Chung et al.'s 'Vector Space of Cycles' reveals brain's hidden recurrent loops

Most statistical methods for directed interactions focus on pairwise effects, ignoring large-scale recurrent organization. Chung et al. address this by modeling directed interactions as edge flows on a simplicial complex, then evolving them under an energy-minimizing dynamical system. This separates transient interaction components from persistent harmonic flows, resulting in a low-dimensional cycle space represented as elements of a Hilbert space. The framework allows projection, averaging, comparison, and population-level inference on cyclic structures without enumerating individual cycles.

Theoretical properties include characterization of the cycle space, variance reduction, and population inference. Simulations show significantly improved recovery of cyclic structure in dense recurrent systems compared to existing methods. Applied to resting-state fMRI from 400 human subjects, the framework uncovers reproducible large-scale cyclic organization that is not detectable through standard edgewise averaging. This provides a scalable statistical framework for studying recurrent interactions in high-dimensional dynamical systems like neural and biological networks.