Hi All,
I recently wanted to calculate the energy barrier for dislocation nucleation with the new NEB package. To do so, I created two configurations: the initial one is right before the dislocation nucleated, the final one is right after a dislocation being nucleated but prior to it propagates away. I used the following script:
With a script like this, you are running the same calculation
8 times, once on each processor until you get to the NEB command.
This includes the 2 initial minimizes. For that part of the
run, nothing will be printed to the screen or to log.lammps.
Instead you should get the same thing printed to each of
the 8 log.lammps.P files. I would take out the NEB command
and see if that is the case. I.e. all 8 procs produce identical
output and dump files and finish cleanly. If that works, then
add the NEB command and see if you get initial output
from NEB to the screen and in the 8 log files.
Note that you can break this into 2 input scripts if you want
to avoid this issue. I.e. run the minimize on 1 proc or in
parallel to create the intiial input files for a 2nd script
that just runs the NEB command.
Steve
Steve,
Thanks for your reply. If i took NEB command out, I’m not sure it finished cleanly as the log.lammps file as follows:
LAMMPS (30 Sep 2010)
Running on 16 partitions of processors
All the log.lammps.P files are identical in the parameters i’m calculating. That seems to indicate if without NEB command, the simulation ends successfully. If that’s the case, what is the next step to dig out the problem?
Note that you can break this into 2 input scripts if you want
to avoid this issue. I.e. run the minimize on 1 proc or in
parallel to create the intiial input files for a 2nd script
that just runs the NEB command.
FOR this, you mean dump a file or restart file to begin another run ? But that should be identical to put them together in one run, right?
Ajing
Well, before you said it was hanging after the 1st minimize, now it
appears to be running through both minimize commands, and hasn't
even reached the NEB command. So I don't know what that means.
If you now add the NEB command back in, you should see
some header info about the NEB run printed to the screen
and to log.lammps. And you should see each of the sub log files
start to record info about the NEB run. I would try it on
2 procs first.
Steve
Steve,
I have sent you my input and output files. Please help me to figure out where the program stops.
Many thanks,
Albert
Please send a tar file, not a Windows rar file. I can't
open the latter.
Steve
Steve,
I have the similar question.
I try to execute NEB calculation. The input script looks like
# start input script
atom_style atomic
boundary p p p
neighbor 2.0 bin
neigh_modify every 1 delay 0 check yes
atom_modify map array sort 0 0.0
variable u uloop 12
read_data begin.dat
pair_style eam/fs
pair_coeff * * Fe_mm.eam.fs Fe
timestep 0.001
fix 1 all neb 1.0
thermo 1
dump myDump all xyz 1 dump.xyz.$u
min_style quickmin
minimize 1.0e-6 1.0e-6 100 1000
neb 1.0e-6 1000 1000 100 end.dat
#end input script
execute program
mpirun -np 12 lmp_openmpi -partition 12x1 -in in.disl > out.disl
However it does not make any minimization. The log files has two
iterations for start configuration (it was minimized before) and 2000
iteration
for intermediate configurations with slightly decreasing energies.
Could you pleas take a look on the script.
Thanks,
German
Posted a 29Nov10 patch that should fix
this issue - it was a communication issue
that hadn't shown up with the smaller
systems we've been running.
Please try out your systems and see
if they now work.
Thanks,
Steve
Steve,
Thank you.
Attached please find tar file with neb run results and input.
I run two replicas on two core processor witn mpi installed. So it should
reprodus energu for configurations (uadruploe of screw dislocations)
desribed in files begin.dat and end.dat.
These two configurations was already minimized by cg style.
However neb continues to move and did not finish minimization in 2.e+4 steps.
Thanks,
German
neb.tar.gz (227 KB)
Steve,
The new patch did not fix the problem I have. Looks like it works with small system (smaller than 1000 atoms) but have trouble in larger system (13300 atoms in my case). Maybe real parallel spatial decomposition of the computational domain may solve this issue? Right now, the NEB package is not actually parallel, just virtual replica within processors, right?
Thanks,
Ajing
Steve, Ajing,
My problem is not relatred to amount of atoms. I have system with 462
atoms and try to reproduce
energy of start and end states which already was minimized.
German
German,
I’m aware that the log file in your case prints out some “nan” , which means sth happened badly, although I could not figure out exactly why.
Cheers,
Ajing
Ajing,
Thank you for your comment.
I withdrow my question.
The difference in the energies is because 1.e-6 energy accuracy is not
enough in cg minimization.
However I don not have "not a numberic" values
Regards
German