EPW Forum

Dear EPW community,

I have a question regarding the choice of nkf and nqf during the initial calculation phase (before restart). From my experience, when I set nkf and nqf larger than nk and nq, the calculation finishes without errors. However, I’m unsure if this is the correct approach.

My understanding is that in the initial step, a fine nkf/nqf mesh might not be necessary, and that the fine mesh should instead be used during the restart stage. Could someone please clarify:
Thank you in advance for your help.

Best regards,
Ershad

Hi Ershad,

In short, correct, for the "initial" calculation, it does not really matter what your nkf and nqf (they can even be smaller) are as long as your nq and nk corresponds to the electronic (nscf) and phonon (ph.x) calculation you performed. The "restart" calculation is where you evaluate functional properties (e.g. mobility, optics, superconductivity, etc) on a fine grid of nqf and nkf, and that is where you need to make sure to converge them. Only thing to remember is that the k-grid and q-grid needs to be commensurate.

The more detailed version is the following: what EPW really does is that 1) it takes the properties (e.g. electronic energy, matrix elements, etc) from a coarse grid (nk, nq), transform those into the Wannier basis, and 2) transform again from the Wannier basis to a fine Bloch basis (nkf, nqf) to evaluate functional properties. The reason why you see we usually have one "initial" calculation and many times multiple "restart" (or to say "production") calculations is that 1) only needs to be done once, and when things are saved in the Wannier basis, one can simply read those in and 2) can then be done repeatedly for e.g. different grids and different properties.

In this case, the point of the initial calculation is really to finish 1), and that's why the nkf and nqf won't matter here. In fact, there is also no physical reason why one can not simply set nkf and nqf to a large number in the "initial" calculation and let it proceed to evaluate converged properties (and after that one can still do restart calculations because the information in the Wannier basis is already saved). It is just that typically both process can take time; and the convergence of the fine grid needs to be examined for many properties, so naturally one need to run the restart calculation multiple times with different parameters.

Hope these are helpful?
Thanks!
Xiao