[Pw_forum] Re [3] : PP-US of Au

[Gabriele Sclauzero]

Il giorno 29/ago/2010, alle ore 16.18, reza shidpoor ha scritto:

> 2 ) The RAM of my system is sufficient for 13 atoms (RAM = 8 Gigabyte DDR3).

How can you be sure? It depends on a lot of factors, besides the number of atoms: cutoffs, number of calculated bands, cell size, diagonalization method, ..., and also pseudopotential type.

Indeed, US PPs have usually many more projector functions than NC PPs.

> 3 ) I tested a case . I used informations related to 01 example ( Si - scf ).
>     I constructed a input file from a imaginary 13-atom Si cluster that is different from
>     previous input file in type of atoms and pseudopotential. Then I run scf calculation both of them.
> 
> Input File related to Au :
> 
>  &control
>     calculation='scf'
>     title='Au13Cluster'
>     restart_mode='from_scratch'
>     outdir='./tmp'
>     prefix='Au13'
>     etot_conv_thr=1.0D-4
>     pseudo_dir = './'
>  /
>  &system
>     ibrav= 2, celldm(1) =  38,nat=  13, ntyp= 1,
>     ecutwfc =40.0, ecutrho=400.0,
>     occupations='smearing',smearing='gaussian',degauss=0.01,
>     nspin=1,
>  /
>  &electrons
>    electron_maxstep=100,
>    diagonalization='cg',
>    conv_thr=1.0D-6,
>    mixing_beta=0.5,
>  /
> &ions
>    ion_dynamics='bfgs'
> /
> &cell
> /
> 
> ATOMIC_SPECIES
>  Au  196.97  Au.pbe-nd-van.UPF
> ATOMIC_POSITIONS {angstrom}
> Au      4.844   4.844   4.844
> Au      0.000   4.844   7.833
> Au      4.844   7.833   0.000
> Au      4.844   7.833   9.688
> Au      9.688   4.844   1.855
> Au      1.855   0.000   4.844
> Au      1.855   9.688   4.844
> Au      4.844   1.855   9.688
> Au      7.833   9.688   4.844
> Au      4.844   1.855   0.000
> Au      9.688   4.844   7.833
> Au      7.833   0.000   4.844
> Au      0.000   4.844   1.855
> K_POINTS {automatic}
> 2 2 2  1 1 1        
> 
> 
> After of several minutes, the result was 
> 
> 
>      Largest allocated arrays     est. size (Mb)     dimensions
>         Kohn-Sham Wavefunctions       110.74 Mb     (  84386,  86)
>         NL pseudopotentials           217.61 Mb     (  84386, 169)
>         Each V/rho on FFT grid         51.50 Mb     (3375000)
>         Each G-vector array             9.18 Mb     (1203541)
>         G-vector shells                 0.03 Mb     (   3661)
>      Largest temporary arrays     est. size (Mb)     dimensions
>         Each subspace H/S matrix        0.11 Mb     (     86,  86)
>         Each <psi_i|beta_j> matrix      0.22 Mb     (    169,  86)
>         Arrays for rho mixing         411.99 Mb     (3375000,   8)
> 
>      Initial potential from superposition of free atoms
>      Check: negative starting charge=   -0.356687
> 
>      starting charge  142.84904, renormalised to  143.00000
> 
>      negative rho (up, down):  0.357E+00 0.000E+00
>      Starting wfc are  117 atomic wfcs
> 
>      total cpu time spent up to now is    255.53 secs
> 
>      per-process dynamical memory:  1035.0 Mb

This estimate is usually a lower bound. I found that the memory requirement at some points of the run can as big as the double of that number. On how many processors was this run done?

> 
>      Self-consistent Calculation
> 
>      iteration #  1     ecut=    51.00 Ry     beta=0.50
>      CG style diagonalization
>      ethr =  1.00E-02,  avg # of iterations =  2.1
> forrtl: severe (174): SIGSEGV, segmentation fault occurred
> Image              PC                Routine            Line        Source             
> pw.x               000000000052089C  addusdens_                 27  addusdens.f90
> pw.x               00000000004DCB76  sum_band_                 153  sum_band.f90
> pw.x               000000000044B7D7  electrons_                287  electrons.f90
> pw.x               000000000040BA6E  MAIN__                     92  pwscf.f90
> pw.x               000000000040B8DC  Unknown               Unknown  Unknown
> libc.so.6          00002B500B54CABD  Unknown               Unknown  Unknown
> pw.x               000000000040B7D9  Unknown               Unknown  Unknown

You may obtain this kind of messages when you hit the maximum available RAM memory or when the code tries to access a memory address not reserved to it. To exclude the latter you could try to recompile with -g -check bounds -traceback and rerun. To exclude the former, try a smaller system or run on more processors on a bigger machine.

> 
> 
>  Now , for input file related Si 
> 
> &control
>     calculation='scf
>     title='Si13Cluster
>     restart_mode='from_scratch
>     outdir='./tmp
>     prefix='Si13
>     etot_conv_thr=1.0D-4
>     pseudo_dir = './
>  /
>  &system
>     ibrav=  2, celldm(1) =  38,nat=  13, ntyp= 1, 
>     ecutwfc =40.0, ecutrho=400.0, 
>     occupations='smearing',smearing='mp',degauss= 0.01, 
>     nspin=1,
>  /
>  &electrons
>    electron_maxstep=100,
>    diagonalization='cg', 
>    conv_thr=1.0D-6,
>    mixing_beta=0.5,
>  /
> &ions
>    
> /
> &cell
> /
> 
> ATOMIC_SPECIES
>  Si  28.086  Si.pz-vbc.UPF
> ATOMIC_POSITIONS {angstrom}
> Si    0.000 0.000 0.000
> Si      0.000   4.844   7.833
> Si      4.844   7.833   0.000
> Si      4.844   7.833   9.688
> Si      9.688   4.844   1.855
> Si      1.855   0.000   4.844
> Si      1.855   9.688   4.844
> Si      4.844   1.855   9.688
> Si      7.833   9.688   4.844
> Si      4.844   1.855   0.000
> Si      9.688   4.844   7.833
> Si      7.833   0.000   4.844
> Si      0.000   4.844   1.855
> K_POINTS {automatic}
> 2 2 2  1 1 1
> 
> And the result was :
> 
> ...............
> ...............
> iteration # 50     ecut=    40.00 Ry     beta=0.50
>      CG style diagonalization
>      ethr =  2.38E-05,  avg # of iterations =  3.8
> 
>      negative rho (up, down):  0.442E-07 0.000E+00
> 
>      total cpu time spent up to now is   7717.21 secs
> 
>      total energy              =     -97.69548088 Ry
>      Harris-Foulkes estimate   =     -97.70290819 Ry
>      estimated scf accuracy    <       0.46657030 Ry
> 
>  iteration # 51     ecut=    40.00 Ry     beta=0.50
>      CG style diagonalization
>      ethr =  2.38E-05,  avg # of iterations =  3.0
> 
>      negative rho (up, down):  0.250E-07 0.000E+00
> 
>      total cpu time spent up to now is   7843.25 secs
> 
>      total energy              =     -97.69672289 Ry
>      Harris-Foulkes estimate   =     -97.69637710 Ry
>      estimated scf accuracy    <       0.33411951 Ry
> ...................
> ...................
> ...................
> and continues : The simulation was done simply !!!!!!

It didn't converge after 51 iterations, this may be suspicious...

> 
> 
> Dear Gabriele
> 
> I have almost waste one month because of this problem and I diddnot solve problem.

Perhaps you are not approaching it in the correct way.

> 
> With regard to 1,2,3 points , Are you agree with me that the source of problem is only and only pseudopotential ?
> 
> Best wishes.

Regards,

Gabriele


> 
> Reza.Shidpour
> Institute for Nanoscienec and Nanotechnology,
> Sharif University of Technology (SUT)
> 
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum


§ Gabriele Sclauzero, EPFL SB ITP CSEA
   PH H2 462, Station 3, CH-1015 Lausanne

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20100829/7a78945f/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 1753 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20100829/7a78945f/attachment.p7s>