Abstract
The brain has been described as the quintessential complex dynamic system (Pushkarskaya et al 2025). Neurons are fundamentally nonlinear information-processing units, and the interconnection of approximately 100 billion of them into complex systems characterized by threshold events, state transitions, feedback loops, and self-organization makes the human brain perhaps the most complex object ever subjected to scientific study.
Of course, progress in studying any complex system requires simplifying assumptions. The most commonly used strategies for analyzing brain data ignore much of its complexity by assuming static organization over time. For example, resting-state functional connectivity analysis of functional magnetic resonance imaging data typically examines linear relationships between measured activity in different brain regions over extended periods of time, implicitly assuming that what is most interesting is constant throughout the epoch. Indeed, it has been argued that we should average across longer epochs of resting-state data (e.g., Birn et al 2013). This can effectively improve the reliability of measures of connectivity, but at the cost of unrealistically assuming static organization over ever longer periods. Functional brain maps based on such connectivity patterns can create an illusion of static brain organization that may be oversimplified. Dynamic functional connectivity analyses, such as sliding-window connectivity analyses, begin to address this problem (Hutchison et al 2013; Peterson et al 2025), but the static methods are so easy to apply that they still predominate in the field.
Approaches that seek to simulate dynamics across the whole brain (Deco and Kringelbach 2025) simultaneously represent the most ambitious effort to embrace the brain’s complexity and the most radical simplifying assumptions—for example, representing brain regions consisting of hundreds of thousands of neurons with a single excitatory and a single inhibitory unit. Despite such (necessary) simplifying assumptions, whole-brain models can give insight into the dynamic organization of brain processes at the largest scale.
Dynamic modeling of complex brain processes is hard; simple, static linear models have a seductive appeal. But the brain is not simple. Ongoing advances in dynamic modeling (Pushkarskaya et al 2025) are encouraging, and dynamic analytic tools should be more broadly deployed to advance our understanding of information processing in the brain.