[Wien] Summary NFS bug

[Javad hashemifar]

> If you observe such large :DIS we need to find out why it happens. I
> cannot imagine that save/restore of vsp helps.

Although I am not as expert as you, but let me to explain some cases
in which after rerunning a converged results, large :DIS occurs.
According to my experience this problem occur in spin polarized cases and its
due to different linearization parameters for up and dn spins. by using in1new
switch the program (after specified iteration) linearize the case.in1
files in each cycle
for up and down spins separately as follows:
1. lapw0
2. write_in1_lapw -up
3. cp case.in1new case.in1
4. lapw1 -up
5. write_in1_lapw -dn
6. cp case.in1new case.in1
7. lapw1 -dn

therefor after achieving convergency, case.in1 files contains only the
linearization
parameters of down spin electrons and hence when you rerun a converged
case by another
runsp_lapw the :DIS jumps ( because at the first iteration lapw1 -up
uses the linearization parameter of down spins). To prevent this jump
the linearization
process after rerunning a case that is converged by using in1new
switch should be start at the first iteration (-in1new 1).
Since the automatic linearization needs to scf2 files, hence in order
to start linearization from the first iteration, the scf2up and scf2dn
files should exist. Therefore I suggest to add these two scf files as
well as in1 file to the list of files that is saved by save_lapw.

Best regards
Javad Hashemifar

======================================
Seyed Javad  Hashemifar,    Ph.D.
[current:]
Tel:+49-203-3794743        Fax:+49-203-3794742
Fachbereich Physik, Universitat Duisburg-Essen
47048 Dusiburg, Germany
[permanent:]
Tel:  +98-311-3912375       Fax: +98-311-3912376
Physics Department, Isfahan University of Technology
84156 Isfahan, Iran
---------------------------------------------------------------------------