ABINIT Discussion Forums

Dear ABINIT friends,

I like to calculate the energy barrier of a hopping oxygen atom on a Pd(111) surface.

I like to perform a minimum energy path calculation. I like to use the string method with 7 images. The initial and final image are already optimised.

13 k-points will be used for every image. Since I want to do the calculation in parallel over images as well as k-points, I need 5*13 = 65 cores.

From scanning over the abinit parameters, I got the impression that specifying ``autoparal 1`` should be sufficient for abinit to figure out the best parallel strategy. Abinit would calculate 5 images in parallel and would use 1 core for every kpoint of each image.

However, if I specify this situation, I got a segmentation fault error. Segmentation should not appear in any case. So I like to report that.

But more important, how do I get this job running?

I have included the input file for a closer inspection: pd.in

To reproduce the segmentation fault, I have also included the PP files and other necessary files. Though I had to append the extension ``.in``.

Hello

I suggest to parallelize manually:
paral_kgb 1
npkpt 13
npband 1
npfft 1
npimage 5

Also, make sure to use a very recent version of ABINIT to have xml pseudopotential files compatible with the string method algo.
best,
Gregory

Dear Gregory and ABINIT friends,

Again, segmentation fault occurred.

<snip>
...
forrtl: severe (174): SIGSEGV, segmentation fault occurred
...
</snip>

So, also this set of parameters does not work for me.

It would be interesting to reproduce the calculation on a different cluster/machine to check if it works in principle.

I think/hope that I'm not the only one doing String Calculations in parallel in 2 dimensions. Can at least someone confirm that it works on his site?

Dear Gregory and ABINIT friends,

Finally it seems to work. There was a bug in ABINIT concerning the format of the PAW data sets. The bug already seemed to be corrected. Version 8.8.4 of ABINIT did not have the correction.

So the xml format of the PAW data sets needed to be converted to abinit format. It was done for Pd and O in the following way:

$cat Pd.GGA_PBE_PAW_input:
<snip>
Pd 46
XC_GGA_X_PBE+XC_GGA_C_PBE scalarrelativistic loggrid 600 80 2.20
5 4 4 0 0 0
5 0 1
4 2 9
0 0 0
c
c
c
v
v
c
c
v
c
v
2
2.5 2.5 2.1 1.0
n
y
2.0
n
y
2.0
n
custom rrkj vanderbiltortho
3 0.50 ultrasoft
2.50
2.50
2.50
2.50
2.10
2.10
2
default
0
</snip>

$ atompaw < Pd.GGA_PBE_PAW_input

$ cat O.GGA_PBE_PAW_input:
$ cat O.GGA_PBE_PAW_input:

<snip>
O 8
XC_GGA_X_PBE+XC_GGA_C_PBE scalarrelativistic loggrid 2001
2 2 0 0 0 0
2 1 4
0 0 0
c
v
v
1
1.41 1.2 1.4 1.2
y
2
n
y
2
n
VANDERBILT
2 2.5 ultrasoft
1.41
1.41
1.41
1.41
2
default
0
</snip>

$ atompaw < O.GGA_PBE_PAW_input

The above input data can (partially) be found at the end of the xml paw data sets.

Two files will be generated:

Pd.GGA_X_PBE+GGA_C_PBE-paw.abinit
O.GGA_X_PBE+GGA_C_PBE-paw.abinit

These are the new PAW data sets.

Once included into the ``files`` file, the STRING calculation can be started with 5*13 CPU's. 13 CPU (is equal to kpoints) for each image.