Page 1 of 1

VASP 5.2.2 Parallel Execution - Help Needed

Posted: Wed Feb 10, 2010 7:23 pm
by sgowtham
We have compiled VASP 5.2.2 on our beowulf linux cluster [ROCKS 5.2.2; Red Hat Enterprise Linux 5.4;Intel 9.1 compilers' mpich 1.2.7p1].

System under consideration is a 50-atom graphite sheet. For the first step, we used the following parameters in INCAR:

Code: Select all

System = Graphene

ISTART  = 1    !  Read WAVECAR
ISMEAR  = 0    !  Gaussian smearing
SIGMA   = 0.1  !  With sigma 0.1 eV

ENCUT   = 750  ! suggestion from GW step
ENCUTGW = 750
LREAL   = .TRUE.
LOPTICS = .TRUE.
Once this calculation was successfully completed using 4 processors, we copied over POSCAR, POTCAR, KPOINTS, CHG, CHGCAR, WAVECAR and WAVEDER files to a new location and used the following parameters in INCAR:

Code: Select all

System  = Graphene

ISTART  = 1    !  Read WAVECAR (will read the one from single point calcn.)
ISMEAR  = 0    !  Gaussian smearing
SIGMA   = 0.1  !  With sigma 0.1 eV
ISIF    = 2
LREAL   = .TRUE.

ENCUT   = 750
ENCUTGW = 100

NBANDS  = 128   # check from previous step
ALGO    = GW    # Self Consistent GW Calculation
LWAVE   = .TRUE.
LOPTICS = .TRUE.
When running the second step with 4 processors, the OUTCAR file [as well as the calculation] stops with the following output:

Code: Select all

vasp.5.2.2 15Apr09 complex 
 executed on ROCKS522_RHAS54_INTE date 2010.02.10  13:56:07
 running on    4 nodes
 distr:  one band on    1 nodes,    4 groups

...

 k-point  1 :  0.00000.00000.0000  plane waves:  157371

 maximum and minimum number of plane-waves per node :  157371 157371

 maximum number of plane-waves:  157371
 maximal index in each direction: 
   IXMAX= 33   IYMAX= 33   IZMAX= 33
   IXMIN=-33   IYMIN=-33   IZMIN=-33

 WARNING: aliasing errors must be expected set NGX to 134 to avoid them
 WARNING: aliasing errors must be expected set NGY to 134 to avoid them
 WARNING: aliasing errors must be expected set NGZ to 134 to avoid them
 aliasing errors are usually negligible using standard VASP settings
 and one can safely disregard these warnings

 serial   3D FFT for wavefunctions
 parallel 3D FFT for charge:
    minimum data exchange during FFTs selected (reduces bandwidth)


 total amount of memory used by VASP on root node 420481. kBytes
========================================================================

   base      :    30000. kBytes
   nonlr-proj:     4957. kBytes
   fftplans  :    71780. kBytes
   grid      :   232800. kBytes
   one-center:      153. kBytes
   HF        :       12. kBytes
   wavefun   :    80779. kBytes
 
 Broyden mixing: mesh for mixing (old mesh)
   NGX = 67   NGY = 67   NGZ = 67
  (NGX  =200   NGY  =200   NGZ  =200)
  gives a total of 300763 points

 charge density for first step will be calculated from the start-wavefunctions


--------------------------------------------------------------------------------------------------------


 Maximum index for non-local projection operator          871
 Maximum index for augmentation-charges          504 (set IRDMAX)


--------------------------------------------------------------------------------------------------------


 initial charge from wavefunction
    FEWALD:  cpu time    0.02: real time    0.03
 Response functions by sum over occupied states:
 ===============================================
   NOMEGA  =    50    number of frequencies
   NOMEGAR =    50    number of frequencies along real axis
   OMEGAMAX=  14.41   maximum frequency
   OMEGATL =  25.64   maximum frequency of tail
   CSHIFT  =  0.375   complex shift used in evaluation of response functions


 Energies w= 
 ============
   0.0   0.0   0.5   0.0   0.9   0.0   1.4   0.0   1.9   0.0
   2.4   0.0   2.8   0.0   3.3   0.0   3.7   0.0   4.2   0.0
   4.6   0.0   5.0   0.0   5.5   0.0   5.9   0.0   6.3   0.0
   6.7   0.0   7.1   0.0   7.5   0.0   7.9   0.0   8.3   0.0
   8.7   0.0   9.1   0.0   9.5   0.0   9.9   0.0  10.3   0.0
  10.7   0.0  11.1   0.0  11.5   0.0  11.9   0.0  12.3   0.0
  12.7   0.0  13.2   0.0  13.6   0.0  14.1   0.0  14.5   0.0
  15.0   0.0  15.5   0.0  16.1   0.0  16.6   0.0  17.2   0.0
  17.8   0.0  18.4   0.0  19.1   0.0  19.8   0.0  20.6   0.0
  21.5   0.0  22.4   0.0  23.4   0.0  24.4   0.0  25.6   0.0
 

IMPORTANT INFORMATION: All symmetrisations will be switched off!
NOSYMM: (Re-)initialisation of all symmetry stuff for point group C_1.

 
 KPOINTS: 1x1x1            this is just a comment-

Automatic generation of k-mesh.
 
 Subroutine IBZKPT returns following result:
 ===========================================
 
 Found      1 irreducible k-points:
 
 Following reciprocal coordinates:
            Coordinates               Weight
  0.000000  0.000000  0.000000      1.000000
 
 Following cartesian coordinates:
            Coordinates               Weight
  0.000000  0.000000  0.000000      1.000000
 

 total amount of memory used by VASP on root node 420481. kBytes
========================================================================

   base      :    30000. kBytes
   nonlr-proj:     4957. kBytes
   fftplans  :    71780. kBytes
   grid      :   232800. kBytes
   one-center:      153. kBytes
   HF        :       12. kBytes
   wavefun   :    80779. kBytes
 
 Basis sets for responsefunctions:
 =================================
 k-point  1 :  0.00000.00000.0000  plane waves:    7689

 maximum and minimum number of plane-waves per node :    7689   7689

 maximum number of plane-waves:    7689
 maximal index in each direction: 
   IXMAX= 12   IYMAX= 12   IZMAX= 12
   IXMIN=-12   IYMIN=-12   IZMIN=-12

 NGX is ok and might be reduce to  50
 NGY is ok and might be reduce to  50
 NGZ is ok and might be reduce to  50
 exchange correlation table for  LEXCH =        2
   RHO(1)=    0.500       N(1)  =     2000
   RHO(2)=  100.500       N(2)  =     4000
Can some one please help us figure out what we are doing wrong? If we need to provide further information, please do let us know.

Thanks for your time and help,
Gowtham
<span class='smallblacktext'>[ Edited ]</span>

VASP 5.2.2 Parallel Execution - Help Needed

Posted: Wed Feb 17, 2010 4:49 pm
by zoowe
It is not a bug, you should post in other forum.

What do you see in stout and sterr files? There should be something in those output. I think the error could relate to memory.

Also, reduce the NOMEGA to test the calculation. Try to use NOMEGA=N*NPROCESSORS (16,24...).
<span class='smallblacktext'>[ Edited Wed Feb 17 2010, 05:50PM ]</span>