next up previous
Next: Basis Sets Up: qchb Previous: qchb

Troubleshooting Calculations

It doesn't converge:
What doesn't converge? The optimization? The SCF? What?
Optimization doesn't converge:
Common solutions: (1) Your guess geometry is so far off, the optimizer gets hopelessly confused; (2) Your guess Hessian is so far off, the optimizer gets hopelessly confused; (3) Why didn't you look at your outputs to realize the energy you're optimizing isn't even right anyway? (For example, it doesn't even converge the SCF or to the right SCF solution). Of course it won't optimize! (4) It's just a floppy molecule and needs some extra steps to get there. This is the case if it looks like (based on your gradients and energies at each step) you're making progress towards a minimum.
Optimization converges to structure with imaginary frequencies:
The following comments apply if you are trying to get a local minimum but wind up with a saddle point. Usually this happens when you start with a high symmetry structure. Optimizers assume you know what you're doing, and if you give them a C2v molecule, they will try to keep it C2v, even if there is no true C2v local minimum, but only a saddle point. If you have very weak imaginaries (say, less than 50i), it might just be that you need to converge tighter (Q-Chem has a sleazy tolerance, try tightening GEOM_OPT_TOL_GRADIENT below 300). This latter problem happens frequently with larger and floppy molecules.
Can't reproduce another program's energy:
Check in the following order: (1) Check the nuclear repulsion energy; this makes sure you have the same geometry. Note, Q-Chem can be off in about the 4th digit after the decimal, but this magically doesn't ever seem to matter... (2) Check the SCF energy; this makes sure you have the same geometry and basis set (and clears up any pure angular momentum vs. Cartesian function confusion -- see below -- this mistake is usually worth a couple of millihartrees or so). (3) Check that both calculations freeze the same core/virtuals (freezing/not freezing core will make a big difference in the total energy; the virtual part would only make a small change on the millihartree or sub-millihartree scale). (4) For MCSCF, you have to make sure the active space is really the same (symmetry issues could make different active spaces even if they have the same number of active orbitals).
The SCF energy is all wrong:
First, make sure you have the right geometry (check the nuclear repulsion energy). Next, check the basis set (remember pure angular momentum vs. Cartesians, see below). Next, see if it looks like the right state (e.g., spin multiplicity, charge). Next, see if it might have landed on the wrong SCF solution (try changing the initial guess, or the convergence options like DIIS, or try to check the SCF stability). Note that ACES II (and sometimes other programs) can be very bad about guessing the wrong initial orbital occupations. If this happens, give the program the orbital occupations yourself, if possible (they can't really be specified in Q-Chem). MOLCAS will guess the occupations wrong more than 90% of the time. ACES II and PSI3 are usually right for closed-shell but often wrong for open-shell.
It complains about not being able to read/write from/to the disk:
It is probably having trouble reading/writing from/to the disk. Is the disk full? Do you have the scratch files set up right? Are the files really there? Is it trying to write a file that goes beyond some operating system limit (e.g., 2GB, 4GB) or a limit on the queue you submitted to?
The frequencies (or ZPVE's) look wrong:
Are you computing frequencies numerically by differences of gradients computed at displaced geometries? If so, maybe you landed on the wrong Hartree-Fock solution at one of the displaced geometries.

next up previous
Next: Basis Sets Up: qchb Previous: qchb
David Sherrill 2004-12-06