Thema der Dissertation:
Systematic identification of relevant features for the statistical modeling of materials properties of crystalline solids
Systematic identification of relevant features for the statistical modeling of materials properties of crystalline solids
Abstract: Over the past two decades, statistical tools are increasingly used to accelerate the search for new materials with desired properties. So far, these tools have been applied on a case-by-case basis, requiring expertise and knowledge in statistical modeling and in relating the essential physics and chemistry to the desired materials property. In this talk, I present a computational method to model the underlying physical relationships based on features used to characterize each material. The method allows us to accurately estimate the properties of new materials, which would otherwise require costly and time-consuming experiments or simulations.
In the first part, I explain the methodology and the concepts I have developed to identify the features related to a property of interest, highlighting the challenges and pitfalls in statistical modeling. In the second part, I present the results of applying the method to estimate elastic properties of inorganic crystalline compounds, finding that it greatly simplifies the construction of statistical models and that the elastic properties can be described by simple analytical expressions. Finally, I highlight a potential screening application by using the same model to estimate the elastic properties of not yet investigated materials and show that the method accurately models the statistical trend of the underlying physical relationship for a wide range of k-nary compounds with different chemical compositions and structures.
Conclusively, these findings suggest that the method can be used as a standardized procedure to automatically identify the set of features related to a property of interest and to provide a means to identify physical relationships from data. Thus, the method has the potential to accelerate materials design and to gain more insights from known materials.
In the first part, I explain the methodology and the concepts I have developed to identify the features related to a property of interest, highlighting the challenges and pitfalls in statistical modeling. In the second part, I present the results of applying the method to estimate elastic properties of inorganic crystalline compounds, finding that it greatly simplifies the construction of statistical models and that the elastic properties can be described by simple analytical expressions. Finally, I highlight a potential screening application by using the same model to estimate the elastic properties of not yet investigated materials and show that the method accurately models the statistical trend of the underlying physical relationship for a wide range of k-nary compounds with different chemical compositions and structures.
Conclusively, these findings suggest that the method can be used as a standardized procedure to automatically identify the set of features related to a property of interest and to provide a means to identify physical relationships from data. Thus, the method has the potential to accelerate materials design and to gain more insights from known materials.
Zeit & Ort
06.07.2022 | 15:00
Hörsaal A (1.3.14)
Fachbereich Physik, Arnimallee 14, 14195 Berlin