I am trying to determine the optical properties of bulk Aluminium and related alloys using Quantum ESPRESSO. I am using both the epsilon.x and the simple.x modules to get wavelength-dependent dielectric functions. However, when comparing to experimental data, the results are not in good agreement, even trying to adjust the inter and intra-smearing parameters. I simulated and compared the results for Gold and Silver alloys with no major problems (results agreed very well with the experimental ones), but with the Aluminium, I faced lots of difficulties. In the literature I found

From a convergence study on the dielectric function, we decided to employ an interpolation k-grid of 64x64x64 and η = γ = 0,1 eV in SIMPLE for each elemental metal considered, with the exception of elemental aluminum for which, because of very slow convergence of ε_inter with respect to k-point sampling...

Are there any other options beyond increasing the k-sampling and the number of unoccupied bands, to achieve better results?

  • 3
    $\begingroup$ +1. But what is the question: "why is aluminum difficult?" or "Is there anything I can do other than icnreasing the k-sampling and number of unoccupied bands?" If you want answers to both questions I think you have to ask both questions (separately). $\endgroup$ Sep 28, 2020 at 20:11
  • 1
    $\begingroup$ Hi @NikeDattani, I guess both... I gonna rephrase the question and add a new, exactly as you proposed! Thanks. $\endgroup$ Sep 29, 2020 at 17:13
  • 1
    $\begingroup$ @AnibalBezerra I would suggest the BerkeleyGW package for your calculation, in which an interpolation tool is provided to obtain the optical properties with dense k-mesh. $\endgroup$
    – Jack
    Apr 14, 2021 at 0:49
  • $\begingroup$ Thanks @Jack I gonna take a look at it! $\endgroup$ Apr 14, 2021 at 17:15
  • 1
    $\begingroup$ @AnibalBezerra were the suggestions by Jack and Phil helpful? $\endgroup$ Oct 19, 2021 at 20:47

1 Answer 1


After a few months working with Aluminum-based structures, I'm confident to respond to my own question. And the answer is NO.

Keeping in mind that I'm using the IPA (independent particle approximation) to evaluate the optical properties since I'm dealing with metals (and the approximation is good enough), Quantum ESPRESSO presents two main packages: epsilon.x and simple.x, both written at IPA level.

The major difference is that simple.x employs the optimal basis method to evaluate a reduced set of basis functions followed by a band interpolation and non-local commutator to speed up the calculations. It allows for an interpolated k-grid with twice as many points per direction.

Even though simple.x enables a denser mesh with respect to epsilon.x, it comes at a cost of lots and lots of ram memory. Even for relatively small supercells with around 50 atoms, it swallows more than 300Gb of ram to perform the calculation for an interpolated mesh of 70x70x70 k-points. Talking to the developers, they said the amount of memory to be the bottleneck of the code, and that lighter algorithms are welcome.

Both epsilon and simples responses are very dependent on the mesh density as well as on the number of unoccupied bands and on the threshold for the self-consistent calculation. For Aluminium, the threshold needed to be as low as 10^-12 Ry. The results agreed very well with the experiments.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .