Dear Soumyadip Ray ,

Spin contamination arises every time you use the unrestricted formalism, so you should not worry about it. As you can see, the scf=maxcycle=512 keyword is not working, as the SCF convergence stopped after 129 cycles (128 in fact, the 129th is for printing, I think), and Gaussian seems to be hardcoded to not allow a larger number of cycles. Try to use scf=(xqc,maxcycle=128). In this way, you do the regular SCF calculation, and if it does not converge, it changes the SCF algorithm to quadratically convergence, and more often than not it lands properly.

In case that keyword also fails, you can try to do a restricted single-point (which should not give any convergence problem) and use that as the initial guess for the unrestricted calculation, with the options guess=save for the restricted and guess=read for the unrestricted (you need to save the checkpoint file in the restricted calculation!)

Hope you find it helpful

Try with lower basis.

Try using stable=opt keyword.

Enrique Manuel Arpa Prasanta Bandyopadhyay

Dear Sir,

Thanks for your help. However, now I am facing a similar problem spin contamination error for transition state optimisation. Should I use the same method?

Soumyadip Ray ,

If it is the same problem, yes, the problem is not the spin contamination per se (that is normal in unrestricted calculations) but the fact that SCF is failing to converge

I am trying with a molecule that consists of metals. After several trial, I failed to optimise it in TS. What should I do if SCF fails to converge? I have tried maxcycle up to 1024. What is the specific command?

Soumyadip Ray ,

As I commented on the first reply, maxcycle has an upper limit of 128 cycles, so it doesn't matter if you write scf=maxcycle=128, scf=maxcycle=1024 or scf=maxcycle=99999. As a first alternative, try again the scf=(xqc,maxcycle=128) option, to see if the quadratically convergent SCF leads to the result without anything else.

In case you are not, try to use pseudopotentials on the metal atom, this may also help. I have attached a simple example in case you don't know how to specify different basis sets and pseudopotentials for different atoms. The def2-sv(p) works well, try to use that one.

And finally, while B3LYP is kind of the jack-of-all-trades functional, it is sometimes not the most appropiate. For organometallic systems, the meta-GGA functional M06-L (written m06l in Gaussian) is a good option, so you may reconsider changing to that one.

Your polarization charge is too high. What SCRF method are you using? Could you put the route section of your Gaussian job?

Elder Mota Thanks Elder, my job is done already. I was using b3lyp/def2-tzvp. But now it is fine.