[DFTB-Plus-User] Question regarding Brilloin-zone sampling

[Hofer, Thomas]

Dear colleagues,


I have got one curious question for which I could not find the answer in the manual and the DFTB+ recipes.


Currently we are testing DFTB+ for the treatment of a class of porous solid-state systems, thereby varying the settings for the Brillouin-zone sampling down until pure gamma point sampling. The weird behaviour is observed for the settings with kmax=1 vs gamma-point sampling. For our cubic systems we use the following setups:


KPointsAndWeights = SupercellFolding {
    1 0 0
    0 1 0
    0 0 1
    0.0 0.0 0.0
}


and


KPointsAndWeights = {
  0.0 0.0 0.0 1.0

}


In both cases the output of dftb shows the following entry:


[...]

K-points and weights:           1:  0.000000  0.000000  0.000000     1.000000

K-points in absolute space:     1:  0.000000  0.000000  0.000000

[...]


implying that Brillouin-zone sampling is identical in both cases.


When executing a single point energy + force calculation of the same input structure indeed the very same energy, forces and partial charges are obtained, supporting this conclusion (possibly the stress tensor is different, but I did not look into this).


When executing a molecular dynamics simulation (with our in-house MD code interfaced to dftb+) using 100% identical settings aside from "KpointsAndWeights", the simulations initially show identical progression as expected.


However, after approx. 3000 MD steps (6 ps) the simulations show divergence, although the initial assessment indicated that both levels should yield identical results. In case of gamma-point sampling the porous structure of the system collapses, whereas  SupercellFolding at kmax=1 provides good results compared to the experimental reference.


Since the only difference in these simulations are the forces obtained from dftb+ (only the forces enter the equations of motion), I am in need for a explanation to describe the observed differences for an upcoming manuscript. Is it possible, that the stress tensor calculated by dftb+, which is used to determine the pressure in MD, is calculated differently by these two settings? This would provide an explanation for the observed differences.


Even more curiously, we did the very same investigation for a different system of the same material class. In this case kmax=1 and gamma-point sampling both yielded effectively the same results down to the overall computing time of the MD runs (as expected).

Any suggestion or pointer in this matter is highly appreciated.

Many thanks and best wishes,
Thomas


/---------------------------------------------------------------------------------------------------------/
assoc. Prof. Dr. Thomas Hofer
Department for Theoretical Chemistry
Institute of General, Inorganic and Theoretical Chemistry
Innrain 80-82
University of Innsbruck
6020 Innsbruck
Austria
Phone: +43-512-507-57111
Email: t.hofer at uibk.ac.at
/---------------------------------------------------------------------------------------------------------/



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20220202/83e0b442/attachment.htm>