Hello, I need simulate an spinel structure (LiMn2O4) using vasp. I need some do some convergence test for structure optimization, like Total energy vs lattice, Total energy vs. k-points and Total energy vs cutoff energy, but I have some confusion with OUTCAR and OSZICAR values.
I'm not sure which value of energy I have take for compare with the parameter that I'm changing (e.g. ENCUT, KPOINTS, lattice in POSCAR).
I appreciate your help.
Thanks in avance!
If you just need the total energy, just take the total energy value in OUTCAR, i.e. the last "free energy" (note between these two words there are two blanks). If you use HP smearing, I would suggest you use the value after "sigma->0".
Thanks for your answer. I have been do some convergence test. For lattice and smearing all is normal, but when I do some test for cutoff energy (ENCUT in vasp) and I graph Total energy vs ENCUT I get a graph like a sawtooth, can you recommended any else to do convergence test for cutoff energy?
Attach the Etotal vs Ecut grahp. Also atach the Etotal vs kpoints that is strange, too.
Could you give more details of your calculations by answering the following questions?
(1) did you include relaxation in the calculation? if you did, which one did you include? atomic, volume or shape relaxation.
(2) what was the k-point mesh that you used for the first figure? Etot vs. Ecut.
(3) did you include magnetism? if you did, what kind? Is it AFM?
(4) the total energy in the figures, is it per atom or per cell?
1. No, yet. Should I included before of convergence test?
2. k-point mesh: 10x10x100 with ENCUT = 600 eV.
3. No, I dit not include magnetism.
4. Is per unit cell. I´m using the space group F d -3 m : origin 2 with 56 atoms per unit cell.
I supose that the pseudopotentials are relevants. I using potUSPP_LDA for Li, Mn and O.
If you need check the input parameters let me know.
From your answers I do not see the reason why the total energy is not converged with respect to cut-off energy. And I guess you use 10x10x10, not 10x10x100.
Do you have a particular reason for using ultrasoft pseudopotentials?
Since you have O, you still need 600-700eV cut-off, if you use PAW.
Oh! I'm sorry was a mistake, the kpoint mesh was 10x10x10.
I trying to reproduce this paper: First-principles studies of cation-doped spinel LiMn2O4 for lithium ion batteries.
All calculation were do with using ultrasoft pseudopotentials.
Article First-principles studies of cation-doped spinel LiMn2O4 for ...
You can change the value of the parameter (ECUT, or KPOINTS) each time, and run an electronic calculation. Only change one parameter at a time and keep the rest fixed. Then, compare the total free energy in the OUTCAR files. Use the optimize parameters and run a full ionic relaxation to get the converged lattice parameters. Hope that helps!
Pegah Mirabedini
do you mean, say for each ENCUT value keep rest of KPOINT, Lattice fixed and relax the system full, until fully converged? it might take long time for each value of ENCUT and similarly for each other parameters, i.e. say 4 ENCUT, and 4 KPOINT mesh grid, and 3 lattice . then there will be 48 total calculations of fully converged values, and then try to fix one set of combination.
It takes ages just for finding parameter combination.
Is this what is convergence test?
What i understood was for each such combination, just run one static calculation keeping NSW=0 or 1. and at the end compare energies and select whichever set/combination give the lowest energy. then do full run of optimisation.
Can anyone kindly clarify how it has to be done?
Regards
SK M Hi! If you are not looking for the relaxed lattice parameters, you don't need to perform a full geometric convergence. You can simply perform an electronic relaxation (i.e. NSW = 0 for VASP) and compare the energies in the OUTCAR file. This shouldn't take long but the timing varies with the size in your system. After finding the optimum parameters, you can use them to perform an ionic convergence.
Pegah Mirabedini
Hi
Thanks for quick reply. Let me write what i understood so far on this query. please correct me if i am wrong.
Say i have a system (crystal) with lattice parameters a, b, c.
anyway, I am not wanting to relax the lattice parameters, i.e. i use ISIF=2.
Now irrespective of PREC tag, if i explicitly give values for ENCUT (example 400, 500, 600 etc.) and NSW=0 and keep all other tags as per the need and set-up 3 jobs (w.r.t 3 ENCUT values) and keep KPOINTS file also same, and run them till electronic relaxation. Then compare the energy from OSZICAR last line. Check whichever is minimum and consider that ENCUT.
Now run another 3 jobs (example for 3 different KPOINTS file, 3x3x1, 4x4x1, 5x5x1) [ the 1 in c-direction is because i keep a vacuum slab in that direction]
and use the previous selected ENCUT value from above test, and do again electronic relaxation (i.e. again NSW=0). Now select the KPOINTS file, whichever gives lowest energy out of 3 jobs.
For the final geometry relaxation use the above ENCUT and KPOINTS keep all other tags same as used for the test.
So, is this the procedure correct? if so, or if not, what other aspects to be considered in general?
Note: In the same above test, if i want Lattice parameters convergence test also. How is that different from ENCUT and KPOINTS test?
Thanks
Regards
Maybe you want to take a look at this ebook:
DFT-book - John Kitchin
kitchingroup.cheme.cmu.edu › dft-book › dft
I am getting this type of energy convergence with K points. I am doubtful this is correct or not. Am I am doing something wrong. Please suggest.
Poonam Sharma except KPOINTS are others remains same in the runs? is your system, 2D or bulk? first need to check this, and then anyway, ignore the first value and see the other energies, they are fluctuating within very narrow range of energies. i would select 4 or 5 as and proceed further. this should be safe. any other opinion by others may please write down for others.
Regards
SK M thanks for the quick response.
yes, other things are the same in the run. It is a bulk system.
Lattice value I took from the literature. It is 5.6 angstrom. I'll try with 4th.
Thank you so much for your response.
- Badges
- Science topic
- Similar topics
- Filing