An assessment of initial guesses for self-consistent field calculations. Superposition of Atomic Potentials: simple yet efficient

Susi Lehtola
DOI: https://doi.org/10.48550/arXiv.1810.11659
IF: 2.552
2018-10-27
Chemical Physics
Abstract:Electronic structure calculations, such as in the Hartree-Fock or Kohn-Sham density functional approach, require an initial guess for the molecular orbitals. The quality of the initial guess has a significant impact on the speed of convergence of the self-consistent field (SCF) procedure. Popular choices for the initial guess include the one-electron guess from the core Hamiltonian, the extended H\"uckel method, and the superposition of atomic densities (SAD). Here, we discuss alternative guesses obtained from the superposition of atomic potentials (SAP), which is easily implementable even in real-space calculations. We also discuss a variant of SAD which produces guess orbitals by purification of the density matrix that could also be used in real-space calculations, as well as a parameter-free variant of the extended H\"uckel method, which resembles the SAP method and is easy to implement on top of existing SAD infrastructure. The performance of the core Hamiltonian, the SAD and the SAP guesses as well as the extended H\"uckel variant is assessed in non-relativistic calculations on a dataset of 259 molecules ranging from the first to the fourth periods by projecting the guess orbitals onto precomputed, converged SCF solutions in single- to triple-{\zeta} basis sets. It is shown that the proposed SAP guess is the best guess on average. The extended H\"uckel guess offers a good alternative, with less scatter in accuracy.
What problem does this paper attempt to address?