At least for the Taylor 1.5 integrator, trying to pass a 2d array for U1s fails to return an ensemble of integrated trajectories (despite the documentation indicating that it should). This is presumably because the integrator code doesn't know how to broadcast correctly.