The Extrapolation in NEB calculation

Dear CP2K users and developers:

I am investigating the oxygen transport
path in perovskite with CI-NEB. In order to consider DFT+U and generate “.model”
file, I have to give up K-point and build a big cell with 250 atoms.

I set 9 REPLICA. My computer doesn’t have
enough memory(My computer only has 256G).

 So I set the NPROC_OF_REPLICA 36 and run with 108 MPI processes
(cp2k.popt). The 9 REPLICA points will run SCF in three groups, like (1, 2, 3),(4,
5, 6),(7, 8, 9). I find that the program can’t use EXTRAPOLATION of each REPLICA,
but with the previous groups…For example, when the program finished the group
(1, 2, 3), the group (4, 5, 6) will use the EXTRAPOLATION method with ASPC,
even if the group (4, ,5, 6) is calculated SCF for the first time in this
mission.

This operation make SCF no convergence when
calculating TS points (it can converge when calculating the single-point energy).

I turned off EXTRAPOLATION and everything
was fine. But without extrapolation, it was too slow. The program has made many
meaningless SCF calculation.

What can I do? Or is there any suggestion to speed up the NEB calculation?

Thanks in advance!

 

Best

Chenxiao wang

Read more here: Source link