Struggling with getting converge for Zr(110) optimisation

[Ohaya lawrence]

Hello everyone, I'm a green hand of CP2K. I try to compare the speed of CP2K and Vasp with  Zr(110) optimization. But  I fail to get converge for  Zr(110) optimization after 11 hours-running. Here are my input and output files.  I want to know why the same system in CP2K is so slow (compare with  2 hours-running in Vasp) and how to get a faster calculation in CP2K.  

[Krack Matthias (PSI)]

Hi

 

You could try a larger smearing value. 300 K correspond to roughly 0.026 eV only. I guess you are using a larger ISMEAR value with VASP.

 

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/e70f11a7-d228-4bfd-91f1-a9c45c024d0bn%40googlegroups.com.

[Lucas Lodeiro]

Hi Ohaya,

As Matthias said, your SIGMA value is too small, if you use 0.2eV in VASP, you have to set 2320K in CP2K, and add more orbitals... and if you delete the CHOLESKY INVERSE line you could get a speed up of 10 %.

Also there are a lot of convergence criteria which are so different between the programs... for example the EDIFF in VASP is 1.0E-4 related to the energy, meanwhile in CP2K the EPS_SCF is 1.0E-5 related to the coefficients of the density matrix... probably with it you are getting more converged energies with CP2K, and this requires more scf cycles.

Just to mention it, for the GEO_OPT, CP2K uses 4 criterias (energy, forces) and VASP uses just one, EDIFFG, the energy or the force... This probably ensures better converged structures in CP2K too.

If you want to compare both programs, you have to check all these criteria, and match them to compare more precisely. 

Finally, you use LBFGs instead of BFGs... I do not know if changing it you get a speed up in the ionic relaxation, but the manual says that LBFGs is for big systems, and yours is small.

Regards 

Lucas Lodeiro

El dom, 13 dic 2020 a las 23:11, Ohaya lawrence (<xunl...@gmail.com>) escribió:

Hello everyone, I'm a green hand of CP2K. I try to compare the speed of CP2K and Vasp with  Zr(110) optimization. But  I fail to get converge for  Zr(110) optimization after 11 hours-running. Here are my input and output files.  I want to know why the same system in CP2K is so slow (compare with  2 hours-running in Vasp) and how to get a faster calculation in CP2K.  

--

[Ohaya lawrence]

OK, I try to change this value. Thank you! 

[Ohaya lawrence]

Hi Lucas

Thank you for your detailed answer. From your answer, there are many values that I need to test. I will do my best.