Efficient Iterative TDDFT: Applications to QMD Simulations of Optical Spectra

Here, we present an efficient algorithm to compute the absorption spectra within linear response time-dependent density functional theory (TD-DFT). The algorithm utilizes the locality of underlying basis functions and Krylov-subspace techniques; it is useful with any compact set of localized basis functions such as atomic orbitals, Wannier functions or a general tight-binding representation. The current realization, implemented in our open-source code pySCF_nao (https://github.com/cfm-mpc/pyscf/tree/nao), employs numerical atomic orbitals and assumes the use of pseudo-potentials to describe inner electrons. Thus, it is compatible with popular DFT codes like SIESTA, OpenMX or Fireball. The efficiency of our algorithm allows combining it with molecular dynamics (MD) simulations, where spectra must be averaged over hundreds of configurations to obtain relevant information, e.g., including the effects of temperature on optical properties. In spite of the widespread use of DFT-based MD simulations, its combination with TD-DFT calculations has remained scarce to date due to the computational cost of such calculations. However, here we present some examples of such combination thanks to the efficiency of our iterative TD-DFT method.