You probably should provide information about the program you are trying to use. Furthermore, you can either do a geometry optimization OR a AIMD simulation. These things are mutually exclusive.

Mrs/Miss Josue,

There is no sense to loss so much computational time for AIMD, even using DFT improvements towards computational costs, for, in fact, one and the same information, which can be obtained by MM/MD using force field approaches accompanied with suitable DFT static computations.

Please find as attachment computations of your system, using all mentioned approaches. The information about H-transfer you have received by first, computations MM/MD-force field (Fig. 1, attachment) and static DFT (Fig. 2). The H-transfer is the most stable state. The static state as shown corresponds to the global minimum.

The Fid. 3 is AIMD (DFT), where you have obtained the same information, however in an increasing in more than 30 times computational cost (CPU) over the approach above, because of at each step of the trajectory you use accurate optimization.

Please bear in mind that the computations by AIMD as you can see are even not finished...

You can compute set of different dispositions obtained by MM/MD, by static DFT, thus receiving a picture of AIMD within the frame of a significantly reduced computational time, first. Second, you have ability to use highly accurate methods for the static DFT computational step, thus increasing accuracy of the information obtained.

Because of you use Gamess you should employ instead the shown theoretical level, for example, GBASIS = MNDO for semi-empirical computations with $MD (Fig. 4) or $GLOBOP for Monte Carlo simulations (Fig. 5). Then you should perform optimization at high level of theory by static DFT methods.