Stepsize restrictions for boundedness and monotonicity of multistep methods (Q421333)

The paper is concerned with nonlinear monotonicity and boundedness properties of linear multistep methods for initial value problems. The initial value problem for systems of ordinary differential equations on a vector space \(\mathbb V\) \[ u'(t)=F(u(t)),\quad t\geq 0, \quad u(0)=u_0 \tag{1} \] is considered. Let \(\|\cdot\|\) be a norm or a seminorm on \(\mathbb V\) (in this paper some of the results are formulated with sublinear functionals instead of seminorms). It is assumed that there exists a constant \(\tau_0\) such that \[ \|v+\tau_0F(v)\|\leq \|v\| \quad \text{for all } v\in \mathbb V. \] Let \(u_n\) (\(n\geq k\)) be approximations of the exact solution at \(t_n=n\Delta t\) which are computed by a linear \(k\)-step method. The starting values \(u_j\) (\(0\leq j<k\)) are computed by a Runge-Kutta method. With arbitrary starting values, the property \[ \|u_n\| \leq \max_{0\leq j<k}\|u_j\| \tag{2} \] for \(n\geq k\) under the stepsize restriction \(\Delta t\leq c\tau_0\), where \(c>0\) is a constant, is referred to as monotonicity for linear multistep methods. Unfortunately, this property does not hold for many linear multistep methods of practical interest. Instead of (2), the authors consider a boundedness condition that is weaker than (2), namely \[ \|u_n\| \leq \mu \cdot \max_{0\leq j<k}\|u_j\| \tag{3} \] for \(n\geq k\) under the stepsize restriction \(\Delta t\leq \gamma\tau_0\), where the stepsize coefficient \(\gamma>0\) and the factor \(\mu\geq 1\) are determined by the multistep method. Using the general framework of \textit{W. Hundsdorfer} et al. [SIAM J. Numer. Anal. 47, No. 5, 3797--3819 (2009; Zbl 1209.65074)], necessary and sufficient conditions for boundedness property (3) are obtained. These conditions are relatively transparent and easy to verify numerically for popular families of linear multistep methods. In addition, conditions that ensure monotonicity (2) for linear multistep methods combined with starting Runge-Kutta procedures are given. Finally, computation of the stepsize coefficients \(\gamma\) for boundedness or monotonicity is discussed in details and illustrated by several numerical examples.