Lyapunov vector

In applied mathematics and dynamical system theory, Lyapunov vectors, named after Aleksandr Lyapunov, describe characteristic expanding and contracting directions of a dynamical system. They have been used in predictability analysis and as initial perturbations for ensemble forecasting in numerical weather prediction. In modern practice they are often replaced by bred vectors for this purpose. In applied mathematics and dynamical system theory, Lyapunov vectors, named after Aleksandr Lyapunov, describe characteristic expanding and contracting directions of a dynamical system. They have been used in predictability analysis and as initial perturbations for ensemble forecasting in numerical weather prediction. In modern practice they are often replaced by bred vectors for this purpose. Lyapunov vectors are defined along the trajectories of a dynamical system. If the system can be described by a d-dimensional state vector x ∈ R d {displaystyle xin mathbb {R} ^{d}} the Lyapunov vectors v ( k ) ( x ) {displaystyle v^{(k)}(x)} , ( k = 1 … d ) {displaystyle (k=1dots d)} point in the directions in which an infinitesimal perturbation will grow asymptotically, exponentially at an average rate given by the Lyapunov exponents λ k {displaystyle lambda _{k}} . If the dynamical system is differentiable and the Lyapunov vectors exist, they can be found by forward and backward iterations of the linearized system along a trajectory. Let x n + 1 = M t n → t n + 1 ( x n ) {displaystyle x_{n+1}=M_{t_{n} o t_{n+1}}(x_{n})} map the system with state vector x n {displaystyle x_{n}} at time t n {displaystyle t_{n}} to the state x n + 1 {displaystyle x_{n+1}} at time t n + 1 {displaystyle t_{n+1}} . The linearization of this map, i.e. the Jacobian matrix   J n {displaystyle ~J_{n}} describes the change of an infinitesimal perturbation h n {displaystyle h_{n}} . That is Starting with an identity matrix Q 0 = I   {displaystyle Q_{0}=mathbb {I} ~} the iterations where Q n + 1 R n + 1 {displaystyle Q_{n+1}R_{n+1}} is given by the Gram-Schmidt QR decomposition of J n Q n {displaystyle J_{n}Q_{n}} , will asymptotically converge to matrices that depend only on the points x n {displaystyle x_{n}} of a trajectory but not on the initial choice of Q 0 {displaystyle Q_{0}} . The rows of the orthogonal matrices Q n {displaystyle Q_{n}} define a local orthogonal reference frame at each point and the first k {displaystyle k} rows span the same space as the Lyapunov vectors corresponding to the k {displaystyle k} largest Lyapunov exponents. The upper triangular matrices R n {displaystyle R_{n}} describe the change of an infinitesimal perturbation from one local orthogonal frame to the next. The diagonal entries r k k ( n ) {displaystyle r_{kk}^{(n)}} of R n {displaystyle R_{n}} are local growth factors in the directions of the Lyapunov vectors. The Lyapunov exponents are given by the average growth rates and by virtue of stretching, rotating and Gram-Schmidt orthogonalization the Lyapunov exponents are ordered as λ 1 ≥ λ 2 ≥ ⋯ ≥ λ d {displaystyle lambda _{1}geq lambda _{2}geq dots geq lambda _{d}} . When iterated forward in time a random vector contained in the space spanned by the first k {displaystyle k} columns of Q n {displaystyle Q_{n}} will almost surely asymptotically grow with the largest Lyapunov exponent and align with the corresponding Lyapunov vector. In particular, the first column of Q n {displaystyle Q_{n}} will point in the direction of the Lyapunov vector with the largest Lyapunov exponent if n {displaystyle n} is large enough. When iterated backward in time a random vector contained in the space spanned by the first k {displaystyle k} columns of Q n + m {displaystyle Q_{n+m}} will almost surely, asymptotically align with the Lyapunov vector corresponding to the k {displaystyle k} th largest Lyapunov exponent, if n {displaystyle n} and m {displaystyle m} are sufficiently large. Defining c n = Q n T h n {displaystyle c_{n}=Q_{n}^{T}h_{n}} we find c n − 1 = R n − 1 c n {displaystyle c_{n-1}=R_{n}^{-1}c_{n}} . Choosing the first k {displaystyle k} entries of c n + m {displaystyle c_{n+m}} randomly and the other entries zero, and iterating this vector back in time, the vector Q n c n {displaystyle Q_{n}c_{n}} aligns almost surely with the Lyapunov vector v ( k ) ( x n ) {displaystyle v^{(k)}(x_{n})} corresponding to the k {displaystyle k} th largest Lyapunov exponent if m {displaystyle m} and n {displaystyle n} are sufficiently large. Since the iterations will exponentially blow up or shrink a vector it can be re-normalized at any iteration point without changing the direction.