Unfortunately there are severe problems with your setup.

You are running the preoptimisation as unrestricted. This is the default since you did not specify anything (neither U- nor RO-). The guess from this calculation will not be satisfactory for a restricted open shell calculation. 

You do not need calcfc or noeigentest since you are not optimising a transition state.

It is better to run the frequency calculation after the optimisation, reading guess and geometry. Compound jobs take much too much time and you do not need the frequencies for a higher level of theory.

Use a finer grid when you are calculating with tight convergence criteria, i.e. int(ultrafinegrid).

You should also consider a more modern and better balanced basis set, like cc-pVDZ -> cc-pVTZ or def2-SVP -> def2-TZVPP to speed up your calculations.

Still having trouble with convergence, add Use=L506 as SCF=QC is not available for restricted open shell and add more cycles. Check your output if the energy oscillates. 

You can also turn off symmetry. Check if you do not need multi configurations.

I think it is better to opt your structure at lower level of theory such as ROPBEPBE/6-31+G(d,p) or ROPBEPBE/6-31+G(d). Then, use this optimized structure as input file for main job. Any way, you need more reliable structure now. Therefore, you should test ways to get it.

Hello:

Did the clac run out of cycles? You haven't provided enough information. Did you achieve any SCF cycles, or did it crash?

Dear Peter,

I tried with 200 and 400 SCF cycles, it is achieved the final cycle, but it is not converged. It is not crashed. Do you have any idea?

Thanks so much

there r two problems in it......first is using ROPBEPBE.  Better try to use UPBEPBE for radical systems. the second one is using guess=read. sometimes the old orbitals read from the checkpoint file is not good enough for higher level calculations. try to avoid it

Thanks so much Vijaya, i will try it. This problem is only appeared in some radicals, and just in RO mode. And this is also met with other basic set (i.e. HSEH1PBE).

Hi,

Does the SCF energy oscillate?

Peter

Dear Vijaya, your solution is good for this problem. However, I wonder if the results from RO- mode are better than U- one by compraring to experiments?

Unfortunately there are severe problems with your setup.

You are running the preoptimisation as unrestricted. This is the default since you did not specify anything (neither U- nor RO-). The guess from this calculation will not be satisfactory for a restricted open shell calculation. 

You do not need calcfc or noeigentest since you are not optimising a transition state.

It is better to run the frequency calculation after the optimisation, reading guess and geometry. Compound jobs take much too much time and you do not need the frequencies for a higher level of theory.

Use a finer grid when you are calculating with tight convergence criteria, i.e. int(ultrafinegrid).

You should also consider a more modern and better balanced basis set, like cc-pVDZ -> cc-pVTZ or def2-SVP -> def2-TZVPP to speed up your calculations.

Still having trouble with convergence, add Use=L506 as SCF=QC is not available for restricted open shell and add more cycles. Check your output if the energy oscillates. 

You can also turn off symmetry. Check if you do not need multi configurations.

Restricted open shell calculations (RO) is used to study the open shell systems by enforcing the alpha and beta orbitals to pairup,  other than the last filled orbital. This is used to study porphyrin like metal systems where unpaired electron tis bound to a particulat metal atom. But in ur case this option wont work or will give wrong results. thats why ur calculation didnt converge.

There is absolutely nothing wrong with doing restricted open shell calculations. Ideally the unrestricted solution should be quite close to the RO solution. RO calculations also fo not suffer from spin contamination. Lifting this restriction might lead to meaningless results if spin contamination is large. Alternatively for radical systems it is always worthwhile thinking about CAS calculations.

Thanks so much Martin, your very detailed answers will be very useful for me. I will try it. 

But I still have a small question: why I worked with the same radicals but I have the problem above only for some cases, not for all? And I tested 5 different basic sets (BMK, M06, PBEPBE, HSEH1PBE, and TPSSTPSS), but for BMK and M06, I have no problem, and for the others three cause sometimes the trouble, especially the PBEPBE?

Thanks you again

The mentioned BMK, M06, PBEPBE, HSEH1PBE, and TPSSTPSS are density functionals and not basis sets. The basis sets you are using are Pople's basis sets 6-31G, etc. 

Not all density functionals perform the same and there are different levels of accuracy. PBEPBE (or short PBE) is a pure functional, while M06 is a meta-hybrid functional. I can recommend the book "A Chemist's Guide to Density Functional Theory"

by Wolfram Koch, Max C Holthausen to get a deeper insight to what that actually means. 

Sometimes calculations fail because the electronic structure is very complicated, sometimes they succeed, but are meaningless. It is important to check orbitals and occupancy and see if different methods give different results.

I do not know what happens in the code of Gaussian, so I really cannot tell you what went wrong at what point. I also do not know enough about your system to make any predictions.

Since optimizer is using Newton Rapson method solving double derivative equations hence if you give exactly linear angle (i.e. 180) in your Z-matrix then the matrix will b singular and it will give you error, better to use angle 179 or near by something instead of 180.

opt=calcfc and geom=allcheck are often used for convergence failure.