hi there, you should first tune your encut and kpoint mesh until satisfactory converged result before making any conclusions. different algo tags (if the encut and k-mesh is setup) should not make much difference in the resulting values (up to the convergence limit you've set). also, the nature of the system matters. metals (conductors) require far more k-points than semiconductors and insulators. I suggest you use prec=high, that automatically increases cutoff by 25% and tune the kpoints.

Dear mr. Chovan,

thank you for you reply.

I performed a test. I calculated the band structure of one hex bulk material with ALGO = Damped and ALGO = Normal. All other parameters were the same. Both calculations successfully converged to EDIFF = 1E-6.

Also you mentioned PREC. I used PREC = Normal.

I attached two pictures of a band structure (zoomed). The first one calculated with ALGO = Damped, the second one with ALGO = Normal.

Why this can happening?

Thank you in advance.

Hi, are you running MD? vasp manual reads "IALGO=53 damped MD with damping term automatically determined by the given time-step (ALGO=D)." . if your results are from MD, then it does not much sense and there is no point of evaluating BS and DOS and making conclusion, since the ALGO tag only affects the way the ions are moved. one need to generate charge density first and then run BS ad DOS with different smearing. !!! READ THE MANUAL CAREFULLY !!! how to perform these calculations. there are whole chapters about it. also, increase cutoff and kpoint mesh for the static runs. the kpoint mesh is not good (dense) enough if you are not getting smooth BS.

https://www.researchgate.net/post/How_can_I_calculate_the_DOS_by_using_HSE06_in_vasp may be of use.

Dear mr. Chovan, i read these instructions: http://cms.mpi.univie.ac.at/wiki/index.php/Si_bandstructure

there is such INCAR example:

## Default

ISMEAR = 0

SIGMA = 0.01

GGA = PE

## HSE

LHFCALC = .TRUE. ; HFSCREEN = 0.2 ; AEXX = 0.25

ALGO = D ; TIME = 0.4 ; LDIAG = .TRUE.

I used the same parameters for hybrid calculation with ALGO = Damped. Actually i prefer to split whole calculation to 3 stages (1 - dft sc, 2 - hse sc, 3 - hse bands with modified KPOINTS).

I'm sorry, but i can't understand you, mr. Chovan. I was sure that ALGO determines the way how orbitals will be optimised, not how ions will be moved.

VASP manual says:

"The ALGO tag is a convenient option to specify the electronic minimisation algorithm" and "IBRION determines how the ions are updated and moved."

Moreover, there is such prase about ALGO: "Mind: All implemented algorithms will result in the same result, i.e. they will correctly calculate the KS groundstate, if they converge." And about ALGO = Damped specifically: "The algorithm has been carefully optimized and should be selected for Hartree-Fock type calculations."

Please do not worry that the bands above is not smooth, this is not a problem really. I know that it's non-smoothness caused by low k-point grid density in band structure calculation. I want to understand, why the bands became splitted. Especially if there are so much tips in manual like "use ALGO = Damped with hybrids".

yes, you are right. I messed that part up. but to my defense, I had it right first time, then rewrote without checking. apology. Probably should read manual carefully myself after the years to refresh.

anyways, I am sure that all algorithms have been tested and should work fine unless you have some kind of exotic system that has not been . I will consult guys here at my workplace if they experienced something similar to your problem (because I haven't) with given calc. setup and will get bk to you.  do you mind sending input files so I can check? this is new to me and will be also educational exercise to crack this thing. 

sorry for such long absence. i solved that problem several montsh ago, but i want to share with those who may face problem like that.

so the problem is in the energies of the new k-points with zero weights, that are used for band structure calculation. it's all right if i use ALGO = Normal. i found out, that ALGO = Damped or ALGO = All make calculation speed increase significantly, but the energies in k-points with zero weights differ from the energies in k-points with non-zero weights. that's why band srtucture may be so bad. but it can be fixed by making several band structure calculations with ALGO = Damped or ALGO = All. i tried setting NSW (make several ionic steps with fixed ions positions), but i found it much less effective than starting a new calculation.

so if you need to use ALGO = D or ALGO = A (e.g., if using ALGO = Normal causes divergence), you just need to have ISTART = 1 and LWAVE=.TRUE. in your INCAR and run this calculation several times (e.g., 3--6).

i hope this message will help you if you face similar problem with band structure calculation within HSE in VASP.

This discussion is really helpful to other people.

Evgeniy Petrov this is interesting discussion.

i wish to understand more on your suggestion:

"... so if you need to use ALGO = D or ALGO = A (e.g., if using ALGO = Normal causes divergence), you just need to have ISTART = 1 and LWAVE=.TRUE. in your INCAR and run this calculation several times (e.g., 3--6). ."

Actually my problem started when i tried to reproduce a published Bandgap calculation of a material with HSE06. Say, published value was 2.71 eV, i got only 1.79 ev. But with the same cation of the Material with Selenium as anion, the published value was 2.37 eV and i got 2.18 eV (which is almost similar) with same settings of INCAR that i used with the previous calculation. This is intriguing.

i am searching for a reason for it which happened to look at your suggestion, might be interesting to test.

Now my question on your suggestion is:

if we use ISTART=1, means the WAVECAR file shall be supplied from previous run. if not supplied it will start fresh (in the first run, say).

Then if we run 3 to 6 times as you mentioned, how does it change the band structure? sorry, i am not expert on the internal algorithm how it actually works.

Hope my question is clear.

regards