jPCA and rotational dynamics

Churchland and colleagues found that during reaching, motor-cortex population activity contains a prominent rotational component: projected into the right two-dimensional plane, the state circulates in a consistent direction across conditions, as if driven by an underlying oscillator. jPCA is the method built to expose this. It fits the population's time-derivative with a purely rotational (skew-symmetric) dynamics matrix, dz/dt = M z with M = -M-transpose, and then projects the data onto the plane spanned by the eigenvectors with the largest imaginary eigenvalues — the plane in which rotation is strongest. The result is a compact, quasi-oscillatory picture that a look-up-table view of tuning does not predict.

Rotational structure was influential as concrete evidence for the dynamical-systems view, but its interpretation should stay measured. Skew-symmetric fitting will always return the most rotational plane it can find, so demonstrating rotation is necessary but not sufficient to prove an intrinsic oscillator; smoothly varying, temporally structured signals can look rotational, and the same data can be reproduced by networks with quite different internal mechanisms. Rotation is best read as a robust, reproducible signature of structured population dynamics rather than as identification of a specific circuit.