Hello Dear Mudar Ahmed Abdulsattar. 

For ab initio methods, DFT is the best choice. However, for some systems, the convergence has a high computational cost! This barrier can be overcome through the change of the basis set or functional.

I was surprised by the question because, usually, the mentioned compounds easily converge. For example, ZnO in wurtzite structure may be optimized over 30 minutes with CRYSTAL09 package using all electron basis set and hybrid functionals. 

Which is the package do you use?

Best wishes.

Dear Renan A. P. Ribeiro

I am using gaussian09 program. However I am not using periodic boundary conditions (PBC), I am trying to simulate tetrahedrally bonded molecules of II-VI compounds in which the problem of convergence occurs.

Dear Mudar A. A. 

I had the same problem with tin compexes. Maybe you shuld try another DFT routine, keeping the same basis. I. e., you should search on Gaussian 09 manual a keyword for Pulay's routine. In my case was enough to converge the calculation.

Good luck!

Dear Matus Dubecky thank you for your support. My input begins with:

%MEM=150MW

%NProcShared=4

% chk = temp.chk

# PBEPBE/6-31G(d) OPT FREQ=RAMAN

Zn3O3H12 molecule

0 1

1 -0.44401875 2.90318265 1.2895197

30 -0.4041988 2.6305834 -0.0011475

8 0.28725325 -1.8671304 -0.0024752

Dear Mudar,

I think that the problem you are experiencing is difficult SCF convergence.

I would try the following:

- use Fermi broadening, adding the keyword SCF(FERMI). This usually helps a lot when metals are involved.

- If Fermi broadening doesn't help, you can try with the quadratically convergent algorithm. It is much more expensive than the standard SCF, but it is usually very robust (and it won't be more expensive than the subsequent frequency computation!). You can use the QC algorithm with SCF=QC.

- A mixed strategy that you might try before resorting to the QC, is to start with a regular (including Fermi broadening or not) SCF and switch to QC if the SCF has not converged in a reasonable number of iteration, for instance, 25. You can do that with the keyword SCF=(XQC,MaxConventionalCycles=25)

Good luck!

Dear Filippo Lipparini: Thank you for your suggestions. I performed some of your suggestions previously. I certainly will try your remaining suggestions as soon as possible.

Let me know how it goes!

You may also wish to increase your memory allocation if possible this may help the convergence.

Dear James McDonagh: Thanks for your remark. However, convergence problems appear even in small molecules that do not need high memory.

You can use WIEN2k code.This code is  efficient and can do the calculations so fast.In addition to that it is friendly code for the beginners.

Mohammed

Thanks Mohammed Abu-Jafar for your suggestion. However, I do not work presently  with periodical systems for which WIEN2K code is usually used.

Dear Mudar Ahmed Abdulsattar,

a bit belated notice: how many optimization steps does your structure make? One of solutions could be to increase the number of geometry optimization steps using the keyword 'opt = (maxcycle=700)', usually 200 or 300 steps are more than enough in geometry optimizations. Also, did you try first to optimize the structure with smaller basis sets, e.g. Lanl2dz or 3-21g, run frequencies for this optimization, and then proceed with larger basis set? Using other density functionals, like B3LYP or PW91, might help as well.

With best regards,

Aleksey.

Dear Aleksey E. Kuznetsov: Thank you for your suggestions. I performed some of your suggestions previously. I tried to increase the number of steps with no success. I also tried to use lower basis states such as 3-21G basis. I certainly will try your remaining suggestions such as lower the theory level as soon as possible.

Also, using different algorithms for geometry optimization like these:

'opt=(maxcycle=500,newton)' or

'opt=(maxcycle=500,bfgs)'  or

'opt=(maxcycle=500,rfo)' or

'opt=(maxcycle=500,steep)'

could be helpful. Another way which I often use in the case of poor geometry convergence: stop the job at some point, calculate frequencies, then if there are any imaginary frequencies change the Cartesian coordinates of the structure slightly, and re-run geometry optimization reading force constants from this frequency calculation. Good luck with the geometry optimization for your complex!

Best regards,

Aleksey.

Dear Aleksey E. Kuznetsov: Thanks for your suggestions. I tried the first one of your suggestions (opt=(maxcycle=500,newton)) with no improvements. Your second suggestion (stopping the job at some point) improved some of the results that converged faster than before. I chose a point where I think the minimum energy should be. Thanks again for your suggestions. 

Excellent, glad it was helpful!

Best regards,

Aleksey.

Dear Dr. Mudar 

You can use another method to perform the convergence in calculation for example scf=conver=n , were n is the number of the basis in 10-n  , to overcome the error in the convergence or to perform the fast convergence we may reduce n, for example if the convergence failed in scf  10-5  we use scf=conver=4 , if failed in 10-4  we use scf=conver=3  ••••••••••••

Thanks Dr. Hamad for your suggestion. However, I must be sure first not to alter the results default accuracy of Gaussian program.