inaccuracy of Pulay stress calculation

[suen]

Dear all,
I have calculated the Pulay stress for my system using the method in the manual(vasp5.3.5):
First, relax the structure adequately(ENCUT=550; PREC = Accurate; EDIFF = 1E-5; EDIFFG = -0.0001;ISIF=3),
then, cp CONTCAR POSCAR, set NSW=1, other tags in INCAR is not change, re-run VASP with different ENCUT, the result is:

Code: Select all

ENCUT     Pressure(kB)
400            -61.14
500            -0.21
520             0.71
550             0.08	
600            -2.52	
900             1.72
1000           1.22

I am confused that the external pressure is not close to zero when ENCUT>550. the ENMAX in my POTCAR is 400, i do not know how to explain the result of my Pulay stress calculation, whether my calculation method is wrong? or i should use ENCUT=1000 to calculate the Pulay stress?

Pulay stress = P(encut=400) - P(encut=1000) = -62.36
Pulay stress = P(encut=400) - P(encut=900) = -62.86
Pulay stress = P(encut=400) - P(encut=600) = -58.62
Pulay stress = P(encut=400) - P(encut=550) = -61.22

the error value is about 4 kB, maybe this is not a ignorable error. whether this difference is normal in vasp ?
Any suggestion will be appreciated !

[Neutrino]

Dear Suen,

First, I do not think that typical VASP PAW POTCAR can afford going to such high energy cutoffs as the one you explored. Try searching for the following line in your OUTCAR

Code: Select all

ERROR FEXCP: supplied Exchange-correletion table is too small

The appearance of this message indicates that the cutoff is too high than what the POTCAR is designed for.

Your calculations shows that the pressure at 500 eV on the structure relaxed with 550 eV is only -0.21 kB. This is the estimate of Pulay stress and as such, 500 eV should be very safe to use.

On the same topic, and this also discussed in the manual, you can calculate the equation of state of the material by performing relaxations at different volumes (using ISIF=4) and fitting the results to Birch - Murnaghan (or a similar) equation of state. This way you will avoid all issues related to Pulay stress and get more physical information about the properties of the material such the Bulk modulus and its derivative. After you get the final volume from the fit, I would recommend doing a final ISIF=4 relaxation on the optimal unit cell volume.


Mostafa

[suen]

Dear Suen,

First, I do not think that typical VASP PAW POTCAR can afford going to such high energy cutoffs as the one you explored. Try searching for the following line in your OUTCAR

Code: Select all

ERROR FEXCP: supplied Exchange-correletion table is too small

The appearance of this message indicates that the cutoff is too high than what the POTCAR is designed for...

Dear Neutrino,

Thank you for your kind reply. Maybe I can use ENCUT=500 in my following MD simulation though it is more expensive. I have checked my OUTCAR file, there is no "ERROR FEXCP" in it, I do not know what does this error mean physically. And yes, fitting the E-V relation is a good method, thanks for this suggestion, i will learn to do that
because of my lack of the knowledge in theoretical physics, it is difficult for me to learn vasp well, do you have any resource about these knowledge that i can learn ? "step by step" will be better

[alex]

Dear Suen,

I still wonder what quantity you are interested in?

Pulay stress orgininates in numerical incompleteness of the basis set and change of the cell volume. The better the basis (higher cut-off energy) the less Pulay stress you'll introduce. A standard procedure is, e.g. repeat the cell size optimisation (three times, double ENMAX as high level cut-off; e.g. for O standard PAW ENMAX = 400 eV, so use 800 eV for cut-off) to end up with a well converged unit cell.

However, I'm not sure if your MD calc afterwards will alter the unit cell structure (I'm not sure if this possible at all with VASP).

My recipie would be: Do the cell optimisation on a high level, then go cheaper with the structure optimisations. Finally, you might do the expensive calculations for some selected examples to convince yourself, everything is good enough.

Hth,

alex

[Neutrino]

Dear Suen,


If you plan to do constant volume MD , then you should not worry too much about Pulay stress and you could simply use the maximum of all ENMAX from the POTCAR's of the elements you are modelling. Pulay stress is a concern when you do variable volume calculations.


For a practical introduction to DFT , I'd recommend reading :
"Density Functional Theory: A Practical Introduction" by D Sholl and J A Steckel.

The book is very accessible.


Mostafa

[suen]

Dear Suen,

I still wonder what quantity you are interested in?

Pulay stress orgininates in numerical incompleteness of the basis set and change of the cell volume. The better the basis (higher cut-off energy) the less Pulay stress you'll introduce. A standard procedure is, e.g. repeat the cell size optimisation (three times, double ENMAX as high level cut-off; e.g. for O standard PAW ENMAX = 400 eV, so use 800 eV for cut-off) to end up with a well converged unit cell.

However, I'm not sure if your MD calc afterwards will alter the unit cell structure (I'm not sure if this possible at all with VASP).

My recipie would be: Do the cell optimisation on a high level, then go cheaper with the structure optimisations. Finally, you might do the expensive calculations for some selected examples to convince yourself, everything is good enough.

Hth,

alex

Dear alex,
I am sorry for replying you so late, now i just want to learn the vasp well, and to calculate some physical properties of materials. Thanks for your suggestion, it is very helpful. In my MD simulation, the unit cell structure will be changed due to the variations of Pressure and Temperature, so i should eliminate the influence on volume from Pulay stress and LDA (GGA).

[suen]

Dear Suen,


If you plan to do constant volume MD , then you should not worry too much about Pulay stress and you could simply use the maximum of all ENMAX from the POTCAR's of the elements you are modelling. Pulay stress is a concern when you do variable volume calculations.


For a practical introduction to DFT , I'd recommend reading :
"Density Functional Theory: A Practical Introduction" by D Sholl and J A Steckel.

The book is very accessible.


Mostafa

Dear Neutrino,
Thanks for your recommends, it seems that the book is a coordinated materials for vasp and it is well suited for beginners, very exciting, thank you!