Fig. 6.

Learned Koopman eigenfunctions for the mean-field model of fluid flow past a circular cylinder at Reynolds number 100. a Reconstruction of trajectory from linear Koopman model with two states; modes for each of the state space variables x are shown along the coordinate axes. b Koopman reconstruction in eigenfunction coordinates y, along with eigenfunctions $y=\phi \left(x\right)$. c Two examples of trajectories that begin off the attractor. The Koopman model is able to reconstruct both given only the initial condition