mailhexu
commented
4 years ago Hi Nick,
I calculate the spin density for non-colinear case in this way: $Gk(z)=(zI-Hk)^-1$, where Hk is from sisl. nk=-1/\pi \int_{-\infty}^{Ef} Sk Gk(z) note that I put the Im to later. n is the integration of nk over the BZ. The submatrix ni of each orbital in n(z) is then a 22 matrix. Mi_y = Im (ni[0,1](1j)+ni[1,0](-1j))/2. The M_y calculated from this has the opposite sign to the siesta result. The M_x and M_z has no problem.
In Sisl I find that this function is applied to the Hamiltonian, which seems to be where the convention is defined? Can I find the documentation of the siesta convention somewhere?
Thanks!
def _mat_spin_convert(M, spin=None): """ Conversion of Siesta spin matrices to sisl spin matrices The matrices from Siesta are given in a format adheering to the following concept: A non-colinear calculation has the following entries (in C-index) for the sparse matrix: H[:, [0, 1, 2, 3]] H11 == H[:, 0] H22 == H[:, 1] H12 == H[:, 2] - 1j H[:, 3] # spin-box Hermitian H21 == H[:, 2] + 1j H[:, 3] Although it really does not make sense to change anything, we do change it to adhere to the spin-orbit case (see below). I.e. what Siesta saves is the -Im[H12], which we now store as Im[H12]. A spin-orbit calculation has the following entries (in C-index) for the sparse matrix: H[:, [0, 1, 2, 3, 4, 5, 6, 7]] H11 == H[:, 0] + H[:, 4] H22 == H[:, 1] + H[:, 5] H12 == H[:, 2] - 1j H[:, 3] # spin-box Hermitian H21 == H[:, 6] + 1j H[:, 7] """ if spin is None: if M.spin.is_noncolinear: M._csr._D[:, 3] = -M._csr._D[:, 3] elif M.spin.is_spinorbit: M._csr._D[:, 3] = -M._csr._D[:, 3] elif spin.is_noncolinear: M._D[:, 3] = -M._D[:, 3] elif spin.is_spinorbit: M._D[:, 3] = -M._D[:, 3]
Best wishes, HeXu
zerothi
Hi He Xu! :) You mentioned something about the convention, now I am intrigued :)