Energy conservation problems during dynamics , bond breaking and forming for collision reaction

Questions
#1

Hi,
I am tryin to interface Q-CHEM to our own molecular dynamics code for collsion reactions. I noticed that there is problem for energy convervation when molecule are approaching closer followed by bond breaking or forming. This is due to potential energy changes more in each steps which is calculating on the fly.
The system contains eight atoms.

The $rem section is given below

$rem
JOBTYPE FORCE
BASIS 6-31G**
EXCHANGE B3LYP
UNRESTRICTED = TRUE
GUI = 2
MAX_SCF_CYCLES 200
SCF_CONVERGENCE 8
SCF_ALGORITHM DIIS_GDM
GEOM_OPT_PRINT 3
SYM_IGNORE TRUE
SCF_GUESS READ
THRESH 14
S2THRESH 12
$end

Thanks,
Prabhash

#2

Two possibilities. One is that you simply need shorter time steps in strongly repulsive parts of the potential surface. The other is the fact that by reading in the previous MOs as a guess, you are technically violating time reversibility, which implies that integrators that one normally expects to be symplectic (like Verlet or velocity Verlet) are no longer. That, too, can lead to failure of energy conservation. I discussed that a long time ago in this paper: