Characterizing dynamics with covariant Lyapunov vectors
We introduce a general, innovative approach to determine, in
both continuous- and discrete-time dynamical systems, an intrinsic set of
directions at each point of phase space that are covariant with the dynamics
as well as invariant under time reversal. The knowledge of these covariant
Lyapunov vectors (CLV) allows for an accurate characterization of the
underlying dynamics. In particular, it is possible to quantify the degree of
hyperbolicity or to compute the complete spectrum of Lyapunov exponents via
ensemble averages over the attractor invariant measure rather than through
time averages.
Moreover, regarding the search for "collective modes", we show that the
statistical properties of CLV differ from those of the vectors obtained from
the standard orthogonalization procedure introduced by Benettin et al. In
particular, we show evidence in both dissipative and Hamiltonian spatially
extended systems that CLV are strictly localized almost everywhere in the
asymptotic spectrum, thus providing a meaningful hierarchical decomposition
of spatiotemporal chaos.