[Pw_forum] negative dr2

Holzwarth, Natalie natalie at wfu.edu
Wed Jul 6 14:58:56 CEST 2011


Dear Dan,

I have seen this error in several cases particularly for transition metals
and
for basis and projector functions generated by our atompaw program.  My
understanding
(based on a discussion with Stefano de Gironcoli) is that the program stops
when the
program detects a certain product of one-center terms to be negative.   In
my experience
these terms can be negative before convergence and I have commented out the
offending
line.   In my tests, this allows the program to converge consistently in the
tests I have done.
This is not necessarily safe or well-tested and I did promise to look into
it further.   In case
you would like to also try, the offending line in 4.2.1 is line 553 in
sct_mod.f90:

  IF (okpaw)         rho_ddot = rho_ddot + paw_ddot(rho1%bec, rho2%bec)

I repeat that commenting out this line may not be a good idea, but it has
worked for me in
limited tests.  (I still hope to do more tests.)


 Sincerely,

      Natalie Holzwarth

N. A. W. Holzwarth                                       email:
natalie at wfu.edu
Department of Physics                                 web:
http://www.wfu.edu/~natalie
Wake Forest University                                 phone:
1-336-758-5510
Winston-Salem, NC 27109 USA                     office: Rm. 300 Olin
Physical Lab

On Wed, Jul 6, 2011 at 5:54 AM, Dan Fors <forsdan at gmail.com> wrote:

> Dear users,
>
> I am running a simple 2-atomic non-magnetic (NM) bcc Cr system (see
> below for the input file)  with v.4.2.1 and I keep getting an error
> message regarding a negative dr2:
>
>     from mix_rho : error #         1
>     negative dr2
>
> This error occurs for values close to as well as away from the
> estimated equilibrium volume. What is the possible the origins of this
> error and how may I eliminate it?
>
> I have tried to change the mixing_beta in the range 0.1 to 0.8, as
> well as changing the mixing mode to 'TF'. I have also tried to set
> nspin = 2 with a zero start magnetisation since an anti-ferromagnetic
> (AFM) start configuration does not give rise to the problem (but
> yields the AFM solution instead of the NM one). However. this also
> gives the error. Finally, I tested to use ibrav = 3 and with only one
> atom in the cell, but I still get the above error.
>
> What I have noticed is that in the AFM case I get a couple of very
> flat bands in the bandstructure just above the Fermi level. If I
> change to a USPP potential or an alternative PAW potential these bands
> disappear, and the above error does not occur. Can the potential I use
> in the present case be the source to the error, and if yes, what is
> the reason behind it?
>
> I would appreciate any help or suggestions.
>
> Dan
>
>
>
> %-------------------------------------
>
> &CONTROL
>        calculation = 'scf',
>        restart_mode = 'from_scratch',
>        prefix = 'Cr',
>        tstress = .true.,
>        tprnfor = .true.,
>        pseudo_dir =
> '/home/forsdan/PROGRAMS/QUANTUM_ESPRESSO/espresso-4.2.1_intel/pseudo/',
>        outdir = '/home/forsdan/FeCr_PROJECT/TEMP_PAW/'
> /
> &SYSTEM
>        ibrav =  1, celldm(1) = 5.378798, nat =  2, ntyp = 1,
>        ecutwfc = 40.0, ecutrho = 180.0, nbnd = 14
>        occupations = 'smearing',
>        smearing = 'mp', degauss = 0.022,
>        nspin = 1,
> /
> &ELECTRONS
>        electron_maxstep = 100,
>        conv_thr =  1.0d-8,
>        diagonalization = 'cg',
>        mixing_mode = 'plain',
>        mixing_beta = 0.3
> /
> ATOMIC_SPECIES
>  Cr  51.996  Cr.pbe-paw_kj_6.UPF
> ATOMIC_POSITIONS {crystal}
>  Cr 0.00 0.00 0.00
>  Cr 0.50 0.50 0.50
> K_POINTS {automatic}
>  12 12 12 0 0 0
>
> %-------------------------------------
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.democritos.it/pipermail/pw_forum/attachments/20110706/dca1903f/attachment.htm 


More information about the Pw_forum mailing list