Geometry Optimization

optking is the program responsible for orchestrating the process of geometry optimization. It can do a number of tasks automatically, such as generating internal coordinates, produce empirical force constant matrix, if necessary, update it, and check if geometry optimization is over. Some or all of the following files are necessary to perform a geometry optimization with optking:

• file11.dat - contains the cartesian geometry and the nuclear forces, produced by cints -deriv1;
• fconst.dat- contains force constants; if absent - empirical force constants will be generated by optking;
• intco.dat- contains internal coordinates in a format readable by a human; if absent - internal coordinates are generated automatically by optking.

The procedure for setting up such a calculation is as follows:

• define internal coordinates if desired (or, do nothing, and optking will do it for you automatically!)
• obtain a set of force constants in an fconst.dat file (or, again, optking can do this automatically for you)
• If analytic gradients are available for your chosen method, set dertype=first. If not, set dertype=none and also set numerical_dertype=first in the default section.
• Run the optimization by setting the opt flag set to true and nopt to the number of geometry optimization steps (say, around 5 to 10). If analytic gradients are not available, then nopt instead gives the number of energy points to compute. This should be the desired number of geometry optimization steps, multiplied by (2*num_symm_coord + 1), where num_symm_coord is the number of totally-symmetric internal coordinates.

The precision with which geometry is optimized depends on the residual forces on the nuclei. By default optking will terminate the job if the residual cartesian gradients in file11.dat are less than 1E-5 in atomic units. It is probably enough for most tasks. Going below this will most likely waste CPU time unless you are doing benchmarks.

An important aspect of a geometry optimization is the accuracy of the first derivatives of energy that PSI3 computes. Depending on how poorly your wavefunction has been convereged, the gradients themselves may not be sufficiently accurate for the requested convergence criterion. After computing first derivatives of the energy, cints runs a simple check of the quality of the energy derivative. It's a good idea to look at cints' output to make sure that the gradients are OK.

Let us take a look at each step involved in optimizing molecular geometry.