[CPMD-list] PBE + NLCC convergence problem

Wed Aug 4 17:28:15 CEST 2004

>>> "OY" == Oleg Yazyev <oleg.yazyev at epfl.ch> writes:

OY> Dear Axel,

AK> and finally, does the same setup work in closed shell cases?

OY> Yes, for closed shell it works fine. At least for water molecule I've
OY> just checked.

ok.

AK> also, did you check, how much this is affected by GC-CUTOFF?

OY> Is starts to work fine if GC-CUTOFF >> 1.e-4,
OY> but this is unacceptable for any application.
OY> Probably, at this level no gradient-correction is left at all!

i agree.

AK> i'm not so sure, if this applies here, but i have seen, that the
AK> line search occasionally gets diverted for some systems in the initial
AK> state of the optimization (too much noise). in those cases, i could get
AK> to convergence, by first using ODIIS until it was 'stuck' and then
AK> continue with PCG MINIMIZE. you should check, however, if the results
AK> are physically reasonable.

OY> Both of the them explodes. The example is isolated CH3 radical, see
OY> below. ODIIS's case even worse!

ok. so as far as i can tell, that leaves two sources of errors:
there is a problem with the pseudopotential generation program
with NLCC and PBE or there is a problem within cpmd. unfortunately
i have not yet looked enough at the cpmd sourcecode to give a
recommendation. it would be helpful to know, if _anybody_ has
used that combination (PBE/NLCC/LSD) successfully at all.

you may want to have a look at the sourcecode yourself, and
compare the relevant sections of lsd_func.F and functionals.F.

good luck,
     axel.

OY> PGC MINIMIZE:
OY>  NFI      GEMAX       CNORM           ETOT        DETOT      TCPU
OY>    1  2.956E-02   3.896E-03      -7.189008    0.000E+00      9.19
OY>  LINE SEARCH : LAMBDA=.703E-01 PREDICTED ENERGY =  5.147515178916
OY>    2  2.075E+01   3.903E+00      -7.249209   -6.020E-02      8.66
OY>  LINE SEARCH : LAMBDA=.211     PREDICTED ENERGY =  3.769833904981
OY>    3  1.377E+04   2.676E+03       3.770053    1.102E+01      8.42
OY>  LINE SEARCH : LAMBDA=.105     PREDICTED ENERGY =  3.769858361282
OY>    4  1.289E+04   2.648E+03       3.769858   -1.947E-04      8.39
OY>  LINE SEARCH : LAMBDA=.105     PREDICTED ENERGY =  3.769858574551
OY>    5  1.276E+04   2.519E+03       3.769858    1.093E-07      8.42
OY>  LINE SEARCH : LAMBDA=.105     PREDICTED ENERGY =  3.769858918988
OY>    6  1.292E+04   2.549E+03       3.769859    2.017E-07      8.28
OY>  LINE SEARCH : LAMBDA=.105     PREDICTED ENERGY =  3.769859484147
OY>    7  1.303E+04   2.530E+03       3.769859    3.286E-07      8.31
OY>  LINE SEARCH : LAMBDA=.105     PREDICTED ENERGY =  3.769860363765
OY>    8  1.260E+04   2.608E+03       3.769859    5.389E-07      8.30

OY> ODIIS:
OY>  NFI      GEMAX       CNORM           ETOT        DETOT      TCPU
OY>    1  2.956E-02   3.896E-03      -7.189008    0.000E+00      3.22
OY>    2  5.204E+03   1.049E+03      -7.383763   -1.948E-01      3.25
OY>    3  3.489E+03   9.662E+02      -7.383541    2.220E-04      3.22
OY>    4  1.876E+03   4.716E+02      -7.375378    8.164E-03      3.24
OY>    5  4.387E+01   2.062E+01      -7.363097    1.228E-02      3.25
OY>    6  2.883E+02   1.893E+02      -7.352987    1.011E-02      3.25
OY>    7  2.128E+01   6.533E+00      -7.344642    8.345E-03      3.25
OY>    8  3.753E+01   1.806E+01      -7.337581    7.060E-03      3.25
OY>    9  5.459E+00   1.071E+00      -7.331482    6.099E-03      3.25
OY>   10  6.486E+01   2.790E+01      -7.326023    5.458E-03      3.28
OY>   11  1.321E+01   3.965E+00      -7.321248    4.776E-03      3.25
OY>   12  1.340E+05   5.053E+04      21.290431    2.861E+01      3.14
OY>  ODIIS| Insufficient progress; reset!
OY>   13  1.652E+04   6.124E+03      22.153249    8.628E-01      3.08
OY>   14  1.074E+06   1.288E+05      11.617793   -1.054E+01      2.90
OY>   15  1.246E+06   1.514E+05      11.626043    8.250E-03      2.88
OY>   16  1.044E+06   1.397E+05      20.808698    9.183E+00      2.96
OY>   17  4.957E+05   1.125E+05      28.292228    7.484E+00      2.97
OY>   18  6.336E+05   1.083E+05      34.496024    6.204E+00      2.97
OY>   19  1.452E+05   4.385E+04      39.732968    5.237E+00      2.96
OY>   20  2.063E+04   7.521E+03      44.416126    4.683E+00      2.94

AK> could you please post some more information about the 
AK> pseudopotentials, you are using.

OY> Sure. For the example above I used BHS pseudos with "typical" cutoffs.
OY> PW cutoff was 60 Ry here. PP headers are:

OY>     ============================================================
OY>     |    Pseudopotential Report                                |
OY>     ------------------------------------------------------------
OY>     |  Atomic Symbol                   :   C                   |
OY>     |  Atomic Number                   :   6                   |
OY>     |  Number of core states           :   1                   |
OY>     |  Number of valence states        :   2                   |
OY>     |  Exchange-Correlation Functional :                       |
OY>     |     Slater exchange :  0.6667                            |
OY>     |     LDA correlation : Ceperley-Alder                     |
OY>     |     Exchange GC     : Perdew-Burke-Ernzerhof             |
OY>     |     Correlation GC  : Perdew-Burke-Ernzerhof             |
OY>     |  Electron Configuration :   N   L  Occupation            |
OY>     |                             1   S    2.0000              |
OY>     |                             2   S    2.0000              |
OY>     |                             2   P    2.0000              |
OY>     |  Full Potential Total Energy =      -37.748221           |
OY>     |  Bachelet-Hamann-Schluter normconserving PP              |
OY>     |     n    l        rc       energy                        |
OY>     |     2    S    1.2000     -0.50506                        |
OY>     |     2    P    1.2000     -0.19451                        |
OY>     |     3    D    0.7159     -0.19451                        |
OY>     |  Nonlinear core correction                               |
OY>     |  Core charge cutoff radius :   0.71590                   |
OY>     |  Number of Mesh Points :   615                           |
OY>     |  Pseudoatom Total Energy =   -5.356782                   |
OY>     ============================================================

OY>     ============================================================
OY>     |    Pseudopotential Report                                |
OY>     ------------------------------------------------------------
OY>     |  Atomic Symbol                   :   H                   |
OY>     |  Atomic Number                   :   1                   |
OY>     |  Number of core states           :   0                   |
OY>     |  Number of valence states        :   1                   |
OY>     |  Exchange-Correlation Functional :                       |
OY>     |     Slater exchange :  0.6667                            |
OY>     |     LDA correlation : Ceperley-Alder                     |
OY>     |     Exchange GC     : Perdew-Burke-Ernzerhof             |
OY>     |     Correlation GC  : Perdew-Burke-Ernzerhof             |
OY>     |  Electron Configuration :   N   L  Occupation            |
OY>     |                             1   S    0.7000              |
OY>     |  Full Potential Total Energy =       -0.474341           |
OY>     |  Bachelet-Hamann-Schluter normconserving PP              |
OY>     |     n    l        rc       energy                        |
OY>     |     1    S    0.5000     -0.37203                        |
OY>     |     2    P    0.3828     -0.37203                        |
OY>     |  Number of Mesh Points :   511                           |
OY>     |  Pseudoatom Total Energy =   -0.367687                   |
OY>     ============================================================

OY> -- 
OY> Best regards,
OY>  Oleg

OY> ______________________________________________________
OY> Oleg Yazyev
OY> Institute of Molecular and Biological Chemistry
OY> Swiss Federal Institute of Technology in Lausanne (EPFL)
OY> EPFL - BCH
OY> CH-1015 Lausanne (Switzerland)
OY> Tel.: +41 21 693 9881
OY> E-mail: oleg.yazyev at epfl.ch
OY> WWW: http://icmbcu001.epfl.ch/yazyev/index.html
OY> _______________________________________________________

--

=======================================================================
Axel Kohlmeyer       e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.