No label defined (Q4353863)

This paper is concerned with the efficient computation of the Poincaré normal form around a critical point of a system of nonlinear ordinary differential equations. Assuming that such a critical point is \({\mathbf x}= \mathbf{0}\), and that the Taylor expansion around this point is \({\mathbf x}'= {\mathbf A} {\mathbf x} +\sum_{k\geq 2} {\mathbf F}_k ({\mathbf x})\) where \({\mathbf F}_k\) are homogeneous polynomials of order \(k\) in \({\mathbf x} =(x_1, \dots, x_s)\), there is a well established methodology that allows to reduce to normal form in their natural consecutive orders \(k=2,3, \dots\). This is carried out for the \(k\)-order terms by means of a near-identity transformation \({\mathbf x}= {\mathbf y} +{\mathbf P}_k ({\mathbf y})\) where \({\mathbf P}_k\) is an homogeneous polynomial of degree \(k\) suitably chosen. However as remarked by the authors this natural order reduction is often very inefficient. They propose order-sequences different from the natural \(k=2,3, \dots\) of near-identity transformations and prove that they are optimal (in the sense that they minimize the number of term updates) under some conditions on the frequencies of the system. Finally some computational remarks on the use of an algebraic algorithm proposed by the authors to compute a matrix representation of the Lie-Poisson bracket are also given.