SCF and Geometry Convergence

During geometry optimization, Jaguar adjusts the convergence criteria for the SCF calculations at each geometry step for efficiency. For the initial iterations of an optimization, the SCF calculations are performed at the Quick accuracy level described in SCF Accuracy Level (unless the input contains a transition metal, in which case the accuracy level is Accurate). However, for the last few geometry iterations, the accuracy level for the SCF calculations is reset to the Accurate level, which uses tighter cutoffs and denser pseudospectral grids than the Quick level.

For optimizations to minimum-energy structures or transition states, the convergence criterion for SCF calculations is chosen to assure accurate analytic gradients. The SCF calculations during an optimization to a minimum-energy structure or transition state do not use the energy convergence criterion used by other SCF calculations.

The geometry is considered to have converged when the energy of successive geometries and the elements of the analytic gradient of the energy and the displacement have met the convergence criteria. Convergence is tested by calculating a score, in which each of the five criteria (on the energy difference, maximum gradient, rms gradient, maximum displacement, rms displacement) adds one point to the score if it is met. The optimization is considered converged if the score is 5 or more (the “standard” criteria). By default, however, an extra point is added if the energy change is less than 0.05 of the threshold and another if it ls less than 0.005 of the threshold, or if the other criteria are met to 0.2 times the threshold (the “flexible” criteria). You can enforce the standard criteria by setting geoconv_mode=standard in the gen section of the input file. Using the standard criteria is recommended for transition state searches.

At the end of a geometry optimization Jaguar performs a simple analysis of the geometry optimization convergence. The output file contains a verdict that indicates the dynamics of the convergence process—if the convergence was monotonic or erratic—and the quality of the converged structure—if the final geometry corresponds to the lowest energy or not.

The analysis places the convergence in one of the following five categories:

Categories 0-2 are considered successful convergence. Categories 3-4 might require additional attention. Category 3 convergence is considered successful for optimizations in solution but borderline otherwise. Category 4 should be considered unsuccessful convergence. No category is assigned if the geometry optimization ran out of iterations.

In most cases only category 4 convergence should be scrutinized for potential problems. You might consider starting from a different initial guess or using different geometry optimization settings. If you choose not to rerun your calculation it might be safer to consider the geometry corresponding to the optimization step with the lowest energy as the converged result.

The geometry optimization analysis is controlled by the input file keyword optverdict. Set it to 0 if you want to disable the geometry optimization analysis. Set it to 2 if you want to perform the analysis after each optimization step.

If you want to test that the converged structure is a minimum, you can set the check_min keyword. There are several options for checking the convergence and finding the minimum if the structure is not actually at a minimum: calculating an analytic Hessian and following its negative eigenvalues, or perturbing various torsions and optimizing the perturbed geometries. Note that this feature is not available with dynamic constraints or for scans or IRC calculations.

For a visual inspection of convergence, you can plot the energies at each step in the QM Monitor Panel, which you can open from the Workflow Action Menu . If you want to inspect the intermediate geometries for the points at which the convergence appears to be diverging, you can save the intermediate geometries by setting the Save intermediate geometries to output structure file option (Optimization tab) from Maestro, or by setting ip472=2 in the gen section.