[CPMD-list] Fe calculations

[Juerg Hutter]

Hi

this bug has been reported before. You can find the bugfix
at

http://www.cpmd.org/pipermail/cpmd-list/2003-February/000687.html


We will soon make a new version available with the
recently found bugs fixed.

Juerg



----------------------------------------------------------
Juerg Hutter                   Phone : ++41 1 635 4491
Physical Chemistry Institute   FAX   : ++41 1 635 6838
University of Zurich           E-mail: hutter at pci.unizh.ch
Winterthurerstrasse 190
CH-8057 Zurich, Switzerland
----------------------------------------------------------


On Fri, 14 Feb 2003, Eyvaz Isaev wrote:

> Dear all,
>
> Now I have my own question. I tried to calculate using
> CPMD_3.7 the total energy for a number of lattice
> parameters for BCC Iron with one atom in the unit
> cell. Input file I have used is attached.
> Surprisingly,  I could not start the job on
> Athlon/1700 based PC with 512Mb core memory. The same
> happened on Athlon/2100 with 1Gb core memory. In both
> cases cpmd.x  stops with error about memory
> allocation:
>
> ****************************************************************
>  PROCESSOR     0 ALLOCATION OF   140810176 WORDS OF
> MEMORY FAILED
>
> ****************************************************************
>
>  PROGRAM STOPS IN SUBROUTINE MEMORY| ALLOCATION FAILED
> (SCR)
>
> The job started for small cutoff energy (15Ry), but
> there was a restriction for k-mesh, i.e. the  6x6x6
> mesh worked, but 8x8x8 mesh did not.  Then I have
> observed another puzzle: "grep 'NFI' Fe.out" showed
> that the total energy converged, but the last value
> was different compared to total energies  for
> preceding iterations.
>
> I have to note that CPMD_3.5 works well with
> E_cut=60Ry and 20x020x20 k-mesh on the same PC.
> Lattice parameter 2.72Ang for BCC Fe, probably, is not
> bad for  the LDA calculation. But there is no magnetic
> moment though the "LSD" key is mentioned. The above
> mentioned difference for the total energy between the
> last and preceding
> iterations takes place, too (The next cited results
> reference to CPMD_3.5).
>
>  ==     NFI=     32      ETOT=    -19.953008   TCPU=
>  397.85 ==
>  ==     NFI=     33      ETOT=    -19.953007   TCPU=
>  221.33 ==
>  ==     NFI=     34      ETOT=    -19.953006   TCPU=
>  497.17 ==
>  ==     NFI=     35      ETOT=    -19.953006   TCPU=
>  236.00 ==
>  ==     NFI=     36      ETOT=    -19.953075   TCPU=
>  443.73 ==
>
>  == DRHOMAX= 1.366E-03  DETOT=     2.240E-06    THL=
> 5.124E-01 ==
>  == DRHOMAX= 1.179E-03  DETOT=     6.589E-07    THL=
> 5.346E-01 ==
>  == DRHOMAX= 1.053E-03  DETOT=     1.330E-06
> THL=-1.930E-03 ==
>  == DRHOMAX= 1.067E-03  DETOT=    -8.935E-08    THL=
> 1.861E+00 ==
>  == DRHOMAX= 9.999E-04  DETOT=    -6.931E-05    THL=
> 9.562E-01 ==
>
>
> I guessed it is due to a rather sharp change of the
> parameter DRHOMAX and put 5.D-4 (instead 1.D-3).
> It turned out that the suggestion is not true:
>
>  ==     NFI=     20      ETOT=    -19.953006   TCPU=
>  139.03 ==
>  ==     NFI=     21      ETOT=    -19.953006   TCPU=
>  120.44 ==
>  ==     NFI=     22      ETOT=    -19.953006   TCPU=
>  136.42 ==
>  ==     NFI=     23      ETOT=    -19.953006   TCPU=
>   72.60 ==
>  ==     NFI=     24      ETOT=    -19.953075   TCPU=
>  218.27 ==
>
>  == DRHOMAX= 6.108E-04  DETOT=    -2.409E-08    THL=
> 1.091E+00 ==
>  == DRHOMAX= 6.504E-04  DETOT=    -6.143E-08    THL=
> 1.074E+00 ==
>  == DRHOMAX= 5.551E-04  DETOT=     1.478E-07
> THL=-1.008E-01 ==
>  == DRHOMAX= 5.225E-04  DETOT=     4.693E-08
> THL=-3.112E+00 ==
>  == DRHOMAX= 3.774E-04  DETOT=    -6.899E-05    THL=
> 3.587E-01 ==
>
> One can see that ETOT(24) and ETOT(23) are different
> (69 a.u.) again and the convergence  for the last
> iteration unexpectedly becomes worser.
>
> Later the problem was fixed using more strong DETOT
> (1.D-5). So, this kind of convergence behaviour makes
> one be careful everytime about convergence.
>
> Could you please tell me what is wrong in my input
> file for total energy calculations and why the job
> faild to start for such a small system?
> Why Fe is nonmagnetic despite "LSD" key presents?
> Besides, may be something is wrong in my notes about
> convergence?
>
> Any suggestions are welcome.
>
> Regards,
> Eyvaz Isaev, PhD
> Senior Scientific Researcher
> Theoretical Physics Department
> Moscow Steel and Alloys Institute
>
> __________________________________________________
> Do you Yahoo!?
> Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
> http://mailplus.yahoo.com