9
$\begingroup$

I'm using SCAN-RVV10 to optimize the hexagonal-BN but the optimized geometry doesn't look true as the calculated interlayer distance is far smaller than the experimental value(exp=3.35 Aangstrom, calc=3.20 Angstrom). (SCAN-RVV10 is known as one of the best methods which take into account missing van der Waals interactions.)

My input parameters:

EDIFF = 1E-6
EDIFFG = -1E-2

ISMEAR = 0 ; SIGMA = 0.05

ENCUT = 600

NSW = 100
ISIF = 3
IBRION = 2

METAGGA  = SCAN
LUSE_VDW = .TRUE.
BPARAM = 15.7
LASPH = .TRUE.

NPAR = 32
KPAR = 16

What am I doing wrong? Is there any published interlayer distance for bulk hBN and/or Graphite interlayer distance calculated using SCAN-RVV10?

$\endgroup$
5
  • $\begingroup$ At what temperature are the experimental lattice parameters measured? $\endgroup$
    – ProfM
    Mar 5 at 16:49
  • $\begingroup$ Probably at room temperature. I got your point but other dispersion correction methods (D3, TS, etc) also find close value to the experimental value. $\endgroup$
    – Savir
    Mar 5 at 17:00
  • $\begingroup$ I have now added an answer. $\endgroup$
    – ProfM
    Mar 5 at 18:45
  • $\begingroup$ @Savir Your system is two-dimensional? If so, you can't use the ISIF=3 strategy. $\endgroup$
    – Jack
    Mar 9 at 12:35
  • $\begingroup$ I recently compared a huge variety of DFT methods for predicting 0 K lattice constants of lithium halides (dispersion turns out to play a big role here). SCAN-rVV10 was not the best, TMTPSS-rVV10 was most accurate for both lattice energies and lattice constants. You might consider trying some alternative functionals. $\endgroup$ Mar 13 at 22:43
9
$\begingroup$

Imagine you have a (hypothetical) method that can exactly capture the van der Waals interaction. Then when you do a geometry optimization, you are neglecting the quantum and thermal fluctuations of the atoms, and you end up with what is called the static lattice approximation. As quantum and thermal fluctuations tend to expand the lattice, then you would expect that an exact method would predict a geometry that underestimates the experimental lattice parameters. This is true if experiments are performed at room temperature, but it is also true even if they are performed at zero temperature because of quantum fluctuations.

Additionally, layered van der Waals compounds tend to have a relatively large thermal expansion coefficient along the stacking direction, so for these compounds you would expect an even larger underestimation between an exact van der Waals calculation and experiment.

Putting this together, if you have an approximate van der Waals method that exactly reproduces the experimental lattice parameter, then the van der Waals approximate method is in fact not that good because when you added the quantum and thermal fluctuations, you would end up overestimating the lattice parameter. On the other hand, if you have a van der Waals approximate method that slightly underestimates the lattice parameter, then that has a good chance of being a better approximation, because adding quantum and thermal fluctuations would take you closer to the experimental value. To confirm this, you should calculate the effects of thermal expansion, for example using the quasiharmonic approximation.

I have no experience in SCAN-RVV10, but underestimating the lattice parameter by 0.15 Angstrom (I am guessing the units here because you did not provide them), actually sounds pretty good.

$\endgroup$
1
  • $\begingroup$ I'll note that if you can find measured linear coefficients of thermal expansion for your system, then you can integrate these from 0 K up to the experimental temperature of lattice parameter measurement to get an approximation to the thermal expansion of the experimental lattice parameters. This won't get the zero-point vibrational expansion, but I expect this will be relatively small. $\endgroup$ Mar 13 at 22:47

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.