid electrolyte batteries: minimizing fire hazards, longer cycle lifetimes, more comprehensive temperature ranges, and enhanced energy density by potential usage of Li metal anodes. In particular, solid electrolytes of the Li2S-P2S5 (LPS) material class have gained substantial attention due to their favorable properties. First, they possess high RT conductivities of up to 10−2 S/cm for crystalline LPS components, which ranks them among the most conductive solid electrolytes. Secondly, they are composed of the earth-abundant elements sulfur and phosphorous enabling applications at large scales. However, this material class’ design of potent SSE is hampered by the poor understanding of structure-property relations. This manifests in massive deviations in reported Li-ion conductivity in different experimental setups and from theory and experiment.
Simulations based on Density Functional Theory (DFT) or classical force elds (FF) have enabled material comprehension e.g. new insights into material properties for decades. Insights at the atomistic level are irreplaceable for a mechanistic understanding of chemical processes. Unfortunately, due to high computational costs, DFT methods are limited to small systems while providing a highly accurate and complete description. At a much reduced computational cost, classical FFs allow to account for such effects. Yet, here the problem is an often reduced accuracy in the description of the potential energy surface (PES). To this end, emerging Machine Learning (ML) methods have shown to be increasingly able to bridge this gap, with good rst-principles accuracy at a much reduced computational cost. However, the basic assumption of locality, implying the neglect of long-range interactions, is problematic in many cases.
To this end, the central topic of this thesis is threefold. First, we intended to systematically identify systems where this locality assumption does not hold. We especially tried to understand when and why the locality assumption holds for polar and ionic systems and when it fails. Second, as we realized that local ML models accurately predict isotropic bulk material properties, we developed a near-universal Gaussian Approximation Potential (GAP) model for the crystalline and amorphous compounds in Li2S-P2S5. We then used the GAP model to systematically investigate the effect of the local anion composition in glassy Li2S-P2S5 compounds.
At the same time we realized that a short-range model can accurately describe isotropic systems, we understood that we need an accurate description of non-local interactions for non-isotropic systems. To this end, we developed the kernel-based charge equilibration scheme called kQEq. The novel kQEq schemes enable the prediction of partial charges based on local environments by
including the ability to predict non-local charge transfer.