Predicting chaotic statistics with unstable invariant tori

Abstract: It has recently been speculated that statistical properties of chaos may be captured by weighted sums over unstable invariant tori embedded in the chaotic attractor of hyperchaotic dissipative systems; analogous to sums over periodic orbits formalized within periodic orbit theory. Using a novel numerical method for converging unstable invariant 2-tori in a chaotic PDE, we identify many quasiperiodic, unstable, invariant 2-torus solutions of a modified Kuramoto-Sivashinsky equation exhibiting hyperchaotic dynamics with two positive Lyapunov exponents. The set of tori covers significant parts of the chaotic attractor and weighted averages of the properties of the tori -- with weights computed based on their respective stability eigenvalues -- approximate statistics for the chaotic dynamics. These results are a step towards including higher-dimensional invariant sets in a generalized periodic orbit theory for hyperchaotic systems including spatio-temporally chaotic PDEs.