Installation of vasp 4.6.36 on Scinet (IBM AIX)

[imcnab]

Hi,

I have unpacked and compiled vasp 4.6.36 on Scinet, TCS, OS is AIX 5.3, which has available the fortran compiler IBM XL fortran for AIX V12.1, running on PowerPC hardware.

When I ran the first test job "Benchmark",
I got the warning IALGO=8 in no longer supported, which is a bummer, because the example output was generated with IALGO=8.

I changed the INCAR to IALGO=38 (and 48) and tried again.

The job appears to work OK for the first 4 cycles of DAV:
thereafter it reports
"BRMIX: very serious problems the old and the new charge density differ."

The job appears to complete, after generating this BRMIX warning on every line, but the OSZICAR produced looks nothing like the one in the tarball.

Is there an updated benchmark test available for 4.6 ?

All help would be most gratefully received.

Thanks for your time... Iain

[admin]

please read the warning that is written in OUTCAR:
IALGO=8 must not be set for copyright reasons, please use IALGO=38 instead

[imcnab]

If you read my initial request for help, you will see that I did read the OUTCAR, I did change the IALGO value in INCAR to 38 (and also 48), and my query is about what happened after.

I also queried whether a new benchmark example output file is available that has been run with IALGO=38 or 48.

It is pointless to give a benchmark that no longer works. Is is by no means obvious that output using different IALGO parameters can be usefully compared to verify an installation.

all help will be gratefully received.
Iain

[forsdan]

The BRMIX problem often indicates that your optimization level during the compilation is too high. So as a first test I would recommend to decrease the optimization and try again. There is also additional discussion in previous threads about possible causes (just search the forum) if it doesn't help.

I also disagree with you about of the benchmark being pointless just because it was performed with IALGO = 8. The different algorithms will converge to the same final energy within the given accuracy provided that they indeed converge (see the manual for considerations for each algorithm). So it still serves as a reference to check if your calculation behaves reasonable.

Cheers,
/Dan

<span class='smallblacktext'>[ Edited Fri Jul 30 2010, 12:37AM ]</span>

[imcnab]

Dear Dan,
thank you very much for the suggestion. I will cut down on the level of optimization and try again, (although the optimization levels used were those given in the makefile for this OS/compiler).

As far as the benchmark being pointless, I think we have a philosophical disagreement. In my view the purpose of a benchmark is not to produce similar convergence using a different algorithm, but to show exactly the same output for input to verify that the installation is successful and known conditions are successfully reproduced. The point being that a complete novice has no way of knowing whether similar output produced with a different algorithm is good enough.

Coming from a background of installing Gaussian (I know, I know, in line for "evil company of the decade" award), where hundreds of test jobs are given, seeing ONE test job, which failed, didn't give me a lot of comfort.

Perhaps the originators of VASP are so annoyed by the patent award which prevents the use of IALGO=8 that they wish all new installers of VASP to see the message, but in my opinion this does NOT a benchmark make.

Thank you very much for taking the time and trouble to help me. I do appreciate it.

best regards... Iain

[imcnab]

Thanks Dan,

compiling with optimization of 2 (not 4 as set in given makefile) appeared to result in a working program. Of course, I am still unsure if the results I get are as expected, because my OSZICAR is significantly different from the OSZICAR.ref provided in the tarball. The final lines are as follows:

MY OSZICAR
5 T= 1918. E= -.90208569E+04 F= -.90225947E+04
E0=-.90218341E+04 EK= 0.17351E+01 SP= 0.26E-02 SK= 0.93E-04

OSZICAR.ref
5 T= 1683.6 E= -.90209039E+04 F= -.90226420E+04
E0= -.90218814E+04 EK= .17410E+01 SP= -.32E-02 SK= .32E-03

Do you consider that my version of vasp has passed the test?
All help would be very gratefully received.

best regards... Iain

[imcnab]

Guys, I could really use some help here.
Should I be worried about the fact that the temperature, EK and SK values are different in my OSZICAR, and in the OSZICAR given in the benchmark test case?
All help would be very gratefully received.

[imcnab]

The OSZICAR output generated with IALGO=38 output is given in the file
R. LORENZ, “COMPUTATIONAL PLATFORMS FOR VASP�
which is available on
http://cms.mpi.univie.ac.at/marsweb/
under "Documentation",
"Computational platforms"
The output from "MY OSZICAR" above is the same as that given
in "Computational platforms" although differing significantly from that which is bundled with VASP4.6 in the tarball under OSZICAR.ref.