WARNING in minimax_rpa

[Jan Wilhelm]

Dear Kun-Han,

this warning is fortunately not evil. It is related to the convergence of 
the frequency integration in RPA using the minimax integration grid. In 
your case, I would just check the convergence of the frequency integration 
by varying the number of frequency points for RPA (keyword 
QUADRATURE_POINTS) from 8, 12, 16 to 20. The best result should be obtained 
with 20 points. I would expect that your RPA correlation energy is already 
well-converged with very few points.

Concerning your question "There is something I could do to increase the 
range of minimax approximation over 60?": The range of the minimax 
approximation Rc is defined as 

Rc = (e_max - e_min) / (e_LUMO - e_HOMO) ,

where e_n are the eigenvalues from the SCF. e_min is the smallest 
eigenvalue and e_max the biggest one. Therefore, you could increase Rc by 
using a bigger basis resulting in an increased e_max. 

Details on the minimax integration for RPA can be found in the (very 
technical) references http://dx.doi.org/10.1016/j.cpc.2014.10.021 and 
http://dx.doi.org/10.1021/ct5001268. Have a look at the right part of Fig. 
2 of the first reference: It shows for which minimax range Rc the minimax 
parameters have been optimized. For 20 quadrature points, you see that the 
lowest bound of Rc is in the order of 60. For fewer points, Mauro optimized 
the minimax grid also for smaller ranges.

In case I was not clear, please let me know. 

Best,

Jan


Am Freitag, 13. Januar 2017 13:32:03 UTC+1 schrieb Kun-Han Lin:
>
> Dear CP2K users and developers,
>
> I was trying to perform a HXX/dRPA computation on a benzene molecule in 
> gas phase using MINIMAX_QUADRATURE. Towards the end of the computation I 
> got the following Warning:
>
>
>
>  *** WARNING in minimax_rpa.F:110 :: The range for the Minimax             
> ***
>  *** approximation is smaller than the lower bound for the required number 
> ***
>  *** of quadrature points. The Minimax parameters have been optimized for 
>  ***
>  *** the lower bound of: 60.00                                             
> ***
>
>   INTEG_INFO| Range for the minimax approximation:                       
> 51.3757
>
>
> There is something I could do to increase the range of minimax 
> approximation over 60?
>
> Best Regards
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170116/62a30ce6/attachment.htm>