Details

Type: Bug

Status: Closed

Priority: Major

Resolution: Fixed

Affects Version/s: 2.0

Fix Version/s: 2.1

Labels:None

Environment:
Eclipse sous Red Hat 5
Description
In a space application using DOP853 i detected what seems to be a bad parameter in the call to the method initializeStep of class AdaptiveStepsizeIntegrator.
Here, DormandPrince853Integrator is a subclass for EmbeddedRungeKuttaIntegrator which perform the call to initializeStep at the beginning of its method integrate(...)
The problem comes from the array "scale" that is used as a parameter in the call off initializeStep(..)
Following the theory described by Hairer in his book "Solving Ordinary Differential Equations 1 : Nonstiff Problems", the scaling should be :
sci = Atol i + y0i * Rtoli
Whereas EmbeddedRungeKuttaIntegrator uses : sci = Atoli
Note that the GraggBulirschStoer integrator uses the good implementation "sci = Atol i + y0i * Rtoli " when he performs the call to the same method initializeStep(..)
In the method initializeStep, the error leads to a wrong step size h used to perform an Euler step. Most of the time it is unvisible for the user.
But in my space application the Euler step with this wrong step size h (much bigger than it should be) makes an exception occur (my satellite hits the ground...)
To fix the bug, one should use the same algorithm as in the rescale method in GraggBulirschStoerIntegrator
For exemple :
final double[] scale= new double[y0.length];;
if (vecAbsoluteTolerance == null) {
for (int i = 0; i < scale.length; ++i)
} else {
for (int i = 0; i < scale.length; ++i)
}
hNew = initializeStep(equations, forward, getOrder(), scale,
stepStart, y, yDotK[0], yTmp, yDotK[1]);
Sorry for the length of this message, looking forward to hearing from you soon
Vincent Morand