12
$\begingroup$

When we present our Density Functional Theory simulation results e.g. lattice parameters, stacking fault energies, band gaps, etc. to people who are experimentalists then the very first question which they ask is about the validity of those numbers. Since the calculations are done at ~0 kelvin, they start doubting those numbers.
I know that in the absolute terms these numbers may not have that much meaning unless they are compared with some reference.
I want to know the take of Matter Modeling community on this.

$\endgroup$
  • 4
    $\begingroup$ You say they are "doubting the numbers because the calculation was done at 0 Kelvin" but are you sure they are not doubting the numbers because they were not done with the universal functional and in a complete basis set? $\endgroup$ – Nike Dattani Sep 19 at 21:14
  • 7
    $\begingroup$ On the same lines as @NikeDattani's question, you should first ask why you have faith in the numbers. You can't convince collaborators if you haven't convinced yourself. No functional/basis set is perfect. Have you tried others to make sure the key findings still hold? Have you considered temperature effects where appropriate (e.g. adsorption energies)? Have you used the right tool for the job (e.g. hybrid functional for band gaps)? Have you considered other important variables (think to yourself what could be different between theory and experiment). This is different than explaining DFT. $\endgroup$ – Andrew Rosen Sep 20 at 0:07
  • 7
    $\begingroup$ Once you have convinced yourself, you can explain to experimentalists why you have faith in what you have computed. Your experimental colleagues are smart, so walk them through the possible limitations and how you have considered them. It doesn't sound like they misunderstand DFT. Rather, they want to know why you think the calculations are valid, and that's reasonable to ask. Also, you say absolute numbers don't have much meaning. That's absolutely true for energies, but the other properties you reported can be taken at face-value, assuming the calculations are sufficiently accurate. $\endgroup$ – Andrew Rosen Sep 20 at 0:09
  • 4
    $\begingroup$ @AndrewRosen I think those comments can be the start of an excellent answer: something along the lines of "first try other functionals and basis sets and convince yourself that the band gap you're getting is the same no matter what functional/basis-set you're using, then use this to help explain to the experimentalists why you're confident with the numbers". $\endgroup$ – Nike Dattani Sep 20 at 1:57
  • 3
    $\begingroup$ @AndrewRosen both of your comments looks like an answer in themselves. It'll be great if you could elaborate a little bit on those points. I'll be tempted to accept your answer, but I will wait for few more perspectives to this. $\endgroup$ – Sufyan Sep 20 at 9:46
2
+100
$\begingroup$

I had the same doubt regarding the validity of my calculations. In some cases, it was astonishing how certain well-known quantities like the bandgap of Silicon could be easily underestimated by DFT. But as seen in most research papers on DFT calculations, there is a difference in experimental and predicted values. This depends on various factors including the type of pseudopotentials, consideration of spin-orbit coupling etc.. But there are ways to work around such innate differences and observe what works for you. You could use different pseudopotentials, and if that's not showing evident change, you can use codes that implement LAPW methods rather than plane-waves. The idea is not to make your calculations as close to the experimental values as possible, it's just to know if the physical properties predicted are in fact due to the inherent physics of the material. There may be cases where the presence of impurities, measurement error..etc can affect your data. That noise is unremovable. So its better to see if your predicted values are in the range allowed for the parameter in consideration.

I hope it's a viable thought.

| cite | improve this answer | |
$\endgroup$

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.