General Question: EPW vs. YAMBO vs. ABINIT

[chrisewolf]

Dear all,

I was reading a bit about the similarities and differences of QE/EPW and QE/YAMBO as well as ABINIT after coming across S. Ponce et al. arXiv:1309.0729v1. I was quite surprised to see that QE/YAMBO was used in that work and not QE/EPW.

The difference between ABINIT and QE/YAMBO as well as QE/EPW in this case is clearly the "Sternheimer formalism" used in ABINIT to avoid summation over a large number of empty bands. Further, if I understand this correctly, EPW only calculates the Fan contribution to the self energy and not the Debye-Wallner term (i.e. ZPM cannot be calculated with EPW)? Both EPW and YAMBO apparently allow for "random grids" in the integration of the electron-phonon self energy.

Last but not least EPW goes through Wannier interpolation which allows to use a rather coarse grid for the initial calculations which are then interpolated on the "fine grid" (nkf, nqf).

I hope that I was correct so far and I am curious what else sets them apart?

Thanks for your contribution in advance, your time is very much appreciated!

Best,
Chris

[sponce]

Hello Chris,

Let me try to answer that:
1) EPW was not used in S. Ponce et al. arXiv:1309.0729v1 because it was not available to me at that time

2) QE/YAMBO and ABINIT have very similar functionalities:
- computes Fan and DW contribution which allow to compute ZPR, temperature dependence, scattering rates etc (basically all properties associated with real and imaginary part of el-ph self-energy).
- can both use "random grid". Actually some quantities needs to be computed on homogeneous grids but one can always perform nscf calculations to arbitrary k-points and DFPT allow for arbitrary q-point by construction.
- the main difference between the two is indeed the fact that ABINIT can use Sterheimer to avoid summation on empty states.

3) EPW computes Fan only terms. It is actually very simple to compute DW terms in EPW but is not done because it would requires a summation over a lot of empty states to be converged which cannot be done easily because of the Wannierzation (very difficult and expansive to Wannierize hundreds of empty states accurately).
Therefore you are correct that EPW should not be use to compute ZPR.
EPW rely on Wannier interpolation to very efficiently sample k/q grids.
EPW should be use to compute quantities related to the imaginary part of the el-ph self-energy.

In conclusion:
- Abinit or Yambo are good for ZPR, temperature-dependence, scattering rates etc.. but requires very expansive calculations to converge the q-points integrals since you need a direct DFPT calculations for each points.
- EPW is good for the scattering rates, mobility, superconductivity etc and is very cheap ( in comparison ) but has the drawback of not being able to do ZPR.

Abinit and Yambo can do more but are computationally more expansive. Also (to my knowledge), quantities that requires k-integrals in additional to q-integrals (like mobilities) are not implemented in those codes since it would be prohibitively expansive.