[Wien] [SPAM?] Re: Counting the numer of electrons / slightly metallic tail crossing EF

Peter Blaha pblaha at theochem.tuwien.ac.at
Fri Feb 18 23:17:56 CET 2011


> When I check TOTAL CORE-CHARGE of Gd after lstart (and scf file),
>
>   TOTAL CORE-CHARGE:                   46.000000
>   TOTAL CORE-CHARGE INSIDE SPHERE:     45.999929
>   TOTAL CORE-CHARGE OUTSIDE SPHERE:     0.000071
>
> MAGNETIC MOMENT IN SPHERE seems quite reasonable,
> but I would like to make sure whether this order of charge leakage
> can be negligible in total energy calculations (with different spin
> configurations).

This is perfectly ok and you don't have to worry.

> But when I run x tetra to draw the DOS plot, and check outputtdn/up files,
> the up+dn sum of the NUMBER OF ELECTRONS UP TO EF
> is about 61 % of the number from :NOE.
>
> Even though I exclude the 5S and 5P electrons (10 e-) from counting in :NOE,
> the number from NUMBER OF ELECTRONS UP TO EF is still smaller.
> Is it normal to have much smaller (like 2/3) values in outputtdn/up
> than that of :NOE? Or those two numbers are of different meaning?

Probably your Emin (-0.5 )in case.int is not ok and you are truncating 
some DOS (F 2s ?)

> 3.
> When I plot the DOS, with settings in int file as
>   -0.500   0.00001   0.8  0.0005     #Emin, DE, Emax, Gauss-Broad

With such DE you are "asking" for numerical inaccuracies.
Stick to the defaults first or modify them only slighly.

For sure "Gauss-broadening" introduces some small tails above EF which 
are of course artifacts due to the broadening.

> this results in slight VBM tail crossing the EF.
>
> I expect from the simple electron countings,
> this system should shows the gap between VBM and CBM,
> suspecting this is solely due to the Gaussian broadening.
> (Even though I use the GGA+U scheme, this tail is still there.)
>
> So, I checked outputtup/dn file, and I find
> I have non-integer occupation number in up spin, like 195.9998, at EF
> and it sustains the decimal value upto the conduction band minimum.
> The down spin shows exact interger number of electrons up to EF.
>
> I would like to know whether the 0.0002 e- deficiency
> is due to the numerical error (thus can be neglected),
> or I should check with the band structure to determine the metallicity
> of the system.
>
> When the number from the outputtup/dn file suggests sort of band gap,
> then could I consider this system as a semi-conductor regardless of the
> metallic states tail in dos1up/dn files?
>
> And, if 0.0002 e- is due to the numerics,
> is there any way to remove this annoying number?


-- 
Peter Blaha
Inst.Materials Chemistry
TU Vienna
Getreidemarkt 9
A-1060 Vienna
Austria
+43-1-5880115671


More information about the Wien mailing list