Density functional theory
From Academic Kids

Density functional theory (DFT) is one of the most popular approaches to quantum mechanical manybody electronic structure calculations of molecular and condensed matter systems.
Contents 
Description of the theory
Traditional methods in electronic structure, like HartreeFock theory are based on the complicated manyelectron wavefunction. The main objective of density functional theory is to replace the manybody electronic wavefunction with the electronic density as the basic quantity. Whereas the manybody wavefunction is dependent on <math>3N<math> variables, three spatial variables for each of the <math>N<math> electrons, the density is only a function of three variables and is a simpler quantity to deal with both conceptually and practically.
Although density functional theory has its conceptual roots in the ThomasFermi model, DFT was not put on a firm theoretical footing until the HohenbergKohn (HK) theorem which demonstrates the existence of a onetoone mapping between the ground state electron density and the ground state wavefunction of a manyparticle system. Moreover, the HK theorem proves that the ground state density minimizes the total electronic energy of the system. Since the HK theorem holds only for the ground state, DFT is also a ground state theorem.
The HohenbergKohn theorem is only an existence theorem, stating that the mapping exists, but does not provide any such exact mapping. It is in these mappings that approximations are made. The most popular such mapping is localdensity approximation (LDA) which gives an approximate mapping from the density of the system to the total energy. The LDA is exact for the uniform electron gas, also known as jellium.
In practise, the HK theorem is not often used to directly make calculations. Instead, the most common presentday implementation of density functional theory is through the KohnSham method. Within the framework of KohnSham DFT, the intractable manybody problem of interacting electrons in a static external potential is reduced to a tractable problem of noninteracting electrons moving in an effective potential. The effective potential includes the external potential and the effects of the Coulomb interactions between the electrons.
In many cases DFT with the localdensity approximation gives quite satisfactory results in comparison to experimental data at relatively low computational costs when compared to other ways of solving the quantum mechanical manybody problem.
DFT has been very popular for calculations in solid state physics since the 1970s. However, it was not considered accurate enough for calculations in quantum chemistry until the 1990s, when the approximations used in the theory were greatly refined. DFT is now a leading method for electronic structure calculations in both fields.
However, there are still systems which are not described very well by LDA. The LDA fails to describe properly intermolecular interactions, especially van der Waals forces (dispersion). Another popularly cited result is the underestimation of the band gap in semiconductors, but this strictly, does not demonstrate a failure since DFT is a ground state theory, and the band gap is an excited state property.
Early models: ThomasFermi model
The first true density functional theory was developed by Thomas and Fermi in the 1920s. They calculated the energy of an atom by representing its kinetic energy as a functional of the electron density, combining this with the classical expressions for the nuclearelectron and electronelectron interactions (which can both also be represented in terms of the electron density).
Although this was an important first step, the ThomasFermi equation's accuracy was limited because it did not attempt to represent the exchange energy of an atom predicted by HartreeFock theory. An exchange energy functional was added by Dirac in 1928.
However, the ThomasFermiDirac theory remained rather inaccurate for most applications because it is difficult to represent kinetic energy with a density functional, and it neglects electron correlation entirely.
Derivation and formalism
As usual in manybody electronic structure calculations, the nuclei of the treated molecules or clusters are seen as fixed (BornOppenheimer approximation), generating a static external potential <math>\,\!V<math> in which the electrons are moving. A stationary electronic state is then described by a wave function <math>\Psi(\vec r_1,\dots,\vec r_N)<math> fulfilling the manyelectron Schrödinger equation
 <math> H \Psi = \left[{T}+{V}+{U}\right]\Psi =
\left[\sum_i^N \frac{\hbar^2}{2m}\nabla_i^2 + \sum_i^N V(\vec r_i)
+ \sum_{i
where <math>\,\!N<math> is the number of electrons and <math>\,\!U<math> is the electronelectron interaction. The operators <math>\,\!T<math> and <math>\,\!U<math> are socalled universal operators as they are the same for any system, while <math>\,\!V<math> is system dependent or nonuniversal. As one can see the actual difference between a singleparticle problem and the much more complicated manyparticle problem just arises from the interaction term <math>\,\!U<math>. Now, there are many sophisticated methods for solving the manybody Schrödinger equation, e.g. there is diagrammatic perturbation theory in physics, while in quantum chemistry one often uses configuration interaction (CI) methods, based on the systematic expansion of the wave function in Slater determinants. However, the problem with these methods is the huge computational effort, which makes it virtually impossible to apply them efficiently to larger complex systems.
Here DFT provides an appealing alternative, being much more versatile as it provides a way to systematically map the manybody problem, with <math>\,\!U<math>, onto a singlebody problem without <math>\,\!U<math>. In DFT the key variable is the particle density <math>n(\vec r)<math> which is given by
 <math>n(\vec r) = N \int{\rm d}^3r_2 \int{\rm d}^3r_3 \cdots \int{\rm d}^3r_N
\Psi^*(\vec r,\vec r_2,\dots,\vec r_N) \Psi(\vec r,\vec r_2,\dots,\vec r_N).<math>
Hohenberg and Kohn proved in 1964 [1] that the relation expressed above can be reversed, i.e. to a given ground state density <math>n_0(\vec r)<math> it is in principle possible to calculate the corresponding ground state wave function <math>\Psi_0(\vec r_1,\dots,\vec r_N)<math>. In other words, <math>\,\!\Psi_0<math> is a unique functional of <math>\,\!n_0<math>, i.e.
 <math>\,\!\Psi_0 = \Psi_0[n_0]<math>
and consequently all other ground state observables <math>\,\!O<math> are also functionals of <math>\,\!n_0<math>
 <math> \left\langle O \right\rangle[n_0] =
\left\langle \Psi_0[n_0] \left O \right \Psi_0[n_0] \right\rangle.<math>
From this follows in particular, that also the ground state energy is a functional of <math>\,\!n_0<math>
 <math>E_0 = E[n_0] =
\left\langle \Psi_0[n_0] \left T+V+U \right \Psi_0[n_0] \right\rangle<math>,
where the contribution of the external potential <math>\left\langle \Psi_0[n_0] \leftV\right \Psi_0[n_0] \right\rangle<math> can be written explicitly in terms of the density
 <math>V[n] = \int V(\vec r) n(\vec r){\rm d}^3r. <math>
The functionals <math>\,\!T[n]<math> and <math>\,\!U[n]<math> are called universal functionals while <math>\,\!V[n]<math> is obviously nonuniversal, as it depends on the system under study. Having specified a system, i.e. <math>\,\!V<math> is known, one then has to minimise the functional
 <math> E[n] = T[n]+ U[n] + \int V(\vec r) n(\vec r){\rm d}^3r <math>
with respect to <math>n(\vec r)<math>, assuming one has got reliable expressions for <math>\,\!T[n]<math> and <math>\,\!U[n]<math>. A successful minimisation of the energy functional will yield the ground state density <math>\,\!n_0<math> and thus all other ground state observables.
The variational problem of minimising the energy functional <math>\,\!E[n]<math> can be solved by applying the Lagrangian method of undetermined multipliers, which was done by Kohn and Sham in 1965 [2]. Hereby, one uses the fact that the functional in the equation above can be written as a fictitious density functional of a noninteracting system
 <math>E_s[n] =
\left\langle \Psi_s[n] \left T_s+V_s \right \Psi_s[n] \right\rangle,<math>
where <math>\,\!T_s<math> denotes the noninteracting kinetic energy and <math>\,\!V_s<math> is an external effective potential in which the particles are moving. Obviously, <math>n_s(\vec r)\equiv n(\vec r)<math> if <math>\,\!V_s<math> is chosen to be
 <math>V_s = V + U + \left(T_s  T\right).<math>
Thus, one can solve the socalled KohnSham equations of this auxiliary noninteracting system
 <math>\left[\frac{\hbar^2}{2m}\nabla^2+V_s(\vec r)\right] \phi_i(\vec r)
= \epsilon_i \phi(\vec r), <math>
which yields the orbitals <math>\,\!\phi_i<math> that reproduce the density <math>n(\vec r)<math> of the original manybody system
 <math>n(\vec r )\equiv n_s(\vec r)=
\sum_i^N \left\phi_i(\vec r)\right^2. <math>
The effective singleparticle potential <math>\,\!V_s<math> can be written in more detail as
 <math>V_s = V + \int \frac{e^2n_s(\vec r\,')}{\vec r\vec r\,'}
{\rm d}^3r' + V_{\rm XC}[n_s(\vec r)],<math>
where the second term denotes the socalled Hartree term describing the electronelectron Coulomb repulsion, while the last term <math>\,\!V_{\rm XC}<math> is called exchange correlation potential. Here, <math>\,\!V_{\rm XC}<math> includes all the many particle interactions. Since the Hartree term and <math>\,\!V_{\rm XC}<math> depend on <math>n(\vec r )<math>, which depends on the <math>\,\!\phi_i<math>, which in turn depend on <math>\,\!V_s<math>, the problem of solving the KohnSham equation has to be done in a selfconsistent way. Usually one starts with an initial guess for <math>n(\vec r)<math>, then one calculates the corresponding <math>\,\!V_s<math> and solves the KohnSham equations for the <math>\,\!\phi_i<math>. From these one calculates a new density and starts again. This procedure is then repeated until convergence is reached.
Approximations
The major problem with DFT is that the exact functionals for exchange and correlation are not known except for the free electron gas. However, approximations exist which permit the calculation of certain physical quantities quite accurately. In physics the most widely used approximation is the local density approximation (LDA), where the functional depends only on the density at the coordinate where the functional is evaluated:
 <math>E_{XC}[n]=\int\epsilon_{XC}(n){\rm d}^3r.<math>
The local spindensity approximation (LSDA) is a straightforward generalization of the LDA to include electron spin:
 <math>E_{XC}[n_\uparrow,n_\downarrow]=\int\epsilon_{XC}(n_\uparrow,n_\downarrow){\rm d}^3r.<math>
Highly accurate formulae for the exchangecorrelation energy density <math>\epsilon_{XC}(n_\uparrow,n_\downarrow)<math> have been constructed from simulations of a freeelectron gas.
Generalized gradient approximations (GGA) are still local but also take into account the gradient of the density at the same coordinate:
 <math>E_{XC}[n_\uparrow,n_\downarrow]=\int\epsilon_{XC}(n_\uparrow,n_\downarrow,\vec{\nabla}n_\uparrow,\vec{\nabla}n_\downarrow){\rm d}^3r.<math>
Using the latter (GGA) very good results for molecular geometries and ground state energies have been achieved. Many further incremental improvements have been made to DFT by developing better representations of the functionals.
Relativistic generalization
The relativistic generalization of the DFT formalism leads to a current density functional theory.
Applications
In practice, KohnSham theory can be applied in two distinct ways depending on what is being investigated. In the solid state, plane wave basis sets are used with periodic boundary conditions. Moreover, great emphasis is placed upon remaining consistent with the idealised model of a 'uniform electron gas', which exhibits similar behaviour to an infinite solid. In the gas and liquid phases, this emphasis is relaxed somewhat, as the uniform electron gas is a poor model for the behaviour of discrete atoms and molecules. Because of the relaxed constraints, a huge variety of exchangecorrelation functionals have been developed for chemical applications. The most famous and popular of these is known as B3LYP [35]. The adjustable parameters of these functionals are generally fitted to a 'training set' of molecules. Unfortunately, although the results obtained with these functionals are usually relatively accurate for most applications, there is no systematic way of improving them (in contrast to some of the traditional wavefunctionbased methods like configuration interaction or coupled cluster method). Hence, in the current DFT approach it is not possible to estimate the error of the calculations without comparing them to other methods or experiment.
References
[1] P. Hohenberg and W. Kohn, Phys. Rev. 136 (1964) B864
[2] W. Kohn and L. J. Sham, Phys. Rev. 140 (1965) A1133
[3] A. D. Becke, J. Chem. Phys. 98 (1993) 5648
[4] C. Lee, W. Yang, and R. G. Parr, Phys. Rev. B 37 (1988) 785
[5] P. J. Stephens, F. J. Devlin, C. F. Chabalowski, and M. J. Frisch, J. Phys. Chem. 98 (1994) 11623
Literature
 Klaus Capelle, A bird'seye view of densityfunctional theory (http://arxiv.org/abs/condmat/0211443)de:Dichtefunktionaltheorie
es:Teoría del Funcional de la Densidad fr:Théorie de la fonctionnelle de la densité ko:밀도범함수이론 ja:密度汎関数理論 pl:Teoria funkcjonałów gęstości