12
$\begingroup$

Even though it is a powerful tool, Quantum ESPRESSO's hardware requirements are very high. For instance, to evaluate optical properties, even for structures with tens of atoms, Quantum ESPRESSO can require more than 100GB of RAM. What about VASP?

Improve this question
$\endgroup$
5
  • 3
    $\begingroup$ What do you mean by hardware requirements? You can run QE on a laptop and it's been even run on an Android phone. $\endgroup$
    – Kevin J. May
    Nov 23, 2020 at 1:31
  • 1
    $\begingroup$ @KevinJ.M., sure, for very small system. QE is a high RAM dependent. $\endgroup$
    – Camps ♦
    Nov 23, 2020 at 11:31
  • $\begingroup$ Exactly @Camps! To evaluate optical properties, for example, even to structures with tens of atoms, Quantum Espresso can take more than 100Gb of RAM. Therefore running on mobile can not be the case! $\endgroup$
    – Anibal Bezerra
    Nov 23, 2020 at 17:33
  • 1
    $\begingroup$ My point is: as it was, the question could not be answered clearly (Nike has now edited it to make it more specific). Also the RAM requirements depend on other parameters in the calculation, such as how much data is stored in on disk (disk_io). And what specific approach is being used to calculate optical properties here? $\endgroup$
    – Kevin J. May
    Nov 24, 2020 at 0:02
  • $\begingroup$ It's unlikely that VASP and QE would have much difference in RAM utilization since both are based on the plane-wave method. If RAM utilization is to be significantly decreased, other methods such as linear atomic orbitals in the SIESTA package or CP2K with Gaussian type orbitals would be a good recommendation. Maybe there are better alternatives as well. $\endgroup$
    – PBH
    Nov 5, 2021 at 14:46

1 Answer 1

Reset to default
7
$\begingroup$

This question is kind of general. I will outline some points related.

  • Firstly, it depends on the theoretical levels that you use to calculate the optical properties. In general, there are three levels:

    • Independent level;
    • RPA (random phase approximation) level;
    • GW&BSE level;

The first one is memory friendly in terms of the empty bands contained. However, the last two are very memory-consuming even for a small system.

  • Secondly, it depends on how large your system and whether includes the spin-orbit coupling or not.

  • Thirdly, you must test the parameters to obtain converged results. All these parameters are closely related to memory consumption. In VASP, the key parameters are:

    • KMESH
    • ENCUT
    • BANDS & ENCUTGW
    • NBANDSV
    • NBANDSO

For your example with tens of atoms, to obtain converged results I believe 100G memory is not enough in VASP. In summary, you can test it and get some feeling.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Is it normal that a calculation for TENs atoms require more that 100GB of RAM? $\endgroup$
    – Camps ♦
    Nov 25, 2020 at 11:41

You must log in to answer this question.

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