Possible bug in Na_sv PAW-PBE pseudopotential

[jat255]

Hello,

I am attempting to converge parameters for the Na_sv pseudopotential for use in high pressure alloy simulations.

I am doing this using the experimentally determined bcc structure, which is stable at ambient conditions. I am using the following INCAR:

Code: Select all

ENCUT=200
EDIFF=1E-6
ISIF = 3
IBRION = 2
NSW = 21
ISPIN = 1

ISMEAR = 1
SIGMA = 0.1

PREC = Accurate
ALGO = NORMAL
#IALGO = 58

LPLANE = .TRUE.
PSTRESS = 0

NPAR = 4

LWAVE = .FALSE.
LCHARGE = .FALSE.

Without exception, the run fails just after the first ionic step, returning the error "Error EDDDAV: Call to ZHEGV failed." Below is a sample of the output:

Code: Select all

 vasp.5.2.11 01Dec10 complex 
 POSCAR found type information on POSCAR  Na
 POSCAR found :  1 types and       1 ions
 LDA part: xc-table for Pade appr. of Perdew
 generate k-points for:   10   10   10
 POSCAR, INCAR and KPOINTS ok, starting setup
 FFT: planning ...(           1 )
 WAVECAR not read
 entering main loop
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1     0.979439119157E+02    0.97944E+02   -0.18066E+03   776   0.441E+02
DAV:   2     0.726616855489E+02   -0.25282E+02   -0.24073E+02  1184   0.875E+01
DAV:   3     0.720825152839E+02   -0.57917E+00   -0.57879E+00   768   0.202E+01
DAV:   4     0.720748686810E+02   -0.76466E-02   -0.76466E-02  1136   0.149E+00
DAV:   5     0.720748486767E+02   -0.20004E-04   -0.20004E-04   816   0.555E-02    0.576E+01
DAV:   6     0.131837706740E+03    0.59763E+02   -0.29583E+02   752   0.998E+01    0.235E+01
DAV:   7     0.135940115374E+03    0.41024E+01   -0.60289E+00   792   0.261E+01    0.109E+01
DAV:   8     0.135902154588E+03   -0.37961E-01   -0.12299E+00   800   0.121E+01    0.130E+00
DAV:   9     0.135807686767E+03   -0.94468E-01   -0.85748E-02   800   0.242E+00    0.286E-01
DAV:  10     0.135809341779E+03    0.16550E-02   -0.80973E-03   968   0.693E-01    0.261E-02
DAV:  11     0.135809240275E+03   -0.10150E-03   -0.10230E-04   880   0.853E-02    0.156E-02
DAV:  12     0.135809246248E+03    0.59725E-05   -0.23990E-05   888   0.411E-02    0.290E-03
DAV:  13     0.135809247933E+03    0.16854E-05   -0.75573E-07   552   0.685E-03    0.114E-03
DAV:  14     0.135809248756E+03    0.82234E-06   -0.50781E-07   408   0.545E-03
   1 F= 0.13580925E+03 E0= 0.13580964E+03  d E =0.135809E+03
 curvature:   0.00 expect dE= 0.000E+00 dE for cont linesearch  0.000E+00
 trial: gam= 0.00000 g(F)=  0.000E+00 g(S)=  0.236E+05 ort = 0.000E+00 (trialstep = 0.100E+01)
 search vector abs. value=  0.236E+05
 bond charge predicted
       N       E                     dE             d eps       ncg     rms          rms(c)
Error EDDDAV: Call to ZHEGV failed. Returncode =   7 1   8

As you can see, the energies are totally unreasonable for an experimentally stable structure, and they blow up as the electronic steps continue.

I have tried a number of suggestions that I found on the forum, changing IBRION, AMIX, IALGO, and POTIM, all to no avail. I have also varied the ENCUT and PSTRESS with no changes.

Suspecting that the problem was the psuedopotential, I attempted using the Na and Na_pv PAW-PBE PPs in order to see if this changed anything. When I did this (using the same INCAR, KPOINTS, and POSCAR as before), both runs converged easily.

Na PAW-PBE: TOTEN = -1.301915 eV in 4 ionic steps
Na_pv PAW-PBE: TOTEN = -1.689522 eV in 5 ionic steps

Obviously, I would like to use the Na_sv PP because I am going to be looking at very high pressures and I want to include as many core electrons as possible in the calculations.

I am wondering, has anyone else seen this error? And if so, is there anyway to correct it? I should note that the PPs that I am using are dated as follows:
PAW_PBE Na_sv 28Sep2000
PAW_PBE Na_pv 05Jan2001
PAW_PBE Na 08Apr2002
My PI is away at a conference, and thus I don't have access to the ftp server to check if these are the most recent potentials, so I figured I would post here.

Any help would be very much appreciated!

- Josh Taillon

[forsdan]

Please note that the sv-potential is much harder than Na_pv and Na, which requires a higher minimum energy cutoff. Try to run it with say ENCUT = 520 eV and upwards.

Cheers,
/Dan

[jat255]

Well don't I certainly feel foolish. I attempted a go at it with an ENCUT of 500 (which is what I meant by "varied" in my original post), which didn't work, but trying again at 700 gave me a convergence within 4 ionic steps.

Sometimes it just takes an outside perspective, it seems!

Thanks for the help!

- Josh