Problems with vdW-DFT

Message

jmorale4 · #1 Post by **jmorale4** » Wed Feb 06, 2013 3:58 am

I have been using the vdW-DFT corrections implemented by J. Klimes, D.R. Bowler, A. Michelides. I have the precompiled kernel

Code: Select all

vdw_kernel.bindat

in the same directory.

Unfortunately I am getting the error:

Code: Select all

?LDA?part:?xc-table?for?Pade?appr.?of?Perdew
?POSCAR,?INCAR?and?KPOINTS?ok,?starting?setup
?WARNING:?small?aliasing?(wrap?around)?errors?must?be?expected
?FFT:?planning?...
?WAVECAR?not?read
?WARNING:?random?wavefunctions?but?no?delay?for?mixing,?default?for?NELMDL
?entering?main?loop
???????N???????E?????????????????????dE?????????????d?eps???????ncg?????rms??????????rms(c)
?DESATURATE:?out?of?range?-5.551115123125783E-017

TACC:?Shutdown?complete.?Exiting.

Does anyone have any idea what the possible source for this error is? Thank you for reading this post beforehand.

jmorale4 · #2 Post by **jmorale4** » Wed Feb 06, 2013 4:01 am

The error section became corrupted, so I include a new version:

LDA part: xc-table for Pade appr. of Perdew
POSCAR, INCAR and KPOINTS ok, starting setup
WARNING: small aliasing (wrap around) errors must be expected
FFT: planning ...
WAVECAR not read
WARNING: random wavefunctions but no delay for mixing, default for NELMDL
entering main loop
N E dE d eps ncg rms rms(c)
DESATURATE: out of range -5.551115123125783E-017

TACC: Shutdown complete. Exiting.

kelum · #3 Post by **kelum** » Wed Feb 06, 2013 11:07 am

Hi,
this is a bit odd but clear, somehow b*exp(0) - b is not calculated as 0 but as -epsilon. I haven't come across this before and I can't reproduce this using either no or high optimisation setting for ifort, is there any special compiler/compiler settings you use? Anyway, to fix it in the code the loop on the line 268 of vdw_nl.F needs to be changed from (subroutine GEN_ALPHA_POINTS)
DO i=1,Nalp-1
to
DO i=2,Nalp-1
That should do it, the first value of the array should be zero so nothing changes with this.
Best,
jik

jmorale4 · #4 Post by **jmorale4** » Thu Feb 07, 2013 10:35 pm

Thank you Jiri. This solution did manage to eliminate the DESATURATE error, but we are having errors of the type:

Error EDDDAV: Call to ZHEGV failed. Returncode = 114 2 128

Which are addressed here.

http://cms.mpi.univie.ac.at/vasp-forum/ ... .php?3.214

We have tried all of the solutions except for the last one (disable the ZHEGV subroutine). Do you think disabling ZHEGV will compromise the calculations used when employing the vdw corrections?

jmorale4 · #5 Post by **jmorale4** » Thu Feb 07, 2013 10:36 pm

Sorry, the last sentence should read:

We have tried all of the solutions except for the last one (disable the ZHEGV subroutine). Do you think disabling ZHEGV will compromise the calculations done with the vdw corrections?

kelum · #6 Post by **kelum** » Fri Feb 08, 2013 1:14 pm

OK, I must say I never got up to the point 4 in the ZHEGV fails (usually I get a crash because of a bad structure). Does the code work correctly if you run a PBE? Do you get this error for every system? If both are yes then I'd say there is probably something going wrong in the vdW part and changing from ZHEGV to the other routine will probably not help. Or if it does the results might not be all right. Then I'd compare to results from a computer where it works (I can send you something if you need).
jik

jmorale4 · #7 Post by **jmorale4** » Fri Feb 08, 2013 11:03 pm

I see. Well, I am trying to use optB86b so that in my INCAR file I have the lines:

LUSE_VDW = .TRUE. # Correlation functionals
AGGAC = 0.0000 #
GGA = MK # Exchange functionals
PARAM1 = 0.1234 #
PARAM2 = 1.0000 #

I would appreciate some of those results where vdw actually works! Thank you!

kelum · #8 Post by **kelum** » Mon Feb 11, 2013 4:14 pm

The INCAR entries are fine.
I made an example calculation for diamond and put the input/output files here:
atrey.karlin.mff.cuni.cz/~kelum/out_files.tar.gz
You only need to add the POTCAR which I used the standard carbon one (PAW_PBE C 08Apr2002).
Does PBE run OK for your system?

brockp · #9 Post by **brockp** » Tue Jan 07, 2014 10:29 pm

We ran into this exact same issue with 5.3.3.18Dec2012

What happens is with the intel compiler 12.1 or 14.0 if you enable CPU dispatch:

-axAVX,SSE4.2 -msse2

The code will compile and run, but if you run on sandybridge (AVX enabled machines) you get that error.

Very very strange thing is is you remove those, but leave the optimization level in effect, which should still enable the vector unit for the CPU you are compiling on (but may then fail on CPUs without that feature) the code runs fine. This is kinda unexpected given that I compiled on sandybridge and thus AVX should be on, and then fail, I took the code to a westmerir chip and it still ran, so I am not sure what ifort is doing.