ApproximateProjection
Requests a spin-projection correction [Noodleman95, Yamaguchi88, Yamaguchi93, Yamanaka94, Hratchian13, Thompson15, Thompson19] be applied to a broken-symmetry unrestricted HF or DFT wavefunction in order to reduce errors arising from spin contamination. Spin contamination occurs when an unrestricted single-determinant wavefunction mixes contributions from higher spin states, which can lead to significant errors in energies and derived properties, particularly for systems with near-degeneracy or strong static correlation.
Approximate projection (AP) provides an efficient, mean-field-cost approach to mitigate these errors by constructing a corrected energy from a linear combination of a contaminated low-spin (LS) solution and a corresponding higher-spin (HS) solution. The method assumes that the spin contamination is dominated by a single higher-spin state and uses the expectation values of the total spin operator, <S2>for the LS and HS solutions to determine the appropriate mixing. In this two-state formulation, the projected energy recovers the correct spin limit in cases where the system is well described by a small number of interacting configurations.
This approach is widely used for the study of open-shell singlet states, diradicals, and transition metal systems, where broken-symmetry solutions are commonly employed to capture static correlation effects. In such cases, AP often provides improved energies, optimized geometries, and vibrational frequencies relative to the underlying spin-contaminated unrestricted results, while avoiding the cost of multi-reference methods [Sheng20, Taka22].
The AP implementation in Gaussian is limited to a two-state projection scheme in which the user specifies a higher-spin reference state to be projected from a contaminated low-spin solution. While most commonly applied to open-shell singlets, the method is applicable to any contaminated unrestricted state paired with an appropriate higher-spin solution. Typically, the two solutions are calculated prior to the AP calculation, and the two converged SCF solutions are read into the AP calculation using Guess=Input. The facility also requires an additional input section consisting of the number of states (always 2 at present), the charge and multiplicity for the high-spin state, and then the low-spin state, on three separate lines. (see the Examples).
The AP model is formally related to projection-after-variation approaches based on Löwdin projection operators [Lowdin55, Thompson14], and is exact for systems that can be described as two electrons in two orbitals. In more complex cases involving multiple near-degenerate configurations or more than one significant contaminating state, the underlying assumptions of the method may break down, and results should be interpreted with caution.
Energy calculations and geometry optimizations using unrestricted Hartree-Fock and DFT methods.
The following input file illustrates the method for specifying the checkpoint files containing the converged and stability-tested wavefunctions for the high-spin and low-spin states:
# UB3LYP/6-311+G(2d,p) ApproximateProjection Guess=Input ------------------------------------------------------------------- AP single point energy 0 1 C H 1 1.0 H 1 1.0 2 109.0 H 1 1.0 2 109.0 3 109.0 H 1 2.8 2 109.0 3 -109.0 2 0 3 0 1 highSpin.chk lowSpin.chk
The AP input section specifies the number of states as 2 and then indicates that the high-spin state is a triplet and the low-spin state is a singet.
A custom basis set can also be specified using the Gen keyword:
# UB3LYP/Gen ApproximateProjection Guess=Input AP single point with custom basis set 0 1 C … @BasisSet highSpin.chk @BasisSet lowSpin.chk
