[CPMD-list] Platform-dependent wave function optimization results

[Vladimir Stegailov]

Dear colleagues,

is it normal to get essentially different WFO processes on different platforms using the same input script? 

I use the 3.11 version and compare two platforms:
1. cpmd compiled with IFC v. 8.0 and libatlas_p4.a 
2. cpmd compiled with IFC v. 9.0 and MKL 8.0.1

The input script is given below.

In the 1st case the WFO process initially gives several "FRIESNER_C| EIGENVECTOR 4  IS VERY BAD!" warnings, but eventually goes well and stops at NFI=179.

In the 2nd case there are no warnings and WFO stops much quicker at NFI=44. 

Is the reason just the difference in FFT libraries? Could it be specific to the FEMD simulations?

I would appreciate any comments!

Vladimir


&CPMD

    FILEPATH

      /home/stegailov/testing/cpmd/md3/

    OPTIMIZE WAVEFUNCTION

    UNIT HESSIAN

    BFGS

    FREE ENERGY FUNCTIONAL

    LANCZOS DIAGONALISATION 

    LANCZOS PARAMETERS

      1   6 10   1.D-18

    TROTTER FACTOR

      0.001

    BOGOLIUBOV CORRECTION OFF 

    GRAM-SCHMIDT ORTHOGONALISATION

    CONVERGENCE

      1.D-6  5.D-6

    MAXSTEP

      5000

    BROYDEN MIXING

      0.15 200   0.01  0   8

    ALEXANDER MIXING

      1.1 

    TEMPERATURE

      400.

    ELECTRON TEMPERATURE

      10000.

    COMPRESS WRITE32

    STRUCTURE BONDS

    ENERGYBANDS

    ELECTROSTATIC POTENTIAL

    RHOOUT

 &END

 

 &SYSTEM

   POINT GROUP

    AUTO

   SYMMETRY

    1

   CELL

    8.064    1.0    1.0   0.0 0.0 0.0   (8.064=2*4.032, 4.032A is the eq lattice const of Al)

   CUTOFF

    15.000

   ANGSTROMS

   STATES

    250

   SCALE

   TESR

     1

   KPOINTS MONKHORST-PACK FULL

   2  2  2   

 &END

 

 &ATOMS

*AL_SGS KLEINMAN-BYLANDER

   LMAX=D

   32

0.0178998 -0.0197471 0.0145112

...

0.500011 0.750491 0.756745

 

 

VELOCITIES

32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

-5.10982 0.101812 -1.43574

...

7.04835 2.96887 2.84766

END VELOCITIES

 

  &END

 

 &BASIS

     PSEUDO AO 2

      0  1   

 &END

 

 &DFT

   NEWCODE

 &END

 

 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cpmd.org/pipermail/cpmd-list/attachments/20070426/103303a3/attachment.html