The numerical approximation of center manifolds in Hamiltonian systems. (Q1421173)

This paper is concerned with the numerical approximation of center manifolds around equilibrium points in Hamiltonian systems: \[ x' = J \nabla H(x), \quad x \in {\mathbb R}^{2n}. \tag{1} \] Although the method proposed here can be applied in a general setting, it turns out useful only for high dimensional systems with a low dimensional invariant subspace. Assuming that \( x=0\) is an equilibrium point of (1) and that \( A = J H_{xx}(0)\) has \(2l\) eigenvalues on the imaginary axis and \( 2m = 2n-2l\) off this axis with \( l <<q\), the theory of center manifold and Darboux's theorem [cf. \textit{R. Abraham} and \textit{J. E. Marsden}, Foundations of Mechanics, 2nd ed. (1978; Zbl 0393.70001)] imply that the center manifold can be locally parameterized by \( x = f (w), w \in {\mathbb R}^{2l}\) where \(f\) is a symplectic mapping. Then, the invariance condition of the manifold: \( f'(w) J_l ( f'(w))^T \nabla H(f(w)) = J \nabla H(f(w))\) together with the symplecticity condition of \(f\) allow them to compute recursively the Taylor expansion of \(f\) up to any order. For the practical implementation, the authors propose to use a mixed block elimination of \textit{W. Govaerts} and \textit{J. D. Pryce} [IMA J. Numer. Anal. 13, 161--180 (1993; Zbl 0778.65017)] for the linear systems involved into the above approach together with a sparse solver for the Hermitian submatrices. The paper ends with an application of the method to a Hamiltonian semidiscretization of the problem: \( u_{tt} + u_{xx} + F(u)=0,\) \(u(t,0)=u(t,1)=0,\) \( x \in [0,1], t \in {\mathbb R}\) with \( F(u)= 2 \pi^2 u + 3 u^2 \) with a uniform grid in the \( x \) variable in the interval \([0,1]\) computing explicitly up to five--order terms.