Information encoding in spherical DFT
Authors:
Sol Samuels,
Chance M. Baxter,
Susan R. Atlas
Abstract:
Spherical density functional theory (DFT) [Theophilou, J. Chem. Phys. 149, 074104 (2018)] is a reformulation of the classic theorems of DFT, in which the role of the total density of a many-electron system is replaced by a set of sphericalized densities, constructed by spherically-averaging the total electron density about each atomic nucleus. In Hohenberg-Kohn DFT and its constrained-search gener…
▽ More
Spherical density functional theory (DFT) [Theophilou, J. Chem. Phys. 149, 074104 (2018)] is a reformulation of the classic theorems of DFT, in which the role of the total density of a many-electron system is replaced by a set of sphericalized densities, constructed by spherically-averaging the total electron density about each atomic nucleus. In Hohenberg-Kohn DFT and its constrained-search generalization, the electron density suffices to reconstruct the spatial locations and atomic numbers of the constituent atoms, and thus the external potential. However, the original proofs of spherical DFT require knowledge of the atomic locations at which each sphericalized density originates, in addition to the set of sphericalized densities themselves. In the present work, we utilize formal results from geometric algebra -- in particular, the subfield of distance geometry -- to show that this spatial information is encoded within the ensemble of sphericalized densities themselves, and does not require independent specification. Consequently, the set of sphericalized densities uniquely determines the total external potential of the system, exactly as in Hohenberg-Kohn DFT. This theoretical result is illustrated through numerical examples for LiF and for glycine, the simplest amino acid. In addition to establishing a sound practical foundation for spherical DFT, the extended theorem provides a rationale for the use of sphericalized atomic basis densities -- rather than orientation-dependent basis functions -- when designing classical or machine-learned potentials for atomistic simulation.
△ Less
Submitted 1 July, 2025;
originally announced July 2025.
