GEOMETRY OPT not converge

[shixun sun]

Dear CP2K Community,

 Recently, I've performing Geometry Optimizations for my model with the input file written by me,  the model is graphene oxide generated randomly by code. However, I found it is hard to converge in opt, though I use PBE-D and dzvp BASIS.  Through the output,  the energy and force are basically decreasing but in the last few steps, it seems to vibrate and precarious. here are the last few steps 

  Total energy:                                             -1749.41509473804945

 ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):            -1749.415094862594287

  Total energy:                                             -1749.37672182140795

 *** WARNING in qs_scf.F:542 :: SCF run NOT converged ***

 ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):            -1749.376720297097563

I attach the output, image of GO without geo-opt and input, 

Can anyone please help me. Any help will be highly appreciated. 

[Fabian Ducry]

Dear Shixun,

There are several things you should do:

 - decrease EPS_DEFAULT in DFT%QS to 1e-10, 1e-6 as you have it is not enough by far

 - decrease EPS_SCF to 1e-7

 - change SCF%MAX_SCF to 30 and add an OUTER_SCF to DFT%SCF

       &OUTER_SCF

         EPS_SCF 1e-7

         MAX_SCF 30

       &END OUTER_SCF

 - change SCF%SCF_GUESS to RESTART

 - increase MGRID%NGRID to 5

I would also recommend to use BFGS instead of LBFGS for small systems (<1000 atoms).

Fabian

[shixun sun]

A lot of thanks for your reply,

Fabian Ducry <fabia...@gmail.com> 于2020年5月5日周二 下午1:59写道：

--
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/cb390eb3-83d5-4da0-9a22-31922a35103b%40googlegroups.com.