Jaguar Geometry Optimization Suggestions
If you have built a structure in Maestro, or suspect that the structure is not very close to the optimized geometry, it can be useful to perform a geometry cleanup. To clean up a geometry in Maestro using molecular mechanics, choose Edit → Minimize → All Atoms.
For flexible molecules, which might adopt one of several possible conformations, you might consider performing a conformational search with MacroModel. See MacroModel Conformational Searches for more information.
If you are performing a geometry optimization and are not starting from a high-quality initial molecular structure, you might want to do a “quick and dirty” calculation to obtain a somewhat better geometry, then perform a more accurate calculation by starting with the results you have generated already. For example, if you wanted to perform an LMP2 geometry optimization, you could start by performing a Hartree-Fock geometry optimization, then restart the calculation using the HF results in an LMP2 geometry optimization. See Restarting Jaguar Jobs and Using Previous Results for a description of restarting calculations and incorporating previous results in a later run.
Whenever you are doing a geometry optimization, make sure that you really do obtain a converged structure; the run ends before converging if you reach the maximum number of iterations allowed (as set in the Optimization tab). If it did not reach convergence, you can restart the run, as described in Restarting Jaguar Jobs and Using Previous Results. The quality of the convergence is analyzed and printed by default (see SCF and Geometry Convergence). If you want to automatically check for convergence, you can set the check_min keyword, and choose from the related options for convergence checking. At the end of the optimization, a procedure for perturbing the structure is performed, either by following negative eigenvalues of the analytical Hessian, or by perturbing along various torsional coordinates and reoptimizing. This feature is intended for geometry optimizations; it is not supported with dynamic constraints or for scans or IRC calculations.
Geometry optimizations can fail to converge properly due to pseudospectral noise. Jaguar provides a means of alleviating this problem by switching to analytic integral evaluation near convergence. The switch is done when convergence reaches a threshold defined by a multiple of the final convergence threshold. To use this approach, you can specify the multiple with the nops_opt_switch keyword, for which a value of 10.0 is appropriate, or select Switch to analytic integrals near convergence option in the Optimization tab, which sets this keyword to 10.0. This approach is not compatible with the use of check_min or with IRC calculations.
To assist with failed convergences in geometry optimizations, you can use the nofail=1 setting. This setting assumes that the geometry corresponding to the minimum energy traversed in the course of the geometry optimization is an acceptable solution to the minimization problem. Note that this geometry and energy pair is not necessarily the geometry and energy pair from the final geometry iteration step. In such cases, the lowest energy and the corresponding geometry are returned as the final result, a warning is printed, and the calculation exits successfully. If, however, the minimum is found, then nofail=1 does nothing.
The nofail=1 option is often a convenient practical compromise for various workflows that perform geometry optimizations as intermediate calculations. Normally, if one such geometry optimization fails, the whole workflow fails because it depends on the result generated in the geometry optimization and its successful exit status. The nofail=1 option makes the workflow that suffers from geometry optimization problems finish successfully, albeit with a possibly lower quality of the final result. In our validation tests with numerous practical examples we have found that the geometry corresponding to the true energy minimum (obtained in a separate calculation) was only slightly different from the geometry returned in a failed geometry optimization that was performed with the nofail=1 option. We have also noticed that this option greatly improves the robustness of geometry optimization-dependent workflows, often at a negligible loss of accuracy. For this reason, for example, our Jaguar pKa workflow uses the nofail=1 setting by default. Of course, you must be careful when using the nofail=1 setting if a true stationary point on the potential energy surface is required (which is often the case for subsequent vibrational frequency calculations)
Sometimes it is important to prevent a chemical reaction from occurring during a geometry optimization, for example stopping a hydrogen migration. You can do this with the stop_rxn keyword, which turns on checking of the connectivity after each geometry optimization step, and restarts the optimization with repulsive potentials or harmonic restraints to prevent the change in bonding.