RI basis optimization

[Wan-Lu Li]

Dear CP2K users,

I have two questions concerning RPA calculations.

1) Can RPA method implemented in CP2K be used in open-shell systems?

2) In my case, I have iron atom which doesn't have the existing RI basis file, so I have to optimize it (firstly I tried to use Ahlrichs-VTZ as the primary basis). Here is my input file, but the convergence behavior looks very strange for single Fe atom (without any energy change). Does anyone have experience with that? Thanks!

Wanlu

[Frederick Stein]

Dear Wanlu,

Generally speaking, you should not mix GTH pseudopotentials with all-electron basis sets like the Ahlrich basis sets. You will need correlation-consistent basis sets for RPA calculations. In the latest trunk, there are optimized cc basis for iron ( BASIS_ccGRB_UZH). I am not aware of already optimized RI basis sets with iron. You can find an example with Titanium in the Supporting Information of the publication http://dx.doi.org/10.1039/C4CP03981E .

To 1) Yes, have a look at the regtest directories (tests/QS/regtest-ri-rpa) for information on how to setup a calculation. Please note, that RPA calculations are not compatible with the GAPW approach, i.e. all-electron calculations are not available.

To 2) You have two options depending on your requirements:

a) Do not specify the RI basis set and use the automatically generate RI basis sets (set the AUTO_BASIS keyword in the DFT section like:   AUTO_BASIS RI_AUX <Basis set size>  (<Basis set size> can have the values SMALL, MEDIUM, LARGE and HUGE). If you just have to carry out a few calculations, this option can be sufficient.

b) Optimize your own RI basis sets with MP2, check tests/QS/regtest-ri-mp2-opt for some examples. This option is recommended for a large amount of calculations or if the automatically generated RI basis sets have an insufficient quality. (check the publication above)

Hope that helps.

Frederick