Dielectric constant of SnSe

Submitted by Vahid on Wed, 07/19/2017 - 15:47


User questions


I have used a 4x12x12 grid, shifted by 0.5 0.5 0.5 for WFN, the same grid with the same shift for k and 0.0 0.0 0.001 for q for WFNq and 4x12x12 with zero shifts for epsilon.inp. I have used two cutoffs for epsilon: 40 and 90 Ry both giving identical results. The input parameters are given in a previous post on this forum (degeneracy_check.x on WFN). The number of bands used is 56.

The plotted dielectric constant, which is the inverse of the Head of the epsilon inverse, as printed in epsilon.out is shown in the attached link.


It seems that epsilon consists of several different curves. Has anyone seen such a behavior and would you know what might merge these curves? Is it a matter of not having enough bands?

I might add that for a cutoff of 90 Ry, the epsmat file produced by the epsilon code was around 250Gb.

Any advice as to how the output can be improved is greatly appreciated.

Thank you,


Vahid Askarpour
Department of Physics and Atmospheric Science
Dalhousie University,
Halifax, NS, Canada

Submitted by Vahid on Thu, 07/20/2017 - 05:09

I realized why I has several curves. I was finding the q on the x-axis by the expression SQRT(q1*q1+q2*q2+q3*q3). Here q's are in reduced coordinates and the lattice is orthorhombic. So I had to transform the q's to cartesian coordinates and then find the length of the q vector.

This actually brings me to another question. SnSe is not isotropic. It was suggested in another post "Yet more questions regarding q/k points and WFN files" that for anisotropic materials, one should apply the small q-shift in all three directions and then average the dielectric constant. That means three separate runs with 0.001 0 0, 0 0.001 0, and 0 0 0.001 shifts for WFNq. Is this correct?