TY - CONF A1 - Ueltzen, Katharina T1 - Unveiling the potential of the Kanamori-Goodenough-Anderson rules for magnetic property prediction N2 - Recently, machine learning of magnetic properties of transition metal compounds has attracted large interest due to environmental and availability issues of rare-earth-based functional magnetic materials. Surprisingly, bond-angle-derived features were not found to be relevant for magnetic structure prediction in previous studies using DFT-computed labels. This contrasts with a well-known magnetism heuristic, the Kanamori-Goodenough-Anderson (KGA) rules of superexchange. We review magnetic interaction trends within the MAGNDATA database of experimentally determined magnetic structures. Observed trends follow the KGA rules „of thumb“ and exceptions can be rationalized. We introduce a new, informative label for predicting magnetic structures that can be extended to magnetic sites and structures of arbitrary complexity. Bond-angle-derived features are found to be highly relevant for magnetic structure prediction. T2 - Faraday Discussion: Data-driven discovery in the chemical sciences CY - Oxford, UK DA - 10.09.2024 KW - Magnetism KW - High-throughput analysis KW - Machine learning KW - Transition metal compounds PY - 2024 AN - OPUS4-62250 LA - eng AD - Bundesanstalt fuer Materialforschung und -pruefung (BAM), Berlin, Germany ER - TY - CONF A1 - Ueltzen, Katharina T1 - Revisiting the Kanamori-Goodenough-Anderson rules for magnetic property prediction N2 - Recently, machine learning of magnetic properties of transition metal compounds has attracted large interest due to environmental and availability issues of rare-earth-based functional magnetic materials. Surprisingly, bond-angle-derived features were not found to be relevant for magnetic structure prediction in previous studies using DFT-computed labels. This contrasts with a well-known magnetism heuristic, the Kanamori-Goodenough-Anderson (KGA) rules of superexchange. We review magnetic interaction trends within the MAGNDATA database of experimentally determined magnetic structures. Observed trends follow the KGA rules „of thumb“ and exceptions can be rationalized. We introduce a new, informative label for predicting magnetic structures that can be extended to magnetic sites and structures of arbitrary complexity. Bond-angle-derived features are found to be highly relevant for magnetic structure prediction. T2 - 60th Symposium on Theoretical Chemistry (STC 2024) CY - Braunschweig, Germany DA - 02.09.2024 KW - Magnetism KW - High-throughput analysis KW - Machine learning KW - Transition metal compounds PY - 2024 AN - OPUS4-62249 LA - eng AD - Bundesanstalt fuer Materialforschung und -pruefung (BAM), Berlin, Germany ER -