8
$\begingroup$

I adapted an example first-principles water simulation to do simulations of liquid SiO2, but did not know what I was doing and did not remove a Grimme dispersion correction term.

It would be great to get some guidance on whether there is likely to be much difference between including and not including a dispersion correction in a non-water system.

It seems like dispersion forces are likely to be minimal in a dense liquid like SiO2 (compared to water) but this is not my forte. Maybe Grimme correction terms act regardless of the system. I have not been able to find any newbie-level guidance online. Could anyone please help me think through this?

$\endgroup$
8
  • 2
    $\begingroup$ It's almost always better to include Grimme's dispersion correction than to not include it, even though for systems governed by covalent or ionic interactions, the benefit may be small or even negligible. And the correction is essentially free. For highly charged systems, however, DFT-D3 and earlier versions of DFT-D may worsen the results, because they do not properly account for the effect of atomic charge on the dispersion interaction. DFT-D4 solves this problem, but not all programs support it. I don't know whether the Si in SiO2 counts as heavily charged, but I guess no. $\endgroup$
    – wzkchem5
    Oct 22 at 10:53
  • $\begingroup$ @wzkchem5, thank you very much for that. I am wondering if I need to redo the simulations, which would be quite an effort. When I started the simulations, my reasoning was also that including it would be better than not including it, although I didn't have actual reasons like you do, being a newbie! What has worried me is that I have recently simulated water at high temperature where using it has led to bad results (very different densities compared to experiments) and NOT using it has led to great results. I wonder though if the water is dissociated and there's a charge like you said. $\endgroup$
    – NTS
    Oct 22 at 10:59
  • 3
    $\begingroup$ You can do a quick test by optimizing the structure of crystalline SiO2 with and without dispersion, and see how much do the lattice constants change. Since liquid SiO2 is basically as dense as solid SiO2 (if I'm not mistaken), this should give you a rough idea of how much the effect of dispersion will be in your simulation, without actually doing a costly liquid simulation $\endgroup$
    – wzkchem5
    Oct 22 at 11:03
  • $\begingroup$ @wzkchem5, that's very helpful advice. Thank you so much! I should be able to figure out how to do that. Really appreciate your comments. I will be looking into the effects of charge. I have been very confused lately about why dispersion correction (DFT-D3) seems to worsen my results for water so will be eager to try DFT-D4 if I can get it. $\endgroup$
    – NTS
    Oct 22 at 11:06
  • 2
    $\begingroup$ @wzkchem5 your most recent comment could probably be expanded to a nice answer. The OP could also eventually add an answer comparing the results with and without dispersion. $\endgroup$
    – Tyberius
    Oct 30 at 0:25
4
$\begingroup$

Following Tyberius's suggestion, I now turn my comment into an answer.

As always, the most reliable way of estimating the effect of the dispersion correction is to actually do two simulations, one with the correction and one without it. But since a liquid simulation is expensive, it's better to test the effect of the correction on a simpler system that is sufficiently similar to your system. The simplest choice is crystalline SiO2, whose density is not far from that of liquid SiO2, but whose structure optimization is very cheap due to the high symmetry. A more accurate choice may be glassy SiO2, with which you lose the advantage of a high symmetry and a small periodic cell, but which is still much cheaper than a liquid simulation, since with a glass you can do structure optimization (quenching) instead of MD.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.