Direct computation of period doubling bifurcation points of large-scale systems of ODEs using a Newton-Picard method (Q4937503)

This paper is concerned with the computation of period doubling bifurcations of systems of ordinary differential equations (ODEs) whose state space have a special structure. The approach proposed here is based on the ideas of \textit{K. Lust} and \textit{D. Roose} [SIAM J. Sci. Comput. 19, No. 4, 1188-1209 (1998; 915.65088)] to compute periodic solutions of certain large-scale systems of ODEs. Denoting by \( x(t;x_0,\lambda)\) the solution of an autonomous system of ODEs \( x'=f(x,\lambda)\) where \(\lambda\) is a real parameter, the computation of non-degenerate period doubling points requires the solution of the boundary value problem \( x(T;x_0,\lambda)-x_0=0\) for some \(x_0\) and \(T\) together with the condition that a real Floquet multiplier crosses the unit circle transversally at \(-1\) and, except for this one and the trivial Floquet multiplier, no other Floquet multiplier has modulus equal to 1. These conditions are supplemented with two scalar equations introduced to eliminate the time translation and to fix the parameter of the orbit. In the standard approach the Newton-type solution of these nonlinear equations involves the computation of the monodromy matrix \( M = \partial x(T)/ \partial x_0\) and for high-dimensional systems this is very expensive. However for problems with only a few unstable multipliers the authors show that by using the orthogonal projector of the subspace of the unstable multipliers it is possible to reformulate the boundary value problem so that Picard iteration can be applied without computing all elements of \(M\), and therefore avoid the above mentioned difficulty. The paper also includes also some remarks on computational aspects and the convergence of the suggested approach together with the results of some model problem which arises in a spatial discretization of one-dimensional reaction-diffusion system.