E cient Predictor for Co-Simulation with Multistep Sub-System Solvers

An industrial model of a dynamic system is usually not just a set of differential equations. External inputs acting on the system are common, such as an external force acting on a body or wind pressing on a car. Update of these inputs needs to be handled by the numerical solver in an efficient way. In dynamical simulation, multistep methods are commonly used. A multistep method uses the solution history in order to predict the future solution. When an input is changed, the history is no longer a good approximation for the future solution which may result in order reductions and simulation failure. In this paper, a modification of the predictor is presented. Modifying the predictor, instead of restarting the method, results in an increased performance of the method. The cost of the modification must be weighed with the cost of restarting the method. Experiments show that the benefit of modifying the predictor outweighs the cost of a restart.