I don't think this is necessarily due to differences in initial conditions. Even in the models were algebraic (i.e. not differential equations), having too many parameters and not enough data would lead to wildly erroneous predictions. You would be effectively fitting the model to noise. When the model does involve derivatives, it might be possible for the system to exhibit chaotic behavior, but that is not a necessity. It could be asymptotically or neutrally stable, but the prediction of the stable points could be wildly wrong due to erroneous parameter choices for the reasons I stated above.
So, as one of the earlier posts indicated, the solution would be to simplify the model by decreasing the number of parameters and getting more data.
Slashdot Mirror
I don't think this is necessarily due to differences in initial conditions. Even in the models were algebraic (i.e. not differential equations), having too many parameters and not enough data would lead to wildly erroneous predictions. You would be effectively fitting the model to noise. When the model does involve derivatives, it might be possible for the system to exhibit chaotic behavior, but that is not a necessity. It could be asymptotically or neutrally stable, but the prediction of the stable points could be wildly wrong due to erroneous parameter choices for the reasons I stated above. So, as one of the earlier posts indicated, the solution would be to simplify the model by decreasing the number of parameters and getting more data.