Cp2k convergence test: is it possible that cutoff of 250 Ry has fairly lower total energy than 350 Ry (with same rel_cutoff 60 Ry)?

Sergey Chulkov sergeya... at gmail.com
Mon Aug 27 06:05:57 CEST 2018


Hi Yun,

The purpose of the convergence test is to find the lowest possible value of 
the plane-wave cutoff that still gives you an accurate total energy. The 
actual accurate total energy may be higher then the "minimal" energy across 
all possible values of the plain-wave cutoff.

In case of sparse real-space grids (small cutoff) there is no uniform 
dependence between the total energy and the value of the plain-wave cutoff. 
The main reason is that different energy components are computed using 
different basis set representations. For example, a smaller plain-wave 
cutoff can potentially give you an underestimated exchange-correlation 
energy, but overestimated electrostatic energy between nuclei and electrons 
which can finally lead to a lower total energy.

For a reasonable guess for the cutoff value you can refer to this page 
(https://www.cp2k.org/faq:cutoff). The relative cutoff of 50 Ry is usually 
a good choice, the relative cutoff of 60 Ry is typically enough in most 
cases. 

Best wishes,
Sergey

On Friday, August 24, 2018 at 3:29:24 PM UTC+1, any... at gmail.com wrote:
>
> Dear CP2K developers and users,
>
> I did the convergence tests with cp2k 3.0 for hBN system, following the 
> instructions of https://www.cp2k.org/howto:converging_cutoff . The energy 
> values and the NG on grids are as follows. I wonder why higher cutoff 
> values give fairly higher energies, for example, the total energy for 
> cutoff 350 ry is higher than 300 ry, I thought in principle, higher cutoff 
> values could have more stable energy (lower value), at least not that 
> higher than the number with lower cutoffs. But here in my case, it's not, 
> can anybody help me to see why?  Is 250Ry and rel_cutoff of 70 Ry here a 
> good choice? 
>
> Also, when one starts a calculation,  should the convergence tests always 
> be taken with different values of cut off and rel_cutoff? or it's safe to 
> just use larger values such as (cutoff 400 ry, rel_cutoff 80ry)?  
>
> Thanks a lot in advance for all your suggestions.
>
> Best regards,
> Yun
>
> # Grid cutoff vs total energy
> # Date: Fri Aug 24 13:38:48 CEST 2018
> # PWD: /home/Convergence_test
> # REL_CUTOFF = 60
> # Cutoff (Ry) | Total Energy (Ha) | NG on grid 1 | NG on grid 2 | NG on 
> grid 3 | NG on grid 4
>      50.00  -833.3666229006     559488   81536       0       0
>     100.00  -824.4110369450     457024  180160    3840       0
>     150.00  -824.3925595993     424576  134912   81536       0
>     200.00  -824.3899074807     345088  210560   81536    3840
>     250.00  -824.3979520649     294272  162752  180160    3840
>     300.00  -824.3920162797   294272  162752  180160    3840
>     350.00  -824.3858376968    271744  185280  119488   64512
>     400.00  -824.3841314539    271744  152832  134912   81536
>     450.00  -824.3831328479     271744  152832  134912   81536
>     500.00  -824.3833590835     236416  184320  138752   81536
>
>
> # Rel Grid cutoff vs total energy
> # Date: Fri Aug 24 14:54:23 CEST 2018
> # PWD: /home/Convergence_test/Rel_cutoff
> # CUTOFF = 250
> # Rel Cutoff (Ry) | Total Energy (Ha) | NG on grid 1 | NG on grid 2 | NG 
> on grid 3 | NG on grid 4
>      10.00  -824.9353843726     24384  212032  184320  220288
>      20.00  -824.4674485372     117312  176960  162752  184000
>      30.00  -824.3980791451    236416  184320  138752   81536
>      40.00  -824.3980825574     271744  152832  134912   81536
>      50.00  -824.3979520649     294272  162752  180160    3840
>      60.00  -824.3979520649     294272  162752  180160    3840
>      70.00  -824.3979509738     345088  210560   81536    3840
>      80.00  -824.3979509738     345088  210560   81536    3840
>      90.00  -824.3979509738     420736  138752   79616    1920
>     100.00  -824.3979509738    424576  134912   81536       0
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20180826/75a91881/attachment.html>


More information about the CP2K-user mailing list