2007/5/5, Andrei Postnikov <[EMAIL PROTECTED]>:
Well, that's how the inversion works. The BZ is cut in two by c=0 plane, and points with c<0 are removed. For c>0, obviously all a and b are included. Your 48 points are 4x4 planes, taken three times (for c=0, 0.5, 1). In case of shifted mesh, you's have 4x4 takes two times, for c=0.25 and 0.75. (It may be more tricky for non-cubic meshes).
Okay, I get it. The part with the inversion is fine. But still, SIESTA calculates the wavefunctions at, for instance, both (0,0.25,0) and (0,-0.25,0). So now, if I shifted the mesh by 3x0.5, my mesh would be 32 points. There still would be equivalent points there. But! when I shift the mesh by 3x0.25, every single point would have no equivalent among others (I'd have all 64), and the sampling would effectively be twice as dense (although without including Gamma and BZ edge). Which means that instead of running a '4x4x4 shifted by 0.5' calculation with 32 points among which like only 8 are actually needed or an unshifted calculation with 48 k-points (27 different points, is it?), I could take a 3x3x3 mesh with a 0.25 shift and get the same 27 points in the IBZ, all inequivalent (at least, by simple orthogonal symmetry), with a '6x6x6' sampling density. Cheaper and more efficient, wouldn't it be? Best regards, Vasilii
