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)?

[anyu...@gmail.com]

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

[Sergey Chulkov]

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

[Yun An]

Dear Sergey,

Thanks a lot for your reply!

Best regards,

Yun

Sergey Chulkov <sergeya...@gmail.com> 于2018年8月27日周一 上午6:06写道：

--
You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/acB9J8aFjRU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.
To post to this group, send email to cp...@googlegroups.com.
Visit this group at https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.