[Pw_forum] discrepancy in molecular binding energy between pwscf and localized basis set program

[Garold Murdachaew]

Dear pwscf users,

I am doing some calculations for relatively weakly interacting species and 
decided to compare to what I obtained with localized wavefunction based 
programs such as molpro.  For example, I did calculations of the binding 
energy of Xenon dimer (at the minimum separation of 4.42 Angstroms).
I obtained (in meV):

 	molpro	pwscf3.2
LDA	-34	-34
PBE	 -4	-11

As can be seen, the LDA calculations agree perfectly but the PBE pwscf 
calculation gives an interaction energy almost three times more negative. 
(I realize that for most pwscf users who are usually only concerned about 
energies on the eV rather than on the meV scale these differences may seem 
insignificant.)

Here is my PBE pwscf input file for the Xe dimer (the monomer and the LDA 
inputs are similar):

&CONTROL
   calculation   = 'scf',
   prefix        = '$HEAD',
   title         = '$HEAD',
   restart_mode  = 'from_scratch',
   outdir        = '$TMPDIR/',
   PSEUDO_DIR    = '$PSEUDODIR/',
/
&SYSTEM
     ibrav     = 0,
     celldm(1) = 25.d0,
     nat       = 2,
     ntyp      = 1,
     ecutwfc   = 30.0d0,
     ecutrho   = 300.0d0,
     occupations = 'smearing',
     degauss     = 0.001D0,
     smearing    = 'mp',
     nbnd        = 12,
/
&ELECTRONS
     conv_thr    = 1.0d-8,
     mixing_mode = 'plain',
     mixing_beta = 0.3d0,
/
ATOMIC_SPECIES
Xe 131.30   Xet2tPBE.RRKJ3
ATOMIC_POSITIONS { angstrom }
Xe     .000000000    .000000000    .000000000  0  0  0
Xe     .000000000    .000000000   4.420000000  0  0  0
K_POINTS { gamma }
CELL_PARAMETERS { cubic }
    1.00000    0.00000    0.00000
    0.00000    1.00000    0.00000
    0.00000    0.00000    1.00000

As can be seen, the molecule is in a cubic box of edge length 25 bohr and 
thus there is sufficient vacuum to minimize the spurious interactions with 
its periodic images.  I played with the convergence setting conv_thr but 
setting it to a factor of 100 tighter had very small effect.  Changing the 
degauss setting had no effect on energy (only Fermi level was changed.) 
However, when I increased  ecutwfc and ecutrho to 70 and 700 respectively, 
I obtained a slight change (-10 meV rather than -11 meV).  Can somebody 
please tell me how I can reproduce my PBE  localized basis sets 
calculation.  Do I need to make some changes to the pseudos?  Or is this 
what I can expect from gradient-corrected calculations (such as PBE) due 
to some sort of errors in estimating the gradients (which are not needed 
in the LDA calculation)?

Thank you for you assistance.

Best regards,
Garold Murdachaew

ps:  Here are the details of the pseudo I used in pwscf:

LDA pseudo: XeLDA.RRKJ3.UPF: nonrel, nc, nlcc=t, valence=8, maxl=2
PBE pseudo: Xet2tPBE.RRKJ3 : nonrel, nc, nlcc=t, valence=8, maxl=2

--
Garold Murdachaew
Postdoctoral Fellow - Condensed Matter Sector
SISSA-ISAS (International School for Advanced Studies)
New Building Room 208
Via Beirut 2-4, 34014 Trieste, Italy

tel   : +39 040 3787 477
fax   : +39 040 3787 528
email : murda at sissa.it