1: \begin{abstract}

2: 	

3: 	Lyapunov exponents are important quantities in the analysis of complex dynamical systems. Most commonly, they are computed with Benettin's algorithm, which propagates linear perturbations along a background trajectory. While many application trust in convergence of the algorithm, they often do not consider the effects of numerical errors. In fact, the convergence of Benettin's algorithm strongly depends on the relation between numerical errors and stepsizes. With the usual constant stepsizes, integration errors accumulate and lead to limits different from the Lyapunov exponents.\par 

4: 	

5: 	In this article, we demonstrate how integration errors can affect the convergence of Benettin's algorithm and suggest stepsize variations and weighted averages to counter persistent errors. Our examples demonstrate that both of the suggested approaches improve the accuracy and the speed of convergence of Benettin's algorithm. More rigorous results are obtained in a theoretical analysis, in which we derive conditions for stepsizes and weights under which Benettin's algorithm convergences in the presence of linear integration errors.\par 

6: 	

7: \end{abstract}

8: