[CP2K-user] [CP2K:11071] Generate the GTH pseudopotential by ATOM

ningz... at gmail.com ningz... at gmail.com
Wed Dec 26 05:19:17 CET 2018


Hi Hutter,
  Thank you so much! I followed your suggestions and I finally get some 
reasonable results though I haven't obtained very good convergence. I try 
to use the GTH parameters of Ce as initial guess to generate the GTH 
parameter of Pr. My parameter is:
&POWELL
     ACCURACY   1.e-14
     STEP_SIZE  0.08
     STEP_SIZE_SCALING  0.90
     MAX_INIT   50
     MAX_FUN    100
     WEIGHT_PSIR0 0.0
     TARGET_POT_SEMICORE      [eV]      0.003000
     TARGET_POT_VALENCE       [eV]      0.000300
     TARGET_POT_VIRTUAL       [eV]      0.003000
     SEMICORE_LEVEL  [eV]  15.0
     WEIGHT_POT_NODE                   10.0
     WEIGHT_POT_SEMICORE               10.0
     WEIGHT_POT_VALENCE                 5.0
     WEIGHT_POT_VIRTUAL                 2.0
  &END
&END ATOM
  I give a large STEP_SIZE at first, and then I decrease the STEP_SIZE by 
order for the following restarts. The optimization results are:
 Reference configuration          1               Method number            
    1
    L    N    Occupation      Eigenvalue [eV]           dE [eV]          
dCharge
    0    1          2.00       -38.8397939312 SC  -0.009395[ 1]    
-0.000503[ X]
    0    2          2.00        -3.3860472487 VA   0.010363[87]    
-0.004646[ 0]
    0    3          0.00         1.6037528306 U1   0.013647[ 0]    
-0.001261[ X]
    0    4          0.00         8.1521110396 U2   0.035468[ 0]    
-0.001273[ 0]
    1    1          6.00       -21.3735479056 SC   0.024748[ 8]    
-0.002697[ X]
    1    2          0.00        -1.0360212280 U1   0.029883[ 2]    
-0.009428[ 0]
    1    3          0.00         3.6087950888 U2   0.064960[ 0]    
-0.005279[ 0]
    2    1          0.00        -1.9516163818 U1   0.002300[ X]    
-0.014541[ 0]
    2    2          0.00         2.1059505831 U2   0.025993[ X]    
 0.000310[ 0]
    3    1          3.00        -1.2120770213 VA   0.000989[ 0]    
-0.005278[ 0]
    3    2          0.00         2.9876786134 U1  -0.004826[ 0]    
 0.001172[ X]
    3    3          0.00         8.1202646230 U2   0.034044[ 0]    
-0.006754[ 0]
    s-states N=    1                   Wavefunction at r=0:        
 0.002169[ 0]
    s-states N=    2                   Wavefunction at r=0:        
 0.018634[ 0]
    s-states N=    3                   Wavefunction at r=0:        
 0.016484[ 0]
    s-states N=    4                   Wavefunction at r=0:        
-0.030807[ 0]

 Number of target values reached:                                       6 
of  20
    It still needs several steps restart. I have a question about the 
"optimal parameter". what is the standard for us to choose the best 
parameter? I just try to find the best parameter which can make the 
objective function lowest for each optimization.
    With my best regards,
    Zhi

在 2018年12月21日星期五 UTC-8上午1:12:02,jgh写道:
>
> Hi 
>
> as said before, there is no unique solution to the problem. 
> I gave you the parameters I use, but you can change them at your 
> will. Use what works best for you. 
> Yes, I also use frequent restarts with adjustments of parameters. 
>
> 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 
> <javascript:> 
> Winterthurerstrasse 190 
> CH-8057 Zürich, Switzerland 
> --------------------------------------------------------------- 
>
> -----cp... at googlegroups.com <javascript:> wrote: ----- 
> To: "cp2k" <cp... at googlegroups.com <javascript:>> 
> From: ning... at gmail.com <javascript:> 
> Sent by: cp... at googlegroups.com <javascript:> 
> Date: 12/21/2018 07:09AM 
> Subject: Re: [CP2K:11071] Generate the GTH pseudopotential by ATOM 
>
> Hi Hutter, 
>   Thank you! I try to change the STEP_SIZE and MAX_FUN. I use the optimal 
> parameter, and it can work. 
>   There is another problem. We will choose a series parameters as initial 
> guess for pseudopotential optimization. Then we use the so-called optimal 
> STEP_SIZE and MAX_FUN. But we always need to restart the optimization using 
> the pseudopotential parameter of previous step as initial guess. Do we 
> still use the same STEP_SIZE and MAX_FUN? I just use the same value. But I 
> think we need to use adaptive STEP_SIZE. I know it is very hard to change 
> the STEP_SIZE for each restart. 
>   By the way, I think we should also test the weight for each item in 
> objective function. 
>   With my best regards, 
>   Zhi   
>
> 在 2018年12月19日星期三 UTC-8上午12:45:36,jgh写道:Hi 
>   
> as I said in the last comment - it is an art - to choose the 
> best parameter. You need some experience to get "optimal" 
> performance/convergence.   
> I usually play with these parameters 
>   
>      STEP_SIZE  0.08   
>      MAX_INIT   50   
>      MAX_FUN    250   
>      STEP_SIZE_SCALING  0.90   
>   
> Start with large STEP_SIZE (0.1) and small MAX_FUN (50-100) 
> and end with small STEP_SIZE (0.00001) and large MAX_FUN(5000). 
>   
> 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 
> Winterthurerstrasse 190 
> CH-8057 Zürich, Switzerland 
> --------------------------------------------------------------- 
>   
> -----cp... at googlegroups.com wrote: ----- 
> To: "cp2k" <cp... at googlegroups.com> 
> From: ning... at gmail.com 
> Sent by: cp... at googlegroups.com 
> Date: 12/18/2018 08:15PM 
> Subject: Re: [CP2K:11058] Generate the GTH pseudopotential by ATOM 
>   
> Hi Hutter, 
>   I find I need to restart the optimization for many times. It always 
> needs 5000 restarting calculations using previous step GTH parameters. This 
> process is so slow. How can I speed up this process? 
>   With my best regards, 
>   Zhi 
>   
> 在 2018年12月12日星期三 UTC-8下午11:56:59,jgh写道:Hi   
>   
> for many atoms you have to restart the optimization frequently   
> until you reach sufficient convergence.   
> For the restart you copy the GTH-PARAMETER to your input.   
> Optimization is a kind of art. Choosing the parameter in the   
> POWELL section is important for good performance. Still, it takes   
> me often several 10'000 steps to reach reasonable results.   
>   
> 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   
> Winterthurerstrasse 190   
> CH-8057 Zürich, Switzerland   
> ---------------------------------------------------------------   
>   
> -----cp... at googlegroups.com wrote: -----   
> To: "cp2k" <cp... at googlegroups.com>   
> From: ning... at gmail.com   
> Sent by: cp... at googlegroups.com   
> Date: 12/12/2018 06:39PM   
> Subject: Re: [CP2K:11046] Generate the GTH pseudopotential by ATOM   
>   
> Hi Hutter,   
>   Thank you! I use your input file to generate the pseudopotential of Ti 
> and I can reproduce your result. But I still have a question, the initial 
> guess is nearly same with the final optimized pseudopotential. How can we 
> get the better initial guess? These parameters include the R^cut and 
> coefficients in non-local part and R^cut for each channel and h^ij 
> coefficients in the local part.   
>   With my best regards,   
>   Zhi   
>   
> 在 2018年12月12日星期三 UTC-8上午12:08:27,jgh写道:Hi   
>     
> the optimization of GTH pseudos is not unique. Depending on your   
> object function, starting point and convergence you can get rather   
> different results. The performance of all these different pseudos   
> should be rather similar though.   
> There are also elements that can be tricky to get and many repeated   
> optimizations and some 'magic' is needed to get good results.   
>     
> This is a generic input I am currently using to optimize potentials:   
>     
> &GLOBAL   
>   PROGRAM_NAME ATOM   
> &END GLOBAL   
> &ATOM   
>   ELEMENT Ti   
>     
>   RUN_TYPE PSEUDOPOTENTIAL_OPTIMIZATION   
>     
>   ELECTRON_CONFIGURATION  [Ne] 3s2 3p6 4s2 3d2   
>   CORE [Ne]   
>   MAX_ANGULAR_MOMENTUM 3   
>     
>   &METHOD   
>      METHOD_TYPE  KOHN-SHAM   
>      RELATIVISTIC DKH(3)   
>      &XC   
>        &XC_FUNCTIONAL   
>          &LIBXC   
>           FUNCTIONAL MGGA_X_SCAN   
>          &END LIBXC   
>          &LIBXC   
>            FUNCTIONAL MGGA_C_SCAN   
>          &END LIBXC   
>        &END XC_FUNCTIONAL   
>      &END XC   
>   &END METHOD   
>   &OPTIMIZATION   
>     EPS_SCF 1.e-7   
>   &END   
>   &AE_BASIS   
>      BASIS_TYPE GEOMETRICAL_GTO   
>   &END AE_BASIS   
>   &PP_BASIS   
>      BASIS_TYPE GEOMETRICAL_GTO   
>   &END PP_BASIS   
>   &POTENTIAL   
>     PSEUDO_TYPE GTH   
>     &GTH_POTENTIAL   
>     4    6    2    0   
>     0.38300965389957       2    8.69526962380773   -0.69130398136282   
>        3   
>     0.32535306838298       2    2.49370385691556    3.69297179681161   
>                                                    -4.49419494638446   
>     0.25460783706014       2   -4.62952832752021    8.87087769105041   
>                                                   -10.49552627066998   
>     0.24470423408872       1   -9.41064468800323   
>     &END   
>     CONFINEMENT_TYPE  BARRIER   
>     CONFINEMENT 200. 4.0 12.0   
>   &END POTENTIAL   
>     
>   &POWELL   
>      ACCURACY   1.e-14   
>      STEP_SIZE  0.08   
>      MAX_INIT   50   
>      MAX_FUN    250   
>      STEP_SIZE_SCALING  0.90   
>      WEIGHT_PSIR0 0.0   
>      TARGET_POT_SEMICORE      [eV]      0.003000   
>      TARGET_POT_VALENCE       [eV]      0.000300   
>      TARGET_POT_VIRTUAL       [eV]      0.003000   
>      WEIGHT_POT_NODE                   10.0   
>      WEIGHT_POT_SEMICORE                2.0   
>      WEIGHT_POT_VALENCE                 5.0   
>      WEIGHT_POT_VIRTUAL                 1.0   
>   &END   
> &END ATOM   
>     
> You can find my recent optimized potentials for PBE, PBE0, SCAN, HF   
> at my github repo (absolutely no warrenty!)   
>     
> https://github.com/juerghutter/GTH   
>     
> 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   
> Winterthurerstrasse 190   
> CH-8057 Zürich, Switzerland   
> ---------------------------------------------------------------   
>     
> -----cp... at googlegroups.com wrote: -----   
> To: "cp2k" <cp... at googlegroups.com>   
> From: ning... at gmail.com   
> Sent by: cp... at googlegroups.com   
> Date: 12/12/2018 02:19AM   
> Subject: [CP2K:11041] Generate the GTH pseudopotential by ATOM   
>     
> Hello,   
>   I need to generate the GTH pseudopotential and hope to use ATOM code in 
> CP2K. But I find it is a little hard for me. The only thing I can do is 
> just changing the initial guess of pseudopotential parameters. But there 
> are so many parameters in GTH pseudopotential. The optimized 
> pseudopotential changes slightly compared with initial guess. Can anyone 
> tell me how to generate the GTH pseudopotential correctly by ATOM? I will 
> be very appreciated!   
>   Best regards,   
>   Zhi     
>   --     
>  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+... at googlegroups.com.   
>  To post to this group, send email to cp... at googlegroups.com.   
>  Visit this group at https://groups.google.com/group/cp2k.   
>  For more options, visit 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+... at googlegroups.com.   
>  To post to this group, send email to cp... at googlegroups.com.   
>  Visit this group at https://groups.google.com/group/cp2k.   
>  For more options, visit 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+... at googlegroups.com. 
>  To post to this group, send email to cp... at googlegroups.com. 
>  Visit this group at https://groups.google.com/group/cp2k. 
>  For more options, visit 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+... at googlegroups.com <javascript:>. 
>  To post to this group, send email to cp... at googlegroups.com <javascript:>. 
>
>  Visit this group at https://groups.google.com/group/cp2k. 
>  For more options, visit https://groups.google.com/d/optout. 
>   
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20181225/29764fad/attachment-0001.html>


More information about the CP2K-user mailing list