Pole-based approximation of the Fermi-Dirac function (Q1047074): Difference between revisions

The authors study a system where the finite temperature single-particle density matrix of the system is given by the Fermi operator \[ \rho=2(1+\exp{(\beta (H-\mu))})^{-1}=1-\tanh{({\beta\over 2}(H-\mu))}, \] where \(H\) is an effective one-particle Hamiltonian, \(\beta=1/(K_{\text{B}}T)\) is the inverse temperature and \(\mu\) the chemical potential. To find accurate and numerical efficient representations of this operator, the authors introduce two alternative rational expansions that have computational cost that scales logarithmical. 1. Use a contour integral and conformal mapping (proposed recently in [\textit{N. Hale, N. J. Higham} and \textit{L. N. Trefethen}, SIAM J. Numer. Anal. 46 (5), 2503--2523 (2008; Zbl 1176.65053)]). Two cases are considered: the Hamiltonian has a gap in its spectrum around \(\mu\) or not. The method uses stepwise application of different conformal mappings. 2. Use of multipole expansion (proposed recently in [\textit{L. Lin, J. Lu, L. Ying, R. Car} and \textit{W. E}, Phys. Rev. B, 79 (11), 115--133 (2009; Zbl 1182.65072)]). The method uses the Matsubara representaion and the fast multiple pole method with dyadic grouping of poles. The results are given without proofs.