[Wien] hyper fine field on different platforms

[Björn Hülsen]

Dear Lyudmila and Gerhard,

thank you for your response. I think I have found the source for the 
discrepancy. The input for both calculations is exactly the same (GGA 
PBE96, same K_max, k-point set, G_max...) and the commands I used for 
the scf cycle are:

runsp_lapw -cc 0.01 -ec 0.01 -in1orig
rm *broy*
runsp_lapw -cc 0.0001 -ec 0.00005 -in1new 1

Both calculations converged very well.
But I noticed a difference in the new .in1 files, please see below:

Pentium4
WFFIL        (WFPRI, SUPWF)
  8.00       10    4 (R-MT*K-MAX; MAX L IN WF, V-NMT
 .49572   5   0      global e-param with N other choices, napw
 0    0.429     0.000 CONT 1
 0   -5.921     0.002 CONT 1
 1    0.566     0.000 CONT 1
 1   -3.423     0.002 CONT 1
 2    0.645     0.000 CONT 1
 .49572   5   0      global e-param with N other choices, napw
 0    0.458     0.000 CONT 1
 0   -4.809     0.002 CONT 1
 1    0.574     0.000 CONT 1
 1   -2.589     0.000 CONT 1
 2    0.621     0.000 CONT 1
 .49572   5   0      global e-param with N other choices, napw
 0    0.413     0.000 CONT 1
 0    0.088     0.000 CONT 1
 1    0.473     0.000 CONT 1
 1   -5.852     0.002 CONT 1
 2    0.636     0.000 CONT 1
K-VECTORS FROM UNIT:4  -11.0       2.0      emin/emax window

Power5
WFFIL        (WFPRI, SUPWF)
  8.00       10    4 (R-MT*K-MAX; MAX L IN WF, V-NMT
 0.49677   5   0      global e-param with N other choices, napw
 0    0.429     0.000 CONT 1
 0   -5.902     0.000 CONT 1
 1    0.566     0.000 CONT 1
 1   -3.421     0.000 CONT 1
 2    0.646     0.000 CONT 1
 0.49677   5   0      global e-param with N other choices, napw
 0    0.458     0.000 CONT 1
 0   -4.807     0.000 CONT 1
 1    0.574     0.000 CONT 1
 1   -2.588     0.000 CONT 1
 2    0.621     0.000 CONT 1
 0.49677   5   0      global e-param with N other choices, napw
 0    0.412     0.000 CONT 1
 0    0.088     0.000 CONT 1
 1    0.473     0.000 CONT 1
 1   -5.843     0.000 CONT 1
 2    0.636     0.000 CONT 1
K-VECTORS FROM UNIT:4  -11.0       2.0      emin/emax window

The numbers for the energies agree but in the Power5 file there is no 
search for the low lying local orbitals. I checked (with diff command) 
that both scripts are exactly the same. I copied the in1 file from the 
Pentium 4 to the Power5 and did another calculation and now the 
agreement for the HFF between the two platforms is much better.

Pentium4
:HFF001: HFF: 230.220           0.000        -320.734         -90.514 
(KGAUSS)
:HFF002: HFF: 446.428           0.000        -655.243        -208.816 
(KGAUSS)
:HFF003: HFF:  26.940           0.000           1.443          28.383 
(KGAUSS)
:ENE  : ********** TOTAL ENERGY IN Ry =        -8471.424448

Power5
:HFF001: HFF: 214.893           0.000        -320.609        -105.716 
(KGAUSS)
:HFF002: HFF: 454.304           0.000        -658.877        -204.572 
(KGAUSS)
:HFF003: HFF:  26.514           0.000           1.382          27.895 
(KGAUSS)
:ENE  : ********** TOTAL ENERGY IN Ry =        -8471.423047

So why is the write_in1 script executed differently on the Power5? Maybe 
a UNIX or AIX expert can comment.

Best regards,
Bjoern

Björn Hülsen
Fritz-Haber-Institut
Faradayweg 4-6
14195 Berlin

phone:  +49 30 8413-4863
email:  huelsen at fhi-berlin.mpg.de
www:    www.fhi-berlin.mpg.de