Geometry optimization is a routine job.
Below are several recommendations for doing it.
The optimization cannot be applied to arbitrary geometry. First, the initial
geometry should be close to the desired. There should be no considerable
overlapping of valance-unbound atoms. The bonds should be of realistic
Second, the optimization does not affect the symmetry. For instance, if a
molecule was drawn in the screen plane, the optimization cannot displace it
from this plane.
Accordingly, one should try to keep to correct scale and conformation when
mouse drawing a molecule. Otherwise, the molecule should be edited using
either the conformation editor or molecular mechanics optimization.
It is recommended not to start drawing from scratch but rather use
a convenient model generated previously (in the Kernel box or
Since quantum mechanical calculations are time consuming, it is important
to correctly select the basis for a task.
Semiempirical calculation often suffices to correct gross geometry violations.
Accordingly, semiempirical optimization is a routine task after mouse drawing
a molecule. In this case, a crude gradient can be selected, e.g., 1e-04.
Nonempirical calculation is required to obtain accurate geometry.
Typical bases can be found in the Standard panel. The minimum basis
provides for sensible bond lengths and valance angles.
If the desired geometry substantially depends on nonvalence interactions, at
least the 2d basis is required with an obligatory account for
electron correlation. Unfortunately such bases are impractical for large
models. In our opinion, if a model is too large to be calculated using
MP2 6-311+G(2d,p), the task should be either abandoned or reduced
to small model systems to be optimized using MP2 6-311+G(2d,p)
and then assembled into a large model, possibly, using molecular
Accurate optimization should be performed until low gradients are reached,
at least, 1e-06 Hartree/Bohr. The optimization of van der Waals molecules
requires gradients of 1e-07 - 1e-08.
Unfortunately, numerical accuracy problems arise in this case.
Such problems are typically indicated by the oscillations
of energy or geometry and procedural convergence failure. In this case, one should
first try to switch the Calc mode from Fast to
Accurate. This decelerates the optimization but can improve the
geometry in many cases. Otherwise, one should consult the GAMESS
documentation and try different optimization methods, HF calculation options,
and coordinate systems.