Why is there a difference in model parameter checking between optimization methods?

Hi, When one uses a typical PK ADVAN and a classical optimization method (FO/FOCE/FOCEI), letting a model parameter (most notably CL) become negative or forcing it to be a negative value will cause NONMEM 7.3 to stop and error out of the optimization routine. This is also the case with IMP and BAYES methods. However, for ITS or SAEM, the NONMEM report clearly shows that the optimization goes through many iteration without stopping. Why is that? An error message appears at the bottom of the report outside of the termination section, which seem to indicate that optimization completed... this is also odd Thanks for your feedback