Trying to calculate a structure using sa.inp script and the energy in the cooling loop increases instead of going down.

In the end there is a message

****&&&& rerun job with smaller timestep (i.e., 0.003)

But making the timestep smaller does not help. Could anyone suggest where to look?

edit: I've also commented out my restrains thinking that maybe my noe's are too whacky, but still structure diverges.

Thanks!