SCF goes wrong in QMMM after 37000 steps

[Csilla Varnai]

Dear All,

I'm running a QMMM simulation for 32 water molecule box with 1 qm  
molecule, NVE. The energy is not conserved, goes up gradually. Between  
steps 37186th and 37270 at several steps the SCF goes wrong. It does  
not converge in 50 steps, the energy increases and then oscillates.  
Afterwards, it seems to be fine until 40000th step.
The QM cell is 10x10x10 Angstrom, density cutoff 300 Ry, time step 0.5  
fs, temperature 300 K. As for the SCF, I'm using the default  
parameters. MAX_ITER is 50. For the other steps, SCF never required  
more than 20 steps except a few ones, in which case it still remained  
below 31. Any idea why the SCF went wrong?
Thanks very much,

Csilla