> Dear Siesta User, > > I am making some calculation on a slab. I enter a Monkhorst Pack grid > that generates 40 k points and it is suggested to think of > shifting this k-point by the program as the following > lines from the .out file suggest it. > I am a bit puzzled by this k-point shifting. To my knowledge this > shift is introduced in a paper of Moreno and Soler.
Dear Alexandre, they were probably not the first to introduce it - the idea seems to have a long tradition - but in any case they describe it quite well. > The shifted k-grid should give > the same results as the non shifted, is not it? A priori not at all, why? At least not exactly. One can argue about "the same results" in the sense that when you converge your results with the number of k-points, it shouldn't ultimately matter whether the k-mesh was shifted or not. And for a given density of k-mesh (number of divisions), you'd have less points in the shifted mesh that in the unshifted one. So the shifted mesh is often slightly more economic, that's all. > Some earlier test did > not show this trend. > Would you follow the recommandation of the out file to add the shift? As a rule of thumb - do it for a system with the gap, and be more careful (make better tests) in case of a metal... Best regards Andrei Postnikov
