[Wien] mixer erro (segmentation fault and ispin)

[Fred Nastos]

Zhang, please try using Intel's Fortran compiler version 8.1.  To corroborate 
what others on the mailing list are saying, I find it to be superior to 8.0 
in many respects.

Fred

On October 8, 2004 08:54 am, Jack London wrote:
> Dear all
>
> I install Wien2k code (WIEN2k_04.6 (Release
> 21/06/2004)) not the latest version in supercomputer
> (LInux cluster: redhat 9.0 and ifc8.0 ). Because I
> have to compare the results on the new supercomputer
> with old results on other machines. This version was
> successfully installed.
>
> But when I run a small job (bulk Pd), "x mixer"
> crashed  in the first iteration with the following
> message.
>
> forrtl: severe (174): SIGSEGV, segmentation fault
> occurred
> Image              PC        Routine            Line
>      Source
> Unknown            4013256E  Unknown
> Unknown  Unknown
> Unknown            4013139C  Unknown
> Unknown  Unknown
> mixer              0808A59D  Unknown
> Unknown  Unknown
> mixer              080720CB  Unknown
> Unknown  Unknown
> mixer              08071B92  Unknown
> Unknown  Unknown
> mixer              0804AA84  Unknown
> Unknown  Unknown
>
> Stack trace terminated abnormally.
>
> And I also can find "  mixer       (12:34:05)  ispin
> set to           1" in case.dayfile.
>
> I did the following methods to fix the problem.
>
> 1). I downloaded the mixer package of the latest
> Wien2k version and compiled it. Then calcualted the
> bulk Pd. It finished the SCF in the 15th cycles, and
>
> :DIS and :ENE converged at 0.0000507 and -10094.096399
>
> Ry, respectively. The total energy in new machine is
> about 0.6 meV difference with our old results in old
> machine(same Wien2k version). And for the slab test,
> the difference of total energy is becoming larger and
> larger with increase layers in slab calculation. (all
> parameters,case.in1, case.in2 and case.klist are the
> same)
>
> 2). I checked the Wien2k mail list and found some
> disscussions about this problem. Based on suggestion
> of Fred Nastos, I found the line
> "FORCE(JATOM)=.false." in scfana.f and cancled it.
> After recompiled MIXER, the porblem was really fixed
> in bulk Pd calculation. But the SCF ONLY run 13 cycles
> and :DIS was converged to 0.0000475. It is very
> different with the 15 cycles in 1). And The total
> energy -10094.096364 Ry is also different with 1).
>
> So, I also did the test in my old machine. We got the
> same :DIS and :ENE (total energy is -10094.096446 Ry)
> with and without that line in scfana.f of the old
> Wienk version. After installed the Mixer package of
> the latest version, :DIS is converged at the 18th
> iteration and total energy is -10094.096456 Ry.
>
> In order to compare each other clearly, I attach a
> file in doc for Windows and other file for linux.
>
> 3), based on L.D. Marks suggestion, "use -traceback in
> your compile lines options", but it didn't work.
>
> forrtl: severe (174): SIGSEGV, segmentation fault
> occurred
> Image              PC        Routine            Line
>      Source
> Unknown            4013256E  Unknown
> Unknown  Unknown
> Unknown            4013139C  Unknown
> Unknown  Unknown
> mixer              08089D8D  Unknown
> Unknown  Unknown
> mixer              080718BB  Unknown
> Unknown  Unknown
> mixer              08071382  Unknown
> Unknown  Unknown
> mixer              0804AA2F  errclr_
>  63  errclr.f
> mixer              08054F69  Unknown
> Unknown  Unknown
> mixer              08049E68  Unknown
> Unknown  Unknown
> Unknown            400D3A67  Unknown
> Unknown  Unknown
> mixer              08049D21  Unknown
> Unknown  Unknown
>
> How to fix the beginning error in MIXER? and Why I get
> the different total on two kinds of linux cluster?
>
> Thanks in advance!
>
> Best regards
>
> Zhang