GGA+U correction of Si bandgap: contradictions with literature's U and J values. why?

[blazing]

Dear all,

Please help me find the reason of contradiction i faced!

By means of GGA+U approach I tried to "enforce" dimond-structure Silicon to demonstrate experimental value for energy gap, i.e. 1.12 eV. Ordinary GGA gives a value around 0.6 eV (i use exc functional by Perdew and Wang, LEXCH = 91 in VASP).

My tests showed, that combination U = 0 eV and J = 4 eV yields experimental value of bandgap (i applied these U and J to p-states of Si). This result is supported by this VASP-maked work:

P. Ramanarayanan et.al., Trans. Mater. Res. Soc. Jpn., 29, 8, 625, 2004.
The work was done using LDA+U (i mean not GGA+U).
One can find this work in Archiv: http://arxiv.org/pdf/cond-mat/0310606v1

On the other hand, there are two more works, which results are in some contrast with mine:
[1] In GGA+U Ge bandgap was tuned to experiment applying U=2eV on p-states. (S. Picozzi et.al., phys. stat. sol. (a) 203, No. 11, 2738– 2745, 2006, http://www3.interscience.wiley.com/jour ... 2/abstract )
[2] In LDA+U Si bandgap was fitted by effecting on p-states by U=2.18 eV (G Cubiotti et al 1999 J. Phys.: Condens. Matter 11 2265-2278, http://www.iop.org/EJ/article/0953-8984 ... 078947ac07)
Interesting to note, in both works nothing was said about J value.

I tried to apply U=2 eV and J=0 or 1 eV, but Si gap was even smaller than in ordinary GGA. But according to [1] and [2] this combinations of U and J are proper one. So, this contradictions are confusing me a lot.
In works [1] and [2] the FLAPW and LMTO methods respectively were used. And according the paper's texts LDA+U approach in both used code was implemented according Anisimov et.al. J. Phys.: Condens. Matter. 9, 767,1997.

At the same time i tried both LDATYPE = 1 and LDATYPE = 2 which are accordig to Lichtenstein (which is, as far as i;m understand, the same as Anisimov et.al. J. Phys.: Condens. Matter. 9, 767, 1997) and Dudarev respectively (references are given in vasp manual). The results were the same: U=0 & J=4 eV -> experimental gap, U=2 & J=0 (or 1) eV -> gap is around 0.55 eV.

I dont know where the differences in results arise from?
So i'm not sure that my calculations is right because of this contradiction.

I use following INCAR

SYSTEM = Si

ISTART = 0
ICHARG = 2

PREC = High
IALGO = 38
VOSKOWN = 1
ISMEAR = -5
SIGMA = 0.2
LORBIT = 11

NEDOS = 4500

ENCUT = 500.00 eV
RWIGS = 1.3

ISPIN = 2
ISYM = 0
ISIF = 7
NELMIN = 7
NSW = 0
IBRION = -1

EDIFF = 1E-04

LDAU = .TRUE.
LDATYPE = 2 (or 1)
LDAUL = 1
LDAUU = 0
LDAUJ = 4
LDAUPRINT = 2


POSCAR:

Si bulk
5.466
0.00 0.50 0.50
0.50 0.00 0.50
0.50 0.50 0.00
2
Direct
0.00 0.00 0.00
0.25 0.25 0.25

And G-centered k-points grid 12x12x12

Am I calculating right or wrong?
Does somebody have any suggestions why this contradictions are occure?

Thanks in advance

Mikhail

<span class='smallblacktext'>[ Edited ]</span>

[Danny]

your results are right. Positive U values decrease the BG for semiconductors, negative increase the BG. Problem is that some ppl might consider this "unphysical" for some reason.

[blazing]

Thank you for your answer!

Did you mean that namely in pseudopotential approach positive U values decrease the BG for semiconductors? It seems like when using the same U such a different results occurs because of different approaches are used, aren't they?