9
$\begingroup$

To verify the reliability of calculation scheme, I want to compare lattice constants, bond length and band gap calculated by the different functionals with the experimental results.

So in order to do it, what are the calculations I should do ?

$\endgroup$

1 Answer 1

8
$\begingroup$

There are two ways to approach this problem. You can either perform a vc-relax calculation which will optimize/relax the atomic positions as well as the cell parameters. Then from the relaxed coordinates, you can check the bond length and the cell parameters. As a second option, you can perform multiple scf calculation at different cell parameters. It is somewhat easier to do for a cubic cell since you only have one free parameter a or celldm(1). However, for hexagonal or other complex cell shape, you might also need celldm(2), celldm(3) etc. since they are not necessarily equal as in the cubic case. In any case, you will perform multiple scf calculation and plot the energy varying the cell parameter. You can fit the energy vs cell parameter data with some kind of equation of states to get the lattice parameter (check ev.x that comes with QE). After having the optimized cell parameters, using them you can run a relax calculation which will give you the bond lengths. If your system is complex and reading the coordinates from the output is not straightforward, then you can use any visualization tools such as XCrySDen or VESTA. They have built-in function to show you the bond lengths, any atomic distances, and many more features.

After having relaxed cell parameters and relaxed atomic positions, now you can run a scf calculation to obtain the band gap. In your QE input file, set calculation='scf' and verbosity='high'. Setting the latter will ensure that if you have many k-points, they will show the KS energy values for each k-points explicitly which you will need to obtain the band gap. Then you can follow this answer in How to find the band gap energy value in Quantum ESPRESSO? to obtain the band gap energy.

Now a few important remarks. You must do convergence testing for parameters such as ecutwfc, ecutrho, K_POINTS, degauss etc before proceeding to any of the above stated calculations. Even though I generally perform only energy convergence tests, here Tyler Sterling suggests checking the force and pressure convergence too. This is important since using non-converged parameters (i.e. using literature values or using suggested minimum values) can give terrible result. I had good results for simple systems without convergence testing but once you are not doing the tutorials with silicon or graphene or other simple materials, convergence testing is must.

The second important point us that you are trying to compare results from different functionals. In this case, you have to repeat the convergence test each time you are changing your functional. The converged values for an LDA functional will be significantly different than for a PBE functional. Even if the functional remains the same but once you change your pseudopotential (such as from ONCV to PAW), the convergence testing needs to be repeated.

$\endgroup$
3
  • 1
    $\begingroup$ If the scheme got validated for its reliability in this way .Will that reliability stay valid in cases of increased supercell size or cutting of surface from bulk ? $\endgroup$ Apr 5 at 9:13
  • 3
    $\begingroup$ @Joyalsunny If you have perfect periodicity in your system, then the supercell size should not be a problem. The results that you will get in the supercell should be some coefficient times the result from the unit cell. But if you have aperiodicity in your system i.e. some isolated defects, doping etc., then you should converge the supercell size too and consider some kind of correction scheme (mp or mt). For surface, I never personally did this but I guess it should not work in general. You need to set a lot of vacuum to make the surface and the amount of vacuum should be converged too. $\endgroup$ Apr 5 at 9:20
  • 4
    $\begingroup$ +1. I suggest converging wrt. forces and pressure since I usually use DFT calculate phonons, which require converging those properties very tightly. One could probably get away with converging only wrt. energy for e.g. structural relaxation, but note that if your pressure is only converged within an amount dP, the you wont be able to relax the unitcell to a target pressure any better than P +- dP. Same for forces! $\endgroup$ Apr 5 at 16:28

You must log in to answer this question.