[Wien] A basic problem about KGEN(shift k-mesh?)

[Peter Blaha]

Of course, different k-meshes will give a different total energy, until the
mesh is "converged", i.e. dense enough so that the integration over the BZ
gives identical results. (But your energies are not even well converged within
the scf cycle). If you want E-tot with higher precision you need to increase the
k-mesh until nothing changes within the desired accuracy.

A few remarks:
A shifted k-mesh usually converges "faster", i.e. needs a less dense k-mesh to
converge.

Usually it is NOT necessary to increase the k-mesh until E-tot does not change.
In particular the electron density in the scf-cycle and the forces (in a structure
optimization) converge (for an insulator !!!) quite fast.
Only whe you compare E-tot between 2 calculations (2 phases,...) you need to be sure
that the numbers are converged.

Usually it is a good strategy to start with a more crude k-mesh, "save_lapw" the results,
"x kgen" creating a denser mesh (more k-points) and continue with another "run_lapw"
(WITHOUT "init_lapw" or "x dstart" !!). When the previous mesh was good, only 3 scf-cycles
will be done until convergence was reached.

Am 23.12.2011 03:20, schrieb W Hx:
> Dear users,
> In "WIEN2k-FAQ: KGEN: Add inversion ? Shift k-mesh ?"
> # *"shift"the k-mesh ? *(only for some lattice types)
>
>   * "Shifting" of k-mesh means that it will add (x,x,x) to all generated k-points, thus shifting them from high symmetry points (lines) to more "general" points with a
>     higher weight. By this procedure (known also as *"special k-point methods"*) one generates an equally "dense" mesh, but with less basis points.
>   * Usually a shift is *recommended *. Only one word of warning: When you are interested in "gaps" of semiconductors, they are often located at Gamma or X (or at some other
>     BZ-border point). With shifted meshes you will NOT have those high-symmetry points in your mesh, thus the gap may seem to be smaller/larger than expected.
>   * Solution to this problem: Do the scf cycle with the shifted mesh, but for a DOS change to a fine unshifted mesh.
>
>  From that we know that shifting of k-mesh generates an equally 'dense' mesh. So we should get nearly the same results for the same case whether shift k-mesh or not.
> I calculated the same case (TiO2 2*2*2 supercell) with shift k-mesh and no shift k-mesh. But the ENE are different.
> grep :ENE 222.scf_mini (shift k-mesh? Yes)
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63765571
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63322019
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63804051
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63815033
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63855011
> grep :ENE 222.scf_mini(shift k-mesh? No)
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63642529
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63182990
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63679565
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63690506
> :ENE : ********** TOTAL ENERGY IN Ry = -32144.63730723
> The energies are different widely. I can not understand.
> For other case i minimized the struct with shift k-mesh and then run a cycle with no shift k-mesh, the energy may be larger or smaller. And the forces are not smaller than 2.
> Can you explain it for me? Thank you for your time and patience.
> yours sincerely,
> Hongxia Wang
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien

-- 
-----------------------------------------
Peter Blaha
Inst. Materials Chemistry, TU Vienna
Getreidemarkt 9, A-1060 Vienna, Austria
Tel: +43-1-5880115671
Fax: +43-1-5880115698
email: pblaha at theochem.tuwien.ac.at
-----------------------------------------