comparing CP2K-DFTB (5.1) and DFTB+ (18.2)
Wei Lai
Dear CP2K users and developers,
I was testing the DFTB functionality of CP2K and comparing it to DFTB+. I was using Li2O (324 atoms) as an example and the CP2K input file is shown in the following (without the full coordinates).
I found CP2K-DFTB results are very sensitive to the choice of Ewald (SPME which is the only Ewald method available for DFTB) parameter (ALPHA in A^-1). For a RCUT of 6 A, an estimate of ALPHA is usually 3.5/6=0.583. These results actually suggest indeed this is a good estimate.
ALPHA Li-Mulliken-charge H0-energy Repulsive energy SCC-energy Tot energy
0.64 0.571831 -414.5978261 1.790186269 8.302889648 -404.5047502
0.583 0.57183 -414.5978042 1.790186269 8.302909311 -404.5047086
0.5 0.57183 -414.5978156 1.790186269 8.302899133 -404.5047302
0.35 0.572372 -414.6135268 1.790186269 8.288678932 -404.5346616
0.22 0.610178 -415.6039561 1.790186269 7.126741326 -406.6870285
0.219 0.611188 -415.6274692 1.790186269 7.090867312 -406.7464156
On the other hand, DFTB+ (using Ewald instead of SPME) results are NOT very sensitive to ALPHA. I am not sure what is the unit of ALPHA in DFTB+ but I tried several values. A value of 0.219 (hand tuned) in CP2K gave similar results to those in DFTB+ with an automatic ALPHA value.
ALPHA Li-Mulliken-charge H0-energy Repulsive energy SCC-energy Tot energy
auto 0.61175023 -415.6390952 1.790189279 7.070521175 -406.7783847
0.338 0.61175319 -415.6391637 1.790189279 7.07041506 -406.7785594
0.64 0.6117653 -415.6394434 1.790189279 7.069981919 -406.7792722
1 0.61178357 -415.6398657 1.790189279 7.069327807 -406.7803486
I am puzzled by this comparison and wondering if you have any insight on this.
Thanks, Wei
&GLOBAL
PROJECT Li2O
RUN_TYPE ENERGY_FORCE
PRINT_LEVEL LOW
&END GLOBAL
&FORCE_EVAL
STRESS_TENSOR ANALYTICAL
METHOD QS
&DFT
&QS
METHOD DFTB
&DFTB
SELF_CONSISTENT T
DO_EWALD T
DISPERSION F
&PARAMETER
SK_FILE Li Li ./Li-Li.skf
SK_FILE Li O ./Li-O.skf
SK_FILE O Li ./O-Li.skf
SK_FILE O O ./O-O.skf
&END PARAMETER
&END DFTB
EPS_DEFAULT 1.0E-12
EXTRAPOLATION ASPC
EXTRAPOLATION_ORDER 2
&END QS
&SCF
SCF_GUESS RESTART ! can be used to RESTART an interrupted calculation
MAX_SCF 100
EPS_SCF 1.0E-7
ADDED_MOS 216
CHOLESKY INVERSE
&SMEAR ON
METHOD FERMI_DIRAC
ELECTRONIC_TEMPERATURE [K] 300
&END SMEAR
&MIXING
METHOD DIRECT_P_MIXING
ALPHA 0.1
&END
&END SCF
&POISSON
PERIODIC XYZ
&EWALD
EWALD_TYPE SPME
ALPHA .64
GMAX 21
RCUT 5
&END EWALD
&END
&END DFT
&SUBSYS
&CELL
PERIODIC XYZ
A 13.9765253067 0.0000000000 0.0000000000
B 0.0000000000 13.9765253067 0.0000000000
C 0.0000000000 0.0000000000 13.9765253067
&END CELL
&COORD
Li 1.1647105 1.1647105 1.1647105
Li 1.1647105 1.1647105 5.8235526
Li 1.1647105 1.1647105 10.4823942
hut...@chem.uzh.ch
the parameter in the EWALD section are not very user friendly
(for historical reasons). In your case you should not keep
RCUT at a small value and change ALPHA.
It is better to use the default (RCUT calculated from EWALD_ACCURACY
and ALPHA internally).
Also you have to make sure that GCUT is big enough for your system
to give converged values.
With all these settings you should see the same behaviour as in DFTB+.
regards
Juerg Hutter
--------------------------------------------------------------
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: "cp2k" <cp...@googlegroups.com>
From: "Wei Lai"
Sent by: cp...@googlegroups.com
Date: 12/19/2018 04:50AM
Subject: [CP2K:11061] comparing CP2K-DFTB (5.1) and DFTB+ (18.2)
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 https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
Wei Lai
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/RfYbOq3j0DM/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.
Wei Lai
Thomas Kühne
Maxime Van den Bossche
I also did a quick comparison now between CP2K and DFTB+
for a simple rocksalt MgO structure (8 atoms in total,
using the OT minimizer in CP2K). The input and output
files for both codes are attached.
What I found so far:
* Without periodicity, the two codes agree quite well
in terms of the H0 and SCC energies (up to ~1e-4 Ha)
as well as the Mulliken charges.
* In the periodic case, so far I haven't managed to
get comparable results with the two codes. The CP2K
results indeed appear sensitive to the choice of the
Ewald alpha parameter, similar to what Wei observed.
For each of the three ALPHA values I tried, the resulting
energies and SCC charges are quite different from DFTB+.
* In contrast to Wei, though, I also get different results
for both codes in non-self-consistent calculations
(but at least here the CP2K results indeed no longer
depend on ALPHA).
It would be great if you could check whether my CP2K
inputs have been appropriate. If so, I hope this example will
Maxime
Thomas Kühne
--
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 https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
<files.zip>
Wei Lai
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/RfYbOq3j0DM/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.
Wei Lai
hut...@chem.uzh.ch
I looked into these problems and I see that this behavior is
from a combination of bugs and hidden constraints on variables.
The bugs were introduced when the MIC was dropped after the implementation
of the k-point code. The Ewald parameters are partly set from the
input and partly set by using default values within the code.
To be on the save side I would suggest to only do DFTB calculations
with large enough computational cells. You can use
the MULTIPLE_UNIT_CELL keyword if you have a small cell.
&TOPOLOGY
MULTIPLE_UNIT_CELL ${NREP} ${NREP} ${NREP}
&END
&CELL
ABC 4.0 4.0 4.0
MULTIPLE_UNIT_CELL ${NREP} ${NREP} ${NREP}
&END CELL
I guess a cell with 10 Angstrom box length should be ok.
Use an ALPHA value that is 1 or larger and adjust GMAX for
your box size to get converged values.
I am working on fixes, they should be ready in the TRUNK version soon
and will be in the next release.
best 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: "cp2k" <cp...@googlegroups.com>
Sent by: cp...@googlegroups.com
Date: 12/22/2018 10:45PM
Subject: Re: [CP2K:11082] comparing CP2K-DFTB (5.1) and DFTB+ (18.2)
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 https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.