Density functional theory with fractional orbital occupations

In contrast to the original Kohn-Sham (KS) formalism, we propose a density functional theory (DFT) with fractional orbital occupations for the study of ground states of many-electron systems, wherein strong static correlation is shown to be described. Even at the simplest level represented by the local density approximation (LDA), our resulting DFT-LDA is shown to improve upon KS-LDA for multi-reference systems, such as dissociation of H2 and N2, and twisted ethylene, while performing similarly to KS-LDA for single-reference systems, such as reaction energies and equilibrium geometries. Because of its computational efficiency (similar to KS-LDA), this DFT-LDA is applied to the study of the singlet-triplet energy gaps (ST gaps) of acenes, which are “challenging problems” for conventional electronic structure methods due to the presence of strong static correlation effects. Our calculated ST gaps are in good agreement with the existing experimental and high-level ab initio data. The ST gaps are shown to decrease monotonically with the increase of chain length, and become vanishingly small (within 0.1 kcal/mol) in the limit of an infinitely large polyacene. In addition, based on our calculated active orbital occupation numbers, the ground states for large acenes are shown to be polyradical singlets.

💡 Research Summary

The paper introduces a novel density‑functional‑theory (DFT) framework that allows fractional occupations of Kohn‑Sham (KS) orbitals, addressing a long‑standing deficiency of conventional KS‑DFT in describing strong static correlation. In the standard KS formalism each orbital is constrained to be either fully occupied (occupation = 1) or empty (occupation = 0). This binary restriction works well for single‑reference systems but fails for multi‑reference cases where several configurations contribute equally to the ground state, such as bond dissociation, twisted ethylene, and extended π‑conjugated systems. The authors propose to relax this constraint by treating the orbital occupations as continuous variables that are optimized together with the KS orbitals under the usual particle‑number constraint. The resulting equations are derived from a variational principle that includes Lagrange multipliers for both the orthonormality of the orbitals and the total‑electron‑number condition.

Implementation is deliberately kept simple: the exchange‑correlation functional is taken to be the Local‑Density Approximation (LDA), exactly as in conventional KS‑LDA. The only modification is that the Kohn‑Sham potential now depends on the set of fractional occupations, and the self‑consistent field (SCF) cycle updates both the orbitals and their occupations until convergence. Because the functional form of the energy remains unchanged, the computational scaling stays at O(N³), identical to ordinary KS‑LDA, and the method can be applied to large molecules without a prohibitive cost increase.

The authors first benchmark the fractional‑occupation DFT‑LDA (FO‑DFT‑LDA) on prototypical multi‑reference problems. For H₂ and N₂ dissociation, the method yields smooth occupation changes from 1 to 0 as the bond is stretched, reproducing the correct dissociation limit and eliminating the artificial “fractional‑spin” error that plagues KS‑LDA. In twisted ethylene, where the π‑bond rotation creates near‑degenerate configurations, FO‑DFT‑LDA correctly predicts the barrier height and the gradual mixing of the two configurations, whereas KS‑LDA severely underestimates the barrier.

In contrast, for single‑reference thermochemical data (reaction energies, equilibrium geometries of a diverse test set) the performance of FO‑DFT‑LDA is essentially indistinguishable from that of KS‑LDA, confirming that the added flexibility does not degrade accuracy when static correlation is weak.

Having demonstrated comparable accuracy and unchanged computational cost, the authors apply FO‑DFT‑LDA to a challenging class of systems: acenes (linear poly‑benzenes) of increasing length. Acenes are known to develop strong static correlation as the number of fused benzene rings grows, making the singlet‑triplet energy gap (ST gap) a stringent test for any electronic‑structure method. The calculated ST gaps decrease monotonically with chain length, approaching a value smaller than 0.1 kcal mol⁻¹ in the limit of an infinite polyacene. This trend matches experimental extrapolations and high‑level ab‑initio results obtained with methods such as CCSDT, DMRG, and FCIQMC.

A further insight comes from the analysis of the natural‑orbital occupation numbers obtained in the SCF solution. For short acenes the occupations are close to 2 (fully occupied) or 0 (empty), but as the chain length increases several frontier orbitals acquire occupations near 1.0 – 1.5, i.e., roughly 0.5 per spin. This pattern signals the emergence of a polyradical singlet ground state: many electrons become effectively unpaired yet remain overall spin‑singlet due to antiferromagnetic coupling. The authors thus provide a clear electronic‑structure picture of the transition from a closed‑shell singlet to a multi‑radical singlet as the π‑system expands.

In summary, the paper demonstrates that allowing fractional orbital occupations within a conventional LDA functional yields a method that (i) captures static correlation in multi‑reference systems without sacrificing the efficiency of standard KS‑LDA, (ii) retains high accuracy for ordinary single‑reference chemistry, and (iii) offers a physically transparent description of polyradical character in extended conjugated molecules. The approach opens a practical route to treat strongly correlated electrons in large molecular systems, bridging the gap between inexpensive DFT and expensive multi‑reference wave‑function techniques.