SCF does not converge and bad condition error using UKS

[Gabriel Reynald]

Dear CP2K community,

     I am a new CP2K's user and recently started to study small non-periodic systems, like nanoclusters. I need to run a geometry optimization, for instance, using gold atoms, and when I do it, it converges fast and in a few steps. But when I try to use Unrestricted Kohn-Sham in the run by adding &DFT&UKS, the SCF simply does not converges and before the end of the cycle, an error saying "Bad condition number R_COND = ... (smaller than the machine working precision)" appears. The input and output are attached below. I tried increasing the cell size and fixing one of the atoms to avoid the Wavelet error, but I did the both calculation using and not using the &CONSTRAINT option, in different cell sizes. All them failed to converge and give the same error.

     Does anybody know how to fix this error and could say what and where am I getting it wrong? I will be very glad for any response, and certainly it will help other beginners.

[Marcella Iannuzzi]

Hi

You are using LSD but the spin up and spin down densities are identical for that system.

That wouldn't cause any problem in general.

With the G-space mixing of the density, however, the mixing is done for the density spin up and for the spin density, i.e. difference between spin up and down. 

In the specific case of the Pulay mixing, when the residual matrix is inverted, since the spin density is always identical to zero, this causes the numerical problem. 

The solution is not to use LSD if not really needed, or change to another mixing procedure (Broyden should work).

Regards

Marcella

[Gabriel Reynald]

Hi Marcella,

I really need to work with UKS, and setting BROYDEN_MIXING it's working! Thank you very much for your answer.