11
$\begingroup$

My Masters in inorganic chemistry is focused in magnetic properties of transition metals; great part of the work is done with DFT methods using softwares like ORCA and NWChem, but since my University has a lack of resources, we are performing computations with not much computational power. That is why we are focusing in results and using functionals like PBE and B3LYP (since many papers are showing good results). Nevertheless, as said by Martin in this thread, today we have "better and more efficient functionals" available, because PBE and B3LYP are very popular, but kind of old. That is why I’d like to ask: what are good, modern, functionals to work with transition metals (mostly Ti…Zn)?

Aditional information:

We are working with systems that have no more than 200 atoms, most of the time, usually dimers.

$\endgroup$
  • 3
    $\begingroup$ You can pass some time reading this question. I hope I can come back to this question soon, but essentially the answer will most likely be: It's not easy. $\endgroup$ – Martin - マーチン Aug 31 '15 at 19:02
  • $\begingroup$ Thank you, Martin, and great text here. I always like your answers. $\endgroup$ – Henrique Junior Sep 1 '15 at 23:09
  • 1
    $\begingroup$ Depends what you mean by 'better'. Getting a 'better answer' for the wrong reasons may be serendipitous and inconsistent. Getting a 'reasonable answer' for a 'better reason' is somewhat preferred by others. For instance, my group likes to use PBE0 for a lot of metals because we simply don't like over-parameterized functionals. In the words of von Neumann, "Give me four parameters and I can fit an elephant." $\endgroup$ – LordStryker Sep 9 '15 at 11:53
  • $\begingroup$ Better for what properties? Also, you realize that many times often the problem is not the functional, right? $\endgroup$ – Greg Sep 15 '15 at 22:09
  • 2
    $\begingroup$ I can't speak for the OP, but I would say the question is about finding reliable functionals that can give accurate ground-state geometries and predict the right spin state. Getting relative thermodynamics is important, and maybe a few comments about common pitfalls (e.g., convergence). $\endgroup$ – Geoff Hutchison Sep 16 '15 at 1:32
6
$\begingroup$

This question is like the snake that eats its own tail. There is no functional good enough for metals. To my limited knowledge, fitting too many parameters -which happens with metals- feels like DFT is not first principle anymore and could be called "empirical" (which is also not correct). You allow a number of parameters to to be fit(1,2,more!!) to a specific system. Those dozens of parameters -at the end- may give you desirable results but you "trained" the set of energies according to your will which is not that realistic. You train your dragon not in the jungle but in a cage. The result is never the same. This "fixing" not only contradicts the spirit of DFT but is also a weak form of empirism. It is like adding more "dark force" to the community. Imagine in what extend many of those "good", modern functionals affect research daily. To end, maybe other comments will come up with accurate functionals about Ti or Zn with papers in good journals etc but without being against DFT, I would just keep a small basket.

$\endgroup$
  • 2
    $\begingroup$ Up-voting for "You train your dragon not in the jungle but in a cage." $\endgroup$ – Todd Minehardt Sep 15 '15 at 22:47
  • $\begingroup$ Having looked through several review articles over the last few days, I disagree with your comment about "no functional good enough for metals." Yes, there are some functionals that have a lot of parameters, but there are clearly modern functionals without "dozens" of parameters, e.g. $\omega$B97-xD $\endgroup$ – Geoff Hutchison Sep 22 '15 at 2:02
4
+100
$\begingroup$

Your best bet with this sort of research is to see what other functionals people have used on similar systems. Try and find validation for the functionals against real world data for similar systems. If you use a functional & basis set which hasn't been benchmarked, for lack of a better term, against the type of system you're modelling and real world data, it becomes hard to trust any results, no matter how computationally expensive. Are you looking at crystals & macromolecules, or individual molecules/dimers? When you talk about a functional, there is also the basis set to be considered.

A paper from Cramer and Truhlar* appears to be well worth a look, its pretty exhaustive in examining DFT and TMs. You are probably better off looking at stuff like this than asking on here, given the limited expertise on here for your specific systems.


* Christopher J. Cramer, Donald G. Truhlar, Phys. Chem. Chem. Phys., 2009, 11, 10757–10816

$\endgroup$
  • 2
    $\begingroup$ Since many people can't access that paper, it would certainly help if you're willing to give a quick summary on Cramer & Truhlar's suggestions. $\endgroup$ – Geoff Hutchison Sep 16 '15 at 1:33
  • 1
    $\begingroup$ I think the OP asked the question because they were looking for some suggestions on what functionals people have used (beyond PBE and B3LYP). Obviously if benchmarking is possible, it's preferable. $\endgroup$ – Geoff Hutchison Sep 16 '15 at 1:34

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.