CUTTOFF and REL_CUTTOFF for large molecular system
1,185 views
Skip to first unread message
Tobias Kraemer
Jun 10, 2014, 1:05:59 PM6/10/14
to cp...@googlegroups.com
Dear CP2K group,
being a relatively new cp2k user, I was wondering if this group could give me some advice on the choice of (REL) CUTOFF values.
The system I am considering is a somewhat extended molecular crystal structure, containing organometallic fragments (basically a rhodium phosphine complex with additional ligand plus large borate counteranions (BArF4-).The full system contains about 600+ atoms in the unit cell, basically 4 cation/anion pairs
(asymmetric units). I think the system cannot be reduced in size, and the full unit cell has to be taken into account.I have attached a cif file which gives you
an idea about the kind of system (not exactly the one used in the calculation below, but very much related)
The basic idea behind determining the CUTOFF is clear to me and I have done the tutorial on bulk Si. I have used the same protocol for my system
(running a series of calculations with varying CUTOFF and REL_CUTOFF values). In contrast to the suggested method in the tutorial, I have
fully converged the SCF cycle) Below are the (first) results of this series. The energy appears to be converging towards higher CUTOFF values (some more calculations are currently running)
(1) I wonder if choosing such a large value for the cutoff is reasonable? It should also increase the accuracy of the calculations, but also the cost.
Also, with respect to the number of Gaussian being mapped onto the finest grid, what would one need to look for?
In the tutorial the decision was based on the fact that "the distribution of Gaussian functions on the grids are reasonable". Could somebody elaborate on this
and give me some guidance how to decide this?
(2) How does the size of the CUTOFF value correlate with the system size, if there is any correlation?
(3) Are there some rule of thumbs as how to choose the REL_CUTOFF with respect to the CUTOFF? For now I have tried 30, 60 and 100 Ry.
But I may need to go to higher numbers in this particular case.
# Grid cutoff vs total energy
# Date: Tue Jun 10 17:26:55 BST 2014
# REL_CUTOFF = 60
# Cutoff (Ry) | Total Energy (Ha) | Charge | SCF | | NG on grid 1 | NG on grid 2 | NG on grid 3 | NG on grid 4
50.00 -4583.1402739393 -8.4046446120 200 1623260 248362 2112 0
100.00 -4477.8479306771 -0.0745436930 128 1264867 519816 89051 0
150.00 -4479.5978060995 -0.0191170935 128 1150415 472845 248362 2112
200.00 -4478.5030451759 0.0008175249 128 905641 531280 432781 4032
250.00 -4478.4529157377 -0.0000917804 128 838757 500556 523109 11312
300.00 -4478.4141842105 -0.0000534995 128 796882 467985 519816 89051
350.00 -4478.3827319182 -0.0000170792 128 748308 474790 454036 196600
400.00 -4478.3597765363 0.0000003328 128 679828 499675 489931 204300
450.00 -4478.3498795935 0.0000000149 128 621687 528728 472845 250474
500.00 -4478.3399690553 0.0000000147 128 567997 416013 600098 289626
550.00 -4478.3338555119 -0.0000000007 128 485909 446314 559882 381629
600.00 -4478.3312330111 0.0000000007 128 442629 463012 531280 436813
650.00 -4478.3291405214 0.0000000006 128 399089 473084 554024 447537
700.00 -4478.3308119206 0.0000000008 128 375655 473218 509402 515459
750.00 -4478.3298871109 0.0000000008 128 365399 473358 500556 534421
800.00 -4478.3266993684 0.0000000007 128 353459 468098 461722 590455
850.00 -4478.3266987930 0.0000000007 128 348184 458373 472494 594683
900.00 -4478.3273848891 0.0000000007 128 336984 459898 467985 608867
950.00 -4478.3277135782 0.0000000007 128 309758 476872 469877 617227
1000.00 -4478.3277705414 0.0000000008 128 291089 482269 455857 644519
1050.00 -4478.3278605666 0.0000000007 128 229036 519272 474790 650636
Some further questions;
(4) Could you recommend some literature (books) which explains the theory of solid state calculations, k-sampling, planewaves?
(5) As a next step I would like to manipulate the structure, i.e. rotate/move ligands around? Is there a good visualizer that can do such
things for periodic systems? The problem I am facing is that I would like to do these things simultaneously for all 4 asymmetric units.
I think it becomes impracticable to do this for each unit separately. I have started using GDIS but I am a little bit uncertain about its
abilities (great program though!!). Has anyone experience with VESTA? I am particularly looking for free software.
Thank you very much for your help, this is very much appreciated. There are a lot of questions right a the beginning, and it is not always easy to
find some answers.
Best
Tobias
being a relatively new cp2k user, I was wondering if this group could give me some advice on the choice of (REL) CUTOFF values.
The system I am considering is a somewhat extended molecular crystal structure, containing organometallic fragments (basically a rhodium phosphine complex with additional ligand plus large borate counteranions (BArF4-).The full system contains about 600+ atoms in the unit cell, basically 4 cation/anion pairs
(asymmetric units). I think the system cannot be reduced in size, and the full unit cell has to be taken into account.I have attached a cif file which gives you
an idea about the kind of system (not exactly the one used in the calculation below, but very much related)
The basic idea behind determining the CUTOFF is clear to me and I have done the tutorial on bulk Si. I have used the same protocol for my system
(running a series of calculations with varying CUTOFF and REL_CUTOFF values). In contrast to the suggested method in the tutorial, I have
fully converged the SCF cycle) Below are the (first) results of this series. The energy appears to be converging towards higher CUTOFF values (some more calculations are currently running)
(1) I wonder if choosing such a large value for the cutoff is reasonable? It should also increase the accuracy of the calculations, but also the cost.
Also, with respect to the number of Gaussian being mapped onto the finest grid, what would one need to look for?
In the tutorial the decision was based on the fact that "the distribution of Gaussian functions on the grids are reasonable". Could somebody elaborate on this
and give me some guidance how to decide this?
(2) How does the size of the CUTOFF value correlate with the system size, if there is any correlation?
(3) Are there some rule of thumbs as how to choose the REL_CUTOFF with respect to the CUTOFF? For now I have tried 30, 60 and 100 Ry.
But I may need to go to higher numbers in this particular case.
# Grid cutoff vs total energy
# Date: Tue Jun 10 17:26:55 BST 2014
# REL_CUTOFF = 60
# Cutoff (Ry) | Total Energy (Ha) | Charge | SCF | | NG on grid 1 | NG on grid 2 | NG on grid 3 | NG on grid 4
50.00 -4583.1402739393 -8.4046446120 200 1623260 248362 2112 0
100.00 -4477.8479306771 -0.0745436930 128 1264867 519816 89051 0
150.00 -4479.5978060995 -0.0191170935 128 1150415 472845 248362 2112
200.00 -4478.5030451759 0.0008175249 128 905641 531280 432781 4032
250.00 -4478.4529157377 -0.0000917804 128 838757 500556 523109 11312
300.00 -4478.4141842105 -0.0000534995 128 796882 467985 519816 89051
350.00 -4478.3827319182 -0.0000170792 128 748308 474790 454036 196600
400.00 -4478.3597765363 0.0000003328 128 679828 499675 489931 204300
450.00 -4478.3498795935 0.0000000149 128 621687 528728 472845 250474
500.00 -4478.3399690553 0.0000000147 128 567997 416013 600098 289626
550.00 -4478.3338555119 -0.0000000007 128 485909 446314 559882 381629
600.00 -4478.3312330111 0.0000000007 128 442629 463012 531280 436813
650.00 -4478.3291405214 0.0000000006 128 399089 473084 554024 447537
700.00 -4478.3308119206 0.0000000008 128 375655 473218 509402 515459
750.00 -4478.3298871109 0.0000000008 128 365399 473358 500556 534421
800.00 -4478.3266993684 0.0000000007 128 353459 468098 461722 590455
850.00 -4478.3266987930 0.0000000007 128 348184 458373 472494 594683
900.00 -4478.3273848891 0.0000000007 128 336984 459898 467985 608867
950.00 -4478.3277135782 0.0000000007 128 309758 476872 469877 617227
1000.00 -4478.3277705414 0.0000000008 128 291089 482269 455857 644519
1050.00 -4478.3278605666 0.0000000007 128 229036 519272 474790 650636
Some further questions;
(4) Could you recommend some literature (books) which explains the theory of solid state calculations, k-sampling, planewaves?
(5) As a next step I would like to manipulate the structure, i.e. rotate/move ligands around? Is there a good visualizer that can do such
things for periodic systems? The problem I am facing is that I would like to do these things simultaneously for all 4 asymmetric units.
I think it becomes impracticable to do this for each unit separately. I have started using GDIS but I am a little bit uncertain about its
abilities (great program though!!). Has anyone experience with VESTA? I am particularly looking for free software.
Thank you very much for your help, this is very much appreciated. There are a lot of questions right a the beginning, and it is not always easy to
find some answers.
Best
Tobias
Dr. Tobias Kraemer
Research Associate
Institute of Chemical Sciences
School of Engineering & Physical Sciences
Heriot-Watt University
Edinburgh EH14 4AS
United Kingdom
hut...@chem.uzh.ch
Jun 11, 2014, 8:46:55 AM6/11/14
to cp...@googlegroups.com
Hi
some info on cutoffs:
- The relative cutoff is the minimal cutoff that is used to
integrate a Gaussian with exponent 1. This scales linearly.
Exponent a -> a*RelCutoff.
40 Rydberg usually gives reasonable results.
60 Rydberg should be close to 'convergence'.
80 Rydberg should always be converged.
!!! Remember, convergence tests always also depend on other
factors, like the Max. Cutoff, number of grids, how subgrids are
calculated.
- Estimate of total cutoff: Max. exponent in your basis set=a
Cutoff = 2*a*Rel_cutoff
This is usually way too high but can be used to work your way down
to smaller cutoffs.
- For larger systems (many atoms and electrons) and dense systems,
CPU time is not dominated by the cutoff
- For some GGA functionals convergence with Cutoff can be slow.
Dirty trick: use one of the smoothing options (XC_GRID)
Better: use higher cutoff for XC calculation only(testing needed)
- Sometimes slow convergence with cutoff points to a problem with
the GTH pseudopotentials and GGA functionals.
There are NLCC pseudos available that solve this issue.
-> more testing needed
regards
Juerg
--------------------------------------------------------------
Juerg Hutter Phone : ++41 44 635 4491
Institut für Chemie C FAX : ++41 44 635 6838
Universität Zürich E-mail: hut...@chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland
---------------------------------------------------------------
-----cp...@googlegroups.com wrote: -----
To: cp...@googlegroups.com
From: Tobias Kraemer <161brun...@gmail.com>
Sent by: cp...@googlegroups.com
Date: 06/10/2014 07:06PM
Subject: [CP2K:5389] CUTTOFF and REL_CUTTOFF for large molecular system
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To post to this group, send email to cp...@googlegroups.com.
Visit this group at http://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
[attachment "input_test.inp" removed by Jürg Hutter/at/UZH]
[attachment "PCCPiBu_NBA.cif" removed by Jürg Hutter/at/UZH]
some info on cutoffs:
- The relative cutoff is the minimal cutoff that is used to
integrate a Gaussian with exponent 1. This scales linearly.
Exponent a -> a*RelCutoff.
40 Rydberg usually gives reasonable results.
60 Rydberg should be close to 'convergence'.
80 Rydberg should always be converged.
!!! Remember, convergence tests always also depend on other
factors, like the Max. Cutoff, number of grids, how subgrids are
calculated.
- Estimate of total cutoff: Max. exponent in your basis set=a
Cutoff = 2*a*Rel_cutoff
This is usually way too high but can be used to work your way down
to smaller cutoffs.
- For larger systems (many atoms and electrons) and dense systems,
CPU time is not dominated by the cutoff
- For some GGA functionals convergence with Cutoff can be slow.
Dirty trick: use one of the smoothing options (XC_GRID)
Better: use higher cutoff for XC calculation only(testing needed)
- Sometimes slow convergence with cutoff points to a problem with
the GTH pseudopotentials and GGA functionals.
There are NLCC pseudos available that solve this issue.
-> more testing needed
regards
Juerg
--------------------------------------------------------------
Juerg Hutter Phone : ++41 44 635 4491
Institut für Chemie C FAX : ++41 44 635 6838
Universität Zürich E-mail: hut...@chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland
---------------------------------------------------------------
-----cp...@googlegroups.com wrote: -----
To: cp...@googlegroups.com
From: Tobias Kraemer <161brun...@gmail.com>
Sent by: cp...@googlegroups.com
Date: 06/10/2014 07:06PM
Subject: [CP2K:5389] CUTTOFF and REL_CUTTOFF for large molecular system
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To post to this group, send email to cp...@googlegroups.com.
Visit this group at http://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
[attachment "input_test.inp" removed by Jürg Hutter/at/UZH]
[attachment "PCCPiBu_NBA.cif" removed by Jürg Hutter/at/UZH]
Tobias Kraemer
Jun 12, 2014, 3:31:50 AM6/12/14
to cp...@googlegroups.com
Dear Jürg,
thank you for your detailed answer. This was very useful information,
in particular how to get an estimate of the total cutoff value,
this is very good to know. A number of things to try out now.
Cheers
Tobias
Reply all
Reply to author
Forward
0 new messages