One of the greatest advantage of density functional schemes under the local density approximation is their reliance on the charge density as the fundamental variable, which greatly simplifies the amount of calculation required. However they consistently underpredict bandgaps (this is to some extent compensated by the use of clusters, see Section 3.7.3 below), and as a ground state theory they are restricted in their application to excited state problems. Various methods have been used to try and overcome this, notably constrained relaxation techniques to determine saddle points (see Section 6). In addition since the wavefunction eigenvalues are a computational tool and do not directly link to experimental eigenvalues, physical interpretation of these states is somewhat dangerous (although there is normally qualitative agreement between the two). Current work suggests Slater transition methods may provide a way of quantitatively relating the two [42].

Density functional theory under the local density approximation is
well known for overbinding of molecules such as O_{2}, but can be
greatly improved through the use of gradient corrections. AIMPRO
actually does better than the standard DFT result, shown in
Table 3.1. These results support the conclusion that we are
obtaining good structural and vibrational data, but need to improve
the accuracy of our total energies. However the magnitude of such
errors normally decreases with system size. The local density
approximation is good for smoothly varying systems, so given that
other DFT/LDA calculations on oxygen/silicon structures have also had
trouble with total energies (see, for example, Chapter 5),
it suggests that part of the problem of innacurate total energies
could be due to the local density approximation.

Source | Binding | O stretch | Bond Length |

Energy (eV) | mode (cm^{-1}) |
(a.u.) | |

AIMPRO, spin | |||

averaged | 8.79 | 1565 | 2.28 |

polarised | 6.41 | 1581 | 2.28 |

Experiment | 5.23 | 1580 | 2.28 |

Quoted DFT | 7.54 | 1610 | 2.31 |

Hartree Fock | 1.43 | 2000 | 2.18 |