Mex mjdgges gives a -Inf + NaN * i eigenvalues, .m different

When I solve my model using the mex mjdgges one of the eigenvals comes up as -Inf + NaN * i. However when I force it to use the .m version the eigenval just comes up as Inf + 0 * i.

Since the Matlab version is so incredibly slow it would certainly be nice if I could use the mex one. I presume this is just a numerical accuracy thing though about which not much could be done?

Any suggestions would be much appreciated in any case.



The issue should have been solved in the current snapshot. However, even in the current version it should not affect the results. See [NaN in the list of eigenvalues)

Great thanks.