Numerical Hopf bifurcation of linear multistep methods for a class of delay differential equations (Q1004435)

This paper is concerned with the numerical solution of parameter dependent systems of delay-differential equations: \( y'(t) = f( y(t), y(t-1), \tau ), t \geq 0 \) with the parameter \( \tau >0\) and \( y \in {\mathbb R}^d\) undergoing a Hopf bifurcation at some \( \tau = \tau^*\) by means of linear multistep methods. In a previous paper, \textit{N. J. Ford} and \textit{V. Wulf} [J. Comput. Appl. Math. 111, No. 1--2, 153--162 (1999; Zbl 0941.65132)] showed that for a scalar equation, strictly stable linear multistep methods with step size \( h= 1/m\), \( m \in {\mathbb{N}} \) yield to discrete schemes with a bifurcation at \( \tau (h)= \tau^* + O(h^p)\) where \(p\) is the order of the method. The present authors extend these results to the case of \(d\)-dimensional systems with \( d>1\). More specifically, under suitable assumptions on the Jacobian matrices \( f_x(x,y,\tau)\), \(f_y(x,y,\tau)\) in some neighborhood of \( (x,y,\tau)=(0,0,\tau^*)\) they extend the results of Ford and Wulf. In addition, some results on the direction of the numerical Hopf bifurcation and the stability of the bifurcating invariant curve are derived. The paper ends presenting some numerical results for a 2-dimensional equation with a known supercritical bifurcation \( \tau^*\) testing the order of accuracy of \( \tau (h) - \tau^* = O(h^p)\) and stability of several well known linear multistep methods.