Pathological neural synchronization drives Parkinson's tremor, epileptic seizures, and essential tremor. Millions of neurons firing in lockstep when they shouldn't. Therapeutic stimulation — deep brain stimulation, transcranial current — disrupts the synchrony. The challenge is knowing when and where to stimulate. Most approaches require a model of the neural dynamics: coupling strengths, intrinsic frequencies, connectivity maps. The model is never exact, and the parameters drift.
Wilson and Bhatt (arXiv:2603.12878, March 2026) find a model-free target. For any population of coupled oscillators trapped in a pathological limit cycle, there exists a single point in phase space — the centroid of the limit cycle — that is both trivially computable from observations and robust across coupling strengths.
The centroid is not special because of its location. It is special because of its geometry. A stimulus pulse directed toward the centroid has the property of spreading the oscillators' phases uniformly around the cycle, regardless of where individual oscillators happen to be when the pulse arrives. The uniformity means desynchronization. And because the centroid is a geometric property of the cycle rather than a dynamic property of the oscillators, it doesn't change when coupling strength varies — the same target works for mild and severe synchronization.
Computing the centroid requires only time-series data: record the population activity, trace the limit cycle in phase space, average the coordinates. No model fitting. No parameter estimation. No connectivity map.
The structural lesson is that the right coordinate system can absorb the complexity of a dynamical system into a single computable quantity. The neural population has millions of degrees of freedom. The limit cycle reduces the relevant dynamics to a closed curve. The centroid reduces the curve to a point. Each reduction discards information — but in this case, only irrelevant information. The therapeutic target lives in the geometry, not the dynamics.