strange result of relaxation and static calculation

Message

mariesto · #1 Post by **mariesto** » Thu Mar 05, 2009 8:52 am

Dear All,
I am a newbie.
I did relaxation of metal-CNT to find the optimized atomic configuration. Then I used the atomic coordinate and lattice parameter of the final step of relaxation for a POSCAR file of new (static) calculation. I did static calculation with the same value of cutoff energy and electron step as relaxation calculation. It was surprising that the total energy energy was much higher than the total energy of relaxation calculation.

I do not understand the reason. Any one can give me a light?

thank you in adavance.

with best regards,

#2 Post by **admin** » Thu Mar 26, 2009 8:10 am

please specify 'much higher'.
did you change any other parameters, like the k-points?
if you relaxed the lattice parameter, it is well possible that the basis sets (NGX, NGY,NGZ) have changed due to the volume change. The new static run starts with the basis set according to the new lattice parameters and ENCUT. Therefore, please compare the basis sets.
If both, k-mesh and basis sets are comparable, there is no reason why the total energy should be higher.

mariesto · #3 Post by **mariesto** » Thu Apr 09, 2009 1:00 am

sorry for late response, I lost my files.
Now I found the same problem . I have the same KPOINTS file.
Below two INCAR files and part of OUTCAR files
1) relaxation file
INCAR
NELMIN = 8 # do a minimum of ten electronic steps
EDIFF = 1E-5 # high accuracy for electronic groundstate
EDIFFG = -0.01 # small tolerance for ions
NSW = 40 # 20 ionic steps should do
MAXMIX = 80 # keep dielectric function between ionic movements
POTIM = 0.2
IBRION = 2 # use RMM-DIIS algorithm for ions
NFREE = 10 # estimated degrees of freedom of the system
ISIF = 4
LGO = Fast # RMM-DIIS algorithm for electrons
LREAL = A # evaluate projection operators in real space
NSIM = 4 # blocked algorithm update, four bands at a time
ISMEAR = 0
SIGMA = 0.05
ENCUT = 550
ISPIN = 2

free energy TOTEN = -370.448432 eV

energy without entropy= -370.448431 energy(sigma->0) = -370.448431

2)static
NELMIN = 8 # do a minimum of ten electronic steps
EDIFF = 1E-5 # high accuracy for electronic groundstate
NSW = 0 # 20 ionic steps should do
LGO = Fast # RMM-DIIS algorithm for electrons
LREAL = A # evaluate projection operators in real space
NSIM = 4 # blocked algorithm update, four bands at a time
ISMEAR = 0
SIGMA = 0.05
ENCUT = 550
ISPIN =2

free energy TOTEN = -370.425069 eV

energy without entropy= -370.425069 energy(sigma->0) = -370.425069

#4 Post by **admin** » Fri Apr 10, 2009 4:12 pm

1) as ISIF=4 allows for the relaxation of the cell shape (keeping V constant) it may very well be that for the static run, NGX, NGZ and NGZ differ considerably from the values set for the initial cell geometry of the scf run (if the cell shape has changed significantly). Please check these parameters in both OUTCAR files. According to the energy difference you show (if due to cell shape changes), I would strongly suggest that you do another scf run (of 5-10 steps, because the starting parameters of the first scf run might not have been well-suited for the final geometry the system relaxed to. --> in general, it is better to do a cell shape (and size) optimization step-wise (continuing from CONTCAR, without using WAVECAR) to make sure that the FFT-grids are not too bad due to cell shape and volume changes. (please find a short discussion of the theory background in the vasp-workshop tutorial file 'accuracy.pdf' you can download this file from our web-page)

2) please also check if NBANDS was the same if you did your runs in a different number of CPUs (NBANDS is an integer multiple of #CPUs, differences in NBANDS also lead to (very small) differences in the total energies)

3) a short remark: the parameter to choose the electronic scf algorithm is ALGO (not LGO).