Separating Stability and Flexibility in Task Switching via Attractor-Landscape Geometry

Everyday cognitive control requires a balance between stability and flexibility. Although these abilities are often treated as conflicting, it remains unclear whether they must be governed by a single trade-off mechanism. To test for this possibility, we model reaction time data from a well-balanced task-switching paradigm using a dynamical systems account in which task sets correspond to attractors in a working memory landscape. In this framework, we formalize stability as the local curvature around an attractor (which determines how strongly the state is restored after perturbation) and flexibility as the barrier height between attractors (which determines switching difficulty). This separation yields a mechanistic decomposition in which stability and flexibility are controlled by distinct geometric properties. We have implemented this shared 3D dynamical landscape model and fit it to task-switching reaction time data using parallel tempering. A unique landscape was estimated for each task condition, representing the item-wise and list-wise statistical properties of each task block. As a next step, we treat the alignment between these task-defined landscapes and each participant’s effective low-dimensional dynamics as an individual-specific parameter. Differences in this alignment should yield distinct effective 2D geometries—producing systematic changes in apparent local curvature and barrier height—while the underlying attractor structure should remain fixed. We will further examine whether the stability–flexibility relationship in our model is participant-dependent. For example, some participants may exhibit a negative coupling (trade-off), whereas others may show positive coupling (co-improvement) or near-independence, depending on how the induced 2D curvature and barrier height co-vary.