next up previous contents
Next: Coulomb Interactions in Supercell Up: Germanium and Silicon - Previous: Germanium and Silicon -

Choice of Slater Determinant

To determine the optimal choice of tex2html_wrap_inline6653 for diamond-structure germanium, a series of LDA calculations were performed using tex2html_wrap_inline6593 -point meshes with different offsets from the origin[33]. The offset from the origin determines the value of tex2html_wrap_inline6653 of the Slater determinant constructed from the orbitals. The results of these calculations are illustrated in figure gif. The results show the convergence of the total energy with supercell size for four different values of tex2html_wrap_inline6653 . For the tex2html_wrap_inline6701 -point ( tex2html_wrap_inline6653 =0) the energy converges slowly as the size of the supercell is increased. A far more rapid convergence is observed when L-point sampling is used. This means the tex2html_wrap_inline6593 -point grid is offset from the origin by tex2html_wrap_inline6707 , where tex2html_wrap_inline6709 and the tex2html_wrap_inline6711 are the primitive reciprocal lattice translation vectors of the supercell. The best convergence is obtained when using an offset equal to the Baldereschi mean value point[53] of the supercell BZ.The Baldereschi mean point was not chosen for the QMC calculations, as wavefunctions at that tex2html_wrap_inline6593 -point are necessarily complex and the use of complex arithmetic would slow down the code considerably. It is possible to construct a real wavefunction using Baldereschi-point states by taking a linear combination of tex2html_wrap_inline6715 and its complex conjugate, tex2html_wrap_inline6717 , as both tex2html_wrap_inline6715 and tex2html_wrap_inline6717 have the same energy eigenvalue. Unfortunately, this combination is no longer a Bloch function, and this results in a local energy function that does not have the periodicity of the supercell.

Although it would be too computationally expensive to repeat all the LDA calculations from figure gif within the QMC formalism, VMC and DMC calculations for supercells with n = 2 and 3 [33] show that QMC calculations follow the LDA trend very closely. The calculations described in chapters gif and gif use Slater determinants where the one-electron orbitals have tex2html_wrap_inline6593 -points chosen on a mesh offset from the origin at the L-point. This corresponds to Monkhorst-Pack (MP)[54] sampling for n even supercells and better than MP sampling for n odd supercells.

  
Figure: Convergence of the total energy with simulation cell size for different tex2html_wrap_inline6593 -point sampling schemes. The black line shows the results of tex2html_wrap_inline6735 sampling. The green line shows tex2html_wrap_inline6737 sampling, where the tex2html_wrap_inline6711 are the primitive reciprocal lattice translation vectors of the supercell. The red line shows tex2html_wrap_inline6741 sampling, and the blue line shows tex2html_wrap_inline6743 sampling, where tex2html_wrap_inline6745 is the Baldereschi mean value point.


next up previous contents
Next: Coulomb Interactions in Supercell Up: Germanium and Silicon - Previous: Germanium and Silicon -

Andrew Williamson
Tue Nov 19 17:11:34 GMT 1996