1 Density functional theory (DFT)

1.1 Introduction
Density functional theory is an alternative to ab initio methods for solving the non-
relativistic, time-independent Schrödinger equation H |Φi = E |Φi. The ab initio ap-
proach is to assume that the wave function of the system may be decomposed into single
electron wave functions. However, for a system with N electrons this means there are 3N
continuous variables necessary to be fitted. Assuming that 3 ≤ p ≤ 10 parameters per
variable are required to yield a fit of reasonable accuray, this means that p3N parameters
must be optimized for an N electron system. Although this is a great oversimplifica-
tion this exponential wall does exist and limits the traditional wave function methods to
molecules with a small number of chemically active electrons: N ≈ O(10). This prob-
lem shall be overcome by DFT, which is expressed in terms of the density n(r) in the
Hohenberg-Kohn formulation and in terms of n(r) and the single-particle wave functions
ψj (r) in the Kohn-Sham formulation.

1.2 Thomas Fermi (TF) theory

Since the early 1920s this theory was used to roughly compute the electronic energy
based on electron density distributions. However, it did not predict chemical binding and
therefore was of almost no use for questions of chemistry and material science. It did,
however, consider interacting electrons moving in an external potential v(r) and provided
an implicit relation between v(r) and the density distribution n(r):

 3/2
3/2 1 2m
n(r) = γ(µ − vef f (r)) ; γ= (1)
3π 2 ~2
n(r′ )
Z
vef f (r) = v(r) + dr′ (2)
|r − r′ |

Here µ is the chemical potential and the first equation is based on the expression
n = γ(µ − v)3/2 for the density of a uniform, non-interacting, degenerate electron gas in
a constant external pressure. The second term in the second equation is the classically
computed electrostatic potential, also referred to as the Hartree potential vH . Despite the
fact that TF theory was a rough solution to the many-electron Schrödinger equation, it
was unclear whether there was a strict connection between them and whether knowledge
of the groundstate density n(r) alone uniquely determined the system. This mystery was
solved by Hohenberg and Kohn.

1.3 The Hohenberg Kohn theorem

The Hohenberg Kohn theorem. The groundstate density n(r) of a bound system of
interacting electrons in some external potential v(r) determines this potential uniquely.
1.3 The Hohenberg Kohn theorem

Proof. The proof is given for a non-degenerate groundstate, but may easily be expanded
to degenerate groundstates as well. Let n(r) be the non-degenerate groundstate density of
N electrons in the potential v1 (r), corresponding to the groundstate P hi1 and the energy
E1 . Then:

Z
E1 = hΦ1 |H1 Φ1 i = v1 (r)n(r)dr + hΦ1 |(T + U)Φ1 i ,

where H1 is the total Hamiltonian corresponding to v1 and T and U are the kinetic
and interaction energy operators. Note that the exeternal potential v1 is a multiplicative
operator with respect to Φ. Now assume that there exists a second potential v2 (r), not
equal to v1 (r) + constant, with groundstate Φ2 , necessarily 6= eiθ Φ1 , which gives rise to
the same n(r). Then

Z
E2 = v2 (r)n(r)dr + hΦ2 |(T + U)Φ2 i .

Since Φ1 is assumed to be non-degenerate, the variational principle (Rayleigh-Ritz

minimal principle) for Φ1 gives rise to the following inequality:

Z
E1 < hΦ2 |H1 Φ2 i = v1 (r)n(r)dr + hΦ2 |(T + U)Φ2 i (3)
Z
= E2 + [v1 (r) − v2 (r)]n(r)dr

Analogously

Z
E2 ≤ hΦ1 |H2 Φ1 i = E1 + [v2 (r) − v1 (r)]n(r)dr (4)

where ≤ is used since non-degeneracy of Φ2 was not assumed. Adding equations (3)
and (4) leads to the contradiction:

E1 + E2 < E1 + E2

This shows that the assumption of the existance of a second potential v2 (r), which is
unequal to v1 (r) + constant but yet yields the the same n(r), must be wrong.

Since n(r) determines both N and v(r) it gives us the full H and N for the electronic
system. Therefore n(r) implicitely determines all properties of the system derivable from
solving the Schrödinger equation for H.
1.3 The Hohenberg Kohn theorem

1.3.1 N- and v-representability

A question that may arise immediately linked to the HK theorem is how one knows, af-
ter having minimized a given density, whether this density is a density arising from an
antisymmetric N-body wave function (N-representability) and whether it actually corre-
sponds to the ground-state density of a potential v(r) (v-representability). It has been
shown that any non-negative function can be written in terms of some Φ(r1 , r2 , · · · , rN ),
so N representability is not a problem. However, for v-representability no general solution
exists, but the constrained search algorithm of Levy and Lieb shows that this is irrelevant
for the proof above.

1.3.2 Applications of DFT

• Since v(r) depends on a set of parameters - lattice constants or nuclei positions -
the energy may be minimized with respect to these quantities yielding molecular
geometries and sizes, lattice constants, charge distributions etc.
• By looking at the change in energy with respect to these parameters one may cal-
culate compressibilities, phonon spectra, bulk moduli (solids) and vibrational fre-
quencies (molecules).
• Comparison of the energy of a composite system (e.g., a molecule) with its con-
stituent systems (e.g., individual atoms) gives dissociation energies
• Calculating the total energy with an electron more or less yield electron affinities
and ionization energies. Note: within the local density and generalized gradient
approximations the (N + 1)st electron is too weakly bound or even unbound. The
asymptotic potential obtained decays exponentially and not as 1/r, so the binding
of negative ions is strongly suppressed. Self-interaction corections or other fully
non-local functionals are needed to improve this behavior.
• One can calculate forces from the derivative of the total energy with respect to the
nuclear coordinates using the Hellmann-Feynman theorem.
It is now clear, that in principle it is sufficient to express the total energy of the system
in terms of the density and minimize this functional to obtain the groundstate energy,
denisty and wave function. The total energy of a typical molecule or solid is given by:
Z
E(n) = T (n) + U(n) + V (n) = T (n) + U(n) + n(r)v(r) d3 r (5)

Here T is the kinetic energy of the electrons, U the potential energy due to the in-
teraction between them and V the potential energy from an external potential v, usually
the electrostatic potential of the nuclei, which are taken to be fixed in space (Born-
Oppenheimer approximation). Only this potential is easy to treat since it is a multiplica-
tive operator. For the others a universal density functional is known at this point, but a
possible treatment are the Kohn-Sham equations.
1.4 The Kohn-Sham equations

1.4 The Kohn-Sham equations

The idea of the Kohn-Sham approach is to reintroduce a special type of wave functions
(single particle orbitals) into the formalism, to treat the kinetic and interaction energy
terms. In this approach the energy is taken to be composed of the following terms:

E(n) = T (n) + U(n) + V (n) = Ts {φi (n)} + UH (n) + Exc (n) + V (n) (6)

Here the kinetic energy is split into two contributions:

T (n) = Ts (n) + Tc (n)

Ts stands for the kinetic energy of non-interacting particles of density n with s denoting
’single particle’ and Tc being the remainder with c denoting ’correlation’. The interaction
energy U is approximated by the classical electrostatic interaction or Hartree energy UH .
The new energy term Exc stands for a correction of these approximation due to exchange
(x) and correlation (c) effects, i.e. is given by Exc = (T − Ts ) + (U − UH ) = Tc + (U − UH ).
It is often decomposed as Exc = Ex + Ec , where Ex is the exchange energy due to the
Pauli principle (antisymmetry) and Ec is due to correlations (Tc is then a part of Ec ).

1.4.1 The kinetic energy of non-interacting particles Tc

For non-interacting particles, the kinetic energy is nothing other than the sum of the
kinetic energies of each particle. Written in terms of wave functions that is:

N
~2 X
Z
Ts (n) = − φ∗i (r)∇2 φi (r)d3 r (7)
2m i

Ts (n) = Ts [{φi (n)}] is an explicit orbital functional, but only an implicit density func-
tional, due to the dependence of the wave functions on the density. Because Ts is defined as
the expectation value of the kinetic energy operator T̂ with the Slater determinant arising
from the density n, i.e. Ts (n) = hΦ(n)|T̂ |Φ(n)i, all consequences of antisymmetrization
(exchange) are described by employing a determinal wave function in defining Ts . Hence,
Tc , the difference between Ts and T is a pure correlation effect.

1.4.2 The Hartree energy UH

The classical electrostatic interaction energy (Hartree energy) as well as the mean field
result UH in terms of density is given by

q2 n(r)n(r′ )
Z Z
3
UH = dr d3 r ′ (8)
2 |r − r′ |
1.4 The Kohn-Sham equations

1.4.3 The exchange energy Ex

Since the exchange energy correction stems purely from the interaction potential, it may
be written in terms of single-particle orbitals as:

q2 X φ∗j (r)φ∗k (r′ )φj (r′ )φk (r)

Z Z
Ex [{φi (n)}] = − d r d3 r ′
3
(9)
2 |r − r′ |
jk

where a single term in the sum corresponds to the energy of exchanging molecule j
located at r with molecule k located at r′ . No exact expression in terms of the density
is known. The exchange energy therefore describes the energy lowering due to antisym-
metrization, i.e. the tendency of electrons with like spin to avoid each other, giving rise to
the so-called ’exchange-hole’. It also corrects the Hartree term (8) for the self-interaction
(see eqn. 10).

1.4.4 The correlation energy Ec

The correlation energy is the difference between the full ground-state energy obtained
from the correct many-body wave function and the one obtained from the Hartree-Fock
or Kohn-Sham Slater determinant. Recalling the interpretation of the wave function as
probability amplitude, the product form of the many-body wave function corresponds
to treating the probability amplitude of a many-electron system as a product of the
probability amplitudes of individual electrons (the orbitals). This is only the same, when
the individual electrons are independent. Clearly this is not the case, so such wave
functions neglect the fact, that due to Coulombic interaction, electrons try to avoid each
other. The correlation energy is therefore the additional energy lowering obtained in a
real system due to the mutual avoidance of the interacting electrons. To understand the
correlation for interaction energy, let us take another look at the corresponding operator
written in two perfectly equivalent ways:

q2 q2 n̂(r)n̂(r′ ) − n̂(r)δ(r − r′ )
X Z Z
3
Û = = dr d3 r ′ (10)
i<j
|r − r′ | 2 |r − r′ |

The operator character is only carried by the density operators n̂ (in occupation num-
ber representation) and the term with the delta function subtracts out the interaction of
a charge with itself (which is taken care of by the i¡j in the other case). The expectation
value of this operator U = hΦ|Û |Φi involves the expectation value of a product of den-
sity operators hΦ|n̂(r)n̂(r′ )|Φi, which in the Hartree term (8) is replaced by a product
of expectation values, each of the form n(r) = hΦ|n̂(r)|Φi. This replacement amounts
to a mean field approximation, which neglects quantum fluctuations. Therefore, the cor-
relation energy accounts for the energy lowering due to quantum fluctuations, i.e. the
’correlation hole’, which arises because electrons with unlike spins try to coordinate their
movement to minimize their Coulomb energy. The other significant part of the correlation
energy is due to the difference Tc between non-interacting and interacting kinetic energies.
1.4 The Kohn-Sham equations

1.4.5 Nomenclature and properties regarding Exc

• Since both exchange and correlation tend to keep electrons apart, the term ’electron
hole’ was coined, describing the region of reduced probability for encountering a
second electron in the vicinity of a given reference electron.

• Both exchange and correlation energies give negative contributions, which leads to
an upper bound of 0 for theR energy correction, the lower one given by the Lieb-
Oxford bound: Exc ≥ −1.68 d2 r n(r)4/3

• In the one electron limit Ec (n(1) ) = 0 and Ex (n(1) ) = −UH (n(1) ), where n(1) is a
one-electron density. Note: this is satisfied by HF, but not by standard LDA and
GGA functionals.

1.4.6 The Kohn-Sham equations

To get the groundstate energy, i.e. to minimize eqn. (6) with respect to the density, once
cannot directly minimize with respect to n, since Ts is written as an orbital functional.
Instead, Kohn and Sham proposed the following scheme for indirect minimization. To
minimize E with respect to density means, we want:

δE[n] δTs [n] δV [n] δUH [n] δExc [n] δTs [n]
0= = + + + = + v(r) + vH (r) + vxc (r)
δn(r) δn(r) δn(r) δn(r) δn(r) δn(r)
(11)

As a consequence of eqn. (5) δV δn

= v(r), which is nothing other than the external
potential due to the fixed nuclei, the lattice or a truly external field. The term δUδnH
yields the Hartree potential introduced in eqn. (1). The term δEδnxc can only be calculated
after an approximation has been chosen, but nevertheless we can call the result vxc . Now
consider this brillian idea: look at a system of non-interacting particles moving in the
arbitrary potential vs (r). For this system the minimization condition is:

δEs [n] δTs [n] δVs [n] δTs [n]

0= = + = + vs (r)
δn(r) δn(r) δn(r) δn(r)

The density solving this Euler equation is ns (r). Comparison with eqn. (11) shows
that both minimizations have the same solution ns (r) = n(r), if we choose:

vs (r) = v(r) + vH (r) + vxc (r) (12)

Therefore we can calculate the density of an interacting many-bondy system in the

potential v(r), described by a many-body Schrödinger equation, by simply solving the
equations of a noninteracting single-body system in the potential vs (r), given by:
1.5 Construction of exchange functionals

~2 2
 
− ∇ + vs (r) φi(r) = ǫi φi (r) (13)
2m

This yields orbitals that reproduce the density n(r) of the original system via:

N
X
n(r) = ns (r) = fi |φi (r)|2 , (14)
i

where fi is the occupation of the ith orbital. Equations (12) to (14) are the celebrated
Kohn-Sham equations. They replace the problem of minimizing E(n) (and originally of
solving the corresponding many-body Schrödinger equation) by that of solving a nonin-
teracting Schrödinger equation. The solution to this nonlinear problem is usually found
by starting with an initial guess for n(r), calculating the corresponding vs (r) and then
solving the differential equation 13 for the φi . From these a new density is calculated
using eqn. (14) and the process is restarted until reasonable convergence is reached.

1.5 Construction of exchange functionals

1.5.1 Local density approximation (LDA)
The general idea of LDA is to take the known result for a homogeneous system and apply
it locally to a non-homogeneous system. The exchange energy of a homogeneous system
is known to be:

 1/3
3q 2 3
ehom
x (n) =− n4/3
4 π

So in the LDA one takes:

 1/3 Z
3q 2 3
ExLDA (n) =− n(r)4/3 d3 r (15)
4 π

Expressions for EcLDA (n) are parametrizations of highly precise Quantum Monte Carlo
(QMC) calculations for the electron liquid.

1.5.2 Gradient expansion approximation (GEA)

In this case one tries to systematically calculate gradient corrections of the form |∇n(r)|,
|∇n(r)|2 , ∇2 n(r) etc. to the LDA to take into account the rate of density variation in
the system. The lowest order correction to the LDA exchange energy is given by the
Weizsäcker term:
1.5 Construction of exchange functionals

10q 2 |∇n(r)|2
Z
ExGEA(2) (n) = ExLDA (n) − 3
dr (16)
432π(3π 2)1/3 n(r)4/3

In practice, low-order gradient corrections almost never improve the LDA results and
higher-order corrections are exceedingly difficult to calculate.

1.5.3 Generalized gradient approximation (GGA)

It was a major breakthrough, when it was realized that instead of power-series-like sys-
tematic gradient expansions one could experiment with more general functions of n(r)
and ∇n(r), which need not proceed order by order. Such functional have the general
form:

Z
GGA
Exc (n) = d3 r f (n(r), ∇n(r)) (17)

The most popular GGAs are PBE (denoting the functional proposed in 1996 by
Perdew, Burke and Ernzerhof) and BLYP (denoting the combination of the exchange
functional by Becke and the correlation functional of Lee, Yang and Parr, both in 1988).
GGAs give reliable results for chemical bonds, but mostly fail for van der Waals or dis-
persion interactions.

1.5.4 Hybrid functionals

One of the most popular functionals in quantum chemistry today is B3LYP, which com-
bines the LYP GGA for correlation with Becke’s three-parameter hybrid functional B3 for
exchange. The latter mixes a fraction of Hartree-Fock exchange into the DFT exchange
functional. This mixing involves a certain amount of empiricism and optimization for
selected classes of molecules.

1.5.5 Meta-GGA
These functionals depend, in addition to the density and its derivatives, also on the Kohn-
Sham kinetic-energy density τ (r).

~2 X
τ (r) = |∇φi (r)|2 (18)
2m i

The exchange-correlation energy is therefore written as Exc [n(r), ∇n(r), τ (r)]. The
additional degree of freedom is then used to satisfy additional constraints on Exc , such as
a self-interaction correlation functional, etc. In recent test, Meta-GGAs have performed
favorably even when compared to the best GGAs.