Problem: In a Geometry Optimization there is no progress: the atomic positions hardly change or oscillate around, while the energy gradients don't go to zero.
Possible Cause: Occupation numbers have not been supplied in input (OCCUPATIONS). For some molecules the procedure gets stuck between two (or more) different electronic configurations, with gradients pointing in different directions for the competing states. In each new SCF procedure, after a geometry update, the occupation numbers are re-determined, by default according to the aufbau principle. As a consequence, the successive SCF procedures may handle different electronic configurations and hence produce contradicting geometry-updates. See the key OCCUPATIONS.
Check: in the output file the occupation numbers that have been in effect during the successive SCF procedures. If they are different, then:
Cure: supply occupation numbers in input.
Alternative cause: SCF convergence not reached, or the criterion was too weak, or the precision of numerical integration in insufficient. Such causes may lead to inconsistencies between the true energy surface properties and the computed gradients. Usually this will only slow down the convergence but not prohibit it. However, if the SCF convergence is really problematic, it might get more serious. Reconsider the SCF strategy parameters, the Numerical Integration precision, or try keeporbitals (a subkey to the key OCCUPATIONS).
Notes: The accuracy of the gradients can be made higher if the DISHUL parameter in the INTEGRATION block key is increased, for example, to dishul=5. One can use the same integration and convergence criteria for successive cycles, for example: INTEGRATION 5 5, and in the SCF key use the option: converge 1e-6 1e-6. The errors then can become more systematic (instead of random).
