Different energy values in different vasp versions?

[exk301]

Hello!

I have a questions about different energy values in two different vasp versions (vasp 5.4.4 v. and vasp6.4.1v)

The vasp test suite was completed and confirmed there is nothing wrong with the build. But my question is :

I was comparing the geometry opitmization energy/single point cal energy on the same systems in both vasp versions.
I tested on Silicon, Ru metal atom. For Silicon, both versions gave me the same energy values(within the error 0.01eV).
However, they did not give me the same energy values for Ru metal atom in spc or geometry optimization giving me more than 1.5eV difference. I think it's not necessary to run the optimization for one metal atom, but i was wondering why both versions cannot give me the same values in both spc/geometry optimization for simple system. I wonder if it's because my INCAR input change should be required, or just somethting else. I exaclty used the same vasp input files for both vasp versions.

I've attached the job output files here.

Thanks for your help in advance!!!

[marie-therese.huebsch]

Hi,

Thank you for asking.

In short:
The issue is that you have compared the results of calculations with ALGO=FAST that converged to different solutions in the electronic minimization (both are sort of incorrect).

Detailed explanation:
Consider the calculations at Ru_atom/spc_test/vasp_5_4_4 and vasp6_4_1
You have selected ALGO=FAST. This means the Davidson (IALGO=38) is used for the initial phase, and then VASP switches to RMM-DIIS (IALGO=48). By comparing the OSZICAR files, you see that while using DAV, both VASP versions go the same way up to numerical fluctuations. Then, vasp 6 needs 43 steps, and vasp 5 requires more than 1700 steps to converge to EDIFF=1e-5.

Code: Select all

vasp_5_4_4/OSZICAR-RMM: ***    -0.165676504364E+01    0.29964E-02   -0.42170E-03   480   0.229E-01    0.503E-01
vasp_5_4_4/OSZICAR-RMM: ***    -0.165909603243E+01   -0.23310E-02   -0.28522E-04   497   0.739E-02    0.510E-01
vasp_5_4_4/OSZICAR-RMM: ***    -0.165909458896E+01    0.14435E-05   -0.17258E-05   537   0.179E-02
vasp_5_4_4/OSZICAR:   1 F= -.16590946E+01 E0= -.15985749E+01  d E =-.121039E+00  mag=     1.8892
--
vasp_6.3.0/OSZICAR-RMM: 627    -0.167744160085E+01   -0.15404E-02   -0.62096E-04   908   0.911E-02    0.713E-01
vasp_6.3.0/OSZICAR-RMM: 628    -0.167786637831E+01   -0.42478E-03   -0.48503E-05   766   0.274E-02    0.726E-01
vasp_6.3.0/OSZICAR-RMM: 629    -0.167786824090E+01   -0.18626E-05   -0.34602E-06   548   0.716E-03
vasp_6.3.0/OSZICAR:   1 F= -.16778682E+01 E0= -.16041025E+01  d E =-.147532E+00  mag=     1.9105
--
vasp6_4_1/OSZICAR-RMM:  41    -0.145424486483E+01   -0.11886E-03   -0.42200E-04   745   0.644E-02    0.112E+00
vasp6_4_1/OSZICAR-RMM:  42    -0.144564768630E+01    0.85972E-02   -0.68538E-04   755   0.674E-02    0.110E+00
vasp6_4_1/OSZICAR-RMM:  43    -0.144563959848E+01    0.80878E-05   -0.51874E-05   626   0.240E-02
vasp6_4_1/OSZICAR:   1 F= -.14456396E+01 E0= -.12767169E+01  d E =-.337845E+00  mag=     1.4640

Note that RMM is fast but can fail, e.g., because it always stays in the space spanned by the initial orbitals. You can read more about it on the Wiki at IALGO=48. When VASP needs more than 20 steps with RMM, this is suspicious. When it is about 40 steps, I would already bet it failed. But with more than 1700 iterations, you definitely haven't converged to any reasonable solution.

I have compared the two VASP versions by choosing the following for the electronic minimization:

Code: Select all

   ALGO = Normal   
   EDIFF = 1E-08     
   NELM = 100 
   #ISYM = 0 

   AMIX     =   0.40;   BMIX     =  0.0001 
   AMIX_MAG =   0.80;   BMIX_MAG =  0.0001

The rest is the same as you have set it.

This yields

Code: Select all

vasp_5_4_4/OSZICAR-DAV:  26    -0.223098966345E+01   -0.55641E-07   -0.19614E-08   120   0.172E-04    0.302E-05
vasp_5_4_4/OSZICAR-DAV:  27    -0.223098971553E+01   -0.52086E-07   -0.16330E-09   120   0.225E-04    0.101E-04
vasp_5_4_4/OSZICAR-DAV:  28    -0.223098971368E+01    0.18561E-08   -0.95783E-10    96   0.448E-05
vasp_5_4_4/OSZICAR:   1 F= -.22309897E+01 E0= -.20943949E+01  d E =-.273190E+00  mag=     4.0000
--
vasp6_4_1/OSZICAR-DAV:  26    -0.223098966385E+01   -0.56250E-07   -0.19734E-08   120   0.172E-04    0.302E-05
vasp6_4_1/OSZICAR-DAV:  27    -0.223098971535E+01   -0.51503E-07   -0.17460E-09   120   0.225E-04    0.101E-04
vasp6_4_1/OSZICAR-DAV:  28    -0.223098971345E+01    0.19027E-08   -0.10744E-09    96   0.448E-05
vasp6_4_1/OSZICAR:   1 F= -.22309897E+01 E0= -.20943949E+01  d E =-.273190E+00  mag=     4.0000

When you set ISYM=0, you get the same result but need around 40 steps for the electronic minimization in this calculation (which is no reason to get suspicious because this is the DAV and not RMM).

Does this answer your question?

Best regards,
Marie-Therese