When a geometry is optimized using MOZYME, and a single-point calculation is run on the final geometry, sometimes the results of the single-point calculation do not match those of the full geometry optimization. This can be very frustrating, and can call into question the whole idea of reproducibility. So what went wrong?
In all cases examined, the same error occurred every time. When a MOZYME calculation is run, it starts by building a Lewis structure for the system. This Lewis structure is based on the topology of the supplied geometry. If the geometry supplied is severely in error, then the Lewis structure will have an error. During the geometry optimization, all, or at least most, of the geometric errors are corrected by the energy minimization procedure. The final geometry will be much better, i.e., will be chemically more realistic, than the starting structure. If a single point calculation is then run on the final geometry, the correct or more correct Lewis structure will be used. Because a more correct structure is used, the computed heat of formation will be lower than that of the previous run, unless the net charge becomes more positive. If the net charge on the system changes in going from the starting geometry to the optimized geometry, that indicates a significant change in bonding, a change that should definitely be investigated before proceeding any further.
The procedure for handling this fault is also the same in each case. First, do not simply add or remove covalent bonds to fix the fault. It's better to regard the failure of MOZYME as an indication of a specific fault in the initial data set (although on rare occasions, the fault might be in the final data set). Run both the starting and optimized geometries using keyword LEWIS, and compare the output files. Go to the bottom of these files, and look for the charges on the various atoms. Examine each case where the charges differ. Some differences might be due to the formation of salt bridges. When a salt bridge forms, a cation and an anion are created; these are natural structures and after identifying them (look at them using a GUI), they can be ignored. What you're looking for are charges the are present in one file but are not present in the other. As each one is found, carefully examine the location of the charge, and determine if it should be charged or uncharged. To assist in the examination of the site, use the Lewis structure that is printed in the output files. At first, this might seem difficult, but after doing it once or twice the procedure becomes quite easy. The action that is taken once the fault is positively identified depends on what caused the fault. If there is an extra hydrogen atom, or a hydrogen atom missing, edit the geometry to suit. If a bond-length is incorrect so that the topology is in error, use CVB to to correct the topology.
Once all errors (normally only one) in the data set are identified and corrected, re-run the job.
When locating, optimizing, or characterizing a transition state, a very specific problem often occurs. Typically, in a transition state, a covalent bond is made or broken. Consider for example, a hydroxyl group adding across a peptide bond (-CO-NH-). The transition state might consist of the four-membered ring -[O-(C=O)-(NH)-H]-. It would be hard to draw a Lewis structure for this system, and the MOZYME technique would normally fail to generate an acceptable starting Lewis structure. At first sight, then, it would appear that MOZYME cannot be used for modeling transition states, but in fact getting transition states to run is straightforward.
As with ground-state systems, identify the charges associated with the transition state. The Lewis structure generator in MOPAC often makes mistakes when trying to process transition states. Assuming that it does make a mistake in the optimized or refined transition state (the commonest scenario), use CVB to make or break bonds so that the topology is simplified. Use the topography printed in the output file to characterize the environment of the transition state, and to identify the changes that would need to be made in order to have a sensible Lewis structure. For example, in the case of the four-membered ring just described, if the topography indicates that the atoms forming the ring are, in fact, connected, then use CVB to break the "-O-(C=O)-" bond and the "-(NH)-H-" bond. This would then result in a simple topography corresponding to the unreacted structure.