kernel differs, recalculating kernel

Message

zsjan · #1 Post by **zsjan** » Sun May 05, 2013 5:40 am

Hello:
when I use the vdw calculations, I met this warning ?kernel differs, recalculating kernel?. Afterwards, the program started to calculate the kernel with a very long time. The vdw_kernel.bindat file I have placed in the same directory. So what can I do to resolve the problem? Thank you.

kelum · #2 Post by **kelum** » Mon May 06, 2013 10:44 am

Hi,
probably you are running on big endian machine, check this post,
http://cms.mpi.univie.ac.at/vasp-forum/ ... hp?2.11952
if so, then you can download the big endian kernel from here
atrey.karlin.mff.cuni.cz/~kelum/vdw_kernel.bindat.big_endian
move it to the place of the original one (and rename to vdw_kernel.bindat)
and then the calculation shouldn't complain.
best,
jiri

zsjan · #3 Post by **zsjan** » Tue May 07, 2013 11:01 am

Unfortunately, the problem is not resolved even when I download this vdw_kernel.bindat.big_endian as vdw_kernel.bindat file. So what is probably the problem?

kelum · #4 Post by **kelum** » Tue May 07, 2013 2:14 pm

hm, strange.
Just to be sure, can you say if you machine is big or little endian? Big would be something like IBM Power. Intel and AMD are little endians.
Could you do a
hexdump -d vdw_kernel.bindat | head
and post what it does? (First line would be all right.)
I get
0000000 00004 00000 00035 00000 00004 00000 00280 00000
so that in the fourth group the number 35 occurs, which is the same number as used in VASP for the kernel interpolation.
Now if you get 35 for one version of the kernel file and even with that file the code complains then one possibility could be that nd is read as 8 byte integer while it is stored as 4 byte in the file or something like that, then one would have to change the declaration of nmesh to integer*4 around line 419 in the code... maybe.
Actually, what is printed on the previous two lines before the "kernel differs..." should be nmesh= and nd=? nmesh is read from the file and nd should be 35.
best,
jiri

zsjan · #5 Post by **zsjan** » Wed May 08, 2013 3:15 am

Yes, I am using intel machine to do the calculations, so it should be little endian?
The output of command hexdump -d vdw_kernel.bindat | head is
0000000 00000 01024 00000 08960 00000 01024 00000 06145
However, the error information in the logfile begins with
nmesh -603355904
nd 35
kernel differs, recalculating kernel
How could be the nmesh is negative?

kelum · #6 Post by **kelum** » Wed May 08, 2013 10:33 am

OK, this is what I get for the big endian version as well, you should use the original kernel. What are the two lines in that case?
And the hexdump of the kernel?

zsjan · #7 Post by **zsjan** » Thu May 09, 2013 10:37 am

When I use the original kernel, the output of hexdump seems to be right:
0000000 00004 00000 00035 00000 00004 00000 00280 00000
However, another problem appears:
kernel read 0.000000000000000E+000 0.242389471443564
kernel read 0.000000000000000E+000 0.242389471443564
0.507104952116469 0.796202846057805 1.11192896856814
1.45673599250183 1.83330250150242 2.24455379819341
2.69368462896985 3.18418400192505 3.71986229070676
4.30488083485548 4.94378426657054 5.64153581503002
6.40355586252091 7.23576405189726 8.14462527247212
0.507104952116469 0.796202846057805 1.11192896856814
1.45673599250183 1.83330250150242 2.24455379819341
2.69368462896985 3.18418400192505 3.71986229070676
4.30488083485548 4.94378426657054 5.64153581503002
6.40355586252091 7.23576405189726 8.14462527247212
9.13719988157784 10.2211985519339 11.4050421708962
12.6979272569059 14.1098974013170 15.6519212905873
17.3359779149375 19.1751496254094 21.1837237622236
23.3773036439213 25.7729297794946 28.3892122451236
31.2464752538692 34.3669150413914 37.7747722942059
41.4965204599610 45.5610714025986 50.0000000000000
9.13719988157784 10.2211985519339 11.4050421708962
12.6979272569059 14.1098974013170 15.6519212905873
17.3359779149375 19.1751496254094 21.1837237622236
23.3773036439213 25.7729297794946 28.3892122451236
31.2464752538692 34.3669150413914 37.7747722942059
41.4965204599610 45.5610714025986 50.0000000000000
kernel read 0.000000000000000E+000 0.242389471443564
?.
So what can I do?

kelum · #8 Post by **kelum** » Fri May 10, 2013 10:39 am

I'm not sure if this is an issue or not, does the calculation crash or not? In any case this looks more like the output of the old version of the vdw_nl routines, not the official ones. I'd suggest to use those in the first place.

harry045 · #9 Post by **harry045** » Mon May 13, 2013 9:43 am

I met this warning ?kernel differs, recalculating kernel?. Afterwards, the program started to calculate the kernel with a very long time. The vdw_kernel.bindat file I have placed in the same directory. So what can I do to resolve the problem???

kelum · #10 Post by **kelum** » Mon May 13, 2013 10:38 am

Could you go through the points above to resolve the issue/get more details?