Predicting critical transitions in assortative spin-shifting networks

Methods to forecast critical transitions, i.e. abrupt changes in systems' equilibrium states have relevance in scientific fields such as ecology, seismology, finance and medicine among others. So far, the bulk of investigations on forecasting methods builds on equation-based modeling methods, which consider system states as aggregates and thus do not account for the different connection strengths in each part of the system. This seems inadequate against the background of studies that insinuate that critical transitions can originate in sparsely connected parts of systems. Here we use agent-based spin-shifting models with assortative network representations to distinguish different interaction densities. Our investigations confirm that signals of imminent critical transitions can indeed be detected significantly earlier in network parts with low link degrees. We discuss the reason for this circumstance on the basis of the free energy principle.