Dear All,<br><br>I am trying to understand how QTL works in performing its spectral decomposition of the density matrix. I could not find QTL - Technical report on the web anywhere so I hope you can help me.<br><br>If we have a d-orbital density matrix, then the matrix is a 5x5 corresponding to matrix elements:<br>
n_mm' = SUM_E<Ef < KS_k | phi_m> < phi_m' | KS_k>.<br><br>Here, KS_k is a Kohn-Sham orbital and phi_m is a predefined orbital. Here with angular momentum m. I presume the quantization axis for m is set by the local rotation matrix? The summation is over energies less than the Fermi energy.<br>
<br>Then if we do a spectral decomposition for the QTL partial density of states we rewrite this matrix as:<br>n_mm' = ( V * D * V^-1 )_mm'<br>Here, D is a diagonal matrix and V is another square matrix whose columns should be the eigenvectors. We are now in a Hilbert space in which the occupations are diagonal.<br>
What I don't understand then is how these new eigenvectors are assigned to be dxy, dxz etc by QTL? Is it via a following projection? With respect to what spatial axis is the z-axis defined (still by the local rotation matrix in the struct file)? <br>
Also, if we set our own z-axis (and maybe x) for QTL does this simply alter the following projection?<br><br>Finally, if it is easy to get the occupation matrix in a diagonal basis, why do codes use the rotationally invariant form?<br>
<br>Many thanks,<br>David.<br>