Frequently, I face molecules that do not converge in geometrical optimization using gaussian09 calculations. Is this a sign that this molecule is unstable?
It is an interesting question, we are working in the framework of approximations, actually only those structures which are converged (in the criterion you have specified for density and energy) are eigenstates of exact hamiltonian. You may do frequency calculations and even if it does not show any imaginary number for ground state, but you cannot say it is a minimum. You should change the geometry or method for SCF convergence.
Actually this is the challenging aspect of optimization. Sometimes for large molecules with heavy atoms those four criteria is not achieved. the last two criteria (RMS Displacement, Max displacement) often do not converged in some optimizations. sometimes, you can ignore them and do frequency analysis to ensure that the structure has no imaginary frequency. However, some molecules are stable under defined method and basis sets and are unstable in others. So, It is dependent on the framework of computations.
To answer your question directly, no, that does not necessarily imply that a given molecule is inherently unstable. All it means is that the optimization algorithm has been unsuccessful with the given geometric constraints and within the selected level of theory (model). Check obvious things that you can control - that bond lengths and non-bonded distances are appropriate, for example - make sure that the "optimization" is not leading to dissociation, for example. Perhaps do a partial optimization with some bond lengths, bond angles, or torsional angles fixed and then re-do a complete optimization. Perhaps change to a lower (or perhaps higher) level of theory, where the shape of the potential energy surface (and, therefore, the outcome of the optimization) may be different. Use your chemical intuition and the flexibility of the software to find an appropriate structure.
To add to what Allen said: there is little correlation between the convergence and the chemical stability of the molecule. The lack of convergence is in most cases a technical problem arising from the impecrfectness of the SCF / geometry optimization algorithm, poor choice of the geometry of the starting point in geometry optimization and so on.
Chemical stability / instability of molecule is something completely different and has to take into account the solvent. If your system is not converging but still looking reasonable from the chemical point of view, there is no telling if it is stable or not.
Actually, estimation of stability is one of the most difficult things to perform in gaussian (and any other QM) software. The problem is that most gasussian optimziations are done in vacuo (or in simple solvent continuum model), with no true partners/molecules to interact with. Hence, most chemical systems will look like stable ones, since all possible degradation pathways will be impossible for them to achieve - unless they are very unstable due to e.g. huge strains in rings. You can try to estimate the chemical stability e.g. by looking at the HOMO-LUMO gap after convergence and that is sometimes useful.
Thanks to all the participants of this question. Particularly, I like to mention the fact that some molecules that I perform optimization operation using gaussian09 software do not converge nor dissociate. In addition, changing the level of theory or basis functions does not change the situation. The program persists on giving one or two (yes) and then returns to four (no) and come back to one or two (yes) indefinitely. The thing that make the matter more confusing is that the atoms are approximately still in there initial places without achieving any convergence.
I should start by saying I am not a computational specialist by any stretch of the imagination (and so I may very well have a fairly superficial or at least incomplete understanding of the process).
But, In response to your last post (some convergence criteria are met, and some are not) I like to think about the optimisation process as walking around a potential energy surface and trying to find the lowest point:
If your calculation fails to converge with respect to the 'force criteria' it can be thought of as the optimisation rattling around the neck of a steep-sided well (i.e. the differences between successive optimisations are too great. Each step causes you to crash into the other side of the well, and by taking smaller steps you might be able to work further down into the well. There is the 'Maxstep' option that can be changed to address this sort of problem)
In the case of failing the 'step-change criteria', this can be thought of as it being difficult finding the global minimum on a very nearly flat, slightly undulating, potential energy surface (there is a 'calcFC' option that might help by determining in which initial direction the optimisation should go).
(so, as has been said before, a calculation failing to converge is not a good indication of real instability)
Did you try to change the optimization algorithm to quadratic convergence? Sometimes it solves the iterative problems when the energy starts to "vibrate" and a clear point can not be observed. Other options could be used, for example, allowing the calculation of force constant "always", using tight criteria, grid=ultrafine in DFT, etc...
Ian is probably pretty close - you probably have a very flat PES so the optimization is wandering around, not making very big changes in the geometry and never getting any closer to the minimum. The opt=calcfc option will help to establish the curvature of the surface and give a better first few steps. If that doesn't help, the opt=calcall option is much more computationally expensive but recalculates the force constants at every step to take the best possible step toward a nearby minimum (if it exists).
I had the same proplem that some time two crateria or more not converged. I tried to chang the basis sets and the level of calculation but it doesn't work. Lastly, l used the commands scf=xqc and opt=tight together and the result was O.K. I.e. all the four crateria were converged.
I am running optimization for a large molecule using Gaussian 09 and the convergence is been difficult. I have optimized the same molecule in another tool and I could see that the point Gaussian fail to converge already has the same energy as the other tools but has much tighter criteria. How do I make it a bit coarse and less tight so it can easily converge?
Sirous Yourdkhani Somayeh F. Rastegar Allen Clabo Joaquim Mª Rius Bartra
Yes, opt=calcfc is the easiest thing to try - otherwise, just change the geometry slightly and try again - you are probably too close to the minimum to take any meaningful steps -