[Wien] the error information of lapw1

[Laurence Marks]

This may be simple, or it may be rather tricky.

First, what is going on. The Cholesky operation is factorizing the
Hamiltonian and what error means is that your Hamiltonian is not
positive definite. "Cholesky INFO = 5988" means is that the 5988 row
is linearly dependent on some other rows, i.e. you could represent it
as a linear sum of some of the other terms.

Often this is caused by an error in your input. Do you have duplicate
atoms? Do you have duplicate terms in case.in1, e.g. 3 or more
linearization terms with the same L? Did you run through the
initialization doing all the steps and paying attention to any
errors/warnings?

If it is not this, which it sometimes can be, you have more work to do.

First, you should try and identify what term it is. If you look in
case.output1_X (for the particular process/k-point of interest) you
will find the total number of plane waves and the number of
APW/LAPW/LO terms. The Hamiltonian has first the plane waves then the
orbital terms for each atom (not just the unique atoms). Almost
certainly it is one of the APW/LAPW/LO terms. Count forward taking
care to account for the multiplicity (1 for L=0, 3 for L=1 etc) and
determine which term is causing the problem. Then, look to see if you
have a legitimate value for the linearization energy. Check in the
relevant case.output1_X for something odd for the specific term, e.g.
U(R)~0, two LO coefficients with C~0 for both, e.g.

LO COEFFICIENT: l,A,B,C   0          0.99965     0.08441     0.00000
LO COEFFICIENT: l,A,B,C   0          1.08884     0.00000    -0.00001

This later case is trapped in 07.3 but not in 7.2 -- the cure is a) below.

If you don't see anything obvious:
a) Change, by hand, the linearization energy of the specific term,
shifting it up or dn by 0.05 or use -in1new (if :DIS is small enough).
b) Change a little RKMAX (e.g. 7 to 7.1) or your RMT's.
c) Change your k-mesh a little, i.e. slightly more/less values.

These are work arounds for "something" which sometimes causes the
Hamiltonian to be not fully positive definite which (as far as I know)
is not currently fully understood.

On 9/13/07, hui li <huilicn at yahoo.com> wrote:
> Dear Sir:
>    I used WIEN2k_07.2 on a PC (AMD 64 Athlon CPU) with
> Redhat Fedora 6.0 operation system.In this PC, there
> had 2GB SDRAm and 4GB swap.The operation system only
> had Fortran77 (f77) compiler. So, I used exectuable
> code directly. My research is to do SCF calculation
> and accquire charge density. But during the
> calculation, the program stopped in the step of lapw1
> with the information as follows:
>      Cholesky INFO =         5988
>      'SECLR4' - POTRF (Scalapack/LAPACK) failed.
> My model just need less then 1.8G memory, and the swap
> (4GB) was free during the calculation. Thank you very
> much.
>                     Best wishes
>                                      Dr. Hui Li
>                      Beijing University of Technology
>
>
>       ___________________________________________________________
> 雅虎邮箱，以安全著称，是值得信赖的邮箱专家！
> http://mail.yahoo.com.cn/
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>


-- 
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
Web: www.numis.northwestern.edu
EMM2007 http://ns.crys.ras.ru/EMMM07/
Commission on Electron Diffraction of IUCR
www.numis.northwestern.edu/IUCR_CED