How to choose ENCUT?

Message

linglin · #1 Post by **linglin** » Tue Sep 09, 2008 7:10 am

When I do a calculation of a system of 12 atoms. The biggest ENMAX of the 12 atoms is 400eV. Then I wanna do a convergence test. The following is the result(The left column is ENCUT and the right column is the correspond total energy.):
400 -125.211722
450 -125.092998
500 -125.060377
550 -125.065888
600 -125.082137
650 -125.096851
700 -125.109178
But I see the energy difference dangles. The manual says 1.3*ENMAX is OK. I doubt whether 550 is sufficient or I should do a test further?

forsdan · #2 Post by **forsdan** » Tue Sep 09, 2008 10:18 am

There is no straight answer to your question. The required cutoff will depend on the quantities you want to calculate, their respective energy scale and the accuracy that is required. Often you also take the energy difference between two systems so the quantity you're looking at may be sufficiently converged (for your purpose) even though the total energy of the cells are not fully converged. Many bulk properties (elastic constants, lattice parameters, magnetization etc) often converge very rapidly and then it's not necessary to have very accurate total energy.

If you intend to look at volume relaxations you must often increase the cutoff by at least 30% (above the default value) in order for the stress tensor to be evaluated correctly so that you obtain the correct equilibirum volume.

So if your system and the properties you intend to evaluate requires highly accurate total energies I would continue the convergence tests. (You might want to consider to use ISMEAR = -5 for these systems.)
Otherwise I would start to evaluate the properties you want in one case and step by step increase the cutoff until you have reached the accuracy you require.

So the principle is to test your way there.

Cheers,
/Dan Fors

<span class='smallblacktext'>[ Edited Tue Sep 09 2008, 12:50PM ]</span>

linglin · #3 Post by **linglin** » Wed Sep 10, 2008 1:50 am

I appreciate your reply!

linglin · #4 Post by **linglin** » Fri Sep 12, 2008 5:01 am

I continue the convergence test because I wanna do a geometry optimization afterwards. Then I get the result as follow:
400 -125.211722
450 -125.092998
500 -125.060377
550 -125.065888
600 -125.082137
650 -125.096851
700 -125.109178
750 -125.114665
800 -125.118107
850 -125.118480
900 -125.118272

I seems choosing 700eV would be sufficient. But the warning occurs: "PSMAXN for non-local potential too small " from ENCUT=650.
I'm wondering whether to choose ENCUT=700eV neglecting the warning or to set ENCUT=600eV before the warning didn't occur?