Can simple exchange heuristics guide us in predicting magnetic properties of solids?

<jats:p> The Kanamori-Goodenough-Anderson rules are a textbook heuristic for predicting magnetism. They connect bond angles to magnetic ordering in some transition-metal compounds. Such domain knowledge is highly important for building predictive machine learning models in scenarios with scarce data. Yet, there has been no statistical, large-scale evaluation of the heuristic. Here, we evaluate this heuristic relying on both bond angles and orbital occupancy on an experimental database of magnetic structures. We observe that the heuristic is satisfied for 79% of larger bond angles (&gt;110°) but not for bond angles around 90°, and we discuss the exceptions. We then demonstrate that integrating this heuristic into machine learning models for predicting magnetic ordering improves predictive quality. Notably, these magnetism models can also predict whether non-collinear magnetic ordering might occur. Furthermore, the heuristic provides a useful benchmark for evaluating theoretical methods that calculate magnetic properties. We showcase this here for a well-known DFT+ <jats:italic toggle="yes">U</jats:italic> dataset. </jats:p>