Cell_OPT convergence is too slow

[rui zou]

Hi everyone!

I am a new user of CP2K. I want to use CP2K to optimize the cell. However, the convergence of this process is too slow. Can you help me check if there is a problem with my input file? Thank  you so much!

Best regards,

Rui Zou

[Krack Matthias (PSI)]

Hi Rui Zou

 

There are two issues which might hamper the convergence process of your cell_opt run:

1. The SCF convergence threshold value is with 5.0E-6 quite large which will cause a low accuracy of the calculated forces. Noisy forces will in turn “confuse” the optimizer.
2. The applied constraints KEEP_ANGLES and KEEP_SYMMETRY might prevent the optimiser from finding an energetically lower lying structure with less symmetry indicated by residual forces.

 

HTH

 

Matthias

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/2ddfaea7-2b59-4877-9d72-7c4bb462b4ddn%40googlegroups.com.

[rui zou]

Hi, Matthias Krack

Thank you so much for your advice. 

In this calculation I used a supercell of 320 atoms. I suspect this also caused the convergence rate to be too slow. When performing CELL_OPT in CP2K, is it necessary to expand the cell first? 

Best regards

Rui Zou

[jtyan...@gmail.com]

Hi Rui Zou

Your system is very big! So maybe it's normal to be slow. (though you can add cores and nodes to run it) 

What Matthias said is right, and I suggest LBFGS rather than BFGS as your minimizer, you can find description of minimizer in cp2k manual.

Also, DIIS sometimes faster than CG, though it may lead you to convergence problem. 

Best regards

JT Yang