[CP2K-user] [CP2K:11065] comparing CP2K-DFTB (5.1) and DFTB+ (18.2)

Thomas Kühne tku... at gmail.com
Sat Dec 22 14:31:53 UTC 2018


Dear Wei Lai, 

have you used explicit diagonalization or the default OT minimizer? 
Do the fully converged SCC-results agree? If not please provide the 
input file. 

Merry Christmas, 
Thomas

> Am 19.12.2018 um 19:09 schrieb Wei Lai <wei.w... at gmail.com>:
> 
> Dear Juerg,
> 
> I have done further testing and may be closer to find out the source of this discrepancy.  Basically, non-SCC results are "consistent" between CP2K and DFTB+.  Fore SCC-DFTB, they differ in whether to include SCC at the first scf (it also seems that DFTB+ applies non-scc scf first before applying SCC energy).  I don't know the coding thoughts of SCC well enough but I am wondering which one of these two is the better way.  Regarding why CP2K results are sensitive to ALPHA I guess there are many local energy minimums.
> 
> Thanks, Wei
> 
> CP2K:
> Parameters	Li-Mulliken-charge	H0-energy	SCC-energy
> NonSCC full scf	0.737779	    -417.18328600	
> SCC-1scf	                0.737779	    -355.69044000	0
> SCC-2scf	                0.720362	    -361.83972460	0.137714872
> 
> DFTB+:			
> NonSCC full scf	0.737791	     -417.18197790	
> SCC-1scf	                0.737791	     -417.18197790	10.2841704
> SCC-2scf	                0.711388	     -417.10728727	9.561284596
> 
> On Wed, Dec 19, 2018 at 9:50 AM Wei Lai <wei.w... at gmail.com <mailto:wei.w... at gmail.com>> wrote:
> Dear Juerg,
> 
> Thanks for your suggestion.  Parameters I chose previously converged the results reasonably.  To be more systematic,  I tried the following combinations without specifying RCUT (I was doing that to make sure it's less than 1/2 of the box size).  Comparison of these with those I presented earlier still suggests strong sensitivity to ALPHA and difference from DFTB+.
> 
> Thanks, Wei
> 
> Accuracy	  GMAX	ALPHA	     Li-Mulliken-charge	H0-energy	SCC-energy
> 1.00E-06	    21	0.583	            0.57183	    -414.5978042	8.302909355
> 1.00E-10	    21	0.583	            0.57183	    -414.5978042	8.302909355
> 1.00E-14	    21	0.583	            0.57183	    -414.5978042	8.302909355
> 1.00E-14	    61	0.583	            0.57183	    -414.5978033	8.302910247
> 1.00E-14	  101	0.583	            0.57183	    -414.5978027	8.302910732
> 1.00E-14	  101	0.35 (default)	    0.572372	    -414.6135268	8.288678932
> 
> 
> 
> 
> On Wed, Dec 19, 2018 at 3:41 AM <hut... at chem.uzh.ch <mailto:hut... at chem.uzh.ch>> wrote:
> Hi
> 
> 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... at chem.uzh.ch <mailto:hut... at chem.uzh.ch>
> Winterthurerstrasse 190
> CH-8057 Zürich, Switzerland
> ---------------------------------------------------------------
> 
> -----cp... at googlegroups.com <mailto:cp... at googlegroups.com> wrote: -----
> To: "cp2k" <cp... at googlegroups.com <mailto:cp... at googlegroups.com>>
> From: "Wei Lai" 
> Sent by: cp... at googlegroups.com <mailto:cp... at googlegroups.com>
> Date: 12/19/2018 04:50AM
> Subject: [CP2K:11061] comparing CP2K-DFTB (5.1) and DFTB+ (18.2)
> 
> 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  
>   -- 
>  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... at googlegroups.com <mailto:cp2k%2Bun... at googlegroups.com>.
>  To post to this group, send email to cp... at googlegroups.com <mailto:cp... at googlegroups.com>.
>  Visit this group at https://groups.google.com/group/cp2k <https://groups.google.com/group/cp2k>.
>  For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.
> 
> 
> -- 
> 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 <https://groups.google.com/d/topic/cp2k/RfYbOq3j0DM/unsubscribe>.
> To unsubscribe from this group and all its topics, send an email to cp2k+uns... at googlegroups.com <mailto:cp2k%2Bun... at googlegroups.com>.
> To post to this group, send email to cp... at googlegroups.com <mailto:cp... at googlegroups.com>.
> Visit this group at https://groups.google.com/group/cp2k <https://groups.google.com/group/cp2k>.
> For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.
> 
> -- 
> 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... at googlegroups.com <mailto:cp2k+uns... at googlegroups.com>.
> To post to this group, send email to cp... at googlegroups.com <mailto:cp... at googlegroups.com>.
> Visit this group at https://groups.google.com/group/cp2k <https://groups.google.com/group/cp2k>.
> For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.



==============================
Thomas D. Kühne
Dynamics of Condensed Matter
Chair of Theoretical Chemistry
University of Paderborn
Warburger Str. 100
D-33098 Paderborn
Germany
tdku... at mail.upb.de
+49/(0)5251/60-5726

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20181222/6ee33d21/attachment.htm>


More information about the CP2K-user mailing list