Kernel polynomial approximations for densities of states and spectral functions (Q1913742)

Efficient and numerically stable calculations of the properties of very large Hamiltonians are facilitated by the use of Chebyshev polynomial expansions and approximations. An optimal kernel polynomial (Jackson) is found that enforces the non-negativity of the density of states, preserves normalization and gives the best corresponding energy resolution. The relationship to classical uniform approximation theory is noted. Other kernels and methods in use are discussed and compared. The kernel polynomial method is demonstrated for electronic structure and magnetic susceptibility calculations. The use of Chebyshev polynomial approximation of Fermi projection operators is also proposed.