2.

Density functional theory

2.1 Introduction

In the first part of this lecture, we have discussed the Hartree-Fock approx-
imation as an approximate solution to the problem of interacting electrons.
It constitutes an effective simple particle theory

h2 2 
* kα ,σ * * *
Hαeff ϕσkα (r) = − ∇ + Veff (r) ϕσkα (r) = εσkα ϕσkα (r) (2.1)
2m
*
with an effective potential that depends on the state ϕσkα (r), i.e. on the
orbital indices kα :
k α ,σ
(r) = Vext (r) + VHartree (r) + Vxkα ,σ (r)
* * * *
Veff (2.2)

Here, the “external” potential is that of the ionic cores

*
X * * X Zn e2
Vext (r) = V(|r − Rn |) = * * , (2.3)
n n |r − Rn |
the Hartree potential arises from the Coulomb interaction of an electron
with the electrostatic potential generated by all the other electrons
Z *0
* e2 3 0 n(r )
VHartree (r) = d r * *0 (2.4)
4πε0 |r − r |
with
D XNe  E Ne
X  2
*  * *  ϕk (* 
n(r) = Φ δ(r − rα )Φ = α
r) (2.5)
α=1 α=1

k ,σ *
The exchange operator Vx α (r) is given by
Z σ *
k α ,σ * e2 X 3*0 σ ∗ *0 σ *0 1 ϕkβ (r)
Vx (r) = − d r ϕkβ (r )ϕkα (r ) * *0 σ * ; (2.6)
4πε0 |r − r | ϕkα (r)
β

32
 * *
it involves an integral over ϕσkα (r) and all other ϕσkβ (r) with the same spin.
The complicated, nonlocal form of the exchange operator is the reason why
the Hartree-Fock equations are difficult to solve for large systems; they are
mostly used in quantum chemistry where molecules with a small number of
atoms and thus of electrons are studied. Solution typically involves N4basis
integrals where Nbasis is the size of the basis set. Hartree Fock also has  a
dε 
well-known unphysical feature which is a diverging velocity vF = dk 
k=kF
at the Fermi surface in all metallic (i.e. ungapped) systems.

Exchange and correlation

The key problem of electronic structure is that the electrons form an inter-
* * * *
acting manybody system whith a wave function Ψ(ri ) = Ψ(r1 , r2 , · · · , rN ).
Since the interactions involve always pairs of electrons, two-body correla-
tion functions are sufficient to determine many properties like for example
hΨ|H|Ψi
the total energy E = hΨ|Ψi = hHi. Explicitly, the joint probability of
* *
finding electrons of spin σ at r and of spin σ 0 at r 0 is given by
DX E
* *0 0 * * *0 * 0
n(r, σ, r , σ ) = δ(r − rα )δ(σ − σα )δ(r − rβ )δ(σ − σβ )
α=β
X Z  * *0 0 * 2
d r3 · · · d rN Ψ(r, σ; r , σ ; r3 , σ3 ; · · · ; rN , σN )
3 3 *
= N(N − 1)
σ3 ,σ4 ,···
(2.7)

for normalized Ψ. For uncorrelated particles, the joint probability is just

the product of probabilities, so that the measure of correlation is

∆n(r, σ; r 0 , σ 0 ) = n(r, σ; r 0 , σ 0 ) − n(r, σ)n(r 0 , σ 0 )

* * * * * *
(2.8)

and thus

n(r, σ, r 0 , σ 0 ) = n(r, σ)n(r 0 , σ 0 ) + ∆n(r, σ; r 0 , σ 0 )

* * * * * *
(2.9)

It is also useful to define the normalized pair distribution

* * * *
* *0 n(r, σ; r 0 , σ 0 )
0 ∆n(r, σ; r 0 , σ 0 )
g(r, σ; r , σ ) = * * = 1 + * * (2.10)
n(r, σ)n(r 0 , σ 0 ) n(r, σ)n(r 0 , σ 0 )
* *
This is 1 for uncorrelated particles; correlation is measured by g(r, σ; r 0 , σ 0 )−
1. All long range correlation is included in the average terms so that the
* * * *
remaining terms ∆n(r, σ; r 0 , σ 0 ) and g(r, σ; r 0 , σ 0 ) − 1 are short range and

33
 * *
vanish at large |r − r 0 |. The Hartree-Fock approximation (HFA) consists of
neglecting all correlations except those required by the Pauli exclusion
principle. The exchange term in the HFA contains the Pauli exclusion
and the self interaction correction, i.e. it cancels a spurious self interac-
tion contained in the Hartree term. Both effects lower the energy which
can be interpreted as the interaction of each electron with a positive ex-
* *
change hole surrounding it. The exchange hole ∆nx (r, σ; r 0 , σ 0 ) is given by
* *
∆n(r, σ; r 0 , σ 0 ) in the HFA, were Ψ is approximated by the single determi-
nant wave function φ; one finds
X 2
* *0 0 * *0 0  σ ∗ * σ *0 
∆nHFA (r, σ; r , σ ) = ∆nx (r, σ; r , σ ) = − δσσ 0  ϕkα (r)ϕkα (r )
i
(2.11)

It is immediately clear that the exchange hole involves only electrons of

the same spin and that the probability for finding two electrons of the
* *
same spin at the same point r = r 0 vanishes (see eq. (2.9)). There are
stringent conditions for the exchange hole: 1) it can never be positive,
∆nx (r, σ; r 0 , σ 0 ) 6 0 (i.e. δx (r, σ; r 0 , σ 0 ) 6 1) and 2) the integral of the
* * * *

* * *
exchange hole density ∆nx (r, σ; r 0 , σ 0 ) over all r 0 is exactly one missing
* *
electron per electron at point r (if one electron is at r, then the same
*
electron cannot be at r 0 ).
The exchange energy can be interpreted as the lowering of the energy due
to each electron interacting with its positive exchange hole,
Z Z
1X
* *0
  3 *
0
3 0 ∆nx (r, σ; r , σ )
Ex = hVint i−EHartree (n) HFA
= d r n(r) d r * *
2 σ |r − r 0 |
(2.12)

Correlation: The energy of a state of many electrons in the Hartree Fock

approximation is the best possible wave function made from a single deter-
minant. Improvement of the wave function to include correlation introduces
extra degrees of freedom in the wave function and therefore always lowers
the energy for any state, ground or excited; this lowering of the energy is
called the correlation energy Ec . (This definition is not the only possible
as a different reference state could be chosen, but it is the one leading to
the smallest possible magnitude of Ec ).
The effects of correlation can be cast in terms of the remaining part of the

34
pair correlation beyond exchange:

∆n(r, σ; r 0 , σ 0 ) ≡ nxc (r, σ; r 0 , σ 0 ) = nx (r, σ; r 0 , σ 0 )+nc (r, σ; r 0 , σ 0 ) (2.13)

* * * * * * * *

As the entire exchange-correlation hole obeys the sum rule that it inte-
* *
grates to 1, the correlation hole nc (r, σ; r 0 , σ 0 ) must integrate to zero, i.e.
it merely redistributes the density of the hole. In general, correlation is
most important for electrons of opposite spin as electrons of the same spin
are automatically kept apart by the exclusion principle.
1

gx

0.8

0.6

0.4

0.2

Figure 2.1: Exchange

*
0 hole gx (r) in the homoge-
0 1 2 3 4
r /rs neous electron gas.

2.2 Foundations of density functional theory

Density functional theory has become the primary tool for calculation of
electronic structure in condensed matter, and is increasingly important for
quantitative studies of molecules and other finite systems. In a famous
1964 paper1 , P. Hohenberg and W. Kohn showed that a special role can
be assigned to the density of particles in the ground state of a quantum
manybody system; the density can be considered as a basic variable, i.e. all
properties of the system can be considered to be unique functionals of the
ground state density. Hohenberg and Kohn formulated density functional
theory as an exact theory of manybody systems of interacting particles in
1
P. Hohenberg and W. Kohn, Inhomogeneous electron gas, Phys. Rev. 136, B 864 (1964).

35
 *
an external potential Vext (r) including any problem of electrons and fixed
nuclei, where the Hamiltonian can be written
h2 X 2 X * 1X e2
Ĥ = − ∇α + Vext (rα ) + * * (2.14)
2m α α
2 |rα − rβ |
i6=j

Density functional theory is based upon the following two theorems first
proved by Hohenberg and Kohn:
Theorem 1: For any system of interacting particles in an external potential
* *
Vext (r), the potential Vext (r) is determined uniquely, except for a constant,
*
by the ground state particle density n0 (r).
Corollary 1: Since the Hamiltonian is thus fully determined except for a
constant shift of the energy, it follows that the manybody wave functions
for all states (ground and excited) are determined. Therefore, all proper-
ties of the system are completely determined given only the ground state
*
density n0 (r).
Theorem 2: A universal functional for the energy E[n] in terms of the den-
* *
sity n(r) can be defined, valid for any external potential Vext (r). For any
*
particular Vext (r), the exact ground state energy of the system is the global
*
minimum value for this functional, and the density n(r) that minimizes
*
the functional is the exact ground state state density n0 (r).
Corollary 2: The functional E[n] alone is sufficient to determine the exact
ground state energy and density. In general, excited states of the electrons
must be determined by other means.
Schematic representation of the Hohenberg-Kohn theorem:
* HK *
Vext (r) ⇐= n0 (r)
⇓ ⇑
* *
Ψi (r) ⇒ Ψ0 (r) (2.15)
Small arrows indicate the usual solution of the Schödinger equation where
* *
the potential Vext (r) determines all states of the system Ψi (r) including
* *
the ground state Ψ0 (r) and ground state density n0 (r). The large arrow
indicates the Hohenberg-Kohn theorem which completes the circle.
Proof of theorem 1: density as a basic variable
We use the expressions:
*
R 3 P * * * *
* hΨ|n̂(r)|Ψi d r2 · · · d3 rN σ |Ψ(r, r2 , r3 , . . . , rN )|2
n(r) = = NR 3 3 * * * * (2.16)
hΨ|Ψi d r1 d r2 · · · d3 rN |Ψ(r1 , r2 , r3 , . . . , rN )|2
36
for the density of particles, with the density operator
N
X
* * *
n̂(r) = δ(r − rα ) (2.17)
α=1

and
Z
hΨ|Ĥ|Ψi
= hĤi = hT̂ i + hV̂int i + d3 r Vext (r)n(r) + EII (2.18)
* *
E=
hΨ|Ψi
for the total energy, where expectation value of the external potential has
been written explicitly as a simple integral over the density function. EII
is the electrostatic nucleus-nucleus interaction. Now we prove theorem 1
by reductio ad absurdum. Suppose that there were two different external
( 1) * (2) *
potentials Vext (r) and Vext (r) which differ by more than a constant and
*
which lead to the same ground state density n(r). The two external poten-
tials lead to two different Hamiltonians Ĥ(1) and Ĥ(2) which have different
ground state wave functions Ψ(1) and Ψ(2) which are hypothesized to have
*
the same ground state density n0 (r). Since Ψ(2) is not the ground state of
Ĥ(1) , it follows that
E(1) = hΨ(1) |Ĥ(1) |Ψ(1) i < hΨ(2) |Ĥ(1) |Ψ(2) i (2.19)

Here a nondegenerate ground state is assumed which simplifies the proof

but is not essential. The last term can be written as
hΨ(2) |Ĥ(1) |Ψ(2) i = hΨ(2) |Ĥ(2) |Ψ(2) i + hΨ(2) |Ĥ(1) − Ĥ(2) |Ψ(2) i
Z
 ( 1) * (2) * 
= E(2) + d3 r Vext (r) − Vext (r) n0 (r) ,
*
(2.20)

so that
Z
 ( 1) * (2) * 
E(1) < E(2) + d3 r Vext (r) − Vext (r) n0 (r) .
*
(2.21)

On the other hand, if we consider E(2) in exactly the same way, we find the
same equation with superscripts 1 and 2 interchanged:
Z
 ( 2) * (1) * 
E(2) < E(1) + d3 r Vext (r) − Vext (r) n0 (r)
*
(2.22)

If we add Eqs. (2.21) and (2.22), we arrive at the contradictory inequality

E(1) + E(2) < E(1) + E(2) . Thus, there cannot be two different external
37
potentials differing by more than a constant which give rise to the same
nondegenerate ground state density; the density uniquely determines the
external potential to within a constant.
The corollary 1 follows since the Hamiltonian is uniquely determined (ex-
cept for a constant) by the ground state density. Then, in principle, the
wave function of any state is determined by solving the Schödinger equation
with this Hamiltonian. Among all solutions consistent with the given den-
sity, the unique ground state wave function is the one that has the lowest
energy. Of course, no prescription has yet been given to solve the problem.
*
Still, the manybody problem in the presence of Vext (r) needs to be solved.
For example, for electrons in materials where the external potential is the
Coulomb potential due to the nuclei, the theorem only requires that the
electron density uniquely determines the positions and types of nuclei.

Proof of theorem 2
The proof of theorem 2 requires proper definition of the meaning of a
functional of the density and restricting the space of densities. We restrict
*
to densities that are V-representable, i.e. densities n(r) that are ground
state densities of the electron Hamiltonian with some external potential
Vext . Within this space of densities, we construct functionals of the density.
Since all properties such as the kinetic energy, etc. are uniquely determined
* *
if n(r) is specified, each such property can be viewed as a functional of n(r),
including the total energy functional
Z
EHK [n] = T [n] + Eint [n] + d3 r Vext (r)n(r) + EII
* *

Z
= FHK [n] + d3 r Vext (r)n(r) + EII
* *
(2.23)

where EII is the interaction energy of the nuclei. The functional FHK [n]
thus defined includes all internal energies, kinetic and potential of the in-
teracting electron system

FHK [n] = T [n] + Eint [n] (2.24)

which must be universal by construction since the kinetic energy and in-
teraction energy of the particles are functionals only of the density. Now
*
consider a system with a ground state density n(1) (r) corresponding to the
(1) *
external potential Vext (r). The Hohenberg-Kohn functional is equal to the

38
expectation value of the Hamiltonian in the unique ground state which has
the wavefunction Ψ(1)

E(1) = EHK [n(1) ] = hΨ(1) |Ĥ(1) |Ψ(1) i (2.25)

*
Now consider a different density n(2) (r) which necessarily corresponds to
a different wave function Ψ(2) . It follows immediately that the energy E(2)
of this state is greater than E(1) since

E(1) = hΨ(1) |Ĥ(1) |Ψ(1) i < hΨ(2) |Ĥ(1) |Ψ(2) i = E(2) (2.26)

Thus the energy given by (2.23) in terms of the Hohenberg-Kohn functional

*
evaluated for the correct ground state density n0 (r) is indeed lower than
*
the value of this expression for any other density n(r). This means that if
the functional FHK [n] were known, then by minimizing the total energy of
*
the system (2.23) with respect to variations in the density n(r) one would
find the exact ground state density and energy. This establishes corollary
2.

2.3 The Kohn-Sham ansatz

The Kohn-Sham approach is to replace the difficult interacting manybody

system with a different auxiliary system that can be solved more easily -
it is an ansatz because there is no unique prescription of how to choose
the simpler auxiliary system. The ansatz assumes that the ground state
density of the original interacting system is equal to that of some chosen
non-interacting system. This leads to independent particle equations for the
non-interacting system that can be considered exactly solvable with all the
difficult manybody terms incorporated into an exchange-correlation func-
tional of the density. By solving the equations one finds the ground state
density and energy of the original interacting system with the accuracy
limited only by the approximations in the exchange-correlation functional.
These approximations, the most important of which are the local den-
sity approximation (LDA) and generalized gradient approximation (GGA)
functionals will be discussed below. The Kohn-Sham ansatz for the ground
state rests on two assumptions:
1) The exact ground state density can be represented by the ground state
density of the auxiliary system of noninteracting particles. This is called
non-interacting-V-representability (see scheme below)
2) The auxiliary Hamiltonian is chosen to have the usual kinetic energy

39
 σ *
operator and an effective local potential Veff (r) acting on an electron of
*
spin σ at point r.
Schematic representation of the Kohn-Sham ansatz:
* HK * KS * HK *
Vext (r) ⇐= n0 (r) ⇐⇒ n0 (r) =⇒0 VKS (r)
⇓ ⇑ ⇑ ⇓
* * * *
Ψi (r) ⇒ Ψ0 (r) ψi=1,··· ,Ne (r) ⇐ ψi (r) (2.27)

HK0 means Hohenberg-Kohn theorem applied to the noninteracting prob-

lem. The KS arrow indicates connection in both directions between the
manybody and independent particle systems.
The actual calculations are performed on the auxiliary independent particle
system defined by the auxiliary Hamiltonian
1
Hσaux = − ∇2 + V σ (r)
*
(2.28)
2
*
V σ (r) will be specified later. For independent electrons, the ground state
*
has one electron in each of the Nσ (σ =↑, ↓) orbitals ψσi (r) with the low-
est eigenvalues εσi of the Hamiltonian (2.28). The density of the auxiliary
system is given by
σ
X N
XX  σ * 2
*
n(r) =
*
n(r, σ) = ψ (r) , (2.29)
i
σ σ i=1

the independent particle kinetic energy Ts is given by

Nσ Nσ Z
1 XX σ 2 σ 1 XX  
* 2
Ts = − hψi |∇ |ψi i = d3 r ∇ψσi (r) (2.30)
2 σ 2 σ
i=1 i=1

and we define the classical Coulomb interaction energy of the electron

*
density n(r) interacting with itself as
Z * *0
1 3 3 0 n(r)n(r )
EHartree [n] = d r d r * *0 (2.31)
2 |r − r |
The Kohn-Sham approach to the full interacting manybody problem is
to rewrite the Hohenberg-Kohn expression for the ground state energy
functional (2.23) as
Z
EKS = Ts [n] + d3 r Vext (r)n(r) + EHartree [n] + EII + Exc [n] (2.32)
* *

40
 *
Here Vext (r) is the external potential due to the nuclei and any other
external fields (assumed to be independent of spin). All manybody effects of
exchange and correlation are grouped into the exchange-correlation energy
Exc . Comparing the Hohenberg-Kohn (2.23) and the Kohn-Sham (2.32)
expressions for the total energy shows that Exc can be written as

Exc [n] = FHK [n] − (Ts [n] + EHartree [n]) (2.33)

or

Exc [n] = hT̂ i − Ts [n] + hV̂int i − EHartree [n] (2.34)

This shows that Exc [n] is just the difference of the kinetic and internal
interaction energies of the interacting manybody system from those of the
fictitious independent-particle system with electron-electron interactions
replaced by the Hartree energy. As the universal functional Exc [n] of (2.32)
is unknown, approximate forms for Exc [n] make the Kohn-Sham method a
valuable approach for the ground state properties of the manybody electron
system.

The Kohn-Sham variational equations

The solution of the Kohn-Sham auxiliary system for the ground state can
be viewed as a problem of minimization with respect to either the density
* σ *
n(r, σ) or the effective potential Veff (r). As Ts is expressed as a functional
of the orbitals but all other terms are considered to be functionals of the
density, one can vary the wave functions and use the chain rule to derive
the variational equation
  *
δEKS δTs δEext δEHartree δExc δn(r, σ)
* = * + + + =0
δψσi ∗ (r) δψσi ∗ (r) δn(r, σ) δn(r, σ) δn(r, σ) δψσi ∗ (r)
* * * *

(2.35)

subject to normalization conditions

0
hψσi |ψσj i = δij δσσ 0 (2.36)
*
We use the expressions (2.29) and (2.30) for nσ (r) and Ts which give
*
δTs 1 2 σ* δnσ (r) σ *
= − ∇ ψi (r) ; ∗ * = ψi (r) (2.37)
δψσi ∗ (r)
* σ
2 δψi (r)
41
and the method of Lagrange multipliers to handle the constraints:
 XX Nσ Z 
δ  2
d3 r ψσi (r) − 1
*
σ ∗ EKS − εσj =0 (2.38)
δψi σ j=1

This variation leads to the Schrödinger-like Kohn-Sham equations

*
HσKS − εσi ψσi (r) = 0 (2.39)

where εσi are eigenvalues and HKS is the effective Hamiltonian (in Hartree
atomic units)
1
HσKS (r) = − ∇2 + VKS
* σ *
(r) (2.40)
2
with
σ * * δEHartree δExc
VKS (r) = Vext (r) + * + *
δn(r, σ) δn(r, σ)
* * σ *
= Vext (r) + VHartree (r) + Vxc (r) (2.41)

These are the well-known Kohn-Sham equations2 ; they have the form of
independent particle equations with a potential that must be found selfcon-
sistently with the resulting density. The equations are independent of any
approximation to the functional Exc [n], and would lead to the exact ground
state density and energy for the interacting system if the exact functional
Exc [n] were known. Furthermore, the Hohenberg-Kohn theorems guarantee
that the ground state density uniquely determines the potential at the min-
σ * σ *
imum, so that there is a unique Kohn-Sham potential Veff (r)|min ≡ VKS (r)
associated with any given interacting electron system.

Exc , Vxc and the exchange-correlation hole

The genius of the Kohn-Sham approach is that by explicitly separating out
the independent-particle kinetic energy and the long-range Hartree terms,
the remaining exchange-correlation functional Exc [n] can reasonably be
approximated as a local or nearly local functional of the density. This
means that the energy Exc can be expressed in the form
Z
Exc [n] = d3 r n(r)εxc ([n], r)
* *
(2.42)
2
W. Kohn and L. J. Sham, Self-consistent equations including exchange and correlation effects, Phys.
Rev. 140, A 1133 (1965).

42