This action might not be possible to undo. Are you sure you want to continue?
approach to the
manybody problem: key
concepts and exact
functionals
Robert van Leeuwen
Theoretical Chemistry, Materials Science Centre,
Nijenborgh 4, 9747 AG, Rijksuniversiteit Groningen
Groningen, The Netherlands
Abstract
We give an overview of the fundamental concepts of density functional
theory. We give a careful discussion of the several density functionals and
their diﬀerentiability properties. We show that for nondegenerate ground
states we can calculate the necessary functional derivatives by means of
linear response theory, but that there are some diﬀerentiability problems
for degenerate ground states. These problems can be overcome by extending
the domains of the functionals. We further show that for every interacting
vrepresentable density we can ﬁnd a noninteracting vrepresentable density
arbitrarily close and show that this is suﬃcient to set up a KohnSham
scheme. We ﬁnally describe two systematic approaches for the construction
of density functionals.
1. Introduction
2. Deﬁnition of the problem
3. Conditions on the electron density and external potentials
4. The HohenbergKohn theorem
5. KohnSham theory by Legendre transforms
6. Deﬁnition of the functional derivative
7. Static linear response of the Schr¨ odinger equation
8. Invertability of the density response function
9. Functional derivatives and vrepresentability
10. The HohenbergKohn theorem for degenerate ground states
11. Linear response for degenerate ground states and functional
diﬀerentiability
12. The Levy and Lieb functionals F
LL
and F
L
13. Diﬀerentiability of F
L
14. Ensemble vrepresentability
15. The KohnSham approach and noninteracting vrepresentability
16. The gradient expansion
17. The optimized potential method and the e
2
expansion
18. Outlook and conclusions
19. Acknowledgments
References
1. Introduction
In this paper we give an overview of the foundations of density functional
theory for stationary systems. In the discussion we try to be as precise
as possible and we therefore carefully discuss several exact density func
tionals and the domains of densities on which they are deﬁned. At the
heart of almost any application of density functional theory lie the Kohn
Sham equations. These equations describe a noninteracting system that
is required to produce the ground state density of an interacting system.
Therefore any discussion of the validity of these equations has to focus
on how well we can approximate a density of an interacting system with
that of a noninteracting system. Closely related to this question is whether
or not the exact functionals that we deﬁned have functional derivatives.
This is because the various potentials, such as the KohnSham potential,
in density functional theory appear as functional derivatives of the energy
with respect to the density. The investigation of the existence of functional
derivatives will form a central theme of this review.
A large part of this work will follow the proofs of the fundamental papers
by Lieb [1] and Englisch and Englisch [2, 3]. In this paper we try to make
these two important works more accessible by providing some background
on the mathematics involved. We further show, from a more physical view
point, how to calculate functional derivatives by response theory and show
that for nondegenerate ground states the static density response function is
invertable. We also show that for every interacting vrepresentable density
there is a noninteracting vrepresentable density arbitrarily close to it and
that this is suﬃcient to set up a KohnSham scheme. We ﬁnally discuss
two systematic approaches for the construction of the exchangecorrelation
functional and provide an outlook and conclusions.
2. Deﬁnition of the problem
Densityfunctional theory in its earliest formulation by Hohenberg, Kohn
and Sham [5, 8] aims at a description of the description of ground state
properties of manyelectron systems in terms of the electron density. One
may wonder why this is possible. Let us therefore investigate this point
more closely. Consider a Hamiltonian of a stationary manybody system
ˆ
H
v
=
ˆ
T +
ˆ
V +
ˆ
W (1)
where
ˆ
T is the kinetic energy of the electrons,
ˆ
V the external potential,
and
ˆ
W the twoparticle interaction. We denote the Hamilton operator
ˆ
H
v
with a subindex v to indicate that we will consider the Hamiltonian as a
functional of the external potential v. The constituent terms are explicitly
given as
ˆ
T =
N
i=1
−
1
2
∇
2
i
(2)
ˆ
V =
N
i=1
v(r
i
) (3)
ˆ
W =
N
i>j
w([r
i
−r
j
[) (4)
where w([r[) = 1/[r[ will in our discussion always be the Coulomb potential.
We are interested in electronic systems, i.e. molecules and solids. For all
these systems the kinetic energy operator
ˆ
T and twoparticle interaction
ˆ
W are identical. They only diﬀer in the form of the external potential v(r)
and the number of electrons N. The properties of all these systems can
therefore be regarded as a functional of the external potential v. This is in
particular the case for the ground state wavefunction [Ψ[v]) and the ground
state energy E[v], which are of course related by the Schr¨odinger equation
(
ˆ
T +
ˆ
V +
ˆ
W)[Ψ[v]) = E[v][Ψ[v]) (5)
At ﬁrst sight we have not gained much by this viewpoint, the problem looks
as diﬃcult as before. However, the problem will have a diﬀerent appearance
once we eliminate the potential in favor of the electron density. One may
wonder what is so particular about the relation between the density and
the potential. Let us therefore look speciﬁcally at the term which describes
the external potential. It can be written as
ˆ
V =
_
d
3
rv(r)ˆ n(r) (6)
where we deﬁned the density operator by
ˆ n(r) =
N
i=1
δ(r −r
i
) (7)
The expectation value of
ˆ
V is given by
¸Ψ[
ˆ
V [Ψ) =
_
d
3
rn(r)v(r) (8)
where n(r) the electron density . The electron density is obtained from
the manybody wavefunction, which in our case will usually be the ground
state wavefunction of Hamiltonian
ˆ
H
v
, by
n(r
1
) = ¸Ψ[ˆ n(r
1
)[Ψ) = N
σ1...σN
_
d
3
r
2
. . . d
3
r
N
[Ψ(r
1
σ
1
, . . . , r
N
σ
N
)[
2
(9)
In this expression σ
i
is the spin variable for electron i. The physical inter
pretation of the density is that n(r)∆V is proportional to the probability
of ﬁnding an electron in an inﬁnitesimal volume ∆V around r. Note that
in the deﬁnition Eq.(9) we sum over the spin coordinate σ
1
so we do not
distinguish between the probabilities of ﬁnding up or down spin electrons
at r. We see from Eq.(8) that the external potential v(r) and the electron
density n(r) are conjugate variables. With this we mean that they occur
as a simple product in the contribution of the external potential to the
energy. It is exactly this property that allows us to prove the Hohenberg
Kohn theorem which establishes a 11correspondence between the density
and the external potential. We can therefore go from a functional E[v] of
the external potential to a functional E[n] of the density. We can ask our
selves whether there is any reason that E[n] would be easier to calculate
than E[v]. A priori there is no reason to expect this. However, we will
see that the determination of E[n] is equivalent to the solution of a set of
oneparticle equations, known as the KohnSham equations, with a poten
tial v
s
[n] that is also a functional of the density. By now we know that
we can ﬁnd practical and useful approximations for this potential v
s
[n] and
that the KohnSham equations have been succesfully applied to the calcu
lation of properties of many electronic systems. However, as in any physical
theory, there are a number of assumptions made in the transformation to
the KohnSham oneparticle equations. The aim of this paper is to discuss
these assumptions and thereby the validity of the KohnSham equations.
3. Conditions on the electron density and the external
potentials
We begin by giving a discussion of the properties of the two key objects in
density functional theory, the density n and the external potential v. The
density has the obvious properties
n(r) ≥ 0 ,
_
d
3
rn(r) = N (10)
These properties follow directly from the deﬁnition of the density and the
usual normalization condition on the wavefunction. If we take into account
that the density is obtained as the density of a bound eigenstate of Hamil
tonian (1) we can derive further conditions. For this we put the physical
constraint on the manybody system that it has a ﬁnite expectation value
of the kinetic energy, i.e.
T[Ψ] =
1
2
N
i=1
σ1...σN
_
d
3
r
1
. . . d
3
r
N
[∇
i
Ψ[
2
< ∞ (11)
At this point it is useful to introduce a new space of functions. We say that
a function f is in H
1
(¹
n
) (¹ denotes the real numbers) if
f =
__
d
n
r([f(r)[
2
+[∇f(r)[
2
_
1/2
< ∞ (12)
The space of functions H
1
(¹
n
) is called a Sobolev space. The supindex 1
refers to the fact that the deﬁnition of the norm contains only ﬁrst order
derivatives. We therefore see that ﬁniteness of the kinetic energy implies
that Ψ is an element of the function space H
1
(¹
3N
). Diﬀerentiation of
Eq.(9) and use of the CauchySchwarz inequality then leads to [1]
(∇
1
n(r
1
))
2
≤ 4Nn(r
1
)
σ1...σN
_
d
3
r
2
. . . d
3
r
N
[∇
1
Ψ[
2
(13)
and hence
1
2
_
d
3
r(∇
_
n(r))
2
≤ T[Ψ] < ∞ (14)
We see that the ﬁniteness of the kinetic energy puts a constraint on the
density. From Eq.(12) we see that
√
n belongs to H
1
(¹
3
). We also see that
if we consider systems with a ﬁnite kinetic energy, then we only need to
consider the following set of densities
o = ¦n [ n(r) ≥ 0,
√
n ∈ H
1
(¹
3
),
_
d
3
rn(r) = N¦ (15)
This set of densities has a property which will be of importance later,
namely o is convex. With this we mean that if n
1
and n
2
are elements of
o, then also λn
1
+ (1 − λ)n
2
is an element of o where 0 ≤ λ ≤ 1. This
property is easily proven using the CauchySchwarz inequality. Now we will
derive constraints on the allowed set of external potentials. In order to do
this we introduce some other function spaces. We say that a given function
f belongs to the space L
p
if
f
p
=
__
d
3
r[f(r)[
p
_
1/p
< ∞ (16)
Note that here we only consider functions on the usual threedimensional
coordinate space ¹
3
. The letter L refers to Lebesque integration, a feature
that assures that the function spaces are complete (complete normed spaces
are also called Banach spaces). We will, however, not go into the detailed
mathematics and refer the interested reader to the literature [4]. We just
note that for continuous functions the integral is equivalent to the usual
(Riemann) integral. Equation (16) deﬁnes a norm on the space L
p
and
we see from Eq.(10) that the density belongs to L
1
. From the condition
of ﬁnite kinetic energy and the use of a Sobolev inequality one can show
that [1]
_
d
3
rn
3
(r) ≤ C
_
d
3
r(∇
_
n(r))
2
< ∞ (17)
where C = 3(π/2)
4/3
. In other words, the ﬁniteness of the kinetic energy
implies that the density is also in the space L
3
. Since we already know
that n ∈ L
1
we ﬁnd that the density is element of the intersection of both
spaces, i.e. n ∈ L
1
∩ L
3
. We also see from the inequality (17) that o is a
subset of L
1
∩ L
3
. Let us now see what this implies for the allowed set of
external potentials when we require the expectation value of the external
potential to be ﬁnite, i.e. we require
¸
¸
¸
¸
_
d
3
rn(r)v(r)
¸
¸
¸
¸
≤ nv
1
< ∞ (18)
If the potential is bounded, i.e. [v(r)[ < M for some ﬁnite number M then
¸
¸
¸
¸
_
d
3
rn(r)v(r)
¸
¸
¸
¸
≤ sup[v(r)[N < ∞ (19)
The space of bounded functions is called L
∞
and has the norm
f
∞
= sup[f(r)[ (20)
The supremum is deﬁned to be the smallest number M such that [f(r)[ ≤ M
almost everywhere. The term ’almost everywhere’ has a precise mathemat
ical meaning for which we refer to the literature [4]. We almost never use
it in the remainder of this paper. We therefore conclude that if v ∈ L
∞
then the expectation value of the external potential is ﬁnite. To show this
we used that n ∈ L
1
. But we also know that n ∈ L
3
and if we make use of
the H¨older inequality
fg
1
≤ f
p
g
q
(21)
with 1/p + 1/q = 1 we obtain
nv
1
≤ n
3
v3
2
(22)
which is ﬁnite if v ∈ L
3
2
. Therefore the most general set of potentials for
which the expectation value ¸Ψ[
ˆ
V [Ψ) is ﬁnite, is the set
L
3
2
+L
∞
= ¦v[v = u +w, u ∈ L
3
2
, w ∈ L
∞
¦ (23)
i.e. the set of potentials that can be written as a sum of a function from L
3
2
and a function from L
∞
. This is also a normed function space with norm
v = inf¦u3
2
+w
∞
[v = u +w¦ (24)
In the remainder of this paper we will always consider the densities to be in
the space L
1
∩L
3
and the potentials in the space L
3
2
+L
∞
. It is important
to note that the Coulomb potential is in the latter set since we can write
1
[r[
=
θ(1 −[r[)
[r[
+
θ([r[ −1)
[r[
(25)
where θ is the Heaviside function, θ(x) = 0 if x ≤ 0 and θ(x) = 1 if
x > 0. One can readily check that the ﬁrst and the second term on the
right hand side are in L
3
2
and L
∞
respectively. One can easily extend this
result to a ﬁnite sum of Coulomb potentials and therefore every molecule
can be described with the potentials in the space L
3
2
+L
∞
. One may ﬁnally
wonder what the condition of ﬁnite Coulombic electronelectron repulsion,
i.e. ¸Ψ[
ˆ
W[Ψ) < ∞, would imply for the density. However, one can show
that if the kinetic energy ¸Ψ[
ˆ
T[Ψ) is ﬁnite then also ¸Ψ[
ˆ
W[Ψ) is ﬁnite [1],
so this does not yield any new constraints on the density.
4. The HohenbergKohn theorem
The basis of densityfunctional theory is provided by the HohenbergKohn
theorem [5]. We will provide a proof of this theorem for nondegenerate
ground states. The case of degenerate ground states will be discussed later.
The HohenbergKohn theorem states that the density n(r) of a nondegen
erate ground state uniquely determines the external potential v(r) up to an
arbitrary constant. This means that the external potential is a welldeﬁned
functional v[n](r) of the density.
In the proof essential use is made of the fact that the density and the po
tential are conjugate variables. For the same reason we can, for instance,
prove that the 2particle interaction is a unique functional of the diago
nal 2particle density matrix. The general mapping between Nparticle
density matrices and Nbody potentials is discussed by De Dominicis and
Martin [6].
Let us consider the subset 1 ⊂ L
3
2
+L
∞
of potentials that yield a normaliz
able nondegenerate ground state. The solution of the Schr¨odinger equation
provides us with a mapping from the external potential to the ground state
wavefunction, v(r) → [Ψ[v]). Since we assume that we are dealing with
nondegenerate ground states [Ψ[v]) is uniquely determined apart from a
trivial phase factor. We have therefore established a map C : 1 → Φ,
where Φ is the set of ground states.
We will ﬁrst prove that the map C is invertible. Suppose that [Ψ
1
) and
[Ψ
2
) ∈ H
1
(¹
3N
) correspond to external potentials v
1
and v
2
∈ L
∞
+ L
3
2
where v
1
,= v
2
+C. We have to show that [Ψ
1
) ,= [Ψ
2
). If we assume that
[Ψ
1
) = [Ψ
2
) = [Ψ) then by subtraction of the Hamiltonian for [Ψ
1
) and
[Ψ
2
) we ﬁnd that
(
ˆ
V
1
−
ˆ
V
2
)[Ψ) = (E
1
−E
2
)[Ψ) (26)
If v
1
−v
2
is not constant in some region then Ψ must vanish in this region
for the above equation to be true. However if v
1
, v
2
∈ L
∞
+ L
3
2
then [Ψ)
cannot vanish on an open set (a set with nonzero measure) by the unique
continuation theorem [1]. So we obtain a contradiction and hence we must
have made a wrong assumption. Therefore [Ψ
1
) ,= [Ψ
2
) and we obtain the
result that diﬀerent potentials ( diﬀering more than a constant) give diﬀer
ent wavefunctions. Consequently we ﬁnd that the map C is invertible.
We now deﬁne the set / as the set of densities which come from a nonde
generate ground state, where we only consider ground state densities from
potentials in the set L
∞
+L
3
2
. The set / is obviously a subset of the previ
ously deﬁned set o. From a given wavefunction in the set of ground states
Φ we can calculate the density according to Eq.(9). This provides us with
a second map D : Φ → / from ground state wavefunctions to ground state
densities. Also this map is invertible. To show this we calculate
E[v
1
] = ¸Ψ[v
1
][
ˆ
T +
ˆ
V
1
+
ˆ
W[Ψ[v
1
]) < ¸Ψ[v
2
][
ˆ
T +
ˆ
V
1
+
ˆ
W[Ψ[v
2
])
= ¸Ψ[v
2
][
ˆ
T +
ˆ
V
2
+
ˆ
W[Ψ[v
2
]) +
_
n
2
(r)(v
1
(r) −v
2
(r))dr
= E[v
2
] +
_
n
2
(r)(v
1
(r) −v
2
(r))dr (27)
Likewise we ﬁnd
E[v
2
] < E[v
1
] +
_
n
1
(r)(v
2
(r) −v
1
(r))dr (28)
Adding both inequalities then yields the new inequality
_
d
3
r(n
2
(r) −n
1
(r))(v
2
(r) −v
1
(r)) < 0 (29)
If we assume that n
1
= n
2
then we obtain the contradiction 0 < 0 and we
conclude that diﬀerent ground states must yield diﬀerent densities. There
fore the map D is also invertible. Consequently the map DC : 1 → / is
also invertible and the density therefore uniquely determines the external
potential. This proves the HohenbergKohn theorem.
Let us now pick an arbitrary density out of the set / of densities of non
degenerate ground states. The HohenbergKohn theorem then tells us that
there is a unique external potential v (to within a constant) and a unique
ground state wavefunction [Ψ[n]) (to within a phase factor) corresponding
to this density. This also means that the ground state expectation value of
any observable, represented by an operator
ˆ
O, can be regarded as a density
functional
O[n] = ¸Ψ[n][
ˆ
O[Ψ[n]) (30)
In particular we can thus deﬁne the HohenbergKohn functional F
HK
on
the set / as
F
HK
[n] = ¸Ψ[n][
ˆ
T +
ˆ
W[Ψ[n]) (31)
With this functional we can deﬁne the energy functional E
v
as
E
v
[n] =
_
n(r)v(r)dr +F
HK
[n] (32)
If n
0
is a ground state density corresponding to external potential v
0
and
n an arbitrary other ground state density then
E
v0
[n] =
_
n(r)v
0
(r)dr +F
HK
[n] = ¸Ψ[n][
ˆ
T +
ˆ
V
0
+
ˆ
W[Ψ[n])
≥ ¸Ψ[n
0
][
ˆ
T +
ˆ
V
0
+
ˆ
W[Ψ[n
0
]) = E
v0
[n
0
] (33)
Therefore
E[v] = inf
n∈A
__
n(r)v(r)dr +F
HK
[n]
_
(34)
and we see that the ground state energy of a manybody system can be
obtained by minimization of a density functional. For application of this
formula we have to know F
HK
on the set /. In practice it is, of course,
impossible to calculate F
HK
exactly on this set of densities. Instead one
would prefer to make an explicit approximation for F
HK
as is usually done
within the KohnSham scheme. Before we go into that let us ﬁrst discuss
some properties of F
HK
. The functional F
HK
is a convex functional, i.e. if
n
1
, n
2
∈ / and if λ
1
n
1
+ λ
2
n
2
∈ / with 0 ≤ λ
1
, λ
2
≤ 1 and λ
1
+ λ
2
= 1
then
F
HK
[λ
1
n
1
+λ
2
n
2
] ≤ λ
1
F
HK
[n
1
] +λ
2
F
HK
[n
2
] (35)
This is readily proven. Suppose that the ground state densities n
1
, n
2
, λ
1
n
1
+
λ
2
n
2
∈ / correspond to the external potentials v
1
, v
2
and v. Then
F
HK
[n] = ¸Ψ[n][
ˆ
T +
ˆ
V +
ˆ
W[Ψ[n]) −
_
d
3
rn(r)v(r)
= λ
1
¸Ψ[n][
ˆ
H[Ψ[n]) +λ
2
¸Ψ[n][
ˆ
H[Ψ[n]) −
_
d
3
rn(r)v(r)
≤ λ
1
¸Ψ[n
1
][
ˆ
T +
ˆ
W[Ψ[n
1
]) +λ
2
¸Ψ[n
2
][
ˆ
T +
ˆ
W[Ψ[n
2
])
+
_
(λ
1
n
1
(r) +λ
2
n
2
(r))v(r)d
3
r −
_
d
3
rn(r)v(r)
= λ
1
F
HK
[n
1
] +λ
2
F
HK
[n
2
] (36)
and we obtain the convexity of F
HK
. Note, however, that the domain /
of F
HK
does not need to be convex, i.e. if n
1
, n
2
∈ / then not necessarily
λ
1
n
1
+λ
2
n
2
∈ / with 0 ≤ λ
1
, λ
2
≤ 1 and λ
1
+λ
2
= 1. We will come back
to this point later when we consider the diﬀerentiability of F
HK
. Let us
ﬁrst collect our results in the form a theorem
Theorem 1 (HohenbergKohn) The density n corresponding to a non
degenerate ground state speciﬁes the external potential v up to a constant
and the ground state wavefunction [Ψ[n]) up to a phase factor. Moreover,
1. Any ground state expectation value corresponding to an observable
ˆ
O
is a functional of the density according to
O[n] = ¸Ψ[n][
ˆ
O[Ψ[n])
2. The ground state energy of a system with a nondegenerate ground
state and an external potential v can be obtained from
E[v] = inf
n∈A
__
n(r)v(r)dr +F
HK
[n]
_
where F
HK
[n] = ¸Ψ[n][
ˆ
T +
ˆ
W[Ψ[n]).
3. F
HK
is convex
We ﬁnally discuss an interesting consequence of the HohenbergKohn the
orem. Suppose that in Eq.(29) we take v
2
= v
1
+ δv where δv is not a
constant. We therefore obtain potential v
2
by a small perturbation from
potential v
1
. By means of perturbation theory we can then calculate the
change in the density which gives
n
2
(r) = n
1
(r) +δn(r) +O(
2
) (37)
where δn can be calculated from the static density response function χ
δn(r) =
_
d
3
r
χ(r, r
)δv(r
) (38)
The properties and explicit form of χ are described in more detail in a later
section. Therefore
(v
2
(r) −v
1
(r))(n
2
(r) −n
1
(r)) =
2
δv(r)δn(r) +O(
3
) (39)
If we insert this expression in Eq.(29) and divide by
2
> 0 then we obtain:
_
d
3
rδv(r)δn(r) +O() < 0 (40)
Now taking the limit → 0 and expressing δn in terms of the response
function we obtain
_
d
3
rd
3
r
δv(r)χ(r, r
)δv(r
) < 0 (41)
This is true for an arbitrary nonconstant potential variation. We therefore
see that the eigenvalues of χ, when we regard χ as an integral operator,
are negative. Moreover we see that the only potential variation that yields
a zero density variation is given by δv = C where C is a constant. This
implies that χ is invertable. We will go more closely into this matter in a
later section where we will prove the same using the explicit form of the
density response function.
5. KohnSham theory by Legendre transforms
The method described in this section goes back to the work of De Do
minicis and Martin [6]. This work discusses the relations between Nbody
potentials and Nparticle density matrices, of which the densitypotential
relation to be discussed here is a special case. The ground state energy E[v]
and wave function [Ψ[v]) are considered to be functionals of the external
potential through solving the timeindependent Schr¨odinger equation
(
ˆ
T +
ˆ
V +
ˆ
W)[Ψ[v]) = E[v][Ψ[v]) (42)
where the twoparticle interaction
ˆ
W is kept ﬁxed. From this equation we
see that the ground state energy as a functional of the external potential v
can also be written as
E[v] = ¸Ψ[v][
ˆ
H
v
[Ψ[v]) (43)
Our goal is now to go from the potential as our basic variable, to a new
variable, which will be the electron density. The deeper reason that this is
possible is that the density and the potential are conjugate variables. With
this we mean that the contribution of the external potential to the total
energy is simply an integral of the potential times the density. We make use
of this relation if we take the functional derivative of the energy functional
E[v] with respect to the potential v:
δE
δv(r)
= ¸
δΨ
δv(r)
[
ˆ
H
v
[Ψ) +¸Ψ[
ˆ
H
v
[
δΨ
δv(r)
) +¸Ψ[
δ
ˆ
H
v
δv(r)
[Ψ)
= E[v]
δ
δv(r)
¸Ψ[Ψ) +¸Ψ[ˆ n(r)[Ψ)
= ¸Ψ[ˆ n(r)[Ψ) = n(r) (44)
where we used the Schr¨odinger equation
ˆ
H
v
[Ψ) = E[v][Ψ) and the nor
malization condition ¸Ψ[Ψ) = 1. Note that the equation above is nothing
but a functional generalization of the wellknown HellmannFeynman the
orem [7]. Now we can go to the density as our basic variable by deﬁning a
Legendre transform
F[n] = E[v] −
_
d
3
rn(r)v(r) = ¸Ψ[v][
ˆ
T +
ˆ
W[Ψ[v]) (45)
where v must now be regarded as a functional of n. The uniqueness of such
a mapping is garanteed by the HohenbergKohn theorem [5]. The set of
densities for which the functional F[n] is deﬁned is the set of socalled v
representable densities. These are ground state densities for a Hamiltonian
with external potential v. The question which constraints one has to put
on a density to make sure that it is vrepresentable is known as the v
representability problem. We postpone a discussion of these matters to
later sections. From δE/δv = n it follows immediately that
δF
δn(r)
= −v(r) (46)
This is our ﬁrst basic relation. In order to derive the KohnSham equations
we deﬁne the following energy functional for a system of noninteracting
particles with external potential v
s
and with ground state wave function
[Φ[v
s
]):
E
s
[v
s
] = ¸Φ[v
s
][
ˆ
T +
ˆ
V
s
[Φ[v
s
]) (47)
with Legendre transform
F
s
[n] = E[v
s
] −
_
d
3
rn(r)v
s
(r) = ¸Φ[v
s
][
ˆ
T[Φ[v
s
]) (48)
and derivatives
δE
s
δv
s
(r)
= n(r) (49)
δF
s
δn(r)
= −v
s
(r) (50)
We see that F
s
[n] in Eq.(48) is the kinetic energy of a noninteracting system
with potential v
s
and density n. For this reason the functional F
s
is usually
denoted by T
s
. In the following we will adopt this notation. Finally we
deﬁne the exchangecorrelation functional E
xc
[n] by the equation
F[n] = T
s
[n] +
1
2
_
d
3
rd
3
r
n(r)n(r
)w([r −r
[) +E
xc
[n] (51)
This equation assumes that the functionals F[n] and T
s
[n] are deﬁned on
the same domain of densities. We thus assume that for a given ground
state density of an interacting system there is a noninteracting system with
the same density. In other words, we assume that the interacting density
is noninteractingvrepresentable. If we diﬀerentiate Eq. (51) with respect
to the density n we obtain
v
s
(r) = v(r) +
_
d
3
r
n(r
)w([r −r
[) +v
xc
(r) (52)
where
v
xc
(r) =
δE
xc
δn(r)
(53)
deﬁnes the exchangecorrelation potential. Now the state [Φ[v
s
]) is a ground
state for a system of noninteracting particles, and can therefore be written
as an antisymmetrized product of singleparticle orbitals ϕ
i
(r). If we now
collect our results we see that we have converted the ground state problem
into the following set of equations
E[v] =
N
i=1
−
1
2
_
d
3
rϕ
∗
i
(r)∇
2
ϕ
i
(r) +
_
d
3
rn(r)v(r)
+
1
2
_
d
3
rd
3
r
n(r)n(r
)w([r −r
[) +E
xc
[n] (54)
(−
1
2
∇
2
+v(r) +
_
d
3
r
n(r
)w([r −r
[) +v
xc
(r))ϕ
i
(r) =
i
ϕ
i
(r) (55)
n(r) =
N
i=1
[ϕ
i
(r)[
2
(56)
The above equations constitute the ground state KohnSham equations [8].
These equations turn out to be of great practical use. If we can ﬁnd a
good approximation for the exchangecorrelation energy, we can calculate
the exchangecorrelation potential v
xc
and solve the orbital equations self
consistently. The density we ﬁnd in this way can then be used to calculate
the ground state energy of the system.
The exchangecorrelation functional is often split up into an exchange func
tional E
x
and a correlation functional E
c
as
E
xc
[n] = E
x
[n] +E
c
[n] (57)
in which the exchange and correlation functionals are deﬁned by
E
x
[n] = ¸Φ
s
[n][
ˆ
W[Φ
s
[n]) −
1
2
_
d
3
rd
3
r
n(r)n(r
)w([r −r
[) (58)
E
c
[n] = ¸Ψ[n][
ˆ
H
v
[Ψ[n]) −¸Φ
s
[n][
ˆ
H
v
[Φ
s
[n]) (59)
In this equation [Φ
s
[n]) is the KohnSham wavefunction and [Ψ[n]) the true
ground state wavefunction of the interacting system with density n. Since
the KohnSham wavefunction is not a ground state wavefunction of the true
system we see immediately from the variational principle that E
c
< 0. We
see also that the form of the exchange functional in terms of the KohnSham
orbitals is identical to that of the exchange energy within the wellknown
HartreeFock approximation. However, since the KohnSham and Hartree
Fock orbitals diﬀer, the value of E
x
[n] is not equal to the HartreeFock
exchange. We ﬁnally remark that splitting up E
xc
into an exchange and
a correlation part has several disadvantages. First of all, this splitting has
only meaning if the ground state of the true or KohnSham system is non
degenerate. We will later see that E
xc
[n] is a welldeﬁned functional even
for degenerate ground states, but that exchange and correlation separately
are illdeﬁned in that case. Secondly, there are many cases, notably molec
ular dissociation cases, where the exchangeonly theory is a bad starting
point for the treatment of correlation eﬀects and for which it is much easier
to ﬁnd good approximations for the combined exchangecorrelation func
tional.
In this section we gave a derivation of the KohnSham equations. However,
in this derivation we made a number of assumptions. We assumed that
for every ground state density of an interacting system there is a noninter
acting system which has the same density in its ground state. Secondly,
we assumed that the density functionals were diﬀerentiable. This assumes
that the values of the functionals change smoothly with changes in the
density. In the following sections we will investigate to which extent these
assumptions are justiﬁed.
6. Deﬁnition of the functional derivative
Let us start by deﬁning what we mean with a functional derivative. The
derivative we will talk about is what in the mathematical literature [9, 10,
11, 12, 13, 14] is refered to as a Gˆ ateaux derivative. Let G : B → ¹ be
a functional from a normed function space B to the real numbers ¹. If
for every h ∈ B there exist a continuous linear functional δG/δf : B → ¹
deﬁned by:
δG
δf
[h] = lim
→0
G[f +h] −G[f]
(60)
then δG/δf is called the Gˆ ateaux derivative in f ∈ B. If the limit exists
but the resulting functional of h is not linear or continuous then this limit
is called a Gˆ ateaux variation. Note that the deﬁnition of the Gˆ ateaux
derivative is very similar to the deﬁnition of the directional derivative in
vector calculus where the linear functional corresponds to the inner product
of the gradient vector with the vector which speciﬁes the direction of the
diﬀerentiation. Often the linear functional δG/δf can be written in the
form
δG
δf
[h] =
_
d
3
rg(r)h(r) (61)
If this is the case we write
g(r) =
δG
δf(r)
(62)
which we will call the functional derivative of G. Note further that although
δG/δf speciﬁes a linear functional when acting on h, the function g(r)
depends in general on f in a nonlinear way. We also see that the set of
functional derivatives on a function space B is equal to the set of continuous
linear functionals on that space. This set is called the dual space of B and
denoted B
∗
. For instance, the dual space of B = L
1
∩L
3
is known to be the
space B
∗
= L
3
2
+ L
∞
. From this we see that if the derivative of a density
functional deﬁned on the set of densities L
1
∩ L
3
exists then its derivative
is in the set L
3
2
+L
∞
.
Let us now mention a straightforward consequence of the deﬁnition of the
Gˆ ateaux derivative. Suppose that the functional G which we assume to
be Gˆ ateaux diﬀerentiable, has a minimum at f
0
, i.e. G[f] ≥ G[f
0
] for all
f ∈ B. Then the function
g() = G[f
0
+h] ≥ G[f
0
] (63)
has a minimum in = 0 and therefore the derivative of g() in = 0
vanishes. Thus
0 =
dg
d
(0) = lim
→0
g() −g(0)
=
δG
δf
(f
0
)[h] (64)
Therefore a necessary condition for a diﬀerentiable functional for having a
minimum at f
0
is that its Gˆ ateaux derivative vanishes at f
0
. We further
prove one other fact that we will use later. If G is a convex functional, i.e.
G[λf
0
+ (1 −λ)f
1
] ≤ λG[f
0
] + (1 −λ)G[f
1
] (65)
for 0 ≤ λ ≤ 1 then the function
g(λ) = G[f
0
+λ(f
1
−f
0
)] (66)
is a continuous function on the interval [0, 1]. To show this we will show
that g(λ) is convex. Take λ
1
, λ
2
and from the interval [0, 1]. Then
g(λ
1
+ (1 −)λ
2
) = G[(f
0
+λ
1
(f
1
−f
0
)) + (1 −)(f
0
+λ
2
(f
1
−f
0
))]
≤ g(λ
1
) + (1 −)g(λ
2
) (67)
Therefore g(λ) is a convex real function on the interval [0, 1] and hence
continuous. The fact that convexity implies continuity is true for functions
on the real axis, but this does not extend to inﬁnitedimensional spaces.
Let us ﬁnally make some remarks on higher order derivatives. If g() =
G[f
0
+ h] deﬁnes a nfold diﬀerentiable function of then we deﬁne the
nth Gˆ ateaux variation of G as
δ
n
G
δf
n
[h, . . . , h] =
d
n
g
d
n
(0) (68)
where δ
n
G/δf
n
has now n arguments h. If this expression deﬁnes a multi
linear continuous functional then we call this the nthorder Gˆ ateaux deriva
tive of G. For more details on this point we refer to reference [9]. Again, if
this expression can be written in the form
δ
n
G
δf
n
[h, . . . , h] =
_
d
3
r
1
. . . d
3
r
n
g(r
1
. . . r
n
)h(r
1
) . . . h(r
n
) (69)
then we call
δ
n
G
δf(r
1
) . . . δf(r
n
)
= g(r
1
. . . r
n
) (70)
the nthorder functional derivative of G. Sofar our discussion has been
rather abstract. Let us therefore apply the deﬁnition and calculate some
functional derivatives.
7. Static linear response of the Schr¨ odinger equation
We now consider the eﬀect of small changes in the external ﬁeld on the
expectation values of physical observables. This is exactly what is studied
in most experimental situations where one switches on and oﬀ an external
ﬁeld and studies how the system reacts to this. We will here study a more
speciﬁc case in which we look at static changes in the external potential
and their accompanying changes in the ground state expectation values. By
investigating this problem we will learn how to take functional derivatives
and how the existence of these derivatives is related to the existence of the
linear response function.
Suppose that we have solved the following ground state problem:
ˆ
H[Ψ
0
) = (
ˆ
T +
ˆ
V +
ˆ
W)[Ψ
0
) = E
0
[Ψ
0
) (71)
where [Ψ
0
) is a nondegenerate ground state [Ψ[v]) of Hamiltonian
ˆ
H with
external potential
ˆ
V . From the wavefunction we can then calculate the
expectation value of any operator
ˆ
O which is a welldeﬁned functional of
the external potential
O[v] = ¸Ψ[v][
ˆ
O[Ψ[v]) (72)
Let us now calculate the functional derivative δO/δv at a given potential
v. According to our deﬁnition in the previous section we have to calculate
the quantity
δO
δv
[δv] = lim
→0
O[v +δv] −O[v]
(73)
To evaluate this limit we have to calculate O[v+δv] which we will do using
static perturbation theory. We therefore make a slight change
δ
ˆ
V =
_
d
3
rˆ n(r)δv(r) (74)
in the external potential of Hamiltonian
ˆ
H, i.e. we change the potential
to
ˆ
V + δ
ˆ
V . The new ground state wavefunction which we will denote by
[Ψ()) satisﬁes
(
ˆ
H +δV )[Ψ()) = E()[Ψ()) (75)
We will solve this equation to ﬁrst order in with the condition [Ψ(0)) =
[Ψ
0
). We note that the solution of Eq.(75) is not unique, because if [Ψ())
is a solution then also [Φ()) = e
iθ()
[Ψ()) is a solution, where θ() is an
arbitrary function of . If we choose θ(0) = 0 then [Φ()) also satisﬁes the
condition [Φ(0)) = [Ψ
0
). The arbitrariness of the phase factor obviously
does not aﬀect the value of any expectation value, i.e.
O() = ¸Ψ()[
ˆ
O[Ψ()) = ¸Φ()[
ˆ
O[Φ()) (76)
which is a unique function of . However, it aﬀects the appearance of our
ﬁrst order expansion, which for both functions looks like
[Ψ()) = [Ψ
0
) +[Ψ
(0)) +O(
2
) (77)
[Φ()) = [Ψ
0
) +([Ψ
(0)) +iθ
(0)[Ψ
0
)) +O(
2
) (78)
where [Ψ
(0)) and θ
(0) are the ﬁrst order derivatives of [Ψ()) and θ() in
= 0. We see that [Ψ()) and [Φ()) diﬀer in ﬁrst order by the amount
iθ
(0)[Ψ
0
), i.e. by an imaginary number times the unperturbed ground
state. This is exactly the freedom we will ﬁnd in our expansion when we
try to obtain the ﬁrst order change in the wavefunction.
Let us expand the wavefunction and energy in Eq.(75) to ﬁrst order in .
We then obtain
(
ˆ
H −E
0
)[Ψ
(0)) = (E
(0) −δ
ˆ
V )[Ψ
0
) (79)
where E
(0) is the ﬁrst order derivative of E() in = 0. To solve this
equation we expand [Ψ
(0)) in an orthonormal set of eigenstates of the
unperturbed Hamiltonian
ˆ
H as
[Ψ
(0)) =
∞
i=0
c
i
[Ψ
i
) (80)
If the unperturbed Hamiltonian has a continuous spectrum then for the
corresponding energy eigenstates the summation in this equation has to be
replaced by an integration. If we insert the expansion (80) into Eq.(79) for
[Ψ
(0)) we ﬁnd the equation
∞
i=0
c
i
(E
i
−E
0
)[Ψ
i
) = (E
(0) −δ
ˆ
V )[Ψ
0
) (81)
where the energies E
i
for i > 0 are the eigenenergies of the excited states
of the unperturbed Hamiltonian. If we multiply this equation from the left
with ¸Ψ
0
[ we obtain for the change in energy
E
(0) = ¸Ψ
0
[δ
ˆ
V [Ψ
0
) =
_
d
3
rn
0
(r)δv(r) (82)
where n
0
is the density of the unperturbed system. We have therefore
shown that
E
(0) = lim
→0
E[v +δv] −E[v]
=
_
d
3
rn
0
(r)δv(r) (83)
As has been derived before we ﬁnd δE/δv(r) = n
0
(r). If we now multiply
Eq.(81) from the left with ¸Ψ
k
[ for k > 0 we ﬁnd
c
k
= −
¸Ψ
k
[δ
ˆ
V [Ψ
0
)
E
k
−E
0
(for k > 0) (84)
Note that these coeﬃcients are welldeﬁned because E
k
> E
0
, since we are
dealing with an isolated nondegenerate ground state. We therefore ﬁnd for
the ﬁrst order change in the wavefunction
[Ψ
(0)) = c
0
[Ψ
0
) −
∞
k=1
[Ψ
k
)¸Ψ
k
[δ
ˆ
V [Ψ
0
)
E
k
−E
0
(85)
The coeﬃcient c
0
remains undetermined by these equations. However, from
the requirement that the perturbed system remains normalized we ﬁnd
0 =
d¸Ψ[Ψ)
d
(0) = ¸Ψ
(0)[Ψ
0
) +¸Ψ
0
[Ψ
(0)) = c
∗
0
+c
0
(86)
and hence c
0
must be purely imaginary. This is exactly the kind of freedom
in the ﬁrst order wavefunction that we noted before. However, as we have
seen this is related to choice of phase and does not aﬀect the expectation
values. We now want to calculate the ﬁrst order change in an arbitrary
expectation value due to a change δv of the external potential. This change
is given by
O
(0) = lim
→0
O[v +δv] −O[v]
= ¸Ψ
(0)[
ˆ
O[Ψ
0
) +¸Ψ
0
[
ˆ
O[Ψ
(0)) (87)
where O
(0) is the derivative of O() in = 0 and where we assumed that
the operator
ˆ
O does not depend on
ˆ
V (as for instance the Hamiltonian
does). From Eq.(85) we obtain
O
(0) = (c
∗
0
+c
0
)¸Ψ
0
[
ˆ
O[Ψ
0
)
−
∞
k=1
¸Ψ
0
[δ
ˆ
V [Ψ
k
)¸Ψ
k
[
ˆ
O[Ψ
0
) +¸Ψ
0
[
ˆ
O[Ψ
k
)¸Ψ
k
[δ
ˆ
V [Ψ
0
)
E
k
−E
0
(88)
From the normalization condition Eq.(86) we see that the ﬁrst term on the
right hand side vanishes and we can rewrite the equation as
O
(0) =
δO
δv
[δv] =
_
d
3
r
δO
δv(r)
δv(r) (89)
where
δO
δv(r)
= −
∞
k=1
¸Ψ
0
[
ˆ
O[Ψ
k
)¸Ψ
k
[ˆ n(r)[Ψ
0
)
E
k
−E
0
+c.c. (90)
For a system with a nondegenerate ground state we therefore have obtained
an explicit expression for the functional derivative δO/δv of the expectation
value O[v], evaluated at potential v(r). We will use this formula later when
we are studying functional derivatives with respect to the density.
8. Invertability of the density response function
In the previous section we obtained an expression for the functional deriva
tive of an arbitrary expectation value. We now make a special choice for
the operator
ˆ
O and we choose
ˆ
O = ˆ n(r
). In that case we obtain from
Eq.(90):
χ(r
, r) =
δn(r
)
δv(r)
= −
∞
k=1
¸Ψ
0
[ˆ n(r
)[Ψ
k
)¸Ψ
k
[ˆ n(r)[Ψ
0
)
E
k
−E
0
+c.c. (91)
where χ(r
, r) is the static density response function. From this expression
we see immediately that χ is real and symmetric, i.e. χ(r, r
) = χ(r
, r).
The density response function relates ﬁrst order changes in the potential to
ﬁrst order changes in the density according to
δn(r) =
_
d
3
r
χ(r, r
)δv(r
) (92)
If we change the potential by a constant then the density does not change.
We therefore must have
_
d
3
r
χ(r, r
) = 0 (93)
which can also be checked immediately from Eq.(91). Because of this prop
erty we also see that the induced density δn(r) automatically integrates
to zero. The density response function has therefore indeed the physical
properties that we expect. On the basis of the HohenbergKohn theorem
we may further expect that the only potential variation that yields a zero
density variation is the constant potential, i.e. if
0 =
_
d
3
r
χ(r, r
)δv(r
) (94)
then δv(r) = C. This readily proven from the properties of the response
function. Suppose Eq.(94) is true for some δv(r
). Then we obtain by
integration with δv(r):
0 =
_
d
3
rd
3
r
δv(r)χ(r, r
)δv(r
) = −2
∞
k=1
[a
k
[
2
E
k
−E
0
(95)
where we deﬁned a
k
by
a
k
=
_
d
3
r¸Ψ
k
[ˆ n(r)[Ψ
0
)δv(r) (96)
Since E
k
− E
0
> 0 equation (95) can only be true if a
k
= 0 for all k ≥ 1.
But this implies that
0 =
∞
k=1
a
k
[Ψ
k
) =
∞
k=1
_
d
3
r[Ψ
k
)¸Ψ
k
[ˆ n(r)[Ψ
0
)δv(r)
=
_
d
3
rδv(r) (1 −[Ψ
0
)¸Ψ
0
[) ˆ n(r)[Ψ
0
)
=
_
d
3
rδv(r) (ˆ n(r) −n
0
(r)) [Ψ
0
) (97)
where n
0
(r) = ¸Ψ
0
[ˆ n(r)[Ψ
0
) is the ground state density. Written in ﬁrst
quantization this implies that
N
i=1
∆v(r
i
)[Ψ
0
) = 0 (98)
where N is the number of electrons and ∆v the potential
∆v(r) = δv(r) −
1
N
_
d
3
rn
0
(r)δv(r) (99)
Now Eq.(98) implies that ∆v(r) = 0 which together with Eq.(99) implies
that δv(r) = C. We have therefore shown that only constant potentials
yield a zerodensity variation, and therefore the density response function
is invertible up to a constant. One should, however, be careful with what
one means with the inverse response function. The response function de
ﬁnes a mapping χ : δ1 → δ/ from the set of potential variations from a
nondegenerate ground state, which we call δ1 and is a subset of L
3
2
+L
∞
,
to the set of ﬁrst order densities variations, which we call δ/, that are
produced by it. We have just shown that the inverse χ
−1
: δ/ → δ1 is
welldeﬁned modulo a constant function. However, there are density vari
ations that can never be produced by a potential variation and which are
therefore not in the set δ/. An example of such a density variation is one
which is identically zero on some ﬁnite volume.
From our analysis we can further derive another property of the static re
sponse function. Since χ(r, r
) is a real Hermitian integral kernel it has an
orthonormal set of eigenfunctions which can be chosen to be real. Let f(r)
be such an eigenfunction, i.e.
_
d
3
r
χ(r, r
)f(r
) = λf(r) (100)
Multiplication from the right with f(r) and subsequent integration yields
λ
_
d
3
rf
2
(r) =
_
d
3
rd
3
r
f(r)χ(r, r
)f(r
) = −2
∞
k=1
[b
k
[
2
E
k
−E
0
≤ 0 (101)
where
b
k
=
_
d
3
r¸Ψ
k
[ˆ n(r)[Ψ
0
)f(r) (102)
We therefore ﬁnd that λ ≤ 0. However, we already know that λ = 0 is only
possible if f(r) is constant. We have therefore obtained the result that if
a nonzero density variation is proportional to the potential that generates
it, i.e. δn = λδv, then the constant of proportionality λ is negative. This is
exactly what one would expect on the basis of physical considerations. In
actual calculations one indeed ﬁnds that the eigenvalues of χ are negative.
Moreover it is found that there is a inﬁnite number of negative eigenvalues
arbitrarily close to zero, which causes considerable numerical diﬃculties
when one tries to obtain the potential variation that is responsible for a
given density variation. We ﬁnally note the invertability proof for the static
response function can be extended to the timedependent case. For a recent
review we refer to [15].
9. Functional derivatives and vrepresentability
We are now ready to tackle the question how to calculate functional deriva
tives with respect to the density. By the HohenbergKohn theorem every
density associated with a nondegenerate ground state uniquely determines
that ground state and the external potential that produced it. Because
the density n determines the external potential v and vice versa we can
parametrize the ground state either by the density or the external poten
tial. The same is, of course, true for the expectation value of any operator
ˆ
O that we calculate from the ground state. We will therefore write
O[n] = ¸Ψ[n][
ˆ
O[Ψ[n]) = ¸Ψ[v][
ˆ
O[Ψ[v]) = O[v] (103)
Since we know how to calculate the functional derivative with respect to
v the functional derivative with respect to n seems rather straightforward.
By the chain rule for diﬀerentiation we obtain
δO
δn(r)
=
_
d
3
r
δO
δv(r
)
δv(r
)
δn(r)
=
_
d
3
r
δO
δv(r
)
χ
−1
(r
, r) (104)
where we used that the density response function has an inverse (modulo a
constant). However, great care must be taken when deriving this equation.
One needs to consider under which conditions the use of this chain rule
is applicable. Let us therefore start with the deﬁnition of the functional
derivative and try to calculate
δO
δn
[δn] = lim
→0
O[n
0
+δn] −O[n
0
]
(105)
where n
0
is the density of a given nondegenerate ground state with external
potential v
0
. Now we run into the following problem. In order to make the
value of O[n
0
+ δn] welldeﬁned we have to make sure that n
0
+ δn ∈
/, i.e. we have to make sure that this perturbed density belongs to a
nondegenerate ground state. Whether or not this is true obviously depends
on the choice we make for δn. A natural choice would be to take δn ∈ δ/,
i.e. we take a δn from the set of ﬁrst order density variations produced
by some potential variations δv ∈ δ1. In this way we know that for every
such δn there is a potential v
0
+δv that generates, to ﬁrst order in , the
required density. However, this still does not imply that n
0
+ δn ∈ /.
This will only be so if we can ﬁnd an additional potential that would make
the higher order terms in disappear and it is not clear how to prove the
existence of such a potential.
Nevertheless, this idea is suﬃcient to make a useful statement about the
functional derivative. Let us therefore take some δn ∈ δ/. Then, by the
invertability of the density response function, there is a unique δv modulo
a constant, such that
δn(r) =
_
d
3
r
χ
0
(r, r
)δv(r
) (106)
where χ
0
is the static density response function belonging to the nondegen
erate ground state [Ψ
0
) with density n
0
. Consider now the set of external
potentials v
0
+ δv. For each we can solve the Schr¨odinger equation and
obtain a ground state density n
(r), which for small enough will corre
spond to a nondegenerate ground state, i.e. n
(r) ∈ /. By construction
n
−n
0
will, to ﬁrst order in , be equal to the chosen δn we started with,
i.e. n
can be written as
n
(r) = n
0
(r) +δn(r) +m
(r) ∈ / (107)
where the term m
(r) satisﬁes
lim
→0
m
(r)
= 0 (108)
Since n
is the density of a nondegenerate ground state the expectation
value O[n
] is now welldeﬁned and we can calculate
lim
→0
O[n
] −O[n
0
]
= lim
→0
O[v
0
+δv] −O[v
0
]
=
_
d
3
r
δO
δv(r)
δv(r) (109)
We have therefore found a parametrized set of densities n
∈ / for which
the limit above is welldeﬁned. Moreover, in this equation δv is uniquely
deﬁned by the δn ∈ δ/ we started with by δv = χ
−1
0
δn, so that we can
write
lim
→0
O[n
0
+δn +m
] −O[n
0
]
=
_
d
3
r
δO
δn(r)
δn(r) (110)
where the functional derivative δO/δn(r) is given (modulo a constant) by
δO
δn(r)
= −
_
d
3
r
_
∞
k=1
¸Ψ
0
[
ˆ
O[Ψ
k
)¸Ψ
k
[ˆ n(r
)[Ψ
0
)
E
k
−E
0
+c.c.
_
χ
−1
0
(r
, r) (111)
Note that this derivative is deﬁned up to a constant, since δn(r) integrates
to zero. The functional derivative δO/δn[δn] in Eq.(110), regarded as a
linear functional acting on δn, is of course independent of this constant.
We have therefore a result that is in accordance with our naive expectation
of Eq.(104). The functional derivative in Eq.(111) is welldeﬁned in terms
of the properties of the unperturbed system and independent of the choice
for δn, and therefore independent of the parametrized path n
∈ / that
we used to approach n
0
. Moreover, n
approaches the straight path n
0
+
δn arbitrarily closely for → 0 and this is exactly the limit that we are
interested in. We will therefore call the δO/δn of Eq.(111) the functional
derivative of O[n]. Let us check that Eq.(111) gives us back some known
results. Let us take
ˆ
O =
ˆ
T +
ˆ
W so that O[n] = F
HK
[n]. If we insert this
operator in Eq.(111) we see that we have to calculate
¸Ψ
0
[
ˆ
T +
ˆ
W[Ψ
k
) = ¸Ψ
0
[
ˆ
H
0
−
ˆ
V
0
[Ψ
k
)
= E
k
δ
k0
−
_
d
3
rv
0
(r)¸Ψ
0
[ˆ n(r)[Ψ
k
) (112)
where
ˆ
H
0
=
ˆ
T +
ˆ
V
0
+
ˆ
W. If we insert this matrix element into Eq.(111) we
obtain:
δF
HK
[n]
δn(r)
= −
_
d
3
r
d
3
r
v
0
(r
)χ
0
(r
, r
)χ
−1
0
(r
, r) = −v
0
(r) (113)
This is exactly the result that we derived earlier using Legendre transforms.
We can, of course, also do this derivation for a noninteracting system where
ˆ
W = 0 and therefore
ˆ
O =
ˆ
T, in which case we obtain
δT
s
δn(r)
= −v
s
[n
0
](r) (114)
where v
s
[n
0
] is the potential that in a noninteracting system generates
density n
0
, i.e. the KohnSham potential corresponding to density n
0
.
10. The HohenbergKohn theorem for degenerate ground
states
Until now we considered only nondegenerate ground states. These ground
states have the simplifying feature that they are determined uniquely (up to
a phase factor) by the external potential and therefore any expectation value
calculated from the ground state wavefunction is a welldeﬁned functional
of the external potential v. However, degenerate ground states do occur
(for instance in open shell atoms) and in that case the external potential
does not generate a unique ground state but a linearly independent set of
q diﬀerent ground states. The expectation value of any operator (except
the total energy) will then depend on which ground state out of the ground
state manifold we choose to compute the expectation value from. This is
in particular true for the density operator, and therefore diﬀerent ground
states [Ψ
i
) out of the ground state multiplet will yield diﬀerent densities
n
i
. If we want to consider degenerate ground states, then the density is
no longer a unique functional of the external potential. However, we will
show that the inverse mapping n → v is welldeﬁned, i.e. every ground
state density determines uniquely the external potential that generated it.
We will ﬁrst generalize this statement somewhat. Instead of purestate
densities, which come from an eigenstate of the Hamiltonian
ˆ
H, we will
consider ensemble densities. To deﬁne this concept we ﬁrst introduce for a
qfold degenerate ground state ¦[Ψ
i
), i = 1 . . . q¦ the density matrix
ˆ
D =
q
i=1
λ
i
[Ψ
i
)¸Ψ
i
[
q
i
λ
i
= 1 (0 ≤ λ
i
≤ 1) (115)
where the ground state wavefunctions [Ψ
i
) are chosen to be orthonormal.
We then deﬁne the ground state expectation value of operator
ˆ
O as
¸O) = Tr
ˆ
D
ˆ
O (116)
where the trace operation for an arbitrary operator is deﬁned as
Tr
ˆ
A =
∞
i=1
¸Φ
i
[
ˆ
A[Φ
i
) (117)
where ¦[Φ
i
)¦ is an arbitrary complete set of states. The trace is independent
of the complete set that we choose, as follows simply by insertion of a
diﬀerent complete set ¦[Ψ
i
)¦
Tr
ˆ
A =
∞
i=1
¸Φ
i
[
ˆ
A[Φ
i
) =
∞
i,j=1
¸Φ
i
[
ˆ
A[Ψ
j
)¸Ψ
j
[Φ
i
)
=
∞
i,j=1
¸Ψ
j
[Φ
i
)¸Φ
i
[
ˆ
A[Ψ
j
) =
∞
j=1
¸Ψ
j
[
ˆ
A[Ψ
j
) (118)
If we choose the complete set to be the set of eigenstates of Hamiltonian
ˆ
H
then we ﬁnd
Tr
ˆ
D
ˆ
O =
∞
i=1
¸Ψ
i
[
ˆ
D
ˆ
O[Ψ
i
) =
q
i=1
λ
i
¸Ψ
i
[
ˆ
O[Ψ
i
) (119)
This deﬁnes the expectation value of an observable
ˆ
O in an ensemble de
scribed by density matrix
ˆ
D. For the particular case of the density operator
we have
n(r) = Tr
ˆ
Dˆ n(r) =
q
i=1
λ
i
¸Ψ
i
[ˆ n(r)[Ψ
i
) =
q
i=1
λ
i
n
i
(r) (120)
We will denote densities n(r) of this type, which are obtained from an
orthonormal set of ground states ¦[Ψ
i
), i = 1 . . . q¦ corresponding to a po
tential v, as ensemble vrepresentable densities, or for short EVdensities.
We further denote the set of all EVdensities generated by potentials in
L
3
2
+ L
∞
as B. A density will be called a pure state vrepresentable den
sity or for short PSVdensity if it can be written as n(r) = ¸Ψ[ˆ n(r)[Ψ) ,
where [Ψ) is a ground state. Obviously PSVdensities are special cases of
EVdensities and the set of PSVdensities is therefore a subset of the set
of EVdensities.
We are now ready to formulate the HohenbergKohn theorem for degen
erate ground states: Every EVdensity determines the external potential
that generated it up to an arbitrary constant. Let us make this statement a
bit more speciﬁc. Suppose that
ˆ
D
1
and
ˆ
D
2
are density matrices belonging
to ground state ensembles for potentials v
1
and v
2
respectively, with corre
sponding ensemble densities n
1
and n
2
. If v
1
,= v
2
+C with C a constant,
then n
1
,= n
2
. The proof is analogous to the proof of the nondegenerate
case.
Suppose v
1
generates the ground state multiplet A
1
= ¦[Φ
i
), i = 1 . . . q
1
¦
and v
2
generates the ground state multiplet A
2
= ¦[Ψ
i
), i = 1 . . . q
2
¦. All
the wavefunctions within these multiplets may, without loss of generality,
be chosen orthonormal. Then none of the wavefunctions in the sets A
1
and
A
2
are equal. This follows from the same argument as used in the proof of
the HohenbergKohn theorem for the nondegenerate case. In particular, as
the sets A
1
and A
2
are only deﬁned to within a unitary transformation no
[Ψ
i
) in A
2
is a linear combination of the [Φ
i
) in A
1
. This then implies that
two ground state ensemble density matrices constructed from the ground
states in A
1
and A
2
are diﬀerent
ˆ
D
1
=
q1
i=1
λ
i
[Φ
i
)¸Φ
i
[ ,=
q2
i=1
µ
i
[Ψ
i
)¸Ψ
i
[ =
ˆ
D
2
(121)
where
λ
i
=
µ
i
= 1. This follows, for instance, by taking the inner
product on both sides with [Ψ
m
) as the [Ψ
i
) are not linear combinations
of the [Φ
i
). We have thus established that the sets of ground state density
matrices for the two diﬀerent potentials v
1
and v
2
are disjoint. We now have
to prove that the density matrices in these sets yield diﬀerent densities.
If
ˆ
H
1
=
ˆ
T +
ˆ
V
1
+
ˆ
W and
ˆ
H
2
=
ˆ
T +
ˆ
V
2
+
ˆ
W then
Tr
ˆ
D
1
ˆ
H
2
> Tr
ˆ
D
2
ˆ
H
2
(122)
This follows directly from
Tr
ˆ
D
1
ˆ
H
2
=
q1
i=1
λ
i
¸Φ
i
[
ˆ
H
2
[Φ
i
) >
q1
i=1
λ
i
¸Ψ
i
[
ˆ
H
2
[Ψ
i
) =
=
q1
i=1
λ
i
E[v
2
] = E[v
2
] =
q2
i=1
µ
i
¸Ψ
i
[
ˆ
H
2
[Ψ
i
) = Tr
ˆ
D
2
ˆ
H
2
(123)
We can now show that
ˆ
D
1
and
ˆ
D
2
yield diﬀerent densities. We proceed
again by reductio ad absurdum. We have
E[v
1
] = Tr
ˆ
D
1
ˆ
H
1
= Tr
ˆ
D
1
(
ˆ
H
2
+
ˆ
V
1
−
ˆ
V
2
)
= Tr
ˆ
D
1
ˆ
H
2
+
_
n
1
(r)(v
1
(r) −v
2
(r))dr
> Tr
ˆ
D
2
ˆ
H
2
+
_
n
1
(r)(v
1
(r) −v
2
(r))dr
= E[v
2
] +
_
n
1
(r)(v
1
(r) −v
2
(r))dr (124)
Likewise we have
E[v
2
] > E[v
1
] +
_
n
2
(r)(v
2
(r) −v
1
(r))dr (125)
which added to the last inequality leads to
_
d
3
r(n
2
(r) −n
1
(r))(v
2
(r) −v
1
(r)) < 0 (126)
This leads again to the contradiction 0 < 0 if we assume that n
1
= n
2
.
Therefore
ˆ
D
1
and
ˆ
D
2
must give diﬀerent densities, which proves the theo
rem.
Within the set of ensemble ground state density matrices corresponding
to the same potential however, two diﬀerent density matrices can yield the
same density. The simplest example is the hydrogen atom with the degener
ate ground state wavefunctions for an electron with spin up and spin down,
i.e. Ψ
↑
= (1s)α and Ψ
↓
= (1s)β where α and β are the spin wavefunc
tions. Both wavefunctions obviously have the same total density. Another
example is provided by the the two degenerate ground state wavefunctions
(1s)
2
2p
+
and (1s)
2
2p
−
of the noninteracting lithium atom, where 2p
±
are
porbitals with angular momentum quantum numbers l = ±1. Both wave
functions lead to the same density
n(r) = 2[ϕ
1s
(r)[
2
+[ϕ
2p
+(r)[
2
= 2[ϕ
1s
(r)[
2
+[ϕ
2p
−(r)[
2
(127)
We have therefore constructed two degenerate ground states with the same
density. As a consequence the ground state expectation value of a given
operator may no longer be considered as a functional of the density (take,
for instance, the expectation values of the spin and angular momentum in
our examples). However, if two diﬀerent ground state density matrices have
the same density then also the energy Tr
ˆ
D
ˆ
H for those diﬀerent density ma
trices is the same. For every EVdensity n we can therefore unambiguously
deﬁne the ensemble version of the F
HK
functional as [16]
F
EHK
[n] = Tr
ˆ
D[n](
ˆ
T +
ˆ
W) (128)
where
ˆ
D[n] is any of the ground state ensemble density matrices corre
sponding to n. We can now deﬁne an extension of the energy functional
E
v
to the set of EVdensities
E
v
[n] =
_
n(r)v(r)dr +F
EHK
[n] = Tr
ˆ
D[n]
ˆ
H (129)
Similarly as for F
HK
we easily can prove
E[v] = inf
n∈B
__
n(r)v(r)dr +F
EHK
[n]
_
(130)
The functional F
EHK
is an extension of F
HK
since
F
EHK
[n] = F
HK
[n] if n ∈ / (131)
This follows directly from the fact that for a nondegenerate ground state
[Ψ[n]) corresponding to n we have
ˆ
D[n] = [Ψ[n])¸Ψ[n][, so
F
EHK
[n] = Tr
ˆ
D[n](
ˆ
T +
ˆ
W) = ¸Ψ[n][
ˆ
T +
ˆ
W[Ψ[n]) = F
HK
[n] (132)
We can furthermore prove that F
EHK
is convex by the same proof as for
F
HK
. Nothing is however known about the convexity of the set of EV
densities B itself which constitute the domain of F
EHK
. Let us collect our
results in the form of a theorem
Theorem 2 (HohenbergKohn) The EVdensity n speciﬁes the exter
nal potential up to a constant. Moreover,
1. The ground state energy of a system with external potential v can be
obtained from
E[v] = inf
n∈B
__
n(r)v(r)dr +F
EHK
[n]
_
where F
EHK
[n] = Tr
ˆ
D[n](
ˆ
T +
ˆ
W) and
ˆ
D[n] is a ground state density
matrix with Tr
ˆ
D[n]ˆ n(r) = n(r).
2. F
EHK
is convex
As we will now demonstrate the subset of PSVdensities of B is not convex.
More precisely, we will now show that there are EVdensities which are
not PSVdensities [1, 17]. As any EVdensity is a convex combination
of PSV densities this then demonstrates the nonconvexity of the set of
PSVdensities.
Consider an atom with total angular momentum quantum number L > 0
which has a 2L + 1degenerate ground state. The external potential v is
the spherically symmetric Coulomb potential of the atomic nucleus and
the degeneracy is a result of rotational invariance of the Hamiltonian of
the system. The ground state wavefunctions then transform among one
another according to a 2L + 1dimensional unitary representation of the
rotation group. We assume that there is no accidental degeneracy. If we
denote the ground state wave functions by ¦[Ψ[n
i
]) = [Ψ
i
), i = 1 . . . 2L+1¦
and the corresponding electron densities by n
i
then the following convex
combination
¯ n =
1
2L + 1
2L+1
i=1
n
i
(133)
is invariant under all rotations and therefore spherically symmetric. How
ever the n
i
are not spherical. In fact, there is no linear combination of the
ground states [Ψ
i
) that leads to a spherically symmetric density. As the n
j
is obtained from [Ψ
j
), which by a unitary transformation can be obtained
from any other [Ψ
i
), and the external potential is invariant under rotations
we ﬁnd that
_
n
i
(r)v(r)dr =
_
n
j
(r)v(r)dr =
_
¯ n(r)v(r)dr (134)
for all 0 ≤ i, j ≤ 2L + 1. Let us now suppose that ¯ n is a ground state
density obtained from a pure state wavefunction [Ψ[¯ n]). This wavefunction
is not a linear combination of the [Ψ
i
) otherwise ¯ n would not be spherically
symmetric. We then ﬁnd
F
EHK
[¯ n] = ¸Ψ[¯ n][
ˆ
H
v
[Ψ[¯ n]) −
_
¯ n(r)v(r)dr
>
2L+1
i=1
1
2L + 1
¸Ψ[n
i
][
ˆ
H[Ψ[n
i
]) −
_
¯ n(r)v(r)dr
=
2L+1
i=1
1
2L + 1
F
EHK
[n
i
] (135)
This then gives
F
EHK
[¯ n] >
2L+1
i=1
1
2L + 1
F
EHK
[n
i
] (136)
But we also know that F
EHK
is convex on the set of EVdensities. This
leads to a contradiction and hence we must conclude that ¯ n can not be a
pure state density of any potential. The density ¯ n is, however, a convex
combination of ground state densities corresponding to the same external
potential and therefore, by deﬁnition, an ensemble vrepresentable density.
We therefore have constructed an EVdensity which is not a PSVdensity.
For an explicit numerical example of such a density we refer to the work of
Aryasetiawan and Stott [18].
11. Linear response for degenerate ground states and
the diﬀerentiability of F
EHK
In the previous sections we learned to take functional derivatives on the
basis of linear response theory. We concentrated on nondegenerate ground
states for which the response functions are welldeﬁned. We will now use
response theory for degenerate ground states to study the diﬀerentiability
of the functional F
EHK
.
It is clear that in the case of degeneracy the expectation value of all ob
servables, except the energy, depends on which particular ground state we
choose to calculate the expectation value from. This poses a clear diﬃculty
for the deﬁnition of general density functionals and their functional deriva
tives. One may, however, argue that if a potential v
0
leads to a degenerate
ground state then we can always ﬁnd an arbitrarily small perturbation of
the potential δv that lifts the degeneracy and therefore the expectation
value
O[v
0
+δv] = ¸Ψ[v
0
+δv][
ˆ
O[Ψ[v
0
+δv]) (137)
of an observable described by an operator
ˆ
O does exist for any > 0, where
[Ψ[v
0
+ δv]) is the ground state wavefunction of the perturbed system.
Therefore also the limit
¯
O[v
0
, δv] = lim
→0
O[v
0
+δv] (138)
is deﬁned, although it will in general depend on the potential variation δv.
One can therefore deﬁne the functional derivative as follows
O
[δv] = lim
→0
O[v
0
+δv] −
¯
O[v
0
, δv]
(139)
However, we will see that these derivatives depend in a nonlinear way on δv
and are therefore only deﬁned as Gˆ ateaux variations rather than Gˆ ateaux
derivatives. We shall especially be interested in the case where
ˆ
O is the
density operator. For the energy the limit in Eq.(138) is independent of δv
and we may wonder if a Gˆ ateaux derivative still exists. Let us investigate
this in more detail. Suppose we have an Hamiltonian
ˆ
H
0
with external
potential v
0
which has a qfold degenerate ground state. Let us now apply
a perturbation δv which lifts the degeneracy. Then there are q eigenstates
¦[Ψ
k
()), k = 1 . . . q¦ with energies E
k
() of the perturbed Hamiltonian
that are continuously connected to degenerate ground states [Ψ
k
(0)) of the
unperturbed Hamiltonian
ˆ
H
0
, i.e.
lim
→0
[Ψ
k
()) = [Ψ
k
(0)) (140)
with
E
k
(0) = ¸Ψ
k
(0)[
ˆ
H
0
[Ψ
k
(0)) = E
0
(141)
Note that since the states [Ψ
k
()) are orthonormal, also the limiting states
states [Ψ
k
(0)) in the degenerate ground state multiplet are orthonormal.
Which particular ground states of the ground state manifold are reached
by the → 0 limit depends obviously on the form of the perturbation δv.
In which way it depends on δv we will investigate now. The ground state
wavefunctions [Ψ
k
()) satisfy the equation
(
ˆ
H
0
+δ
ˆ
V +
2
2
δ
ˆ
U)[Ψ
k
()) = E
k
()[Ψ
k
()) (142)
where, for the later discussion, we also added a onebody potential δ
ˆ
U
which is of second order in . It turns out that this term inﬂuences the
ﬁrst order density response. We note, like in the nondegenerate case, that
if [Ψ
k
()) is a solution to this equation then also [Φ
k
()) = e
iθ
k
()
[Ψ
k
()) is
a solution, where θ
k
() is an arbitrary function of . This freedom will, as
in the nondegenerate case, not aﬀect any of the expectation values. If we
expand the Schr¨odinger equation (142) to ﬁrst order in we obtain
(
ˆ
H
0
−E
0
)[Ψ
k
(0)) = (E
k
(0) −δ
ˆ
V )[Ψ
k
(0)) (143)
where E
k
(0) and [Ψ
k
(0)) are the ﬁrst order derivatives of E
k
() and [Ψ
k
())
with respect to in = 0. In order to solve this equation for [Ψ
k
(0)) we
expand this quantity in the q connected ground states and all the other
eigenstates of
ˆ
H
0
, i.e.
[Ψ
k
(0)) =
∞
i=1
c
k
i
[Ψ
i
) (144)
where [Ψ
i
) = [Ψ
i
(0)) for i = 1 . . . q and the states ¦[Ψ
i
), i > q¦ are eigen
states of
ˆ
H
0
with eigenenergies E
i
> E
0
. If we insert this expansion into
the ﬁrst order Schr¨odinger equation (143) we obtain the equation
∞
i=1
c
k
i
(E
i
−E
0
)[Ψ
i
) = (E
k
(0) −δ
ˆ
V )[Ψ
k
(0)) (145)
If we multiply this equation from the left with ¸Ψ
j
[ where 1 ≤ j ≤ q we
obtain
0 = E
k
(0)δ
jk
−¸Ψ
j
(0)[δ
ˆ
V [Ψ
k
(0)) (146)
This is the equation tells us exactly how δv picks out the ground states in
the degenerate multiplet. They are exactly the ones that diagonalize δ
ˆ
V
within the qdimensional space of degenerate ground state functions. This
equation also tells us that
E
k
(0) = ¸Ψ
k
(0)[δ
ˆ
V [Ψ
k
(0)) =
_
d
3
rn
k
(r)δv(r) (147)
If the state perturbed state with the lowest energy has label k = 1 then we
see that the functional derivative of the ground state energy is given as
δE
δv(r)
= n
1
(r) (148)
However, since n
1
implicitely depends on the perturbation δv which picked
out a particular ground state, this is not a proper Gˆ ateaux derivative but
only a Gˆ ateaux variation. So even the energy functional has no functional
derivative in the proper sense. Let us investigate the consequences for
diﬀerent expectation values. We therefore ﬁrst have to determine the ﬁrst
order wavefunction. If we multiply Eq.(145) from the right with ¸Ψ
i
[ where
i > q we obtain the coeﬃcient c
k
i
for i > q and the following expression for
[Ψ
k
(0)):
[Ψ
k
(0)) =
q
i=1
c
k
i
[Ψ
i
) −
∞
i=q+1
[Ψ
i
)¸Ψ
i
[δ
ˆ
V [Ψ
k
(0))
E
i
−E
0
(149)
However, the coeﬃcients c
k
i
for 1 ≤ i ≤ q are not determined by Eq.(143).
We know from the orthonormality of the states [Ψ
k
()) that
0 =
d¸Ψ
i
[Ψ
k
)
d
(0) = ¸Ψ
i
(0)[Ψ
k
(0)) +¸Ψ
i
(0)[Ψ
k
(0)) = c
i∗
k
+c
k
i
(150)
i.e. the coeﬃcients c
k
i
form an antiHermitian matrix. In particular we ﬁnd
that the coeﬃcients c
k
k
are purely imaginary. The value of these diagonal
elements is undetermined by the Schr¨odinger equation as they are related
to the arbitrary phase θ
k
() which we can choose for each [Ψ
k
()). As in the
nondegenerate case the value of any expectation value does not depend on
the coeﬃcients c
k
k
. This leaves us with the determination of the oﬀdiagonal
terms c
k
i
for i ,= k. These coeﬃcients can be found by expanding Eq.(142)
to second order in
2
. This yields
(
ˆ
H
0
−E
0
)[Ψ
k
(0)) = 2(E
k
(0) −δ
ˆ
V )[Ψ
k
(0)) + (E
k
(0) −δ
ˆ
U)[Ψ
k
(0)) (151)
where [Ψ
k
(0)) and E
k
(0) are the second order derivatives of [Ψ
k
()) and
E
k
() in = 0. If we multiply this equation from the left with ¸Ψ
l
(0)[ where
1 ≤ l ≤ q and use Eq.(146) and Eq.(149) then we obtain
0 = 2(E
k
(0) −E
l
(0))c
k
l
+E
k
(0)δ
lk
−¸Ψ
l
(0)[δ
ˆ
U[Ψ
k
(0))
+ 2
∞
i=q+1
¸Ψ
l
(0)[δ
ˆ
V [Ψ
i
)¸Ψ
i
[δ
ˆ
V [Ψ
k
(0))
E
i
−E
0
(152)
If we in this equation take k = l we obtain an expression for the energy to
second order in :
E
k
(0) = 2
_
d
3
rd
3
r
δv(r)χ
k
(r, r
)δv(r
) +
_
d
3
rn
k
(r)δu(r) (153)
where n
k
is the density corresponding to [Ψ
k
(0)) and χ
k
the following
function
χ
k
(r, r
) =
∞
i=q+1
¸Ψ
k
(0)[ˆ n(r)[Ψ
i
)¸Ψ
i
[ˆ n(r
)[Ψ
k
(0))
E
i
−E
0
+c.c. (154)
If, on the other hand, we take l ,= k we obtain our desired equation for the
coeﬃcients c
k
i
:
c
k
l
=
1
2
¸Ψ
l
(0)[δ
ˆ
U[Ψ
k
(0))
E
k
(0) −E
l
(0)
+
1
E
k
(0) −E
l
(0)
∞
i=q+1
¸Ψ
l
(0)[δ
ˆ
V [Ψ
i
)¸Ψ
i
[δ
ˆ
V [Ψ
k
(0))
E
i
−E
0
(155)
We see that indeed c
k∗
l
= −c
l
k
as expected. We have therefore completely
determined the ﬁrst order change in the wavefunction. Let us see what this
implies for the observables. The ﬁrst order change in the expectation value
of an observable
ˆ
O is given by
O
(0) = ¸Ψ
k
(0)[
ˆ
O[Ψ
k
(0)) +c.c. (156)
If we insert our expression for the ﬁrst order change in the wavefunction
we obtain
O
(0) =
_
d
3
rζ
k
(r)δv(r) +
1
2
_
d
3
rη
k
(r)δu(r)
+
_
d
3
rd
3
r
ξ
k
(r, r
)δv(r)δv(r
) (157)
where we deﬁned the following functions
ζ
k
(r) =
∞
i=q+1
¸Ψ
k
(0)[
ˆ
O[Ψ
i
)¸Ψ
i
[ˆ n(r)[Ψ
k
(0))
E
i
−E
0
+c.c. (158)
η
k
(r) =
q
l=1(l=k)
¸Ψ
k
(0)[
ˆ
O[Ψ
l
)¸Ψ
l
[ˆ n(r)[Ψ
k
(0))
E
k
(0) −E
l
(0)
+c.c. (159)
ξ
k
(r, r
) =
q
l=1
∞
i=q+1
¸Ψ
k
(0)[
ˆ
O[Ψ
l
)¸Ψ
l
[ˆ n(r)[Ψ
i
)¸Ψ
i
[ˆ n(r
)[Ψ
k
(0))
(E
k
(0) −E
l
(0))(E
i
−E
0
)
(160)
If we choose
ˆ
O = ˆ n(r
) then we obtain an expression for the ﬁrst order
change in the density
δn
k
(r
) =
_
d
3
rχ
k
(r
, r)δv(r) +
1
2
_
d
3
rη
k
(r
, r)δu(r)
+
_
d
3
rd
3
r
ξ
k
(r
, r, r
)δv(r)δv(r
) (161)
where the functions χ
k
, η
k
and ξ
k
are obtained by inserting ˆ n(r
) for
ˆ
O in
Eqns.(158), (159) and (160). If we denote the perturbed state with label
k = 1 as the one with the lowest energy then δn
1
represents the change
in ground state density. We see that this density change does not depend
linearly on the ﬁrst order change δv in the potential and even the second
order change δu of the potential contributes to its value. Let us now take
δu = 0 and write
δn
1
(r
) =
_
d
3
rχ
1
(r
, r)δv(r)
+
_
d
3
rd
3
r
ξ
1
(r
, r, r
)δv(r)δv(r
) (162)
We can ask ourselves the question whether a given ﬁrst order variation δn
1
uniquely determines the ﬁrst order density change δv. One can show from
the HohenbergKohn theorem for degenerate states that this is indeed the
case. If we in Eq.(126) take v
2
= v
1
+ δv(r) where δv is not a constant
function we obtain
2
_
d
3
rδn
1
(r)δv(r) +O(
3
) < 0 (163)
Dividing by
2
and taking the limit → 0 then yields
_
d
3
rδn
1
(r)δv(r) < 0 (164)
We see that the ﬁrst order density change δn
1
can not be zero if δv is not
a constant. In contrast to the nondegenerate case Eq.(164) does not imply
that the function χ
1
is invertable, only the relation in Eq.(162) is invertable
where the inverse only exists on the set of vrepresentable density variations
within the set B.
Let us now see what the linear response theory can tell us about the diﬀer
entiability of F
EHK
. We consider again a system with a degenerate ground
state and external potential v
0
. Let us take out of the set of ground state
ensemble densities for this system a particular density n
0
. We may then
wonder whether the following limit
δF
EHK
δn
[δn] = lim
→0
F
EHK
[n
0
+δn] −F
EHK
[n
0
]
(165)
exists. For this limit to be welldeﬁned we have to make sure that n
0
+
δn ∈ B and we run into the same problem as for the nondegenerate case.
However, in the nondegenerate case we could solve this diﬃculty using
response theory. We will do the same for the degenerate case, but we will
see that some diﬃculties remain. Suppose we look at the perturbed system
with potential v
0
+ δv in which the degeneracy is lifted. For this system
there is a welldeﬁned ground state density n
= n[v
0
+δv]. From response
theory we known that we can write n
as
n
(r) = n
1
(r) +δn
1
(r) +m
(r) (166)
where
n
1
(r) = lim
→0
n[v
0
+δv] (167)
0 = lim
→0
m
(r)
(168)
and where δn
1
is explicitly given in Eq.(162). Now by construction the
density n
is in B for all values of . We can now consider the limit
δF
EHK
δn
[δn
1
] = lim
→0
F
EHK
[n
1
+δn
1
+m
] −F
EHK
[n
1
]
= lim
→0
F
EHK
[v
0
+δv] −
¯
F
EHK
[v
0
, δv]
(169)
where
¯
F
EHK
[v
0
, δv] is deﬁned as in Eq.(138). The latter limit is readily
calculated by inserting
ˆ
T +
ˆ
W =
ˆ
H
0
−
ˆ
V
0
for the operator
ˆ
O in Eqns.(158)
and (160). We obtain for the ﬁrst order change of F
EHK
δF
EHK
= −
_
d
3
rd
3
r
v
0
(r
)χ
1
(r
, r)δv(r)
−
_
d
3
rd
3
r
d
3
r
v
0
(r
)ξ
1
(r
, r, r
)δv(r)δv(r
)
= −
_
d
3
rv
0
(r)δn
1
(r) (170)
and we thus obtain
lim
→0
F
EHK
[n
1
+δn
1
+m
] −F
EHK
[n
1
]
= −
_
d
3
rv
0
(r)δn
1
(r) (171)
We see that this limit is linear in δn
1
and that furthermore v
0
is independent
of δn
1
. Therefore we can write for the functional derivative of F
EHK
in n
1
:
δF
EHK
δn(r)
(n
1
) = −v
0
(r) (172)
up to an arbitrary constant. However, we see that this derivative only exists
for some special set of ground state densities corresponding to potential v
0
.
The derivative exists for those ground state densities that correspond to
pure states that can be obtained in the → 0 limit for a perturbed system
with potential v
0
+δv. It is not clear how to take the functional derivative
at an arbitrary ensemble density n
0
for potential v
0
. We saw that there
exist EVdensities that are not PSVdensities. If we consider an ensemble
corresponding to such a density and change the external potential to v
0
+δv
where δv lifts the degeneracy then for > 0 the ensemble will change into a
pure state and the density will change abruptly. We must therefore conclude
that for general EVdensities the functional derivative of F
EHK
does not
exist. This poses not only a theoretical problem but also a practical one. As
we will discuss later, it is known from numerical investigations that there
are ground state densities of interacting systems that are not pure state
densities for a noninteracting system and therefore there is a clear need for
establishing a KohnSham scheme for arbitrary EVdensities. Fortunately
it turns out that one can deﬁne an extension of the functional F
EHK
to a
larger domain of densities which can be shown to be diﬀerentiable at the
set of all EVdensities. This functional is the Lieb functional F
L
and will
be studied in the next section.
12. The Levy and Lieb functionals F
LL
[n] and F
L
[n]
The functionals F
HK
and F
EHK
have the unfortunate mathematical dif
ﬁculty that their domains of deﬁnition / and B, although they are well
deﬁned, are diﬃcult to characterize, i.e. it is diﬃcult to know if a given
density n belongs to / or B. It is therefore desirable to extend the domains
of deﬁnition of F
HK
and F
EHK
to an easily characterizable (preferably
convex) set of densities. This can be achieved using the constrained search
procedure introduced by Levy [19]. We deﬁne the LevyLieb functional F
LL
as
F
LL
[n] = inf
Ψ→n
¸Ψ[
ˆ
T +
ˆ
W[Ψ) (173)
where the inﬁmum is searched over all normalized antisymmetric Nparticle
wave functions in H
1
(¹
3N
) yielding density n. As shown earlier such a den
sity is always in the convex set o which is again a subspace of L
1
∩L
3
. One
can furthermore show, as has been done by Simon [1], that the inﬁmum is
always a minimum, i.e. there is always a minimizing wave function. As this
is the ﬁrst important result that we obtain for F
LL
we put it in the form
of a theorem
Theorem 3 For any n ∈ o there is a [Ψ[n]) ∈ H
1
(¹
3N
) such that
F
LL
[n] = ¸Ψ[n][
ˆ
T +
ˆ
W[Ψ[n])
Let us discuss some properties of F
LL
. The functional F
LL
is an extension
of the HohenbergKohn functional F
HK
, which is deﬁned on /, to the larger
set o, i.e
F
LL
[n] = F
HK
[n] if n ∈ / (174)
This is readily derived. Suppose n is some ground state density correspond
ing to some external potential v and ground state [Ψ[n]) then
_
n(r)v(r)dr +F
HK
[n] = ¸Ψ[n][
ˆ
H[Ψ[n]) = inf
Ψ→n
¸Ψ[
ˆ
H[Ψ) =
_
n(r)v(r)dr + inf
Ψ→n
¸Ψ[
ˆ
T +
ˆ
W[Ψ) =
_
n(r)v(r)dr +F
LL
[n] (175)
We deﬁne a corresponding energy functional
E
v
[n] =
_
n(r)v(r)dr +F
LL
[n] (176)
If n
0
is the ground state density for potential v with corresponding ground
state wave function Ψ[n
0
] then
E
v
[n] = inf
Ψ→n
¸Ψ[
ˆ
H[Ψ) ≥ ¸Ψ[n
0
][
ˆ
H[Ψ[n
0
]) = E
v
[n
0
] (177)
Minimizing E
v
over the set o therefore yields the ground state density n
0
corresponding to external potential v. The functional F
LL
has however the
inconvenient property that it is not convex.
Theorem 4 The functional F
LL
is not convex
To show this we take the example of a previous section where we presented
a density ¯ n which did not correspond to a ground state wavefunction. It
was a convex combination of 2L + 1 degenerate ground state densities n
i
with corresponding ground states [Ψ[n
i
]) for an external potential v. Then
we ﬁnd
_
¯ n(r)v(r)dr +F
LL
[¯ n] = inf
Ψ→¯ n
¸Ψ[
ˆ
H[Ψ)
>
1
2L + 1
2L+1
i=1
¸Ψ[n
i
][
ˆ
H[Ψ[n
i
]) =
1
2L + 1
2L+1
i=1
F
LL
[n
i
] +
_
¯ n(r)v(r)dr
(178)
and we ﬁnd
F
LL
[¯ n] >
1
2L + 1
N
i=1
F
LL
[n
i
] (179)
which proves the nonconvexity of F
LL
. This is somewhat unfortunate as
convexity is an important property which can be used to derive diﬀeren
tiability of functionals. We will therefore now deﬁne a diﬀerent but related
convex functional with the same domain o. This is the Lieb functional F
L
deﬁned as
F
L
[n] = inf
ˆ
D→n
Tr
ˆ
D(
ˆ
T +
ˆ
W) (180)
where the inﬁmum is searched over all Nparticle density matrices
ˆ
D =
i=1
λ
i
[Ψ
i
)¸Ψ
i
[
i=1
λ
i
= 1 [Ψ
i
) ∈ H
1
(¹
3N
) (181)
which yield the given density n(r) = Tr
ˆ
Dˆ n(r) where ¦[Ψ
i
)¦ is an orthonor
mal set. Also for this case one can prove the inﬁmum to be a minimum,
i.e. there is a minimizing density matrix. We put the result again in the
form of a theorem
Theorem 5 For every n ∈ o there is a density matrix
ˆ
D[n] such that
F
L
[n] = Tr
ˆ
D[n](
ˆ
T +
ˆ
W)
For the proof of this theorem we again refer to Lieb [1]. The functional F
L
is an extension of F
EHK
to the larger set o, that is
F
L
[n] = F
EHK
[n] if n ∈ B (182)
This follows directly from the fact that if n ∈ B then there is a potential v
which generates a ground state ensemble density matrix
ˆ
D[n] which yields
n. So
_
n(r)v(r)dr +F
EHK
[n] =
_
n(r)v(r)dr + Tr
ˆ
D[n](
ˆ
T +
ˆ
W)
= Tr
ˆ
D[n]
ˆ
H = inf
ˆ
D→n
Tr
ˆ
D
ˆ
H
=
_
n(r)v(r)dr +F
L
[n] (183)
We can again deﬁne an energy functional
E
v
[n] =
_
n(r)v(r)dr +F
L
[n] (184)
which by a similar proof as for F
LL
assumes its minimum at the ground
state density corresponding to potential v. We further have the following
relations
F
L
[n] = F
LL
[n] if n ∈ / (185)
and
F
L
[n] < F
LL
[n] if n ∈ B and n ,∈ / (186)
The ﬁrst relation follows from the fact that if the density n is a pure state
vrepresentable density then the minimizing density matrix for F
L
is a pure
state density matrix. The second relation also easily follows. We take n to
be an EVdensity which is not a PSV density. There is a ground state
ensemble density matrix
ˆ
D[n] for which we have
_
n(r)v(r)dr+F
L
[n] =
_
n(r)v(r)dr+Tr
ˆ
D[n](
ˆ
T+
ˆ
W) = ¸Ψ
i
[
ˆ
H[Ψ
i
) (187)
where [Ψ
i
) is any of the ground states in the degenerate ground state multi
plet. Any wave function yielding density n can not be a linear combination
of these ground state wavefunctions otherwise n would be pure state v
representable. Therefore its expectation value with the Hamiltonian must
be larger, i.e
¸Ψ
i
[
ˆ
H[Ψ
i
) < inf
Ψ→n
¸Ψ[
ˆ
H[Ψ) =
_
n(r)v(r)dr +F
LL
[n] (188)
which proves our statement.
We will now demonstrate another important property of F
L
, which is its
convexity.
Theorem 6 The functional F
L
is convex
This is easily shown. If n = λ
1
n
1
+λ
2
n
2
with λ
1
+λ
2
= 1 and 0 ≤ λ
1
, λ
2
≤ 1
then we have
λ
1
F
L
[n
1
] +λ
2
F
L
[n
2
] = λ
1
inf
ˆ
D1→n1
Tr
ˆ
D
1
(
ˆ
T +
ˆ
W)
+ λ
2
inf
ˆ
D2→n2
Tr
ˆ
D
2
(
ˆ
T +
ˆ
W)
= inf
ˆ
D1,
ˆ
D2→n1,n2
Tr(λ
1
ˆ
D
1
+λ
2
ˆ
D
2
)(
ˆ
T +
ˆ
W)
≥ inf
ˆ
D→n
Tr
ˆ
D(
ˆ
T +
ˆ
W) = F
L
[n] (189)
We therefore now have established that F
L
is a convex functional on a
convex space. This is important information which enables us to derive
the Gˆ ateaux diﬀerentiability of the functional F
L
at the set B of ensemble
vrepresentable densities. We will discuss this feature of F
L
in the next
section.
Having obtained some desirable convexity properties of F
L
we try to obtain
some analytic properties of this functional. An obvious question to ask is
whether this functional is continuous. To be more precise, suppose that
a series of densities n
k
approaches a given density n in some sense, for
instance n − n
k

1
→ 0 and n − n
k

3
→ 0 for k → ∞ in L
1
∩ L
3
. Does
this imply that [F
L
[n
k
] − F
L
[n][ → 0. ? It turns out that this question is
not easily answered. However, one can prove a weaker statement. Suppose
n is an EVdensity corresponding to potential v of Hamiltonian
ˆ
H
v
. Then
F
L
[n] +
_
d
3
rn(r)v(r) = Tr
ˆ
D[n]
ˆ
H
v
≤ Tr
ˆ
D[n
k
]
ˆ
H
v
= F
L
[n
k
] +
_
d
3
rn
k
(r)v(r) (190)
and therefore
F
L
[n] ≤ F
L
[n
k
] +
_
d
3
rv(r)(n
k
(r) −n(r)) (191)
Further, since n
k
→ n in the norms on L
1
∩ L
3
for each > 0 there is an
integer M such that n −n
k

1
≤ and n −n
k

3
≤ for k > M. Now we
can split up v ∈ L
3
2
+L
∞
as v = u +w where u ∈ L
3
2
and w ∈ L
∞
and we
have
[
_
d
3
rv(r)(n
k
(r) −n(r))[ ≤ u3
2
n −n
k

3
+w
∞
n −n
k

1
(192)
So if n − n
k
]
1
≤ /C and n − n
k
]
3
≤ /C, where C = u3
2
+ w
∞
then
F
L
[n] ≤ F
L
[n
k
] + (193)
for k suﬃciently large. We therefore see that if we take → 0 from above
then F
L
[n
k
] approaches F
L
[n] from above and n
k
→ n in the norms on
L
1
∩L
3
. This, of course, does not imply that F
L
[n] is continuous in n. For
that we would have to prove that [F
L
[n] −F
L
[n
k
][ ≤ rather than F
L
[n] −
F
L
[n
k
] ≤ . What we have proven is a weaker form of continuity, known
as semicontinuity. Because the limit point is a lower bound, the functional
with the property in Eq.(193) is called lower semicontinuous. This can
also be characterized diﬀerently. If we deﬁne inf F[n
m
] by inf F
L
[n
m
] =
inf¦F
L
[n
k
][k ≥ m¦ then lower semicontinuity implies
F
L
[n] ≤ lim
m→∞
inf F
L
[n
m
] (194)
Since we required that n is an EVdensity we have proven that F
L
is lower
semicontinuous on the set of EVdensities. It turns out that one can prove
that F
L
is lower semicontinuous on all densities in o (see theorem 4.4 of
Lieb [1]). However, the proof of this is not simple and we will therefore not
try to reproduce it here. Since the result is important we present it here in
the form of a theorem
Theorem 7 Suppose n
k
and n ∈ o and n
k
→ n for k → ∞ in the norms
on L
1
∩ L
3
. Then
F
L
[n] ≤ lim
k→∞
inf F
L
[n
k
]
In other words F
L
is lower semicontinuous on the set o.
One can prove an even stronger theorem in which we only need weak con
vergence of the series n
k
. We will, however, not need that property in
the remainder of this review. The notion of lower semicontinuity is an
important property for convex functionals which will allow us to make sev
eral other useful statements about other properties of F
L
. One particular
consequence of lower semicontinuity that we will use, is that for a lower
semicontinuous convex functional G : B → ¹ the following set of points
epi(G) = ¦(n, r) ∈ B ¹[G[n] ≤ r¦ (195)
is closed and convex [11]. This set is called the epigraph of G and is simply
the set of points that lie above the graph of G. The closedness property
means that if a series of points (n
k
, r
k
) in the epigraph of G converges to a
point (n, r) then this point also lies in the epigraph. This also implies that
the set of interior points of epi(G), i.e. the set of points that lie strictly
above G, is an open convex set (this means that for every point in this set
there is a neighborhood that contains it). We will need this property in the
next section.
13. Diﬀerentiability of F
L
In this section we will prove that the Lieb functional is diﬀerentiable on the
set of EVdensities and nowhere else. The functional derivative at a given
EVdensity is equal to −v where v is the external potential that generates
the EVdensity at which we take the derivative. To prove existence of the
derivative we use the geometric idea that if a derivative of a functional G[n]
in a point n
0
exists, then there is a unique tangent line that touches the
graph of G in a point (n
0
, G[n
0
]). To discuss this in more detail we have
to deﬁne what we mean with a tangent. The discussion is simpliﬁed by
the fact that we are dealing with convex functionals. If G : B → ¹ is a
diﬀerentiable and convex functional from a normed linear space B to the
real numbers then from the convexity property it follows that for n
0
, n
1
∈ B
and 0 ≤ λ ≤ 1 that
G[n
0
+λ(n
1
−n
0
)] −G[n
0
] ≤ λ(G[n
1
] −G[n
0
]) (196)
From the fact that G is diﬀerentiable we then ﬁnd
G[n
1
] −G[n
0
] ≥ lim
λ→0
G[n
0
+λ(n
1
−n
0
)] −G[n
0
]
λ
=
δG
δn
[n
1
−n
0
] (197)
Therefore
G[n
1
] ≥ G[n
0
] +
δG
δn
[n
1
−n
0
] (198)
Now δG/δn is a continuous linear functional which will be identiﬁed with
a tangent. This is the basis of the following deﬁnition. If for a convex
functional G there is a continuous linear functional L : B → ¹ such that
G[n
1
] ≥ G[n
0
] +L[n
1
−n
0
] (199)
then L is called a subgradient or tangent functional in n
0
. The geometric
idea behind this deﬁnition is illustrated in Figure 1.
F[n]
n
n
0
F[n
0
]+L[nn
0
]
Figure 1: The convex functional F has a unique tangent
functional L in the point n
0
For the Lieb functional F
L
we will now prove the following statement:
Theorem 8 The functional F
L
has a unique tangent functional for every
EVdensity and nowhere else. Moreover the tangent functional at an EV
density n can be identiﬁed with −v where v is the potential that generates
this density.
In order to show this we ﬁrst deﬁne the energy functional
E[v] = inf
ˆ
D
Tr
ˆ
D
ˆ
H
v
(200)
where the inﬁmum is searched over all density matrices of the form given
in Eq.(181). This is an extension of the previously deﬁned functional in
Eq.(130) to all potentials in L
3
2
+ L
∞
. One can further show that if an
inﬁmum exists the the minimizing density matrix satisﬁes the Schr¨odinger
equation [1]. Let us ﬁrst suppose that n
0
is an EVdensity corresponding
to potential v of Hamiltonian
ˆ
H
v
. Then
F
L
[n
0
] +
_
d
3
rn
0
(r)v(r) = Tr
ˆ
D[n
0
]
ˆ
H
v
≤ Tr
ˆ
D[n]
ˆ
H
v
= F
L
[n] +
_
d
3
rn(r)v(r) (201)
and therefore
F
L
[n] ≥ F
L
[n
0
] −
_
d
3
rv(r)(n(r) −n
0
(r)) (202)
Therefore the functional L
L[n] = −
_
d
3
rv(r)n(r) (203)
deﬁnes a linear and continuous functional (continuity follows from Eq.(192)
) and is therefore a tangent functional. Now we have to show its uniqueness.
Suppose L
is a diﬀerent tangent functional. Since F is deﬁned on B =
L
1
∩L
3
the tangent functional must represent an element in the dual space
B
∗
= L
3
2
+L
∞
and can be written as
L
[n] = −
_
d
3
r˜ v(r)n(r) (204)
with ˜ v ∈ L
3
2
+L
∞
and ˜ v ,= v +C. Now since we assumed that L
was also
tangent functional we have
F
L
[n] ≥ F
L
[n
0
] −
_
d
3
r˜ v(r)(n(r) −n
0
(r)) (205)
and hence
F
L
[n] +
_
d
3
rn(r)˜ v(r) ≥ F
L
[n
0
] +
_
d
3
rn
0
(r)˜ v(r) (206)
We take the inﬁmum over all densities in o on the left hand side. Since
we know that there is ground state density matrix corresponding to n
0
we
obtain from Eq.(206):
E[˜ v] = inf
n∈S
¦F
L
[n] +
_
d
3
rn(r)˜ v(r)¦ ≥ Tr
ˆ
D[n
0
]
ˆ
H
˜ v
(207)
Now there are two cases, either the Hamiltonian with potential ˜ v is able to
support a bound ground state or it is not. If it does support a ground state
then
Tr
ˆ
D[n
0
]
ˆ
H
˜ v
> E[˜ v] (208)
since
ˆ
D[n
0
] is not a ground state density matrix for ˜ v. Together with
Eq.(207) this immediately leads to the contradiction E[˜ v] > E[˜ v]. If the
Hamiltonian with potential ˜ v does not support a bound ground state then
there is no normalized density matrix for which the inﬁmum of E[˜ v] as in
Eq.(200) is attained. We then again obtain Eq.(208) and the same contra
diction E[˜ v] > E[˜ v]. We therefore conclude that the EVdensity n
0
has a
unique tangent functional equal to −v where v is the potential that yields
ground state density n
0
.
Let us now suppose that n
0
is not an EVdensity. Let us further suppose
that there is a tangent functional at n
0
, i.e. that for some ˜ v Eq. (205)
and therefore also Eq.(206) are satisﬁed. The constrained search for F
L
[n
0
]
always has a minimizing density matrix, which we call
ˆ
D[n
0
]. With this
density matrix then also Eq.(207) is true. Since n
0
is not an EVdensity
ˆ
D[n
0
] can not be a ground state density matrix for the Hamiltonian with
potential ˜ v. With the same arguments as before we ﬁnd that Eq.(208) must
be true which again leads to the contradiction E[˜ v] > E[˜ v]. Therefore there
is no tangent functional at n
0
if n
0
is not an EVdensity. We have there
fore proven our statement, there is a unique tangent functional at every
EVdensity and nowhere else.
As a next step we will show that this implies that the functional F
L
Gˆ ateaux
diﬀerentiable at every EVdensity and nowhere else. For the application
of the next theorem it is desirable to extend the domain of F
L
to all of
L
1
∩ L
3
. We follow Lieb [1] and deﬁne
F
L
[n] =
_
inf
ˆ
D→n
Tr
ˆ
D(
ˆ
T +
ˆ
W) if n ∈ o
+∞ if n ∈ L
1
∩ L
3
and n ,∈ o
(209)
The reader may seem surprised with the appearance of +∞ in this def
inition. However, inﬁnite values are welldeﬁned in the theory of convex
functionals [11] and they are usually introduced to deal in a simple way
with domain questions. With this deﬁnition the functional F
L
is a convex
lower semicontinuous functional of the whole space L
1
∩ L
3
. We are now
ready to introduce the following key theorem which we will use to prove
diﬀerentiability of F
L
at the set of EVdensities:
Theorem 9 Suppose F : B → ¹ is a lower semicontinous convex func
tional which is ﬁnite at a convex subset V ⊂ B of a normed linear space
B. If F has a unique continuous tangent functional L : B → ¹ at n
0
∈ V
then F is Gˆ ateaux diﬀerentiable at n
0
and L = δF/δn.
The fact that uniqueness of a tangent functional implies diﬀerentiability is
clear from a geometric point of view. In Figure 2 we display an example of
a functional (in this picture the functional is simply a function) which is not
diﬀerentiable at n
0
. Consequently, there is no unique tangent functional
at that point. In the example there are in fact inﬁnitely many tangent
functionals at n
0
of which there are two drawn in the plot.
F[n]
n n
0
L
1
L
2
Figure 2: The convex functional F has no unique tangent
functional in the point n
0
. Both L
1
and L
2
are tangent
functionals
In the case of inﬁnitely many dimensions one should, however, be careful
in drawing conclusions from simple plots and we have to resort to formal
proofs. Since the proof of the previous theorem is important for our discus
sion and not diﬃcult to follow we will present it here. Since V is a convex
set we have that if n
0
∈ V and n
1
∈ V then n
0
+ λ(n
1
− n
0
) ∈ V for
0 ≤ λ ≤ 1 and since F is convex we obtain
F[n
0
+λ(n
1
−n
0
)] ≤ F[n
0
] +λ(F[n
1
] −F[n
0
]) (210)
Furthermore since F has a unique continuous tangent functional at n
0
we
have
F[n
0
+λ(n
1
−n
0
)] ≥ F[n
0
] +λL[n
1
−n
0
] (211)
Note that n
1
−n
0
need not be in V but is always in B since that is a linear
space. If we combine the two inequalities we ﬁnd that
F[n
1
] −F[n
0
] ≥
F[n
0
+λ(n
1
−n
0
)] −F[n
0
]
λ
≥ L[n
1
−n
0
] (212)
Now since F is convex the function g(λ) = F[n
0
+λ(n
1
−n
0
)] −F[n
0
] is a
convex real function on the interval [0, 1] and hence continuous. Moreover,
as the equation above shows, the function g(λ)/λ has a ﬁnite upper and
lower bound and therefore the limit λ → 0 of g(λ)/λ exists. We therefore
ﬁnd
F[n
1
] −F[n
0
] ≥ F
[n
0
, n
1
] ≥ L[n
1
−n
0
] (213)
where
F
[n
0
, n
1
] = lim
λ→0
F[n
0
+λ(n
1
−n
0
)] −F[n
0
]
λ
(214)
It remains to show that F
[n
0
, n
1
] is continuous and linear and equal to L.
Now F
has the property that for n
2
∈ V
F
[n
0
, n
0
+(n
2
−n
0
)] = lim
λ→0
F[n
0
+λ(n
2
−n
0
)] −F[n
0
]
λ
= F
[n
0
, n
2
]
(215)
Therefore from Eq.( 213) we ﬁnd
F[n
0
+(n
1
−n
0
)] ≥ F[n
0
] +F
[n
0
, n
1
] (216)
This means that the straight line
L = ¦(n
0
+(n
1
−n
0
), F[n
0
] +F
[n
0
, n
1
]) ∈ B ¹[ ∈ [0, 1]¦ (217)
lies below the graph of functional F. Now since F is convex and lower
semicontinuous the set of points above the graph of F (the interior points
of epi(F)) form an open convex set which is nonempty since F is ﬁnite
on V [11]. Now a famous theorem of functional analysis, the HahnBanach
theorem, tells us that if we have an open convex set A (the interior of epi(F)
in our case) and an aﬃne subspace L (which is the line L in our case) that
does not intersect A then there is a hyperplane H that contains L in which
the hyperplane H has the form
H = ¦(n, r) ∈ B ¹[L
[n] +αr = β¦ (218)
where L
: B → ¹ is a continuous linear functional and α and β are real
numbers (We can take α ,= 0 since F
[n
0
, n
1
] is ﬁnite and we will have
no vertical line L or vertical hyperplane H). We will not prove the Hahn
Banach theorem as it is geometrically intuitive and can be found in most
textbooks on functional analysis [20, 21]. We will just use its consequences.
Now we know that our line L is contained in H. The coeﬃcient β is then
determined by the fact that (n
0
, F[n
0
]) ∈ Hand we ﬁnd β = L
[n
0
]+αF[n
0
]
and therefore
H = ¦(n, r) ∈ B ¹[L
[n −n
0
] +α(r −F[n
0
]) = 0¦ (219)
Now from the fact that (n
0
+ (n
1
− n
0
), F[n
0
] + F
[n
0
, n
1
]) ∈ H we ﬁnd
that
L
[n
1
−n
0
] +αF
[n
0
, n
1
] = 0 (220)
From this equation and Eq.(213) we then see immediately that −L
/α is a
continuous tangent functional at n
0
which then must be equal to L since
that was the only continuous tangent functional at n
0
. Consequently
F
[n
0
, n
1
] = L[n
1
−n
0
] (221)
and thus
δF
δn
[n
1
−n
0
] = lim
λ→0
F[n
0
+λ(n
1
−n
0
)] −F[n
0
]
λ
= L[n
1
−n
0
] (222)
We therefore obtained what we wanted to prove, F is Gˆ ateaux diﬀeren
tiable at n
0
and δF/δn = L. Now we can apply this theorem to the Lieb
functional. If we take F = F
L
, B = L
1
∩ L
3
, V = o and use that F
L
has a
unique tangent functional L = −v at every EVdensity and nowhere else,
we obtain:
Theorem 10 The functional F
L
is Gˆ ateaux diﬀerentiable for every EV
density in the set o and nowhere else. Moreover the functional derivative
at an EVdensity is equal to −v where v is the potential that generates this
density.
This leaves us with the question which densities in set o actually are
EVdensities. A useful result on this point is obtained in the next section.
14. Ensemble vrepresentability
We have seen that the Lieb functional F
L
is diﬀerentiable at the set of E
Vdensities in o and nowhere else. For this reason it is desirable to know a
bit more about these densities. The question therefore is: which densities
are ensemble vrepresentable? In this section we will prove a useful result
which will enable us to put the KohnSham approach on a rigorous basis.
Theorem 11 The set of EVdensities is dense in the set o with respect
to the norm on L
1
∩ L
3
.
This statement means the following. Suppose we take an arbitrary
density n
0
from the set o, then for every > 0 we can ﬁnd an EVdensity
n such that n
0
− n
1
≤ and n
0
− n
3
≤ . In other words, for every
density in the set o there is an EVdensity arbitrarily close to it. This
can also be phrased diﬀerently. For every density n
0
in the set o there is a
series of EVdensities n
k
such that
lim
k→∞
n
0
−n
k

p
= 0 (223)
where the subindex for the norm takes the values p = 1 and p = 3. In order
to establish this result we need to use a theorem due to Bishop and Phelps.
For the clarity of the discussion we split the theorem into two parts which
both yield interesting results for our functional F
L
. The ﬁrst part gives
some insight in the set of potentials 1 that generate a bound ground state.
From the relation
E[v] = inf
n∈S
¦F
L
[n] +
_
d
3
rn(r)v(r)¦ (224)
we obtain immediately that for any n ∈ o and any v ∈ L
3
2
+L
∞
F
L
[n] ≥ −
_
d
3
rn(r)v(r) +E[v] (225)
Now the functional on the right hand side of the inequality sign is, for
a given v, a linear functional of n. The inequality sign tells us that this
functional lies below the graph of F
L
[n]. A linear functional with this
property is called F
L
bounded. Let us give a general deﬁnition of these
linear functionals. Let F be a functional F : B → ¹ from a normed
function space (a Banach space) B to the real numbers. Let B
∗
be the
dual space of B, i.e. the set of continuous linear functionals on B. Then
L ∈ B
∗
is said to be Fbounded if there is a constant C such that for all
n ∈ B
F[n] ≥ L[n] +C (226)
where the constant C may depend on L but not on n. We therefore see
from Eq.(225) that every v ∈ L
3
2
+ L
∞
deﬁnes an F
L
bounded functional
where L is deﬁned as in Eq.(203). The geometric picture of an Fbounded
functional is displayed in Figure 3.
F[n]
n
L
1
L
2
Figure 3: Both L
1
and L
2
are Fbounded by the convex
functional F
Note that the tangent functionals of F
L
are special cases of F
L
bounded
functionals, they are the F
L
bounded functionals that also touch the graph
of F
L
. Therefore the set of tangent functionals is a subset of the set of F
L

bounded functionals. This can be illustrated with an example. Consider
the function f(x) = exp (x) on the real axis. This function is convex and
all tangent functions are of the form g
α
(x) = αx + β where α > 0 and
β(α) ≤ 1. The constant function g
0
(x) = β with β ≤ 0 is an fbounded
function, but not a tangent. However, we can always ﬁnd a tangent with a
slope arbitrary close to zero, i.e. arbitrary close to the slope of the constant
fbounded function g
0
. The following theorem (due to Bishop and Phelps)
ensures that a similar situation occurs for the case of general Banach spaces:
Theorem 12 (BishopPhelps I) Let F : B → ¹ be a lower semicontin
uous convex functional on a real Banach space B. The functional F can take
the value +∞ but not everywhere. Then the continuous tangent functionals
to F are B
∗
norm dense in the set of Fbounded functionals in B
∗
.
This means that if L
0
is some Fbounded functional then we can ﬁnd a
set of tangent functionals L
k
such that
lim
k→∞
L
0
−L
k

B
∗ = 0 (227)
where the limit is taken in the norm on B
∗
. We will not prove this theorem
here. The proof is clearly described with a geometric interpretation in
reference [22]. Let us apply this theorem to our functional F
L
. We know
that every tangent functional of F
L
can be identiﬁed with −v, where v is
a potential that yields a bound ground state, i.e. v ∈ 1. Now we know
that every element −v
0
for an arbitrary v
0
∈ L
3
2
+ L
∞
corresponds to an
F
L
bounded functional. Therefore for such a v
0
there is a series of v
k
∈ 1
such that
lim
k→∞
v
0
−v
k
 = 0 (228)
where the norm is the L
3
2
+L
∞
norm described in Eq.(24). Therefore every
potential v
0
∈ L
3
2
+ L
∞
can be approximated to arbitrary accuracy by a
potential that yields a bound ground state. This may seem counterintuitive
at ﬁrst sight as one can imagine v
0
to be a repulsive potential, until one
realizes that one may put the system in big box with a wall of ﬁnite height.
This system will have a bound state if the size of the box is chosen to be
big enough. The particles will then spread out over the box in order to
minimize the repulsion between them. If we choose v
k
= v
0
+w
k
where w
k
describes a series of boxes of increasing size but with decreasing height of
the potential wall then we see that we have created a series of potentials
in 1 that approaches v
0
to an arbitrary accuracy. Let us now discuss the
second part of the BishopPhelps theorem
Theorem 13 (BishopPhelps II) Let F : B → ¹ be a lower semicon
tinuous convex functional on a real Banach space B. The functional F can
take the value +∞ but not everywhere. Suppose n
0
∈ B with F[n
0
] < ∞
and let L
0
∈ B
∗
be an Fbounded functional. Then for every ≥ 0 there
exists n
∈ B and a functional L
∈ B
∗
such that
1. L
−L
0

B
∗ ≤
2. F[n] ≥ F[n
] +L
[n −n
] for all n.
3. n
−n
0

B
≤ F[n
0
] −L
0
[n
0
] −inf
n∈B
¦F[n] −L
0
[n]¦
For the details of the proof we again refer to [22]. The ﬁrst two points of
this theorem are equivalent to the previous theorem. They say that any
Fbounded functional L
0
can be approximated to arbitrary accuracy by a
tangent functional L
. The inequality in the third point of this theorem
allows us to make statements about distances between elements in B, which
in our case will be densities. Note that the right hand side of this inequality
has the geometric meaning of being the diﬀerence of the vertical distance of
the functionals L
0
and F in n
0
and the shortest possible distance between
L
0
and F. Let us now apply this theorem to the functional F
L
and show
that that every density in o is arbitrarily close to an EVdensity. We ﬁrst
need some preliminaries. From Eq.(224) we see that we can write
F
L
[n] = sup
v∈L
3
2 +L
∞
¦E[v] −
_
d
3
rn(r)v(r)¦ (229)
If the supremum is attained for some v then n is an EVdensity. This
follows because then there is a density matrix
ˆ
D[n] that yields density n
(see Theorem 5) such that
Tr
ˆ
D[n]
ˆ
H
v
= F
L
[n] +
_
d
3
rn(r)v(r) = E[v] = inf
ˆ
D
Tr
ˆ
D
ˆ
H
v
(230)
The density matrix
ˆ
D[n] must therefore be a ground state density matrix.
If n is not an EVdensity then the supremum is not attained for any v. In
any case, for every integer k and any density n
0
∈ o we can always ﬁnd
some v
k
such that
E[v
k
] −
_
d
3
rn
0
(r)v
k
(r) ≥ sup
v∈L
3
2 +L
∞
¦E[v] −
_
d
3
rn
0
(r)v(r)¦ −
1
k
= F
L
[n
0
] −
1
k
(231)
where we note that the series v
k
does not converge to any v if n
0
is not an
EVdensity. Furthermore, for any n we have
E[v
k
] = inf
n∈S
¦F
L
[n] +
_
d
3
rn(r)v
k
(r)¦ ≤ F
L
[n] +
_
d
3
rn(r)v
k
(r) (232)
which in combination with the previous inequality yields
F
L
[n] +
_
d
3
rn(r)v
k
(r) ≥ E[v
k
] ≥ F
L
[n
0
] +
_
d
3
rn
0
(r)v
k
(r) −
1
k
(233)
We are now ready to apply the BishopPhelps theorem to the Lieb func
tional F
L
. We take = 1 , n
0
∈ o and let −v
k
correspond to the Fbounded
functional L
0
of the theorem. According to the theorem we can then ﬁnd
a tangent functional −w
k
(the L
of the theorem) such that
v
k
−w
k
 ≤ 1 (234)
in the norm on L
3
2
+L
∞
. From our previous investigations we know that
the tangent −w
k
touches the graph of F
L
in an EVdensity n
k
, i.e.
F
L
[n] ≥ F
L
[n
k
] −
_
d
3
rw
k
(r)(n(r) −n
k
(r)) (235)
where w
k
is the potential that generates density n
k
. The BishopPhelps
theorem then tells us that
n
k
−n
0

p
≤ F
L
[n
0
] +
_
d
3
rn
0
(r)v
k
(r)
− inf
n∈S
¦F
L
[n] +
_
d
3
rn(r)v
k
(r)¦ (236)
where the normindex has the values p = 1 and p = 3. Note that the
inﬁmum in this equation can be taken over o rather than L
1
∩L
3
since F
L
is deﬁned to be +∞ outside o. Then from Eq.(233) we see immediately
that
inf
n∈S
¦F
L
[n] +
_
d
3
rn(r)v
k
(r)¦ ≥ F
L
[n
0
] +
_
d
3
rn
0
(r)v
k
(r) −
1
k
(237)
and therefore
n
k
−n
0

p
≤
1
k
(238)
for p = 1 and p = 3. Since this equation is true for any k we see that any
density n
0
∈ o can be approximated to any accuracy by the EVdensity
n
k
in the norms on L
1
∩ L
3
. This proves the theorem in the beginning of
this section.
15. The KohnSham approach and noninteracting v
representability
We have now come to the discussion of the central equations which form
the basis of almost any practical application of density functional theory:
the KohnSham equations. Kohn and Sham [8] introduced an auxiliary
noninteracting system of particles with the property that it yields the same
ground state density as the real interacting system. In order to put the
KohnSham procedure on a rigorous basis we introduce the functional
T
L
[n] = inf
ˆ
D→n
Tr
ˆ
D
ˆ
T (239)
We see that this is simply the Lieb functional with the twoparticle inter
action omitted. All the properties of the functional F
L
carry directly over
to T
L
. The reason is that all these properties were derived on the basis
of the variational principle in which we only required that
ˆ
T +
ˆ
W is an
operator that is bounded from below. This is, however, still true if we omit
the Coulomb repulsion
ˆ
W. We therefore conclude that T
L
is a convex lower
semicontinuous functional which is diﬀerentiable for any density n that is
ensemble vrepresentable for the noninteracting system and nowhere else.
We refer to such densities as noninteracting EVdensities and denote the
set of all noninteracting EVdensities by B
0
. Let us collect all the results
for T
L
in a single theorem:
Theorem 14 T
L
is a convex lower semicontinuous functional with the fol
lowing properties:
1. For any n ∈ o there is a minimizing density matrix
ˆ
D[n] with the
property T
L
[n] = Tr
ˆ
D[n]
ˆ
T.
2. T
L
is Gˆ ateaux diﬀerentiable at the set of noninteracting EVdensities
and nowhere else.
3. The functional derivative at a noninteracting EVdensity n is given
by:
δT
L
δn(r)
= −v
s
[n](r)
where the potential v
s
generates the density n in a noninteracting
system.
From the last point in this theorem we see that if we want to know if
a given density n from the set o can be obtained as an EVdensity of
a noninteracting system we may try to calculate the derivative of T
L
for
this density. If it exists then the derivative yields the potential that we
were looking for. Now Kohn and Sham [8] assumed that for any density of
an interacting system there is a noninteracting system that has the same
density as its ground state. We can now ask the question whether the sets
of interacting and noninteracting EVdensities are equal. This is currently
not known. However, we can make a number of useful conclusions. First of
all, if we apply the BishopPhelps theorem to the functional T
L
we obtain
the following result:
Theorem 15 The set of noninteracting EVdensities is dense in the set
o with respect to the norm on L
1
∩ L
3
.
This means that for any density n ∈ o there is an noninteracting EV
density arbitrarily close. We can in particular choose n to be an interacting
EVdensity, i.e. n ∈ B and ﬁnd a noninteracting EVdensity arbitrarily
close. We also know from the BishopPhelps theorem applied to F
L
that the
set of interacting EVdensities is dense in o and therefore for any n ∈ o, in
particular n ∈ B
0
, we can ﬁnd an interacting EVdensity arbitrarily close.
We therefore conclude
Theorem 16 The set B
0
of noninteracting EVdensities is dense in the
set B of interacting EVdensities, and vice versa.
If we combine this result with previous theorems we obtain the following
important consequence for the KohnSham scheme:
Theorem 17 Suppose n ∈ B is an interacting EVdensity. Then for every
> 0 there is a noninteracting EVdensity n
∈ B
0
such that
1. n −n

p
≤ for p = 1 and p = 3.
2. The density n
is a ground state ensemble density of a noninteracting
system with potential
v
s,
(r) = −
δT
L
δn(r)
(n
)
i.e. we can always set up a KohnSham scheme that produces a given in
teracting EVdensity to arbitrary accuracy.
This theorem tells us that in practice we can always set up a KohnSham
scheme. We also see that if we want to prove that the sets B and B
0
of in
teracting and noninteracting EVdensities are equal, then we have to show
that the potentials v
s,
in this theorem approach some potential v
s,0
for
→ 0 in some smooth way. This, however, has not been proven until now.
In numerical calculations (see the discussion at the end of this section) one
has indeed always succeeded in obtaining a KohnSham potential for given
interacting EVdensities obtained from accurate Conﬁguration Interaction
(CI) calculations. In these calculations convergence to a given KohnSham
potential is sometimes diﬃcult to obtain, but seems to happen in a rather
smooth way. One might therefore expect that the sets of interacting and
noninteracting EVdensities are in fact equal. In order to encourage further
work in this ﬁeld we put it here as a conjecture
Conjecture 1 The sets B and B
0
of interacting and noninteracting EV
densities are equal, i.e. B = B
0
.
Let us describe a couple of cases for which we know this conjecture to be
true. Consider a system of two particles. Let the interacting system have
density n. Then we can construct a noninteracting KohnSham system with
ground state wavefunction
Ψ
KS
(r
1
σ
1
, r
2
σ
2
) =
1
√
2
ϕ(r
1
)ϕ(r
2
)(α(σ
1
)β(σ
2
) −β(σ
1
)α(σ
2
)) (240)
where α and β are the usual spin functions with α(
1
2
) = β(−
1
2
) = 1 and
α(−
1
2
) = β(
1
2
) = 0. This wavefunction has a density n(r) = 2[ϕ(r)[
2
and
the KohnSham orbital ϕ satisﬁes
_
−
1
2
∇
2
+v
s
[n](r)
_
ϕ(r) = ϕ(r) (241)
If we now choose ϕ(r) =
_
ϕ(r)/2 then for an interacting EVdensity n
the KohnSham Hamiltonian with potential
v
s
[n](r) =
1
2
∇
2
√
n
√
n
+ (242)
has orbital ϕ as an eigenfunction. Because the density is positive the orbital
ϕ =
_
n/2 has no nodes and must be a ground state orbital. We therefore
have constructed a noninteracting system with ground state n, which is
even a pure state density.
Another case where the conjecture is true is for lattice systems. The lattice
system is obtained by discretizing the manybody Hamiltonian on a grid,
i.e. we replace the diﬀerential equation by a diﬀerence equation as one
might do to solve the Schr¨odinger equation numerically. Then any inter
acting EVdensity on the grid is also a noninteracting EVdensity. This
was proven by Chayes, Chayes and Ruskai [23]. This work may be used
to prove the general conjecture above if one can prove that the continuum
limit can be taken in some smooth way.
Let us ﬁnally discuss the exchangecorrelation functional which is the cen
tral object in any application of density functional theory. We deﬁne the
exchangecorrelation functional E
xc
by
F
L
[n] = T
L
[n] +
1
2
_
d
3
r
1
d
3
r
2
n(r
1
)n(r
2
)
[r
1
−r
2
[
+E
xc
[n] (243)
Since both T
L
and F
L
are deﬁned on the set o the exchangecorrelation
functional E
xc
is also deﬁned on that set. Since F
L
and T
L
are diﬀerentiable
respectively on the sets B and B
0
and nowhere else, the functional E
xc
is
diﬀerentiable on B ∩ B
0
and nowhere else. The derivative of Eq.(243) on
that set is given by
−v(r) = −v
s
(r) +
_
d
3
r
n(r
)
[r −r
[
+
δE
xc
δn(r)
(244)
If we deﬁne the exchangecorrelation potential by
v
xc
(r) =
δE
xc
δn(r)
(245)
then we see that the KohnSham potential can be written as
v
s
(r) = v(r) +
_
d
3
r
n(r
)
[r −r
[
+v
xc
(r) (246)
We see that on the set B ∩ B
0
we have obtained the usual KohnSham
equations
_
ˆ
T +
ˆ
V +
ˆ
V
H
[n] +
ˆ
V
xc
[n]
_
[Φ
i
) = E
0
[Φ
i
) (247)
where i = 1 . . . q runs over the q degenerate ground states [Φ
i
) of the Kohn
Sham system. The density must now be calculated from a ground state
ensemble
ˆ
D
s
[n] =
q
i=1
µ
i
[Φ
i
)¸Φ
i
[ (248)
of the KohnSham system and is explicitly given as
n(r) = Tr
ˆ
D
s
[n]ˆ n(r) =
q
i=1
µ
i
¸Φ
i
[ˆ n(r)[Φ
i
) (249)
These equations can be written more explicitely in terms of the KohnSham
orbitals deﬁned to be the eigenstates of the single particle equation
_
−
1
2
∇
2
+v
s
(r)
_
ϕ
i
(r) =
i
ϕ
i
(r) (250)
It is then easily seen that any Slater determinant
[Φ) = [ϕ
i1
. . . ϕ
iN
) (251)
built out of KohnSham orbitals is an eigenstate of the KohnSham system
with eigenvalue E
0
=
i1
+ . . . +
iN
. However, not every eigenstate is
necessarily a Slater determinant. For instance, if two Slater determinants
have the same energy eigenvalue, then a linear combination of them also has
the same eigenvalue. This leads to a subtle point for KohnSham theory of
degenerate states. In Eq.(249) we require that the density is representable
by a ground state ensemble of a noninteracting system. However, we did
not require that it must be representable by an ensemble of ground state
Slater determinants. That one should be careful at this point, has been
shown by Englisch and Englisch [3] and Lieb [1]. They constructed an
explicit example of a pure state density of a noninteracting system that
can not be obtained from a single Slater determinant. We will therefore in
general have
[Φ
i
) =
q
j=1
α
ij
[D
j
) (252)
were [D
j
) is a Slater determinant. We further know that every [Φ
i
) must
be a ground state for potential v
s
. This means that all orbitals below the
highest occupied level must be occupied. If this were not the case then we
could lower the energy by transferring an electron from the highest level to
a lower level which would lower the energy and therefore [Φ
i
) would not be
a ground state for potential v
s
. This also means that if two determinants
[D
i
) and [D
j
) diﬀer, then this diﬀerence must be due to diﬀerent orbitals
in the highest level. Because both determinants are eigenfunctions of the
KohnSham Hamiltonian it is easily seen that these diﬀerent orbitals must
have the same KohnSham eigenvalue. This eigenvalue is the KohnSham
orbital energy of the highest occupied state which we will denote by µ.
From these considerations we ﬁnd that the density is of the general form
n(r) =
i<µ
[ϕ
i
(r)[
2
+
k
=
l
=µ
β
kl
ϕ
k
(r)ϕ
∗
l
(r) (253)
where β
kl
is an Hermitian matrix with eigenvalues between 0 and 1. This
means that we can diagonalize this matrix with a unitary matrix U
ij
. We
then introduce new orbitals for the highest occupied level:
˜ ϕ
i
(r) =
j
U
ij
ϕ
j
(r) (254)
Since these orbitals are linear combinations of degenerate orbitals, they are
again eigenfunctions of the KohnSham singleparticle Hamiltonian for the
highest occupied level. We can now write the density as
n(r) =
i<µ
[ϕ
i
(r)[
2
+
k
=µ
n
k
[ ˜ ϕ
k
(r)[
2
(255)
with occupation numbers n
k
between 0 and 1. We see that for the case of a
degenerate ground state we have to solve Eq.(250) together with Eq.(255)
in which also the coeﬃcients n
k
must be determined. When do we need
fractional occupation numbers in practice? In practice one sometimes ﬁnds
that when solving the KohnSham equations with occupations equal to
one that the converged solution has a hole below the highest level. This
happens for instance for the iron atom within the local density approxima
tion [24]. This is an indication that the true KohnSham density could be
an EVdensity. In practice the lowest energy state is then obtained by a
procedure called ”evaporation of the hole”. In this method one increases
the occupation of the hole while descreasing the occupation of the highest
level until both levels are degenerate [24, 25]. The matter has been in
vestigated in detail by Schipper et al. [26, 27]. These authors calculated
by CImethods accurate charge densities for a couple of molecules. Sub
sequently they constructed the KohnSham potentials that generate these
densities. A particularly interesting case is the C
2
molecule [26]. When
one tries to construct a KohnSham potential that produces the ground
state density of the C
2
molecule using occupation numbers equal to one,
i.e. for a pure state, one ﬁnds that the solution corresponds to a state with
a hole below the highest occupied level. This is then not a proper Kohn
Sham state since it is not the state of lowest energy for the corresponding
potential, but an excited state. If one then tries to construct an ensem
ble solution by the technique of ”evaporation of the hole”, one does ﬁnd a
proper ground state ensemble for the KohnSham system. From this we can
conclude that the ground state density of the C
2
molecule is a noninteract
ing EVdensity but not a noninteracting PSVdensity. This means that
the extension of density functionals to EVdensities is not only of theoret
ical, but also practical interest. As an illustration we display in Figure 4,
the exchangecorrelation potentials of the C
2
molecule for four bonding dis
tances. The potentials are plotted along the C−C bond axis as functions of
the distance z from the bond midpoint. The corresponding bond distances
R(C−C) are displayed in the inset. The potentials v
PS−hole
xc
correspond to
the exchangecorrelation potentials of the improper KohnSham solutions
with a hole. The corresponding state is a pure state consisting of a sin
gle Slater determinant. The potentials v
E
xc
are the KohnSham potentials
corresponding to a ground state ensemble. Both v
PS−hole
xc
and v
E
xc
give
an accurate representation of the true density of the molecule, although
the accuracy attained with v
E
xc
is a bit higher. The potential v
E
xc
has the
features found in many diatomic molecules, i.e. the usual well around the
atom, the small intershell peaks and a plateau around the bond midpoint.
The potential v
PS−hole
xc
on the other hand, is heavily distorted as compared
to v
E
xc
. This may be explained by the fact that the wavefunction of the C
2
molecule has a strong multideterminantial character. With this we mean
that the simple HartreeFock approximation has a relatively small coeﬃ
cient in the CIexpansion of the wave function. The distortions in v
PS−hole
xc
therefore appear to be a price we have to pay for producing the density of
such a strongly multideterminantial state with a single Slater determinant.
The use of an ensemble formulation of KohnSham theory seems therefore
especially relevant in cases of strong electron correlations. Other examples
are provided by neardegeneracy situations such as in avoided crossings [25]
and the H
2
+ H
2
reaction [27]. A recent discussion on KohnSham theory
for ensembles can be found in Ref. [28] which also gives the beryllium series
as an explicit example of systems which are ensemble vrepresentable but
not pure state vrepresentable.
distances to be EV
s
representable, (or class b, as
described in Sect. 1).
Applying the results of Levy [1] and Lieb [2] to
noninteracting systems, we know that there are many
densities that are not PSV
s
representable but that are
EV
s
representable. It is thus certainly possible that a
given interacting ground state density does not belong to
the set of PS noninteracting ground state densities, but
does belong to the set of ensemblerepresentable non
interacting densities. Our results prove this not to be
only an academic possibility, but EV
s
representability is
in fact called for to handle cases with strong electron
correlation, i.e., essentially multideterminantal character
of the interacting groundstate wavefunction.
The KS solution for the lowest singlet states of CH
2
and C
2
has been constructed from the ab initio CI q. To
obtain the ensemble solution in the cases when an at
tempt to construct the PS leads to a nonAufbau solu
tion, the procedure of ``evaporation of the hole below
the Fermi level'' has been employed [15, 17, 37]. Already
for CH
2
, for which the wave function has a relatively
weak multideterminantal character, calculation in the
TZ basis yields the ensemble solution and only with the
extended QZ basis the purestate onedeterminantal KS
solution has been obtained.
With the example of C
2
, the possibility has been
demonstrated of an essentially accurate ensemble KS
solution with accidental degeneracy (EV
s
represent
ability) for a q that is PSV representable for the inter
acting system. A variety of KS solutions has been
obtained depending on the R(C±C) distance, ranging
from a singledeterminantal purestate at R(C±C) 1:8
a.u. through an ensemble with a weak accidental de
generacy at R
e
to an ensemble with a strong degeneracy
Fig. 3a±d The exchangecorre
lation potential m
xc
along the
bond axis of C
2
. m
E
xc
(solid lines)
is the potential corresponding
to the ensemble solution;
m
PSÀhole
xc
(dotted lines) corre
sponds to the singledetermi
nant solutions with a hole
below the Fermi level. a R
e
, b
2.8 a.u., c 3.0 a.u., d 4.0 a.u.
341
Figure 4: Exchangecorrelation potentials producing the accu
rate interacting ground state density of the C
2
molecule at various
bonding distances. The potential v
E
xc
is the exchangecorrelation
potential for a ground state ensemble. The potential v
PS−hole
xc
is
an exchangecorrelation potential that reproduces the same den
sity for a single Slater determinant with a hole.
16. The gradient expansion
Until now we only considered the formal framework of density functional
theory. However, the theory would be of little use if we would not be able to
construct good approximate functionals for the exchangecorrelation energy
and exchangecorrelation functional. Historically the ﬁrst approximation
for the exchangecorrelation functional to be used was the local density
approximation. In this approximation the exchangecorrelation functional
is taken to be
E
LDA
xc
[n] =
_
d
3
r
xc
(n(r)) (256)
where
xc
(n) is the exchangecorrelation energy per volume unit for a homo
geneous electron gas with density n. We therefore treat the inhomogeneous
system locally as an electron gas. This simple and crude approximation
turns out to be surprisingly succesful for realistic and very inhomogeneous
systems although one would expect that the LDA would only be accurate
for systems with slowly varying densities. The LDA therefore seems a suit
able starting point for more accurate approximations. In this section we
will show that the LDA is in fact the ﬁrst term in a systematic expan
sion of the exchangecorrelation functional in terms of spatial derivatives
with respect to the density. This expansion is commonly referred to as the
gradient expansion [5, 29, 30] and has the following form
E
xc
[n] = E
LDA
xc
[n] +
_
d
3
rg
1
(n(r))(∇n(r))
2
+
_
d
3
rg
2
(n(r))(∇
2
n(r))
2
+. . . (257)
where the functions g
i
(n) are uniquely determined by the density response
functions (of arbitrary order) of the homogeneous electron gas. The gradi
ent expansion presents an, in principle, exact way to construct the exchange
correlation functional for solids, provided the series converges. We will come
back to the question of convergence. To derive the gradient expansion we
start out by expanding the exact exchangecorrelation energy functional
around its homogeneous electron gas value n
0
as follows
E
xc
[n] = E
xc
[n
0
]+
∞
m=1
1
m!
_
d
3m
rK
(m)
xc
(n
0
; r
1
. . . r
m
)δn(r
1
) . . . δn(r
m
) (258)
where d
3
r = d
3
r
1
. . . d
3
r
m
and n(r) = n
0
+δn(r). We further deﬁned
K
(m)
xc
(n
0
; r
1
. . . r
m
) =
δ
m
E
xc
δn(r
1
) . . . δn(r
m
)
[
n=n0
(259)
The ﬁrst term E
xc
[n
0
] in Eq.(258) is the exchangecorrelation energy of a
homogeneous system with constant density n
0
and is therefore a function of
n
0
rather than a functional. We will therefore write this term as E
xc
(n
0
).
This function is by now wellknown from extensive investigations of the
homogeneous electron gas [31]. Since the electron gas has translational,
rotational and inversion symmetry the functions functions K
(m)
xc
satisfy
K
(m)
xc
(n
0
; r
1
. . . r
m
) = K
(m)
xc
(n
0
; r
1
+a. . . r
m
+a) (260)
K
(m)
xc
(n
0
; r
1
. . . r
m
) = K
(m)
xc
(n
0
; Rr
1
. . . Rr
m
) (261)
K
(m)
xc
(n
0
; r
1
. . . r
m
) = K
(m)
xc
(n
0
; −r
1
. . . −r
m
) (262)
where a is an arbitrary translation vector and where R is an arbitrary
rotation matrix. Furthermore, the function K
(m)
xc
has full permutational
symmetry, i.e.
K
(m)
xc
(n
0
; r
1
. . . r
m
) = K
(m)
xc
(n
0
; r
P(1)
. . . r
P(m)
) (263)
for an arbitrary permutation P of the indices 1 . . . m. If we choose a = −r
1
in Eq. (260) we see that
K
(m)
xc
(n
0
; r
1
. . . r
m
) = K
(m)
xc
(n
0
; 0, r
2
−r
1
. . . r
m
−r
1
)
= L
(m)
(n
0
; r
2
−r
1
. . . r
m
−r
1
) (264)
where the latter equation deﬁnes the function L
(m)
as a function of m−1
variables. Let us give some speciﬁc examples. The function K
(1)
xc
is given
by
K
(1)
xc
(n
0
; r
1
) =
δE
xc
δn(r
1
)
[
n=n0
= v
xc
(n
0
) (265)
and we see that this function is simply the exchangecorrelation potential
of the electron gas which, due to translational invariance, does not depend
on r
1
. Since δn(r) integrates to zero the term with K
(1)
xc
in the expansion
Eq.(258) does not contribute to E
xc
[n]. The ﬁrst nontrivial term is therefore
K
(2)
xc
(n
0
; r
1
, r
2
) =
δ
2
E
xc
δn(r
1
)δn(r
2
)
[
n=n0
= L
(2)
(n
0
; r
2
−r
1
) (266)
which leads to an expansion of the form
E
xc
[n] = E
xc
(n
0
) +
1
2
_
d
3
r
1
d
3
r
2
L
(2)
(n
0
; r
2
−r
1
)δn(r
1
)δn(r
2
) +. . . (267)
The function L
(2)
(often denoted by K
xc
in the literature) has been the
subject of many investigations [32, 33]. Let us now go back to the more
general case. If we introduce the Fourier transform of a function f as follows
˜
f(q
1
. . . q
m
) =
_
d
3m
rf(r
1
. . . r
m
)e
−iq1·r1−...−iqm·rm
(268)
then
˜
K
(m)
xc
(n
0
; q
1
. . . q
m
) = (2π)
3
δ(q
1
+. . . +q
m
)
˜
L
(m)
(n
0
; q
2
. . . q
m
) (269)
The expansion for E
xc
then becomes
E
xc
[n] = E
xc
(n
0
)+
∞
m=1
1
m!
_
d
3m
q
(2π)
3m
˜
K
(m)
xc
(n
0
; q
1
. . . q
m
)δ˜ n(q
1
) . . . δ˜ n(q
m
) (270)
where d
3m
q = d
3
q
1
. . . d
3
q
m
. This can also be written in terms of the
functions L
(m)
as follows
E
xc
[n] = E
xc
(n
0
)+
∞
m=1
1
m!
_
d
3
q
2
(2π)
3
. . .
d
3
q
m
(2π)
3
˜
L
(m)
(n
0
; q
2
. . . q
m
)
δ˜ n(−q
2
−. . . −q
m
)δ˜ n(q
2
) . . . δ˜ n(q
m
) (271)
The gradient expansion is then obtained by expanding L
(m)
in powers of
q
i
around q
i
= 0 and subsequently transforming back to real space. If we
do this our ﬁnal expansion will then still depend on our reference density
n
0
which is undesirable for application of the functional to general systems.
However, we will show that we can get rid of this dependence on n
0
by
doing inﬁnite resummations. The key formula that will enable us to do
these resummations is the following ﬁrst order change of K
(m)
xc
:
δK
(m)
xc
(r
1
. . . r
m
) =
_
d
3
rK
(m+1)
xc
(n
0
; r, r
1
. . . r
m
)δn(r) (272)
If we take δn(r) = δn
0
to be a constant change we obtain
∂K
(m)
xc
∂n
0
(n
0
; r
1
. . . r
m
) =
_
d
3
rK
(m+1)
xc
(n
0
; r, r
1
. . . r
m
) (273)
Note that by taking δn(r) = δn
0
we violate the condition that δn(r) inte
grates to zero. Nevertheless Eq.(273) can alternatively be derived directly
from the properties of the response functions [30] and is related to a gener
alized form of the wellknown compressibility sumrule of the electron gas.
An important special case is obtained for m = 1:
∂v
xc
∂n
0
=
∂K
(1)
xc
∂n
0
(n
0
) =
_
d
3
r
2
K
(2)
xc
(n
0
; r
1
, r
2
)
=
_
d
3
r
2
L
(2)
(n
0
; r
2
−r
1
) =
˜
L
(2)
(q = 0) (274)
We see that the q = 0 value of
˜
L
(2)
can be directly calculated from the
knowledge of v
xc
(n
0
). For m ≥ 2 relation (273) directly implies the follow
ing relation between L
(m)
and L
(m+1)
∂L
(m)
∂n
0
(n
0
; q
2
. . . q
m
) = L
(m+1)
(n
0
; −q
2
−. . . −q
m
, q
2
. . . q
m
) (275)
The importance of this formula will become clear if we work out, as an
explicit example, the lowest order in the gradient expansion. From the
symmetry properties of K
(m)
xc
one ﬁnds that its Fourier transform has the
following expansion in powers of q
i
:
˜
K
(m)
xc
(n
0
; q
1
. . . q
n
) = (2π)
3
δ(q
1
+. . . +q
m
)
(K
(m)
0
+K
(m)
1
P
(m)
1
(q
1
. . . q
m
) +K
(m)
2
P
(m)
2
(q
1
. . . q
m
)+. . .)(276)
in which the K
(m)
i
are coeﬃcients and P
(m)
i
are symmetric polynomials.
This is a direct consequence of the permutational symmetry of K
xc
. The
ﬁrst polynomial P
(m)
1
(taking into account the δfunction) has the explicit
form
P
(m)
1
(q
1
. . . q
m
) = q
2
1
+. . . +q
2
m
(277)
and is invariant under permutation of the indices, i.e. it transforms accord
ing to a onedimensional representation of the permutation group. More
details on the group theoretical treatment of these reponse functions can be
found in the work of Svendsen and von Barth [34]. From the properties of
˜
K
(m)
xc
we ﬁnd that for m ≥ 2 the function L
(m)
has the following expansion
L
(m)
(n
0
; q
2
. . . q
m
) = L
(m)
0
(n
0
) +L
(m)
1
(n
0
)p
(m)
(q
2
. . . q
m
) +. . . (278)
where the polynomial p
(m)
is given by
p
(m)
(q
2
. . . q
m
) = (−q
2
−. . . −q
m
)
2
+q
2
2
+. . . +q
2
m
= 2
m
i=2
q
2
i
−2
m
i>j≥2
q
i
q
j
(279)
It is easily seen that this polynomial has the property
p
(m+1)
(−q
2
−. . . −q
m
, q
2
. . . q
m
) = p
(m)
(q
2
. . . q
m
) (280)
If we use this property together with Eq.(278) and Eq.(275) we obtain for
the coeﬃcients in the expansion of L
(m)
the following equations
∂L
(m)
0
∂n
0
= L
(m+1)
0
(n
0
) (281)
∂L
(m)
1
∂n
0
= L
(m+1)
1
(n
0
) (282)
where m ≥ 2. Together with
˜
L
(2)
(q = 0) =
∂v
xc
∂n
0
(n
0
) =
∂
2
xc
∂n
2
0
(n
0
) (283)
where
xc
(n
0
) is the exchange energy per volume unit of the electron gas,
this yields
L
(m)
0
(n
0
) =
∂
m
xc
∂n
m
0
(n
0
) (284)
L
(m)
1
(n
0
) =
∂
m−2
L
(2)
1
∂n
m−2
0
(n
0
) (285)
It is these two equations that will allow us to eliminate the dependence on
n
0
in the lowest order gradient expansion. If we insert the explicit form of
L
(m)
of Eq.(278) into Eq.(271) and Fourier transform back to real space we
obtain
E
xc
[n] = E
xc
(n
0
)+
∞
m=1
1
m!
L
(m)
0
(n
0
)
_
d
3
r(δn(r))
m
+
∞
m=2
1
m!
L
(m)
1
(n
0
)m(m−1)
_
d
3
r(∇δn(r))
2
(δn(r))
m−2
+. . .(286)
Since ∇δn(r) = ∇n(r) this can be rewritten with help of Eq.(284) and
Eq.(285) as
E
xc
[n] = E
xc
(n
0
)+
∞
m=1
1
m!
∂
m
xc
(n
0
)
∂n
m
0
_
d
3
r(δn(r))
m
+
∞
m=2
1
(m−2)!
∂
m−2
L
(2)
1
(n
0
)
∂n
m−2
0
_
d
3
r(∇n(r))
2
(δn(r))
m−2
+. . .(287)
We see that the ﬁrst two terms simply give the expansion of the LDA
functional around the constant density n
0
,i.e.
E
LDA
xc
[n] =
_
d
3
r
xc
(n
0
+δn(r)) (288)
whereas the third term involves an expansion of the coeﬃcient L
(2)
1
(n
0
+
δn(r)) around n
0
. We obtain
E
xc
[n] = E
LDA
xc
[n] +
_
d
3
rL
(2)
1
(n(r))(∇n(r))
2
+. . . (289)
and we see that we succeeded in eliminating the dependence on n
0
in the
gradient expansion. Morever we see that the local density approximation
naturally appears as the ﬁrst term in the gradient expansion. Furthermore
we ﬁnd that the coeﬃcient g
1
(n) in our ﬁrst equation (257) is completely
determined by the qexpansion of the function
˜
L
(2)
(n
0
; q) to order q
2
. We
also see that the replacement n
0
→ n(r) requires a resummation over re
sponse functions of arbitrary order. The dependence on n
0
can also be
removed to higher order in powers of q. For an explicit example up to
order q
4
we refer again to Svendsen and von Barth [34].
Let us now address the question of convergence of the gradient expan
sion. This question has been investigated for the gradient expansion of the
exchangeenergy functional for which a comparison with exact exchange
energies is possible. For this case the analytic form of
˜
L
(2)
(n
0
; q) is known
from the impressive work of Engel and Vosko [35, 36] and some higher order
gradient coeﬁcients have been determined by Svendsen, Springer and von
Barth [34, 37]. One ﬁnds [37] that the gradient expansion performs very
well for metallic systems, but that the quality deteriorates as soon as the
system acquires an energy gap. This may not be so surprising if one real
izes that the appearance of a gap drastically changes the low qbehavior of
the response functions K
(m)
xc
which determine the gradient coeﬃcients [38].
This means that the standard gradient expansion can not deal with insu
lators or ﬁnite systems (which may be modelled as insulators by repeating
them periodically with a large lattice constant). Further progress along the
lines of the gradient expansion may be obtained by study of the socalled
gapped electron gas [39]. Currently the most fruitful approach to the con
struction of simple and accurate functionals is the socalled Generalized
Gradient Approximation (GGA) [40, 41]. In this approach one speciﬁes a
form of the paircorrelation function of the manyelectron system and de
termines the parameters in this function by sumrules and information from
the straightforward gradient expansion. These functionals have lead to
large improvements in molecular binding energies as compared to the local
density approximation [42]. However, since the approach is not systematic
it is diﬃcult to improve the quality of the current GGA functionals.
17. The optimized potential method and the e
2
expansion
In this section we will describe a second systematic method to construct
density functionals, namely perturbation theory starting from the Kohn
Sham Hamiltonian [43, 44]. The method is based on traditional pertur
bation theory and is comparable in computational cost. For this reason
the method is less suited to the calculation of properties of large systems.
However, the method has the theoretical advantage that it can be used as a
benchmark to test the quality of diﬀerent approximate density functionals.
We consider the Hamiltonian
ˆ
H
=
ˆ
H
s
+(
ˆ
H −
ˆ
H
s
) =
ˆ
H
s
+(
ˆ
W −
ˆ
V
Hxc
) (290)
where
ˆ
W =
N
i>j
e
2
[r
i
−r
j
[
(291)
ˆ
V
Hxc
=
N
i=1
v
H
(r
i
) +v
xc
(r
i
) (292)
where e
2
is the square of the electronic charge and v
H
is the usual Hartree
potential
v
H
(r) = e
2
_
d
3
r
n(r
)
[r −r
[
(293)
and v
xc
the exchangecorrelation potential. The perturbation term is simply
the diﬀerence between the true and the KohnSham Hamiltonian and we
are interested in the case = 1 although this makes expansion in powers
of rather doubtful. We will come back to the point of convergence later.
We can now do standard perturbation theory and obtain the ground state
energy
E() = E
s
+E
(1)
+
2
E
(2)
+. . . (294)
where E
s
is ground state energy of the auxiliary KohnSham system, i.e.
simply a sum over orbital energies and the terms E
(i)
is the energy to order
i in powers of . The ﬁrst two terms are explicitely given by [43, 44]
E
(1)
= ¸Φ
s
[
ˆ
W −
ˆ
V
Hxc
[Φ
s
) (295)
E
(2)
=
∞
i=1
[¸Φ
s
[
ˆ
W −
ˆ
V
Hxc
[Φ
s,i
)[
2
E
s
−E
s,i
(296)
where E
s,i
and Φ
s,i
are the excited state energies and wave functions of
the KohnSham system. We now note that the energies E
(i)
are implicit
functionals of the density through their dependence on the KohnSham
potential, orbitals, and energies, i.e.
E
(i)
[n] = E
(i)
[¦ϕ
k
[n]¦, ¦
k
[n]¦, v
xc
[n]] (297)
This follows directly from the HohenbergKohn theorem applied to a non
interacting system. The density n uniquely determines the KohnSham
potential v
s
(up to a constant) and therefore the also the orbitals (up to
a phase factor) and eigenvalues (up to constant). The arbitrariness with
respect to a constant shift and with respect to the phase factor cancels out
in the energy expression and therefore the ith order energy becomes a pure
density functional. We therefore have the following series of implications
n(r) → v
s
(r) → ¦ϕ
k
(r),
k
¦ → E
(i)
(298)
Note that the perturbation theory that we constructed is not yet in a form
that we can use in practical calculations. This is because the perturbing
Hamiltonian contains the exchangecorrelation potential which is unknown
from the start and has to be determined selfconsistently from the pertur
bation series. However, the equations can be simpliﬁed if we expand the
energies and potentials in powers of the interaction strength e
2
and take
= 1. This leads to the following set of equations [44]
E[n] = T
s
[n] +
_
d
3
rn(r)v(r)
+
e
2
2
_
d
3
rd
3
r
n(r)n(r
)
[r −r
[
+
∞
i=1
e
2i
E
(i)
xc
(299)
v
Hxc
(r) = v
H
(r) +
∞
i=1
e
2i
v
(i)
xc
(r) (300)
v
(i)
xc
(r) =
δE
(i)
xc
δn(r)
(301)
The e
2
and e
4
terms in the expansion of the exchangecorrelation energy
have the explicit form
e
2
E
(1)
xc
[n] = ¸Φ
s
[
ˆ
W[Φ
s
) −
e
2
2
_
d
3
rd
3
r
n(r)n(r
)
[r −r
[
(302)
and
e
4
E
(2)
xc
[n] =
∞
i=1
[¸Φ
s
[
ˆ
W −
ˆ
V
H
−e
2
ˆ
V
(1)
xc
[Φ
s,i
)[
2
E
s
−E
s,i
(303)
The term of order e
2
can be written explicitly in terms of the KohnSham
orbitals as follows:
E
(1)
xc
[n] = −
1
2
N
k,l=1
_
d
3
rd
3
r
ϕ
∗
k
(r)ϕ
l
(r)ϕ
∗
l
(r
)ϕ
k
(r
)
[r −r
[
(304)
We see that this expression has exactly the same form as the usual expres
sion of the exchange energy within the HartreeFock approximation. The
diﬀerence, however, is that the orbitals in this expression are KohnSham
orbitals which, in contrast to the HartreeFock orbitals, are eigenfunctions
of a singleparticle Hamiltonian with a local potential. The numerical value
of E
(1)
xc
[n] and the HartreeFock exchange will therefore in general diﬀer
from each other. However, because of the similarity to the HartreeFock
deﬁnition of exchange we will deﬁne the exchange functional within density
functional theory to be E
x
[n] = E
(1)
xc
[n]. Corresponding to the exchange
functional there is a local exchange potential deﬁned as
v
x
(r) =
δE
x
δn(r)
=
δE
(1)
xc
δn(r)
= v
(1)
xc
(r) (305)
We see from Eq.(303) that we need to know this potential in order to
calculate the e
4
contribution to the exchangecorrelation energy. This is
a general feature of the present perturbation theory. In order to calculate
E
(i)
xc
we need to calculate v
(i−1)
xc
ﬁrst. Let us therefore start by calculating
v
x
. For this we have to calculate the ﬁrst order change δE
x
in the exchange
functional due to a change δn in the density. Since densities and potentials
are in 11correspondence this task amounts to calculating the change in
E
x
due to a change δv
s
in the KohnSham potential. This is readily done
by perturbation theory. The change δϕ
k
and δ
k
of the orbitals and orbital
energies of the KohnSham system due to a small change δv
s
in the potential
is given by solution of the equation
_
−
1
2
∇
2
+v
s
(r) −
k
_
δϕ
k
(r) = (δ
k
−δv
s
(r))ϕ
k
(r) (306)
This gives
δ
k
=
_
d
3
rϕ
∗
k
(r)ϕ
k
(r)δv
s
(r) (307)
δϕ
k
(r) = −
_
d
3
r
G
k
(r, r
)ϕ
k
(r
)δv
s
(r
) (308)
G
k
(r, r
) =
l=k
ϕ
l
(r)ϕ
∗
l
(r
)
l
−
k
(309)
From these equations we obtain the following functional derivatives
δ
k
δv
s
(r)
= ϕ
∗
k
(r)ϕ
k
(r) (310)
δϕ
k
(r)
δv
s
(r
)
= −G
k
(r, r
)ϕ
k
(r
) (311)
The density change δn is given by
δn(r) =
N
k=1
ϕ
∗
k
(r)δϕ
k
(r) +c.c. =
_
d
3
r
χ
s
(r, r
)δv
s
(r
) (312)
where we deﬁned the static density response function χ
s
of the KohnSham
system by
χ
s
(r, r
) =
δn(r)
δv
s
(r
)
= −
N
k=1
ϕ
∗
k
(r)G
k
(r, r
)ϕ
k
(r
) +c.c. (313)
We can now readily derive the following integral equation for v
x
:
Λ
x
(r) =
δE
x
δv
s
(r)
=
_
d
3
r
δE
x
δn(r
)
δn(r
)
δv
s
(r)
=
_
d
3
r
χ
s
(r, r
)v
x
(r
) (314)
where the inhomogeneity Λ
x
is given by
Λ
x
(r) = −
N
k=1
_
d
3
r
δE
x
δϕ
k
(r
)
G
k
(r
, r)ϕ
k
(r) +c.c. (315)
Since both Λ
x
and χ
s
are given as explicit functionals of the KohnSham
orbitals and orbital energies, the exchange potential can be found from
simultaneous solution of the equations
k
ϕ
k
(r) =
_
−
1
2
∇
2
+v(r) +v
H
(r) +e
2
v
x
(r)
_
ϕ
k
(r) (316)
Λ
x
(r) =
_
d
3
r
χ
s
(r, r
)v
x
(r
) (317)
Once we have solved these equations and determined v
x
we can go on
and calculate E
(2)
xc
from Eq.(303). To calculate the e
6
contribution to the
exchangecorrelation energy we ﬁrst have to evaluate v
(2)
xc
. This potential
is the solution of the integral equation
Λ
(2)
xc
(r) =
δE
(2)
xc
δv
s
(r)
=
_
d
3
r
χ
s
(r, r
)v
(2)
xc
(r
) (318)
Since E
(2)
xc
is an explicit functional of the orbitals, orbital energies and v
x
the inhomogeneity Λ
(2)
xc
can be calculated from
Λ
(2)
xc
(r) =
_
∞
k=1
_
d
3
r
δE
(2)
xc
δϕ
k
(r
)
δϕ
k
(r
)
δv
s
(r)
+c.c.
_
+
∞
k=1
_
d
3
r
δE
(2)
xc
δ
k
δ
k
δv
s
(r)
+
_
d
3
r
δE
(2)
xc
δv
x
(r
)
δv
x
(r
)
δv
s
(r)
(319)
All terms in this equation are explicitly known, except for the term δv
x
/δv
s
.
However, for this term we can ﬁnd an integral equation by diﬀerentiation
of Eq.(317) with respect to v
s
:
δΛ
x
(r
1
)
δv
s
(r
2
)
=
_
d
3
r
3
δχ
s
(r
1
, r
3
)
δv
s
(r
2
)
v
x
(r
3
) +
_
d
3
r
3
χ
s
(r
1
, r
3
)
δv
x
(r
3
)
δv
s
(r
2
)
(320)
Since both Λ
x
and χ
s
are explicitly known in terms of orbitals and orbital
energies their derivatives with respect to v
s
are also explicitly known in
terms of these quantities (see reference [44] for more details) and therefore
Eq.(320) determines δv
x
/δv
s
uniquely. So we see that in order to deter
mine v
(2)
xc
we have to solve two integral equations. For realistic systems
these equations have only be solved approximately (for an explicit solution
of Eq.(320) see reference [34])). From v
(2)
xc
one could go on along the same
lines to determine E
(3)
xc
and subsequently v
(3)
xc
. The determination of these
higher order energies and potentials becomes more involved. Nevertheless,
the perturbation series represents an explicit construction procedure for the
exact exchangecorrelation energy and potential, provided that the series
converges. Before we go on to discuss the convergence properties of this
series, let us brieﬂy review the ﬁrst order equations from a diﬀerent view
point. It is readily seen that one can write the total energy to order e
2
,
which we denote by E
1
[n], as
E
1
[n] = T
s
[n] +
_
d
3
rn(r)v(r) +
e
2
2
_
d
3
rd
3
r
n(r)n(r
)
[r −r
[
+e
2
E
x
[n]
= ¸Φ
s
[
ˆ
H[Φ
s
) (321)
This is just the expectation value of the true Hamiltonian
ˆ
H of the system
with the KohnSham wave function [Φ
s
). Because of the 11correspondence
between the density n and the potential v
s
we can also regard E
1
as a
functional of the potential v
s
, i.e.
E
1
[v
s
] = ¸Φ
s
[v
s
][
ˆ
H[Φ
s
[v
s
]) (322)
We may now try to ﬁnd an approximation to the true total energy of the
system by choosing a local potential v
s
that minimizes the energy expression
E
1
[v
s
]. This means that we have to solve the variational equation
0 =
δE
1
δv
s
(r)
=
N
k=1
_
d
3
r
δE
1
δϕ
k
(r
)
δϕ
k
(r
)
δv
s
(r)
+c.c. (323)
If one works out this expression one obtains equations that are identical
to Eqns.(316) and (317). These equations were ﬁrst derived by Talman
and Shadwick [45]. Since in our procedure we optimized the energy of
a Slater determinant wavefunction under the constraint that the orbitals
in the Slater determinant come from a local potential, the method is also
known as the Optimized Potential Method (OPM). We have therefore ob
tained the result that the OPM and and the expansion to order e
2
are
equivalent procedures. The OPM has many similarities to the Hartree
Fock approach. Within the HartreeFock approximation one minimizes the
energy of a Slater determinant wavefunction under the constraint that the
orbitals are orthonormal. One then obtains oneparticle equations for the
orbitals that contain a nonlocal potential. Within the OPM, on the other
hand, one adds the additional requirement that the orbitals must satisfy
singleparticle equations with a local potential. Due to this constraint the
OPM total energy E
1
will in general be higher than the HartreeFock en
ergy E
HF
, i.e. E
1
≥ E
HF
. We refer to [46, 47] for an application of the
OPM method for molecules.
We ﬁnally make some comments on the calculation of functional derivatives
in this section. We stress this point since careless use of the chain rule for
diﬀerentiation has led to wrong results in the literature [48]. As an example
we consider the exchange functional E
x
. When we regard this functional
as an explicit functional of the orbitals then the functional derivative with
respect to ϕ
k
is given by
_
δE
x
δϕ
k
(r)
_
e
= −
N
i=1
_
d
3
r
ϕ
∗
i
(r)ϕ
i
(r
)
[r −r
[
ϕ
∗
k
(r
) (324)
where we used the subindex e to indicate that we regard the functional
as an explicit orbital functional. This functional derivative represents the
change in E
x
due to a change δϕ
k
in orbital ϕ
k
, while keeping all other
orbitals ﬁxed. Moreover, we regard ϕ
k
and ϕ
∗
k
as independent variables. If
we regard E
x
as a density functional we must require that all orbitals are
eigenfunctions of a noninteracting Hamiltonian with a local potential v
s
. It
is clear that we can we can never ﬁnd a change δv
s
in a local potential that
induces a change in only one orbital while keeping the other orbitals ﬁxed.
If a potential changes one orbital ϕ
k
to ϕ
k
+δϕ
k
then all other orbitals will
change too. The change in E
x
regarded as a functional of v
s
is then given
by the functional derivative
δE
x
δv
s
(r)
=
N
k=1
_
d
3
r
_
δE
x
δϕ
k
(r
)
_
e
δϕ
k
(r
)
δv
s
(r)
+
N
k=1
_
d
3
r
_
δE
x
δϕ
∗
k
(r
)
_
e
δϕ
∗
k
(r
)
δv
s
(r)
(325)
We see that in this case the explicit orbital derivatives only occur in the
sum in which all the orbital changes must be taken into account. We may
also introduce an implicit derivative (δE
x
/δϕ
k
)
i
, which can be given the
meaning of giving the change in E
x
if we know that there is a potential
change δv
s
that changes ϕ
k
to ϕ
k
+ δϕ
k
. This implicit orbital derivative,
which we denote by subidex i is expressed in terms of the explicit orbital
derivatives as follows
_
δE
x
δϕ
k
(r)
_
i
=
N
i=1
_
d
3
r
_
δE
x
δϕ
i
(r
)
_
e
δϕ
i
(r
)
δϕ
k
(r)
+
N
i=1
_
d
3
r
_
δE
x
δϕ
∗
i
(r
)
_
e
δϕ
∗
i
(r
)
δϕ
k
(r)
(326)
The derivatives δϕ
i
/δϕ
k
appear now to take into account that, if orbital ϕ
k
changes to ϕ
k
+δϕ
k
by some potential change δv
s
, then the other orbitals
ϕ
i
change to ϕ
i
+ δϕ
i
. By using the implicit orbital derivative we regard
E
x
as a functional of the potential, or equivalently of the density, and we
can therefore use the chain rule
_
δE
x
δϕ
k
(r)
_
i
=
_
d
3
r
δE
x
δn(r
)
_
δn(r
)
δϕ
k
(r)
_
i
=
_
d
3
r
v
x
(r
)
_
δn(r
)
δϕ
k
(r)
_
i
(327)
In this equation (δn/δϕ
k
)
i
is also an implicit derivative given by:
_
δn(r
)
δϕ
k
(r)
_
i
=
N
i=1
_
d
3
r
_
ϕ
∗
i
(r
)
δϕ
i
(r
)
δϕ
k
(r)
+ϕ
i
(r
)
δϕ
∗
i
(r
)
δϕ
k
(r)
_
(328)
which should be compared to the explicit derivative
_
δn(r
)
δϕ
k
(r)
_
e
= ϕ
∗
k
(r) (329)
for k = 1 . . . N. We stress that Eq.(327) is only true when using the im
plicit derivatives. If in Eq.(327) we replace the implicit derivatives by the
explicit ones of Eq.(324) and Eq.(329) we obtain a result that is wrong. As
pointed out this is due to the fact that a change in just one orbital can
not be induced by a change in a local potential which in density functional
theory is in 11correspondence with the density. Most of the confusion can
be avoided by regarding all functionals as functionals of the potential v
s
and by calculating the change in the functionals by means of perturbation
theory. In this way one can avoid use of the implicit orbital derivatives as
in Eq.(326).
Let us ﬁnally come back to the question of convergence of the perturbation
series. The perturbation series presented in this section is very similar to
MøllerPlesset perturbation theory starting from the HartreeFock approx
imation. For the MøllerPlesset perturbation theory it is known that it is
in general divergent [49, 50, 51]. However, it is wellknown that when car
ried out to low orders this perturbation theory gives reasonable answers.
The MøllerPlesset perturbation series has therefore all the features of an
asymptotic series [52]. Since the perturbation series in this section is very
similar to the MøllerPlesset series it will in general also diverge. This
has indeed been found in a perturbation theory on the basis of some ap
proximate KohnSham Hamiltonians [53]. Nevertheless, it has been found
for the method presented in this section that low orders in perturbation
theory give good results [54] and we conclude that our series give at least
an asymptotic expansion for the exchangecorrelation energy and potential.
18. Outlook and conclusions
In this work we have given on overview of the mathematical foundations
of stationary density functional theory. We discussed in great detail the
question of diﬀerentiability of the functionals and showed that the Kohn
Sham theory can be put on a solid basis for all practical purposes, since
the set of noninteracting EVdensities is dense in the set of interacting
EVdensities. The question whether these two sets are in fact identical is
still an open question. We further discussed two systematic approaches for
the construction of the exchangecorrelation functional and potential.
What can we say about future developments within density functional the
ory? There have been many extensions of density functional theory in
volving spins, relativistic eﬀects, temperature, superconductivity and time
dependent phenomena. The last few years we have, for instance, seen many
applications of response properties, rather than ground state properties, us
ing timedependent density functional theory. For these extended density
functional theories it is, of course, more diﬃcult to provide a rigorous the
oretical basis. This is, however, not particular to density functional theory,
but applies to any method that deals with manybody systems, especially if
one is interested in phenomena like superconductivity or interactions with
laser ﬁelds. Nevertheless, also for these complicated cases simple density
functionals have provided encouraging results although there is still a clear
need for more accurate density functionals. In view of the success of den
sity functional theory for ground state calculations it seems worthwhile to
explore these new areas.
19. Acknowledgments
I would like to thank Evert Jan Baerends, Pieter Schipper and Oleg Grit
senko for providing me with Fig.4.
References
[1] E.H.Lieb Int.J.Quant.Chem. 24, 243 (1983)
[2] H.Englisch and R.Englisch Phys.Stat.Solidi B123, 711 (1984)
[3] H.Englisch and R.Englisch Phys.Stat.Solidi B124, 373 (1984)
[4] G.de Barra Introduction to Measure Theory Van Nostrand Reinhold,
London (1974)
[5] P.Hohenberg and W.Kohn Phys.Rev., 136B, 864 (1964)
[6] C.De Dominicis and P.C.Martin J.Math.Phys., 5, 14 (1964)
[7] H.Hellmann, Einf¨ uhrung in die Quantenchemie, Leipzig (1937) and
R.P.Feynman Phys.Rev 56, 340 (1939)
[8] W.Kohn and L.J.Sham Phys.Rev. 140, A1133 (1965)
[9] H.Jeggle Nichtlineare Funktionalanalysis Teubner, Stuttgart (1979).
[10] L.B.Rall Nonlinear functional analysis and applications Academic
Press, New York, (1971)
[11] I.Ekeland and R.Temam Convex analysis and variational problems
NorthHolland Publishing, Amsterdam (1976)
[12] M.M.Vainberg Variational method and method of monotone operators
in the theory of nonlinear equations John Wiley, New York (1973)
[13] M.C.Joshi and R.K.Bose Some topics in nonlinear functional analysis
Wiley Eastern, New Delhi (1985).
[14] A.H.Siddiqi Functional analysis with applications McGrawHill, New
Delhi (1986)
[15] R.van Leeuwen Int.J.Mod.Phys. B15,1969 (2001)
[16] H.Englisch and R.Englisch Physica 121A, 253 (1983)
[17] M.Levy Phys.Rev. A26, 1200 (1982)
[18] F.Aryasetiawan and M.J.Stott Phys.Rev. B38, 2974 (1988)
[19] M.Levy Proc.Nat.Acad.Sci.USA 76, 6062 (1979)
[20] D.Werner Funktionalanalysis, Springer, Berlin (1995)
[21] W.Rudin Functional Analysis, McGrawHill, New Delhi (1973)
[22] R.B.Israel Convexity in the Theory of Lattice Gases Princeton Univer
sity Press, Princeton, New Jersey (1979)
[23] J.T.Chayes, L.Chayes and M.B.Ruskai J.Stat.Phys. 38, 497 (1985)
[24] F.W.Averill and G.S.Painter Phys.Rev. B15, 2498 (1992)
[25] S.G.Wang and W.H.E.Schwarz J.Chem.Phys. 105, 4641 (1996)
[26] P.R.T.Schipper, O.V.Gritsenko and E.J.Baerends Theor.Chem.Acc.
99, 329 (1998)
[27] P.R.T.Schipper, O.V.Gritsenko and E.J.Baerends J.Chem.Phys. 111,
4056 (1999)
[28] C.A.Ullrich and W.Kohn Phys.Rev.Lett. 87, 093001 (2001)
[29] D.J.W.Geldart in
Topics in Current Chemistry 180, 31, ed.R.F.Nalewajski, Springer,
Berlin (1996)
[30] Y.Osaka J.Phys.Soc.Japan 36,376 (1974)
[31] R.M.Dreizler and E.K.U.Gross Density Functional Theory: An Ap
proach to the ManyBody Problem Springer, Berlin (1990)
[32] D.C.Langreth and S.H.Vosko Adv.Quant.Chem. 21, 175 (1990)
[33] E.Engel Phys.Rev. A51, 1159 (1995)
[34] P.S.Svendsen and U.von Barth Phys.Rev. B54, 17402 (1996)
[35] E.Engel and S.H.Vosko Phys.Rev. B42, 4940 (1990)
[36] M.L.Glasser Phys.Rev. B51, 7283 (1995)
[37] M.Springer, P.S.Svendsen and U.von Barth Phys.Rev. B54, 17392
(1996)
[38] M.Rasolt Phys.Rev. B36, 5041 (1987)
[39] J.Rey and A.Savin Int.J.Quant.Chem. 69, 581 (1998)
[40] K.Burke, J.P.Perdew and Y.Wang in Electronic Density Func
tional Theory: Recent Progress and New Directions eds.J.F.Dobson,
G.Vignale and M.P.Das, Plenum, New York (1997)
[41] J.P.Perdew, K.Burke and Y.Wang Phys.Rev. B54, 16533 (1996)
[42] H.L.Schmider and A.D.Becke J.Chem.Phys. 109, 8188 (1998)
[43] A.G¨orling and M.Levy Phys.Rev. A50, 196 (1994)
[44] E.Engel and R.M.Dreizler J.Comp.Chem. 20, 31 (1999)
[45] J.D.Talman and W.F.Shadwick Phys.Rev. A14, 36 (1976)
[46] S.Ivanov, S.Hirata and R.J.Bartlett Phys.Rev.Lett. 83, 5455 (1999)
[47] A.G¨orling Phys.Rev.Lett. 83, 5459 (1999)
[48] R.K.Nesbet and R.Colle Phys.Rev. A61, 012503 (1999)
[49] O.Christiansen, J.Olsen, P.Jørgensen, H.Koch and P.
˚
A. Malmqvist
Chem.Phys.Lett. 261, 369 (1996)
[50] J.Olsen, P.Jørgensen, T.Helgaker and O.Christiansen J.Chem.Phys.
112, 9736 (2000)
[51] F.H.Stillinger J.Chem.Phys. 112, 9711 (2000)
[52] C.M.Bender and S.A.Orszag Advanced Mathematical Methods for Sci
entists and Engineers, McGrawHill, Singapore (1978)
[53] M.Warken Chem.Phys.Lett. 237, 256 (1995)
[54] A.FaccoBonetti, E.Engel, R.N.Schmid and R.M.Dreizler
Phys.Rev.Lett. 86, 2241 (2001)
3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19.
Conditions on the electron density and external potentials The HohenbergKohn theorem KohnSham theory by Legendre transforms Deﬁnition of the functional derivative Static linear response of the Schr¨dinger equation o Invertability of the density response function Functional derivatives and vrepresentability The HohenbergKohn theorem for degenerate ground states Linear response for degenerate ground states and functional diﬀerentiability The Levy and Lieb functionals FLL and FL Diﬀerentiability of FL Ensemble vrepresentability The KohnSham approach and noninteracting vrepresentability The gradient expansion The optimized potential method and the e2 expansion Outlook and conclusions Acknowledgments References
1.
Introduction
In this paper we give an overview of the foundations of density functional theory for stationary systems. In the discussion we try to be as precise as possible and we therefore carefully discuss several exact density functionals and the domains of densities on which they are deﬁned. At the heart of almost any application of density functional theory lie the KohnSham equations. These equations describe a noninteracting system that is required to produce the ground state density of an interacting system. Therefore any discussion of the validity of these equations has to focus on how well we can approximate a density of an interacting system with that of a noninteracting system. Closely related to this question is whether or not the exact functionals that we deﬁned have functional derivatives. This is because the various potentials, such as the KohnSham potential, in density functional theory appear as functional derivatives of the energy with respect to the density. The investigation of the existence of functional derivatives will form a central theme of this review. A large part of this work will follow the proofs of the fundamental papers by Lieb [1] and Englisch and Englisch [2, 3]. In this paper we try to make these two important works more accessible by providing some background on the mathematics involved. We further show, from a more physical viewpoint, how to calculate functional derivatives by response theory and show that for nondegenerate ground states the static density response function is invertable. We also show that for every interacting vrepresentable density there is a noninteracting vrepresentable density arbitrarily close to it and that this is suﬃcient to set up a KohnSham scheme. We ﬁnally discuss two systematic approaches for the construction of the exchangecorrelation functional and provide an outlook and conclusions.
2.
Deﬁnition of the problem
Densityfunctional theory in its earliest formulation by Hohenberg, Kohn and Sham [5, 8] aims at a description of the description of ground state properties of manyelectron systems in terms of the electron density. One may wonder why this is possible. Let us therefore investigate this point more closely. Consider a Hamiltonian of a stationary manybody system ˆ ˆ ˆ ˆ Hv = T + V + W (1)
ˆ ˆ where T is the kinetic energy of the electrons, V the external potential, ˆ the twoparticle interaction. We denote the Hamilton operator Hv ˆ and W with a subindex v to indicate that we will consider the Hamiltonian as a functional of the external potential v. The constituent terms are explicitly
e. They only diﬀer in the form of the external potential v(r) W and the number of electrons N . i. However. This is in particular the case for the ground state wavefunction Ψ[v] and the ground state energy E[v]. It can be written as ˆ V = d3 rv(r)ˆ (r) n (6) where we deﬁned the density operator by N n(r) = ˆ i=1 δ(r − ri ) (7) ˆ The expectation value of V is given by ˆ ΨV Ψ = d3 rn(r)v(r) (8) where n(r) the electron density . which are of course related by the Schr¨dinger equation o ˆ ˆ ˆ (T + V + W )Ψ[v] = E[v]Ψ[v] (5) At ﬁrst sight we have not gained much by this viewpoint. The electron density is obtained from the manybody wavefunction. For all ˆ these systems the kinetic energy operator T and twoparticle interaction ˆ are identical. We are interested in electronic systems. the problem looks as diﬃcult as before. which in our case will usually be the ground . One may wonder what is so particular about the relation between the density and the potential. Let us therefore look speciﬁcally at the term which describes the external potential. the problem will have a diﬀerent appearance once we eliminate the potential in favor of the electron density. molecules and solids. The properties of all these systems can therefore be regarded as a functional of the external potential v.given as N ˆ T ˆ V ˆ W = i=1 N − 1 2 2 i (2) = i=1 N v(ri ) w(ri − rj ) (3) = i>j (4) where w(r) = 1/r will in our discussion always be the Coulomb potential.
. The physical interpretation of the density is that n(r)∆V is proportional to the probability of ﬁnding an electron in an inﬁnitesimal volume ∆V around r. It is exactly this property that allows us to prove the HohenbergKohn theorem which establishes a 11correspondence between the density and the external potential. We can therefore go from a functional E[v] of the external potential to a functional E[n] of the density.σN d3 r2 . by n(r1 ) = Ψˆ (r1 )Ψ = N n σ1 . By now we know that we can ﬁnd practical and useful approximations for this potential vs [n] and that the KohnSham equations have been succesfully applied to the calculation of properties of many electronic systems. as in any physical theory. For this we put the physical . . 3. . known as the KohnSham equations. The density has the obvious properties n(r) ≥ 0 .(9) we sum over the spin coordinate σ1 so we do not distinguish between the probabilities of ﬁnding up or down spin electrons at r. If we take into account that the density is obtained as the density of a bound eigenstate of Hamiltonian (1) we can derive further conditions. . We can ask ourselves whether there is any reason that E[n] would be easier to calculate than E[v]. d3 rN Ψ(r1 σ1 .. We see from Eq. With this we mean that they occur as a simple product in the contribution of the external potential to the energy. there are a number of assumptions made in the transformation to the KohnSham oneparticle equations. with a potential vs [n] that is also a functional of the density. rN σN )2 (9) In this expression σi is the spin variable for electron i. The aim of this paper is to discuss these assumptions and thereby the validity of the KohnSham equations. Note that in the deﬁnition Eq. Conditions on the electron density and the external potentials We begin by giving a discussion of the properties of the two key objects in density functional theory.ˆ state wavefunction of Hamiltonian Hv . d3 rn(r) = N (10) These properties follow directly from the deﬁnition of the density and the usual normalization condition on the wavefunction. we will see that the determination of E[n] is equivalent to the solution of a set of oneparticle equations. the density n and the external potential v.(8) that the external potential v(r) and the electron density n(r) are conjugate variables. . However. . A priori there is no reason to expect this. However..
In order to do this we introduce some other function spaces. . We say that a function f is in H 1 (Rn ) (R denotes the real numbers) if 1/2 f = dn r(f (r)2 +  f (r)2 <∞ (12) The space of functions H 1 (Rn ) is called a Sobolev space. i. We say that a given function f belongs to the space Lp if 1/p f p = d3 rf (r)p <∞ (16) Note that here we only consider functions on the usual threedimensional coordinate space R3 .constraint on the manybody system that it has a ﬁnite expectation value of the kinetic energy. With this we mean that if n1 and n2 are elements of S. .σN d3 r 1 . The supindex 1 refers to the fact that the deﬁnition of the norm contains only ﬁrst order derivatives. T [Ψ] = 1 2 N i=1 σ1 .(9) and use of the CauchySchwarz inequality then leads to [1] ( and hence 1 n(r1 )) 2 ≤ 4N n(r1 ) σ1 . a feature .. The letter L refers to Lebesque integration. We therefore see that ﬁniteness of the kinetic energy implies that Ψ is an element of the function space H 1 (R3N ). .(12) we see that n belongs to H 1 (R3 ). then we only need to consider the following set of densities S = {n  n(r) ≥ 0. . d3 rn(r) = N } (15) This set of densities has a property which will be of importance later. then also λn1 + (1 − λ)n2 is an element of S where 0 ≤ λ ≤ 1. √ n ∈ H 1 (R3 ). From Eq.. d 3 r N  1 Ψ 2 (13) 1 n(r))2 ≤ T [Ψ] < ∞ (14) d3 r( 2 We see that the ﬁniteness of the √ kinetic energy puts a constraint on the density. d 3 r N  i Ψ 2 <∞ (11) At this point it is useful to introduce a new space of functions.. Diﬀerentiation of Eq. This property is easily proven using the CauchySchwarz inequality.. namely S is convex. We also see that if we consider systems with a ﬁnite kinetic energy.e. Now we will derive constraints on the allowed set of external potentials.σN d3 r 2 .
n ∈ L1 ∩ L3 . We almost never use it in the remainder of this paper.e. We also see from the inequality (17) that S is a subset of L1 ∩ L3 . is the set L 2 + L∞ = {vv = u + w.(10) that the density belongs to L1 . In other words. Equation (16) deﬁnes a norm on the space Lp and we see from Eq.that assures that the function spaces are complete (complete normed spaces are also called Banach spaces). w ∈ L∞ } 3 3 (23) . u ∈ L 2 .e. We therefore conclude that if v ∈ L∞ then the expectation value of the external potential is ﬁnite. not go into the detailed mathematics and refer the interested reader to the literature [4]. We will. i. Since we already know that n ∈ L1 we ﬁnd that the density is element of the intersection of both spaces. Therefore the most general set of potentials for ˆ which the expectation value ΨV Ψ is ﬁnite. v(r) < M for some ﬁnite number M then d3 rn(r)v(r) ≤ sup v(r)N < ∞ The space of bounded functions is called L∞ and has the norm f ∞ (19) = sup f (r) (20) The supremum is deﬁned to be the smallest number M such that f (r) ≤ M almost everywhere. Let us now see what this implies for the allowed set of external potentials when we require the expectation value of the external potential to be ﬁnite. To show this we used that n ∈ L1 . i. We just note that for continuous functions the integral is equivalent to the usual (Riemann) integral.e. The term ’almost everywhere’ has a precise mathematical meaning for which we refer to the literature [4]. i. however. we require d3 rn(r)v(r) ≤ nv 1 <∞ (18) If the potential is bounded. the ﬁniteness of the kinetic energy implies that the density is also in the space L3 . But we also know that n ∈ L3 and if we make use of the H¨lder inequality o fg 1 ≤ f p g q (21) with 1/p + 1/q = 1 we obtain nv 3 1 ≤ n 3 v 3 2 (22) which is ﬁnite if v ∈ L 2 . From the condition of ﬁnite kinetic energy and the use of a Sobolev inequality one can show that [1] d3 rn3 (r) ≤ C d3 r( n(r))2 < ∞ (17) where C = 3(π/2)4/3 .
This means that the external potential is a welldeﬁned functional v[n](r) of the density.e.i. We will provide a proof of this theorem for nondegenerate ground states. However. prove that the 2particle interaction is a unique functional of the diagonal 2particle density matrix. One can easily extend this result to a ﬁnite sum of Coulomb potentials and therefore every molecule 3 can be described with the potentials in the space L 2 +L∞ . the set of potentials that can be written as a sum of a function from L 2 and a function from L∞ . 4. would imply for the density. One can readily check that the ﬁrst and the second term on the 3 right hand side are in L 2 and L∞ respectively. The HohenbergKohn theorem The basis of densityfunctional theory is provided by the HohenbergKohn theorem [5]. This is also a normed function space with norm v = inf{ u 3 2 3 + w ∞ v = u + w} (24) In the remainder of this paper we will always consider the densities to be in 3 the space L1 ∩ L3 and the potentials in the space L 2 + L∞ . For the same reason we can. 3 Let us consider the subset V ⊂ L 2 +L∞ of potentials that yield a normalizable nondegenerate ground state. ΨW Ψ < ∞. for instance. Since we assume that we are dealing with nondegenerate ground states Ψ[v] is uniquely determined apart from a trivial phase factor. We have therefore established a map C : V → Φ. One may ﬁnally wonder what the condition of ﬁnite Coulombic electronelectron repulsion. In the proof essential use is made of the fact that the density and the potential are conjugate variables. where Φ is the set of ground states. so this does not yield any new constraints on the density. The HohenbergKohn theorem states that the density n(r) of a nondegenerate ground state uniquely determines the external potential v(r) up to an arbitrary constant. ˆ i. The general mapping between N particle density matrices and N body potentials is discussed by De Dominicis and Martin [6]. . one can show ˆ ˆ that if the kinetic energy ΨT Ψ is ﬁnite then also ΨW Ψ is ﬁnite [1]. It is important to note that the Coulomb potential is in the latter set since we can write θ(1 − r) θ(r − 1) 1 = + r r r (25) where θ is the Heaviside function.e. θ(x) = 0 if x ≤ 0 and θ(x) = 1 if x > 0. The case of degenerate ground states will be discussed later. v(r) → Ψ[v] . The solution of the Schr¨dinger equation o provides us with a mapping from the external potential to the ground state wavefunction.
Suppose that Ψ1 and 3 Ψ2 ∈ H 1 (R3N ) correspond to external potentials v1 and v2 ∈ L∞ + L 2 where v1 = v2 + C. Let us now pick an arbitrary density out of the set A of densities of nondegenerate ground states. So we obtain a contradiction and hence we must have made a wrong assumption. This proves the HohenbergKohn theorem. Consequently the map DC : V → A is also invertible and the density therefore uniquely determines the external potential. Therefore Ψ1 = Ψ2 and we obtain the result that diﬀerent potentials ( diﬀering more than a constant) give diﬀerent wavefunctions. Consequently we ﬁnd that the map C is invertible. We have to show that Ψ1 = Ψ2 . If we assume that Ψ1 = Ψ2 = Ψ then by subtraction of the Hamiltonian for Ψ1 and Ψ2 we ﬁnd that ˆ ˆ (V1 − V2 )Ψ = (E1 − E2 )Ψ (26) If v1 − v2 is not constant in some region then Ψ must vanish in this region 3 for the above equation to be true. The HohenbergKohn theorem then tells us that . The set A is obviously a subset of the previously deﬁned set S. where we only consider ground state densities from 3 potentials in the set L∞ + L 2 . This provides us with a second map D : Φ → A from ground state wavefunctions to ground state densities. From a given wavefunction in the set of ground states Φ we can calculate the density according to Eq. We now deﬁne the set A as the set of densities which come from a nondegenerate ground state. However if v1 . Also this map is invertible. To show this we calculate E[v1 ] = = = ˆ ˆ ˆ ˆ ˆ ˆ Ψ[v1 ]T + V1 + W Ψ[v1 ] < Ψ[v2 ]T + V1 + W Ψ[v2 ] ˆ ˆ ˆ Ψ[v2 ]T + V2 + W Ψ[v2 ] + E[v2 ] + n2 (r)(v1 (r) − v2 (r))dr (27) n2 (r)(v1 (r) − v2 (r))dr Likewise we ﬁnd E[v2 ] < E[v1 ] + n1 (r)(v2 (r) − v1 (r))dr (28) Adding both inequalities then yields the new inequality d3 r(n2 (r) − n1 (r))(v2 (r) − v1 (r)) < 0 (29) If we assume that n1 = n2 then we obtain the contradiction 0 < 0 and we conclude that diﬀerent ground states must yield diﬀerent densities.We will ﬁrst prove that the map C is invertible. Therefore the map D is also invertible. v2 ∈ L∞ + L 2 then Ψ cannot vanish on an open set (a set with nonzero measure) by the unique continuation theorem [1].(9).
Suppose that the ground state densities n1 . Before we go into that let us ﬁrst discuss some properties of FHK . n2 ∈ A and if λ1 n1 + λ2 n2 ∈ A with 0 ≤ λ1 . The functional FHK is a convex functional. λ2 ≤ 1 and λ1 + λ2 = 1 then FHK [λ1 n1 + λ2 n2 ] ≤ λ1 FHK [n1 ] + λ2 FHK [n2 ] (35) This is readily proven.e. of course. n2 . represented by an operator O. For application of this formula we have to know FHK on the set A.there is a unique external potential v (to within a constant) and a unique ground state wavefunction Ψ[n] (to within a phase factor) corresponding to this density. if n1 . In practice it is. This also means that the ground state expectation value of ˆ any observable. λ1 n1 + λ2 n2 ∈ A correspond to the external potentials v1 . v2 and v. Instead one would prefer to make an explicit approximation for FHK as is usually done within the KohnSham scheme. can be regarded as a density functional ˆ O[n] = Ψ[n]OΨ[n] (30) In particular we can thus deﬁne the HohenbergKohn functional FHK on the set A as ˆ ˆ FHK [n] = Ψ[n]T + W Ψ[n] (31) With this functional we can deﬁne the energy functional Ev as Ev [n] = n(r)v(r)dr + FHK [n] (32) If n0 is a ground state density corresponding to external potential v0 and n an arbitrary other ground state density then Ev0 [n] = ≥ Therefore E[v] = inf n∈A ˆ ˆ ˆ n(r)v0 (r)dr + FHK [n] = Ψ[n]T + V0 + W Ψ[n] ˆ ˆ ˆ Ψ[n0 ]T + V0 + W Ψ[n0 ] = Ev0 [n0 ] n(r)v(r)dr + FHK [n] (33) (34) and we see that the ground state energy of a manybody system can be obtained by minimization of a density functional. i. Then FHK [n] = = ≤ ˆ ˆ ˆ Ψ[n]T + V + W Ψ[n] − d3 rn(r)v(r) d3 rn(r)v(r) ˆ ˆ λ1 Ψ[n]HΨ[n] + λ2 Ψ[n]HΨ[n] − ˆ ˆ ˆ ˆ λ1 Ψ[n1 ]T + W Ψ[n1 ] + λ2 Ψ[n2 ]T + W Ψ[n2 ] . impossible to calculate FHK exactly on this set of densities.
Therefore (v2 (r) − v1 (r))(n2 (r) − n1 (r)) = 2 δv(r)δn(r) + O( 3 ) (39) .(29) we take v2 = v1 + δv where δv is not a constant. i. ˆ 1.e. The ground state energy of a system with a nondegenerate ground state and an external potential v can be obtained from E[v] = inf n(r)v(r)dr + FHK [n] n∈A ˆ ˆ where FHK [n] = Ψ[n]T + W Ψ[n] . n2 ∈ A then not necessarily λ1 n1 + λ2 n2 ∈ A with 0 ≤ λ1 .+ = (λ1 n1 (r) + λ2 n2 (r))v(r)d3 r − λ1 FHK [n1 ] + λ2 FHK [n2 ] d3 rn(r)v(r) (36) and we obtain the convexity of FHK . 3. Let us ﬁrst collect our results in the form a theorem Theorem 1 (HohenbergKohn) The density n corresponding to a nondegenerate ground state speciﬁes the external potential v up to a constant and the ground state wavefunction Ψ[n] up to a phase factor. that the domain A of FHK does not need to be convex. We therefore obtain potential v2 by a small perturbation from potential v1 . λ2 ≤ 1 and λ1 + λ2 = 1. r )δv(r ) (38) The properties and explicit form of χ are described in more detail in a later section. if n1 . By means of perturbation theory we can then calculate the change in the density which gives n2 (r) = n1 (r) + δn(r) + O( 2 ) (37) where δn can be calculated from the static density response function χ δn(r) = d3 r χ(r. Any ground state expectation value corresponding to an observable O is a functional of the density according to ˆ O[n] = Ψ[n]OΨ[n] 2. FHK is convex We ﬁnally discuss an interesting consequence of the HohenbergKohn theorem. however. We will come back to this point later when we consider the diﬀerentiability of FHK . Moreover. Suppose that in Eq. Note.
of which the densitypotential relation to be discussed here is a special case. Moreover we see that the only potential variation that yields a zero density variation is given by δv = C where C is a constant. are negative.(29) and divide by d3 rδv(r)δn(r) + O( ) < 0 Now taking the limit function we obtain 2 > 0 then we obtain: (40) → 0 and expressing δn in terms of the response d3 rd3 r δv(r)χ(r. From this equation we see that the ground state energy as a functional of the external potential v can also be written as ˆ E[v] = Ψ[v]Hv Ψ[v] (43) Our goal is now to go from the potential as our basic variable. The deeper reason that this is possible is that the density and the potential are conjugate variables. r )δv(r ) < 0 (41) This is true for an arbitrary nonconstant potential variation. With this we mean that the contribution of the external potential to the total energy is simply an integral of the potential times the density. This implies that χ is invertable. 5. This work discusses the relations between N body potentials and N particle density matrices. when we regard χ as an integral operator. We make use of this relation if we take the functional derivative of the energy functional E[v] with respect to the potential v: δE δv(r) = ˆ δΨ ˆ ˆ δΨ + Ψ δ Hv Ψ Hv Ψ + ΨHv  δv(r) δv(r) δv(r) . We therefore see that the eigenvalues of χ. We will go more closely into this matter in a later section where we will prove the same using the explicit form of the density response function. to a new variable. The ground state energy E[v] and wave function Ψ[v] are considered to be functionals of the external potential through solving the timeindependent Schr¨dinger equation o ˆ ˆ ˆ (T + V + W )Ψ[v] = E[v]Ψ[v] (42) ˆ where the twoparticle interaction W is kept ﬁxed. KohnSham theory by Legendre transforms The method described in this section goes back to the work of De Dominicis and Martin [6]. which will be the electron density.If we insert this expression in Eq.
(48) is the kinetic energy of a noninteracting system with potential vs and density n. For this reason the functional Fs is usually . In order to derive the KohnSham equations we deﬁne the following energy functional for a system of noninteracting particles with external potential vs and with ground state wave function Φ[vs ] : ˆ ˆ Es [vs ] = Φ[vs ]T + Vs Φ[vs ] (47) with Legendre transform Fs [n] = E[vs ] − and derivatives δEs δvs (r) δFs δn(r) = = n(r) −vs (r) (49) (50) ˆ d3 rn(r)vs (r) = Φ[vs ]T Φ[vs ] (48) We see that Fs [n] in Eq. Note that the equation above is nothing but a functional generalization of the wellknown HellmannFeynman theorem [7]. From δE/δv = n it follows immediately that δF = −v(r) δn(r) (46) This is our ﬁrst basic relation. We postpone a discussion of these matters to later sections. The question which constraints one has to put on a density to make sure that it is vrepresentable is known as the vrepresentability problem. Now we can go to the density as our basic variable by deﬁning a Legendre transform F [n] = E[v] − ˆ ˆ d3 rn(r)v(r) = Ψ[v]T + W Ψ[v] (45) where v must now be regarded as a functional of n. These are ground state densities for a Hamiltonian with external potential v. The set of densities for which the functional F [n] is deﬁned is the set of socalled vrepresentable densities.= = E[v] δ ΨΨ + Ψˆ (r)Ψ n δv(r) Ψˆ (r)Ψ = n(r) n (44) ˆ where we used the Schr¨dinger equation Hv Ψ = E[v]Ψ and the noro malization condition ΨΨ = 1. The uniqueness of such a mapping is garanteed by the HohenbergKohn theorem [5].
we assume that the interacting density is noninteractingvrepresentable. These equations turn out to be of great practical use. we can calculate the exchangecorrelation potential vxc and solve the orbital equations selfconsistently. The density we ﬁnd in this way can then be used to calculate the ground state energy of the system.denoted by Ts . The exchangecorrelation functional is often split up into an exchange functional Ex and a correlation functional Ec as Exc [n] = Ex [n] + Ec [n] (57) . Now the state Φ[vs ] is a ground state for a system of noninteracting particles. (51) with respect to the density n we obtain vs (r) = v(r) + where vxc (r) = d3 r n(r )w(r − r ) + vxc (r) δExc δn(r) (52) (53) deﬁnes the exchangecorrelation potential. In the following we will adopt this notation. If we can ﬁnd a good approximation for the exchangecorrelation energy. We thus assume that for a given ground state density of an interacting system there is a noninteracting system with the same density. and can therefore be written as an antisymmetrized product of singleparticle orbitals ϕi (r). Finally we deﬁne the exchangecorrelation functional Exc [n] by the equation F [n] = Ts [n] + 1 2 d3 rd3 r n(r)n(r )w(r − r ) + Exc [n] (51) This equation assumes that the functionals F [n] and Ts [n] are deﬁned on the same domain of densities. If we now collect our results we see that we have converted the ground state problem into the following set of equations N E[v] = i=1 − 1 2 d3 rϕ∗ (r) i 2 ϕi (r) + d3 rn(r)v(r) (54) (55) (56) + (− 1 2 2 1 2 d3 rd3 r n(r)n(r )w(r − r ) + Exc [n] d3 r n(r )w(r − r ) + vxc (r))ϕi (r) = N i ϕi (r) + v(r) + n(r) = i=1 ϕi (r)2 The above equations constitute the ground state KohnSham equations [8]. If we diﬀerentiate Eq. In other words.
in which the exchange and correlation functionals are deﬁned by Ex [n] Ec [n] = = 1 ˆ d3 rd3 r n(r)n(r )w(r − r ) (58) Φs [n]W Φs [n] − 2 ˆ ˆ Ψ[n]Hv Ψ[n] − Φs [n]Hv Φs [n] (59) In this equation Φs [n] is the KohnSham wavefunction and Ψ[n] the true ground state wavefunction of the interacting system with density n. 12. 11. Since the KohnSham wavefunction is not a ground state wavefunction of the true system we see immediately from the variational principle that Ec < 0. 6. notably molecular dissociation cases. We assumed that for every ground state density of an interacting system there is a noninteracting system which has the same density in its ground state. where the exchangeonly theory is a bad starting point for the treatment of correlation eﬀects and for which it is much easier to ﬁnd good approximations for the combined exchangecorrelation functional. However. The derivative we will talk about is what in the mathematical literature [9. Let G : B → R be a a functional from a normed function space B to the real numbers R. in this derivation we made a number of assumptions. there are many cases. We see also that the form of the exchange functional in terms of the KohnSham orbitals is identical to that of the exchange energy within the wellknown HartreeFock approximation. Deﬁnition of the functional derivative Let us start by deﬁning what we mean with a functional derivative. However. we assumed that the density functionals were diﬀerentiable. First of all. In the following sections we will investigate to which extent these assumptions are justiﬁed. 14] is refered to as a Gˆteaux derivative. this splitting has only meaning if the ground state of the true or KohnSham system is nondegenerate. Secondly. but that exchange and correlation separately are illdeﬁned in that case. We will later see that Exc [n] is a welldeﬁned functional even for degenerate ground states. We ﬁnally remark that splitting up Exc into an exchange and a correlation part has several disadvantages. If for every h ∈ B there exist a continuous linear functional δG/δf : B → R deﬁned by: G[f + h] − G[f ] δG [h] = lim (60) →0 δf . Secondly. 10. This assumes that the values of the functionals change smoothly with changes in the density. 13. the value of Ex [n] is not equal to the HartreeFock exchange. since the KohnSham and HartreeFock orbitals diﬀer. In this section we gave a derivation of the KohnSham equations.
Note that the deﬁnition of the Gˆteaux a a derivative is very similar to the deﬁnition of the directional derivative in vector calculus where the linear functional corresponds to the inner product of the gradient vector with the vector which speciﬁes the direction of the diﬀerentiation. This set is called the dual space of B and denoted B ∗ .e. For instance. i. G[f ] ≥ G[f0 ] for all a f ∈ B. the function g(r) depends in general on f in a nonlinear way. Suppose that the functional G which we assume to a be Gˆteaux diﬀerentiable. Thus 0= = 0 and therefore the derivative of g( ) in g( ) − g(0) δG dg (0) = lim = (f0 )[h] →0 d δf (63) = 0 (64) Therefore a necessary condition for a diﬀerentiable functional for having a minimum at f0 is that its Gˆteaux derivative vanishes at f0 . i. has a minimum at f0 . Often the linear functional δG/δf can be written in the form δG [h] = d3 rg(r)h(r) (61) δf If this is the case we write g(r) = δG δf (r) (62) which we will call the functional derivative of G. the dual space of B = L1 ∩ L3 is known to be the 3 space B ∗ = L 2 + L∞ . Let us now mention a straightforward consequence of the deﬁnition of the Gˆteaux derivative. We further a prove one other fact that we will use later. From this we see that if the derivative of a density functional deﬁned on the set of densities L1 ∩ L3 exists then its derivative 3 is in the set L 2 + L∞ . We also see that the set of functional derivatives on a function space B is equal to the set of continuous linear functionals on that space.e. Note further that although δG/δf speciﬁes a linear functional when acting on h. If the limit exists a but the resulting functional of h is not linear or continuous then this limit is called a Gˆteaux variation.then δG/δf is called the Gˆteaux derivative in f ∈ B. Then the function g( ) = G[f0 + h] ≥ G[f0 ] has a minimum in vanishes. G[λf0 + (1 − λ)f1 ] ≤ λG[f0 ] + (1 − λ)G[f1 ] for 0 ≤ λ ≤ 1 then the function g(λ) = G[f0 + λ(f1 − f0 )] (66) (65) . If G is a convex functional.
h] = n (0) δf n d (68) where δ n G/δf n has now n arguments h. . . . If this expression deﬁnes a multilinear continuous functional then we call this the nthorder Gˆteaux derivaa tive of G. δf (rn ) (70) the nthorder functional derivative of G. 1]. rn ) δf (r1 ) . . This is exactly what is studied in most experimental situations where one switches on and oﬀ an external ﬁeld and studies how the system reacts to this. . . . Sofar our discussion has been rather abstract. . . . . . By investigating this problem we will learn how to take functional derivatives and how the existence of these derivatives is related to the existence of the linear response function. . Then g( λ1 + (1 − )λ2 ) = G[ (f0 + λ1 (f1 − f0 )) + (1 − )(f0 + λ2 (f1 − f0 ))] ≤ g(λ1 ) + (1 − )g(λ2 ) (67) Therefore g(λ) is a convex real function on the interval [0. For more details on this point we refer to reference [9]. if this expression can be written in the form δn G [h. Let us therefore apply the deﬁnition and calculate some functional derivatives. . Suppose that we have solved the following ground state problem: ˆ ˆ ˆ ˆ HΨ0 = (T + V + W )Ψ0 = E0 Ψ0 (71) . Again. h] = δf n then we call d3 r1 .is a continuous function on the interval [0. . Take λ1 . . d3 rn g(r1 . Static linear response of the Schr¨dinger equation o We now consider the eﬀect of small changes in the external ﬁeld on the expectation values of physical observables. . To show this we will show that g(λ) is convex. h(rn ) (69) δn G = g(r1 . Let us ﬁnally make some remarks on higher order derivatives. 7. . but this does not extend to inﬁnitedimensional spaces. λ2 and from the interval [0. rn )h(r1 ) . 1] and hence continuous. The fact that convexity implies continuity is true for functions on the real axis. 1]. If g( ) = G[f0 + h] deﬁnes a nfold diﬀerentiable function of then we deﬁne the nth Gˆteaux variation of G as a dn g δn G [h. We will here study a more speciﬁc case in which we look at static changes in the external potential and their accompanying changes in the ground state expectation values.
We therefore make a slight change ˆ δV = d3 rˆ (r)δv(r) n (74) ˆ in the external potential of Hamiltonian H. However.ˆ where Ψ0 is a nondegenerate ground state Ψ[v] of Hamiltonian H with ˆ external potential V . which for both functions looks like Ψ( ) = = Ψ0 + Ψ (0) + O( 2 ) (77) 2 Φ( ) Ψ0 + (Ψ (0) + iθ (0)Ψ0 ) + O( ) (78) where Ψ (0) and θ (0) are the ﬁrst order derivatives of Ψ( ) and θ( ) in = 0. We then obtain ˆ ˆ (H − E0 )Ψ (0) = (E (0) − δ V )Ψ0 (79) . If we choose θ(0) = 0 then Φ( ) also satisﬁes the condition Φ(0) = Ψ0 . where θ( ) is an arbitrary function of .(75) is not unique. i. According to our deﬁnition in the previous section we have to calculate the quantity O[v + δv] − O[v] δO [δv] = lim (73) →0 δv To evaluate this limit we have to calculate O[v + δv] which we will do using static perturbation theory. it aﬀects the appearance of our ﬁrst order expansion.e. This is exactly the freedom we will ﬁnd in our expansion when we try to obtain the ﬁrst order change in the wavefunction.(75) to ﬁrst order in . We note that the solution of Eq. From the wavefunction we can then calculate the ˆ expectation value of any operator O which is a welldeﬁned functional of the external potential ˆ O[v] = Ψ[v]OΨ[v] (72) Let us now calculate the functional derivative δO/δv at a given potential v. Let us expand the wavefunction and energy in Eq. The arbitrariness of the phase factor obviously does not aﬀect the value of any expectation value. by an imaginary number times the unperturbed ground state.e. because if Ψ( ) is a solution then also Φ( ) = eiθ( ) Ψ( ) is a solution. We see that Ψ( ) and Φ( ) diﬀer in ﬁrst order by the amount i θ (0)Ψ0 . i. The new ground state wavefunction which we will denote by ˆ to V Ψ( ) satisﬁes ˆ (H + δV )Ψ( ) = E( )Ψ( ) (75) We will solve this equation to ﬁrst order in with the condition Ψ(0) = Ψ0 . we change the potential ˆ + δ V .e. ˆ ˆ O( ) = Ψ( )OΨ( ) = Φ( )OΦ( ) (76) which is a unique function of . i.
where E (0) is the ﬁrst order derivative of E( ) in = 0. To solve this equation we expand Ψ (0) in an orthonormal set of eigenstates of the ˆ unperturbed Hamiltonian H as ∞ Ψ (0) = i=0 ci Ψi (80) If the unperturbed Hamiltonian has a continuous spectrum then for the corresponding energy eigenstates the summation in this equation has to be replaced by an integration. If we insert the expansion (80) into Eq.(81) from the left with Ψk  for k > 0 we ﬁnd ck = − ˆ Ψk δ V Ψ0 Ek − E 0 (for k > 0) (84) Note that these coeﬃcients are welldeﬁned because Ek > E0 . If we multiply this equation from the left with Ψ0  we obtain for the change in energy ˆ E (0) = Ψ0 δ V Ψ0 = d3 rn0 (r)δv(r) (82) where n0 is the density of the unperturbed system. If we now multiply Eq. from the requirement that the perturbed system remains normalized we ﬁnd 0= d ΨΨ (0) = Ψ (0)Ψ0 + Ψ0 Ψ (0) = c∗ + c0 0 d (86) . We have therefore shown that E (0) = lim →0 E[v + δv] − E[v] = d3 rn0 (r)δv(r) (83) As has been derived before we ﬁnd δE/δv(r) = n0 (r). We therefore ﬁnd for the ﬁrst order change in the wavefunction ∞ Ψ (0) = c0 Ψ0 − k=1 ˆ Ψk Ψk δ V Ψ0 Ek − E 0 (85) The coeﬃcient c0 remains undetermined by these equations.(79) for Ψ (0) we ﬁnd the equation ∞ i=0 ˆ ci (Ei − E0 )Ψi = (E (0) − δ V )Ψ0 (81) where the energies Ei for i > 0 are the eigenenergies of the excited states of the unperturbed Hamiltonian. However. since we are dealing with an isolated nondegenerate ground state.
r ) = χ(r . We now want to calculate the ﬁrst order change in an arbitrary expectation value due to a change δv of the external potential. From Eq. Ek − E 0 (91) where χ(r . r) = =− δv(r) ∞ k=1 Ψ0 ˆ (r )Ψk Ψk ˆ (r)Ψ0 n n + c. χ(r. r) is the static density response function.and hence c0 must be purely imaginary. From this expression we see immediately that χ is real and symmetric. evaluated at potential v(r). r).(90): δn(r ) χ(r .e. This is exactly the kind of freedom in the ﬁrst order wavefunction that we noted before.c. as we have seen this is related to choice of phase and does not aﬀect the expectation values. .(85) we obtain O (0) = ˆ (c∗ + c0 ) Ψ0 OΨ0 0 ∞ ˆ ˆ ˆ ˆ Ψ0 δ V Ψk Ψk OΨ0 + Ψ0 OΨk Ψk δ V Ψ0 − Ek − E 0 k=1 (88) From the normalization condition Eq.c. i. However. We will use this formula later when we are studying functional derivatives with respect to the density.(86) we see that the ﬁrst term on the right hand side vanishes and we can rewrite the equation as O (0) = where δO =− δv(r) δO [δv] = δv d3 r δO δv(r) δv(r) (89) ∞ k=1 ˆ Ψ0 OΨk Ψk ˆ (r)Ψ0 n + c. This change is given by O (0) = lim →0 O[v + δv] − O[v] ˆ ˆ = Ψ (0)OΨ0 + Ψ0 OΨ (0) (87) where O (0) is the derivative of O( ) in = 0 and where we assumed that ˆ ˆ the operator O does not depend on V (as for instance the Hamiltonian does). In that case we obtain from Eq. We now make a special choice for ˆ ˆ ˆ the operator O and we choose O = n(r ). 8. Ek − E 0 (90) For a system with a nondegenerate ground state we therefore have obtained an explicit expression for the functional derivative δO/δv of the expectation value O[v]. Invertability of the density response function In the previous section we obtained an expression for the functional derivative of an arbitrary expectation value.
On the basis of the HohenbergKohn theorem we may further expect that the only potential variation that yields a zero density variation is the constant potential. r ) = 0 (93) which can also be checked immediately from Eq. r )δv(r ) (94) then δv(r) = C. Written in ﬁrst n quantization this implies that N ∆v(ri )Ψ0 = 0 i=1 (98) . r )δv(r ) = −2 k=1 ak 2 Ek − E 0 (95) where we deﬁned ak by ak = d3 r Ψk ˆ (r)Ψ0 δv(r) n (96) Since Ek − E0 > 0 equation (95) can only be true if ak = 0 for all k ≥ 1. The density response function has therefore indeed the physical properties that we expect.(94) is true for some δv(r ). This readily proven from the properties of the response function. Then we obtain by integration with δv(r): ∞ 0= d3 rd3 r δv(r)χ(r.(91). We therefore must have d3 r χ(r. But this implies that ∞ ∞ 0 = k=1 ak Ψk = k=1 d3 rΨk Ψk ˆ (r)Ψ0 δv(r) n = = d3 rδv(r) (1 − Ψ0 Ψ0 ) n(r)Ψ0 ˆ d3 rδv(r) (ˆ (r) − n0 (r)) Ψ0 n (97) where n0 (r) = Ψ0 ˆ (r)Ψ0 is the ground state density. Because of this property we also see that the induced density δn(r) automatically integrates to zero.The density response function relates ﬁrst order changes in the potential to ﬁrst order changes in the density according to δn(r) = d3 r χ(r. if 0= d3 r χ(r. r )δv(r ) (92) If we change the potential by a constant then the density does not change. Suppose Eq.e. i.
i. r )f (r ) = −2 bk = k=1 bk 2 ≤ 0 (101) Ek − E 0 (102) d3 r Ψk ˆ (r)Ψ0 f (r) n We therefore ﬁnd that λ ≤ 0. r )f (r ) = λf (r) (100) Multiplication from the right with f (r) and subsequent integration yields ∞ λ where d3 rf 2 (r) = d3 rd3 r f (r)χ(r. which we call δA. The response function deﬁnes a mapping χ : δV → δA from the set of potential variations from a 3 nondegenerate ground state. that are produced by it. Let f (r) be such an eigenfunction. then the constant of proportionality λ is negative. We ﬁnally note the invertability proof for the static response function can be extended to the timedependent case. Moreover it is found that there is a inﬁnite number of negative eigenvalues arbitrarily close to zero. However. For a recent review we refer to [15]. which causes considerable numerical diﬃculties when one tries to obtain the potential variation that is responsible for a given density variation.e. there are density variations that can never be produced by a potential variation and which are therefore not in the set δA. we already know that λ = 0 is only possible if f (r) is constant. An example of such a density variation is one which is identically zero on some ﬁnite volume. δn = λδv. however. d3 r χ(r. From our analysis we can further derive another property of the static response function.e.(98) implies that ∆v(r) = 0 which together with Eq. We have just shown that the inverse χ−1 : δA → δV is welldeﬁned modulo a constant function.(99) implies that δv(r) = C. be careful with what one means with the inverse response function. We have therefore obtained the result that if a nonzero density variation is proportional to the potential that generates it. Since χ(r. In actual calculations one indeed ﬁnds that the eigenvalues of χ are negative. One should. to the set of ﬁrst order densities variations. which we call δV and is a subset of L 2 + L∞ . This is exactly what one would expect on the basis of physical considerations. r ) is a real Hermitian integral kernel it has an orthonormal set of eigenfunctions which can be chosen to be real. However. and therefore the density response function is invertible up to a constant.where N is the number of electrons and ∆v the potential ∆v(r) = δv(r) − 1 N d3 rn0 (r)δv(r) (99) Now Eq. We have therefore shown that only constant potentials yield a zerodensity variation. . i.
Whether or not this is true obviously depends on the choice we make for δn. to ﬁrst order in . Nevertheless. of course. i. Let us therefore start with the deﬁnition of the functional derivative and try to calculate δO O[n0 + δn] − O[n0 ] [δn] = lim →0 δn (105) where n0 is the density of a given nondegenerate ground state with external potential v0 .9. This will only be so if we can ﬁnd an additional potential that would make the higher order terms in disappear and it is not clear how to prove the existence of such a potential. In this way we know that for every such δn there is a potential v0 + δv that generates.e. Then. A natural choice would be to take δn ∈ δA. The same is. Now we run into the following problem. We will therefore write ˆ ˆ O[n] = Ψ[n]OΨ[n] = Ψ[v]OΨ[v] = O[v] (103) Since we know how to calculate the functional derivative with respect to v the functional derivative with respect to n seems rather straightforward. Let us therefore take some δn ∈ δA. However. Because the density n determines the external potential v and vice versa we can parametrize the ground state either by the density or the external potential. r) δv(r ) (104) where we used that the density response function has an inverse (modulo a constant). by the .e. we have to make sure that this perturbed density belongs to a nondegenerate ground state. Functional derivatives and vrepresentability We are now ready to tackle the question how to calculate functional derivatives with respect to the density. One needs to consider under which conditions the use of this chain rule is applicable. we take a δn from the set of ﬁrst order density variations produced by some potential variations δv ∈ δV. this idea is suﬃcient to make a useful statement about the functional derivative. i. great care must be taken when deriving this equation. the required density. By the HohenbergKohn theorem every density associated with a nondegenerate ground state uniquely determines that ground state and the external potential that produced it. true for the expectation value of any operator ˆ O that we calculate from the ground state. However. In order to make the value of O[n0 + δn] welldeﬁned we have to make sure that n0 + δn ∈ A. By the chain rule for diﬀerentiation we obtain δO = δn(r) d3 r δO δv(r ) = δv(r ) δn(r) d3 r δO −1 χ (r . this still does not imply that n0 + δn ∈ A.
and therefore independent of the parametrized path n ∈ A that . The functional derivative δO/δn[δn] in Eq. since δn(r) integrates to zero. to ﬁrst order in . i.c.e. Consider now the set of external potentials v0 + δv. χ−1 (r . n can be written as n (r) = n0 (r) + δn(r) + m (r) ∈ A where the term m (r) satisﬁes lim m (r) =0 (108) (107) →0 Since n is the density of a nondegenerate ground state the expectation value O[n ] is now welldeﬁned and we can calculate lim O[n ] − O[n0 ] = lim O[v0 + δv] − O[v0 ] = d3 r δO δv(r) (109) δv(r) →0 →0 We have therefore found a parametrized set of densities n ∈ A for which the limit above is welldeﬁned. i.(104). Moreover.(111) is welldeﬁned in terms of the properties of the unperturbed system and independent of the choice for δn. there is a unique δv modulo a constant. By construction n − n0 will. For each we can solve the Schr¨dinger equation and o obtain a ground state density n (r).e. regarded as a linear functional acting on δn. is of course independent of this constant. so that we can 0 write δO O[n0 + δn + m ] − O[n0 ] = d3 r δn(r) (110) lim →0 δn(r) where the functional derivative δO/δn(r) is given (modulo a constant) by δO =− δn(r) ∞ d3 r k=1 ˆ Ψ0 OΨk Ψk ˆ (r )Ψ0 n + c.invertability of the density response function. in this equation δv is uniquely deﬁned by the δn ∈ δA we started with by δv = χ−1 δn. r) (111) 0 Ek − E 0 Note that this derivative is deﬁned up to a constant.(110). We have therefore a result that is in accordance with our naive expectation of Eq. be equal to the chosen δn we started with. r )δv(r ) (106) where χ0 is the static density response function belonging to the nondegenerate ground state Ψ0 with density n0 . n (r) ∈ A. which for small enough will correspond to a nondegenerate ground state. The functional derivative in Eq. such that δn(r) = d3 r χ0 (r.
then the density is no longer a unique functional of the external potential. This is in particular true for the density operator.(111) the functional derivative of O[n]. The expectation value of any operator (except the total energy) will then depend on which ground state out of the ground state manifold we choose to compute the expectation value from. The HohenbergKohn theorem for degenerate ground states Until now we considered only nondegenerate ground states. r) = −v0 (r) 0 (113) This is exactly the result that we derived earlier using Legendre transforms. r )χ−1 (r . Let us check that Eq. However. If we insert this operator in Eq. of course. Moreover. n approaches the straight path n0 + δn arbitrarily closely for → 0 and this is exactly the limit that we are interested in. we will . These ground states have the simplifying feature that they are determined uniquely (up to a phase factor) by the external potential and therefore any expectation value calculated from the ground state wavefunction is a welldeﬁned functional of the external potential v. We will therefore call the δO/δn of Eq. and therefore diﬀerent ground states Ψi out of the ground state multiplet will yield diﬀerent densities ni .(111) we see that we have to calculate ˆ ˆ Ψ0 T + W Ψk = = ˆ ˆ Ψ0 H0 − V0 Ψk Ek δk0 − d3 rv0 (r) Ψ0 ˆ (r)Ψk n (112) ˆ ˆ ˆ ˆ where H0 = T + V0 + W . degenerate ground states do occur (for instance in open shell atoms) and in that case the external potential does not generate a unique ground state but a linearly independent set of q diﬀerent ground states. also do this derivation for a noninteracting system where ˆ ˆ ˆ W = 0 and therefore O = T . 10. If we want to consider degenerate ground states. the KohnSham potential corresponding to density n0 . in which case we obtain δTs = −vs [n0 ](r) δn(r) (114) where vs [n0 ] is the potential that in a noninteracting system generates density n0 . However.(111) gives us back some known ˆ ˆ ˆ results.we used to approach n0 . We can.(111) we obtain: δFHK [n] =− δn(r) d3 r d3 r v0 (r )χ0 (r . If we insert this matrix element into Eq. i.e. Let us take O = T + W so that O[n] = FHK [n].
To deﬁne this concept we ﬁrst introduce for a qfold degenerate ground state {Ψi . q} the density matrix q q ˆ D= i=1 λi Ψi Ψi  λi = 1 i (0 ≤ λi ≤ 1) (115) where the ground state wavefunctions Ψi are chosen to be orthonormal. The trace is independent of the complete set that we choose.j=1 ˆ Φi AΨj Ψj Φi ∞ = i. we will consider ensemble densities. ˆ We then deﬁne the ground state expectation value of operator O as ˆˆ O = TrDO where the trace operation for an arbitrary operator is deﬁned as ∞ (116) ˆ TrA = i=1 ˆ Φi AΦi (117) where {Φi } is an arbitrary complete set of states. We will ﬁrst generalize this statement somewhat. we have q q ˆn n(r) = TrDˆ (r) = i=1 λi Ψi ˆ (r)Ψi = n λi ni (r) i=1 (120) . every ground state density determines uniquely the external potential that generated it.e. which come from an eigenstate of the Hamiltonian H. i = 1 . as follows simply by insertion of a diﬀerent complete set {Ψi } ∞ ∞ ˆ TrA = i=1 ∞ ˆ Φi AΦi = i.j=1 ˆ Ψj Φi Φi AΨj = j=1 ˆ Ψj AΨj (118) ˆ If we choose the complete set to be the set of eigenstates of Hamiltonian H then we ﬁnd ∞ q ˆˆ TrDO = i=1 ˆˆ Ψi DOΨi = i=1 ˆ λi Ψi OΨi (119) ˆ This deﬁnes the expectation value of an observable O in an ensemble deˆ For the particular case of the density operator scribed by density matrix D. .show that the inverse mapping n → v is welldeﬁned. Instead of purestate ˆ densities. . i.
q1 } and v2 generates the ground state multiplet A2 = {Ψi . with corresponding ensemble densities n1 and n2 . n where Ψ is a ground state. Suppose that D1 and D2 are density matrices belonging to ground state ensembles for potentials v1 and v2 respectively. This follows from the same argument as used in the proof of the HohenbergKohn theorem for the nondegenerate case. be chosen orthonormal. In particular. q} corresponding to a potential v. without loss of generality. . Suppose v1 generates the ground state multiplet A1 = {Φi . . A density will be called a pure state vrepresentable density or for short PSVdensity if it can be written as n(r) = Ψˆ (r)Ψ . as ensemble vrepresentable densities. ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ If H1 = T + V1 + W and H2 = T + V2 + W then ˆ ˆ ˆ ˆ TrD1 H2 > TrD2 H2 This follows directly from q1 q1 (122) ˆ ˆ TrD1 H2 = i=1 ˆ λi Φi H2 Φi > i=1 ˆ λi Ψi H2 Ψi = . as the sets A1 and A2 are only deﬁned to within a unitary transformation no Ψi in A2 is a linear combination of the Φi in A1 . This then implies that two ground state ensemble density matrices constructed from the ground states in A1 and A2 are diﬀerent q1 q2 ˆ D1 = i=1 λi Φi Φi  = i=1 ˆ µi Ψi Ψi  = D2 (121) where λi = µi = 1. which are obtained from an orthonormal set of ground states {Ψi . then n1 = n2 . Let us make this statement a ˆ ˆ bit more speciﬁc. by taking the inner product on both sides with Ψm as the Ψi are not linear combinations of the Φi . Obviously PSVdensities are special cases of EVdensities and the set of PSVdensities is therefore a subset of the set of EVdensities. We further denote the set of all EVdensities generated by potentials in 3 L 2 + L∞ as B. for instance. or for short EVdensities. The proof is analogous to the proof of the nondegenerate case. Then none of the wavefunctions in the sets A1 and A2 are equal. We are now ready to formulate the HohenbergKohn theorem for degenerate ground states: Every EVdensity determines the external potential that generated it up to an arbitrary constant.We will denote densities n(r) of this type. i = 1 . . q2 }. We have thus established that the sets of ground state density matrices for the two diﬀerent potentials v1 and v2 are disjoint. If v1 = v2 + C with C a constant. This follows. . i = 1 . . We now have to prove that the density matrices in these sets yield diﬀerent densities. i = 1 . All the wavefunctions within these multiplets may. .
for instance. where 2p± are porbitals with angular momentum quantum numbers l = ±1. i. As a consequence the ground state expectation value of a given operator may no longer be considered as a functional of the density (take.e. two diﬀerent density matrices can yield the same density. Within the set of ensemble ground state density matrices corresponding to the same potential however. Both wavefunctions obviously have the same total density. the expectation values of the spin and angular momentum in our examples). However. ˆ ˆ Therefore D1 and D2 must give diﬀerent densities. We have E[v1 ] = = > = Likewise we have E[v2 ] > E[v1 ] + n2 (r)(v2 (r) − v1 (r))dr (125) ˆ ˆ ˆ ˆ ˆ ˆ TrD1 H1 = TrD1 (H2 + V1 − V2 ) ˆ ˆ TrD1 H2 + ˆ ˆ TrD2 H2 + E[v2 ] + n1 (r)(v1 (r) − v2 (r))dr n1 (r)(v1 (r) − v2 (r))dr n1 (r)(v1 (r) − v2 (r))dr (124) which added to the last inequality leads to d3 r(n2 (r) − n1 (r))(v2 (r) − v1 (r)) < 0 (126) This leads again to the contradiction 0 < 0 if we assume that n1 = n2 . Both wavefunctions lead to the same density n(r) = 2ϕ1s (r)2 + ϕ2p+ (r)2 = 2ϕ1s (r)2 + ϕ2p− (r)2 (127) We have therefore constructed two degenerate ground states with the same density.q1 q2 = i=1 λi E[v2 ] = E[v2 ] = i=1 ˆ ˆ ˆ µi Ψi H2 Ψi = TrD2 H2 (123) ˆ ˆ We can now show that D1 and D2 yield diﬀerent densities. if two diﬀerent ground state density matrices have . The simplest example is the hydrogen atom with the degenerate ground state wavefunctions for an electron with spin up and spin down. Another example is provided by the the two degenerate ground state wavefunctions (1s)2 2p+ and (1s)2 2p− of the noninteracting lithium atom. Ψ↑ = (1s)α and Ψ↓ = (1s)β where α and β are the spin wavefunctions. We proceed again by reductio ad absurdum. which proves the theorem.
so ˆ ˆ ˆ ˆ ˆ FEHK [n] = TrD[n](T + W ) = Ψ[n]T + W Ψ[n] = FHK [n] (132) We can furthermore prove that FEHK is convex by the same proof as for FHK . Nothing is however known about the convexity of the set of EVdensities B itself which constitute the domain of FEHK . FEHK is convex . Let us collect our results in the form of a theorem Theorem 2 (HohenbergKohn) The EVdensity n speciﬁes the external potential up to a constant. We can now deﬁne an extension of the energy functional Ev to the set of EVdensities Ev [n] = ˆ ˆ n(r)v(r)dr + FEHK [n] = TrD[n]H (129) Similarly as for FHK we easily can prove E[v] = inf n(r)v(r)dr + FEHK [n] (130) n∈B The functional FEHK is an extension of FHK since FEHK [n] = FHK [n] if n ∈ A (131) This follows directly from the fact that for a nondegenerate ground state ˆ Ψ[n] corresponding to n we have D[n] = Ψ[n] Ψ[n]. For every EVdensity n we can therefore unambiguously deﬁne the ensemble version of the FHK functional as [16] ˆ ˆ ˆ FEHK [n] = TrD[n](T + W ) (128) ˆ where D[n] is any of the ground state ensemble density matrices corresponding to n. 2. 1.ˆˆ the same density then also the energy TrDH for those diﬀerent density matrices is the same. The ground state energy of a system with external potential v can be obtained from E[v] = inf n(r)v(r)dr + FEHK [n] n∈B ˆ ˆ ˆ ˆ where FEHK [n] = TrD[n](T + W ) and D[n] is a ground state density ˆ n matrix with TrD[n]ˆ (r) = n(r). Moreover.
We assume that there is no accidental degeneracy. 17]. More precisely. . j ≤ 2L + 1. which by a unitary transformation can be obtained from any other Ψi . and the external potential is invariant under rotations we ﬁnd that ni (r)v(r)dr = nj (r)v(r)dr = n(r)v(r)dr ¯ (134) for all 0 ≤ i. The ground state wavefunctions then transform among one another according to a 2L + 1dimensional unitary representation of the rotation group. 2L + 1} and the corresponding electron densities by ni then the following convex combination 2L+1 1 ni (133) ¯ n= 2L + 1 i=1 is invariant under all rotations and therefore spherically symmetric. Consider an atom with total angular momentum quantum number L > 0 which has a 2L + 1degenerate ground state. Let us now suppose that n is a ground state ¯ density obtained from a pure state wavefunction Ψ[¯ ] . If we denote the ground state wave functions by {Ψ[ni ] = Ψi . . The external potential v is the spherically symmetric Coulomb potential of the atomic nucleus and the degeneracy is a result of rotational invariance of the Hamiltonian of the system. In fact.As we will now demonstrate the subset of PSVdensities of B is not convex. This wavefunction n is not a linear combination of the Ψi otherwise n would not be spherically ¯ symmetric. However the ni are not spherical. We then ﬁnd FEHK [¯ ] n = > i=1 2L+1 Ψ[¯ ]Hv Ψ[¯ ] − n ˆ n 2L+1 n(r)v(r)dr ¯ n(r)v(r)dr ¯ 1 ˆ Ψ[ni ]HΨ[ni ] − 2L + 1 1 FEHK [ni ] 2L + 1 2L+1 = i=1 (135) This then gives FEHK [¯ ] > n i=1 1 FEHK [ni ] 2L + 1 (136) . As any EVdensity is a convex combination of PSV densities this then demonstrates the nonconvexity of the set of PSVdensities. there is no linear combination of the ground states Ψi that leads to a spherically symmetric density. i = 1 . we will now show that there are EVdensities which are not PSVdensities [1. As the nj is obtained from Ψj .
One may. argue that if a potential v0 leads to a degenerate ground state then we can always ﬁnd an arbitrarily small perturbation of the potential δv that lifts the degeneracy and therefore the expectation value ˆ O[v0 + δv] = Ψ[v0 + δv]OΨ[v0 + δv] (137) ˆ of an observable described by an operator O does exist for any > 0. 11. Linear response for degenerate ground states and the diﬀerentiability of FEHK In the previous sections we learned to take functional derivatives on the basis of linear response theory. δv] = lim O[v0 + δv] →0 (138) is deﬁned. however. It is clear that in the case of degeneracy the expectation value of all observables. For the energy the limit in Eq. δv] (139) However. a convex ¯ combination of ground state densities corresponding to the same external potential and therefore. we will see that these derivatives depend in a nonlinear way on δv and are therefore only deﬁned as Gˆteaux variations rather than Gˆteaux a a ˆ derivatives. This poses a clear diﬃculty for the deﬁnition of general density functionals and their functional derivatives. by deﬁnition. Therefore also the limit ¯ O[v0 . Let us investigate a . except the energy. This leads to a contradiction and hence we must conclude that n can not be a ¯ pure state density of any potential. however. an ensemble vrepresentable density. We shall especially be interested in the case where O is the density operator. The density n is. where Ψ[v0 + δv] is the ground state wavefunction of the perturbed system. For an explicit numerical example of such a density we refer to the work of Aryasetiawan and Stott [18]. One can therefore deﬁne the functional derivative as follows O [δv] = lim →0 ¯ O[v0 + δv] − O[v0 .(138) is independent of δv and we may wonder if a Gˆteaux derivative still exists. We concentrated on nondegenerate ground states for which the response functions are welldeﬁned. We therefore have constructed an EVdensity which is not a PSVdensity.But we also know that FEHK is convex on the set of EVdensities. although it will in general depend on the potential variation δv. We will now use response theory for degenerate ground states to study the diﬀerentiability of the functional FEHK . depends on which particular ground state we choose to calculate the expectation value from.
∞ Ψk (0) = i=1 ck Ψi i (144) where Ψi = Ψi (0) for i = 1 . . where θk ( ) is an arbitrary function of . If we insert this expansion into the ﬁrst order Schr¨dinger equation (143) we obtain the equation o ∞ i=1 ˆ ck (Ei − E0 )Ψi = (Ek (0) − δ V )Ψk (0) i (145) . This freedom will. Suppose we have an Hamiltonian H0 with external potential v0 which has a qfold degenerate ground state. It turns out that this term inﬂuences the ﬁrst order density response. Then there are q eigenstates {Ψk ( ) . q and the states {Ψi . Which particular ground states of the ground state manifold are reached by the → 0 limit depends obviously on the form of the perturbation δv. If we expand the Schr¨dinger equation (142) to ﬁrst order in we obtain o ˆ ˆ (H0 − E0 )Ψk (0) = (Ek (0) − δ V )Ψk (0) (143) where Ek (0) and Ψk (0) are the ﬁrst order derivatives of Ek ( ) and Ψk ( ) with respect to in = 0. The ground state wavefunctions Ψk ( ) satisfy the equation ˆ ˆ ( H0 + δ V + 2 2 ˆ δ U )Ψk ( ) = Ek ( )Ψk ( ) (142) ˆ where. lim Ψk ( ) = Ψk (0) →0 (140) with ˆ Ek (0) = Ψk (0)H0 Ψk (0) = E0 (141) Note that since the states Ψk ( ) are orthonormal. k = 1 . . i. We note. Let us now apply a perturbation δv which lifts the degeneracy. i. as in the nondegenerate case. for the later discussion.e. .e. q} with energies Ek ( ) of the perturbed Hamiltonian that are continuously connected to degenerate ground states Ψk (0) of the ˆ unperturbed Hamiltonian H0 . not aﬀect any of the expectation values. In order to solve this equation for Ψk (0) we expand this quantity in the q connected ground states and all the other ˆ eigenstates of H0 . . i > q} are eigenˆ states of H0 with eigenenergies Ei > E0 .ˆ this in more detail. like in the nondegenerate case. also the limiting states states Ψk (0) in the degenerate ground state multiplet are orthonormal. that if Ψk ( ) is a solution to this equation then also Φk ( ) = eiθk ( ) Ψk ( ) is a solution. we also added a onebody potential δ U which is of second order in . In which way it depends on δv we will investigate now.
Let us investigate the consequences for diﬀerent expectation values. i We know from the orthonormality of the states Ψk ( ) that 0= d Ψi Ψk (0) = Ψi (0)Ψk (0) + Ψi (0)Ψk (0) = ci∗ + ck k i d (150) i. This yields ˆ ˆ ˆ (H0 − E0 )Ψk (0) = 2(Ek (0) − δ V )Ψk (0) + (Ek (0) − δ U )Ψk (0) (151) . We therefore ﬁrst have to determine the ﬁrst order wavefunction. This leaves us with the determination of the oﬀdiagonal k terms ck for i = k. These coeﬃcients can be found by expanding Eq.If we multiply this equation from the left with Ψj  where 1 ≤ j ≤ q we obtain ˆ (146) 0 = Ek (0)δjk − Ψj (0)δ V Ψk (0) This is the equation tells us exactly how δv picks out the ground states in ˆ the degenerate multiplet. The value of these diagonal k elements is undetermined by the Schr¨dinger equation as they are related o to the arbitrary phase θk ( ) which we can choose for each Ψk ( ) .(143). the coeﬃcients ck for 1 ≤ i ≤ q are not determined by Eq. They are exactly the ones that diagonalize δ V within the qdimensional space of degenerate ground state functions.(145) from the right with Ψi  where i > q we obtain the coeﬃcient ck for i > q and the following expression for i Ψk (0) : q ∞ ˆ Ψi Ψi δ V Ψk (0) ck Ψi − Ψk (0) = (149) i Ei − E 0 i=q+1 i=1 However. So even the energy functional has no functional a derivative in the proper sense. this is not a proper Gˆteaux derivative but a only a Gˆteaux variation.(142) i to second order in 2 . the coeﬃcients ck form an antiHermitian matrix.e. This equation also tells us that ˆ Ek (0) = Ψk (0)δ V Ψk (0) = d3 rnk (r)δv(r) (147) If the state perturbed state with the lowest energy has label k = 1 then we see that the functional derivative of the ground state energy is given as δE = n1 (r) δv(r) (148) However. If we multiply Eq. since n1 implicitely depends on the perturbation δv which picked out a particular ground state. As in the nondegenerate case the value of any expectation value does not depend on the coeﬃcients ck . In particular we ﬁnd i that the coeﬃcients ck are purely imaginary.
Ei − E 0 (154) If. (156) If we insert our expression for the ﬁrst order change in the wavefunction we obtain O (0) = + d3 rζk (r)δv(r) + 1 2 d3 rηk (r)δu(r) (157) d3 rd3 r ξk (r. r ) = i=q+1 Ψk (0)ˆ (r)Ψi Ψi ˆ (r )Ψk (0) n n + c.where Ψk (0) and Ek (0) are the second order derivatives of Ψk ( ) and Ek ( ) in = 0.(149) then we obtain 0 = ˆ 2(Ek (0) − El (0))ck + Ek (0)δlk − Ψl (0)δ U Ψk (0) l ∞ ˆ ˆ Ψl (0)δ V Ψi Ψi δ V Ψk (0) + 2 Ei − E 0 i=q+1 (152) If we in this equation take k = l we obtain an expression for the energy to second order in : Ek (0) = 2 d3 rd3 r δv(r)χk (r.c. we take l = k we obtain our desired equation for the coeﬃcients ck : i ck l = + ˆ 1 Ψl (0)δ U Ψk (0) 2 Ek (0) − El (0) ∞ ˆ ˆ 1 Ψl (0)δ V Ψi Ψi δ V Ψk (0) Ek (0) − El (0) i=q+1 Ei − E 0 (155) We see that indeed ck∗ = −cl as expected.(146) and Eq. We have therefore completely l k determined the ﬁrst order change in the wavefunction. If we multiply this equation from the left with Ψl (0) where 1 ≤ l ≤ q and use Eq. Let us see what this implies for the observables.c. r )δv(r)δv(r ) . on the other hand. The ﬁrst order change in the expectation value ˆ of an observable O is given by ˆ O (0) = Ψk (0)OΨk (0) + c. r )δv(r ) + d3 rnk (r)δu(r) (153) where nk is the density corresponding to Ψk (0) and χk the following function ∞ χk (r.
Ei − E 0 ˆ Ψk (0)OΨl Ψl ˆ (r)Ψk (0) n + c.c. r )δv(r)δv(r ) ˆ where the functions χk . If we denote the perturbed state with label k = 1 as the one with the lowest energy then δn1 represents the change in ground state density. One can show from the HohenbergKohn theorem for degenerate states that this is indeed the case. Let us now take δu = 0 and write δn1 (r ) = + d3 rχ1 (r .c. If we in Eq. r)δu(r) (161) d3 rd3 r ξk (r . r)δv(r) d3 rd3 r ξ1 (r . Ek (0) − El (0) (158) = l=1(l=k) q ∞ (159) ξk (r. ηk and ξk are obtained by inserting n(r ) for O in ˆ Eqns. We see that this density change does not depend linearly on the ﬁrst order change δv in the potential and even the second order change δu of the potential contributes to its value.(158). r )δv(r)δv(r ) (162) We can ask ourselves the question whether a given ﬁrst order variation δn1 uniquely determines the ﬁrst order density change δv. r.(126) take v2 = v1 + δv(r) where δv is not a constant function we obtain 2 d3 rδn1 (r)δv(r) + O( 3 ) < 0 → 0 then yields (163) Dividing by 2 and taking the limit d3 rδn1 (r)δv(r) < 0 (164) .where we deﬁned the following functions ∞ ζk (r) ηk (r) = i=q+1 q ˆ Ψk (0)OΨi Ψi ˆ (r)Ψk (0) n + c. (159) and (160). r ) = l=1 ˆ Ψk (0)OΨl Ψl ˆ (r)Ψi Ψi ˆ (r )Ψk (0) n n (160) (Ek (0) − El (0))(Ei − E0 ) i=q+1 ˆ If we choose O = n(r ) then we obtain an expression for the ﬁrst order ˆ change in the density δnk (r ) = + d3 rχk (r . r)δv(r) + 1 2 d3 rηk (r . r.
(162). From response theory we known that we can write n as n (r) = n1 (r) + δn1 (r) + m (r) where n1 (r) 0 = = lim n[v0 + δv] →0 (166) (167) (168) lim m (r) →0 and where δn1 is explicitly given in Eq. We will do the same for the degenerate case. r)δv(r) . We consider again a system with a degenerate ground state and external potential v0 . However.(164) does not imply that the function χ1 is invertable. Let us take out of the set of ground state ensemble densities for this system a particular density n0 . We can now consider the limit δFEHK [δn1 ] δn = = lim →0 FEHK [n1 + δn1 + m ] − FEHK [n1 ] ¯ FEHK [v0 + δv] − FEHK [v0 .(138). Now by construction the density n is in B for all values of . δv] (169) lim →0 ¯ where FEHK [v0 .(162) is invertable where the inverse only exists on the set of vrepresentable density variations within the set B. but we will see that some diﬃculties remain. Suppose we look at the perturbed system with potential v0 + δv in which the degeneracy is lifted. in the nondegenerate case we could solve this diﬃculty using response theory.(158) and (160). In contrast to the nondegenerate case Eq. For this system there is a welldeﬁned ground state density n = n[v0 + δv]. For this limit to be welldeﬁned we have to make sure that n0 + δn ∈ B and we run into the same problem as for the nondegenerate case. only the relation in Eq. We obtain for the ﬁrst order change of FEHK δFEHK = − d3 rd3 r v0 (r )χ1 (r . δv] is deﬁned as in Eq. We may then wonder whether the following limit FEHK [n0 + δn] − FEHK [n0 ] δFEHK [δn] = lim →0 δn (165) exists.We see that the ﬁrst order density change δn1 can not be zero if δv is not a constant. Let us now see what the linear response theory can tell us about the diﬀerentiability of FEHK . The latter limit is readily ˆ ˆ ˆ ˆ ˆ calculated by inserting T + W = H0 − V0 for the operator O in Eqns.
We saw that there exist EVdensities that are not PSVdensities. If we consider an ensemble corresponding to such a density and change the external potential to v0 + δv where δv lifts the degeneracy then for > 0 the ensemble will change into a pure state and the density will change abruptly. The derivative exists for those ground state densities that correspond to pure states that can be obtained in the → 0 limit for a perturbed system with potential v0 + δv. are diﬃcult to characterize. we see that this derivative only exists for some special set of ground state densities corresponding to potential v 0 . r )δv(r)δv(r ) − d3 rv0 (r)δn1 (r) (170) FEHK [n1 + δn1 + m ] − FEHK [n1 ] =− d3 rv0 (r)δn1 (r) (171) We see that this limit is linear in δn1 and that furthermore v0 is independent of δn1 . We must therefore conclude that for general EVdensities the functional derivative of FEHK does not exist. It is not clear how to take the functional derivative at an arbitrary ensemble density n0 for potential v0 . i. This functional is the Lieb functional FL and will be studied in the next section. It is therefore desirable to extend the domains of deﬁnition of FHK and FEHK to an easily characterizable (preferably . The Levy and Lieb functionals FLL [n] and FL [n] The functionals FHK and FEHK have the unfortunate mathematical difﬁculty that their domains of deﬁnition A and B. although they are welldeﬁned. As we will discuss later. r. Therefore we can write for the functional derivative of FEHK in n1 : δFEHK (n1 ) = −v0 (r) δn(r) (172) up to an arbitrary constant.e. Fortunately it turns out that one can deﬁne an extension of the functional FEHK to a larger domain of densities which can be shown to be diﬀerentiable at the set of all EVdensities.− = and we thus obtain lim →0 d3 rd3 r d3 r v0 (r )ξ1 (r . However. it is diﬃcult to know if a given density n belongs to A or B. it is known from numerical investigations that there are ground state densities of interacting systems that are not pure state densities for a noninteracting system and therefore there is a clear need for establishing a KohnSham scheme for arbitrary EVdensities. This poses not only a theoretical problem but also a practical one. 12.
Theorem 4 The functional FLL is not convex . i. As this is the ﬁrst important result that we obtain for FLL we put it in the form of a theorem Theorem 3 For any n ∈ S there is a Ψ[n] ∈ H 1 (R3N ) such that ˆ ˆ FLL [n] = Ψ[n]T + W Ψ[n] Let us discuss some properties of FLL . Suppose n is some ground state density corresponding to some external potential v and ground state Ψ[n] then ˆ ˆ n(r)v(r)dr + FHK [n] = Ψ[n]HΨ[n] = inf ΨHΨ = Ψ→n ˆ ˆ n(r)v(r)dr + inf ΨT + W Ψ = Ψ→n n(r)v(r)dr + FLL [n] (175) We deﬁne a corresponding energy functional Ev [n] = n(r)v(r)dr + FLL [n] (176) If n0 is the ground state density for potential v with corresponding ground state wave function Ψ[n0 ] then ˆ ˆ Ev [n] = inf ΨHΨ ≥ Ψ[n0 ]HΨ[n0 ] = Ev [n0 ] Ψ→n (177) Minimizing Ev over the set S therefore yields the ground state density n0 corresponding to external potential v. which is deﬁned on A. i. The functional FLL is an extension of the HohenbergKohn functional FHK .e FLL [n] = FHK [n] if n ∈ A (174) This is readily derived. that the inﬁmum is always a minimum. As shown earlier such a density is always in the convex set S which is again a subspace of L1 ∩ L3 .e. One can furthermore show. as has been done by Simon [1].convex) set of densities. The functional FLL has however the inconvenient property that it is not convex. to the larger set S. there is always a minimizing wave function. We deﬁne the LevyLieb functional FLL as ˆ ˆ FLL [n] = inf ΨT + W Ψ (173) Ψ→n where the inﬁmum is searched over all normalized antisymmetric N particle wave functions in H 1 (R3N ) yielding density n. This can be achieved using the constrained search procedure introduced by Levy [19].
We put the result again in the form of a theorem ˆ Theorem 5 For every n ∈ S there is a density matrix D[n] such that ˆ ˆ ˆ FL [n] = TrD[n](T + W ) For the proof of this theorem we again refer to Lieb [1]. that is FL [n] = FEHK [n] if n ∈ B (182) This follows directly from the fact that if n ∈ B then there is a potential v ˆ which generates a ground state ensemble density matrix D[n] which yields . i. The functional FL is an extension of FEHK to the larger set S.To show this we take the example of a previous section where we presented a density n which did not correspond to a ground state wavefunction. We will therefore now deﬁne a diﬀerent but related convex functional with the same domain S. Then we ﬁnd ˆ n(r)v(r)dr + FLL [¯ ] = inf ΨHΨ ¯ n Ψ→¯ n 1 > 2L + 1 and we ﬁnd 2L+1 ˆ Ψ[ni ]HΨ[ni ] = i=1 1 2L + 1 2L+1 FLL [ni ] + i=1 n(r)v(r)dr ¯ (178) FLL [¯ ] > n 1 2L + 1 N FLL [ni ] i=1 (179) which proves the nonconvexity of FLL . This is the Lieb functional F L deﬁned as ˆ ˆ ˆ FL [n] = inf TrD(T + W ) (180) ˆ D→n where the inﬁmum is searched over all N particle density matrices ˆ D= i=1 λi Ψi Ψi  λi = 1 i=1 Ψi ∈ H 1 (R3N ) (181) ˆn which yield the given density n(r) = TrDˆ (r) where {Ψi } is an orthonormal set.e. It ¯ was a convex combination of 2L + 1 degenerate ground state densities n i with corresponding ground states Ψ[ni ] for an external potential v. This is somewhat unfortunate as convexity is an important property which can be used to derive diﬀerentiability of functionals. Also for this case one can prove the inﬁmum to be a minimum. there is a minimizing density matrix.
We take n to be an EVdensity which is not a PSV density.e ˆ ˆ Ψi HΨi < inf ΨHΨ = Ψ→n n(r)v(r)dr + FLL [n] (188) which proves our statement. Any wave function yielding density n can not be a linear combination of these ground state wavefunctions otherwise n would be pure state vrepresentable. We will now demonstrate another important property of FL . There is a ground state ˆ ensemble density matrix D[n] for which we have n(r)v(r)dr+FL [n] = ˆ ˆ ˆ ˆ n(r)v(r)dr+TrD[n](T + W ) = Ψi HΨi (187) where Ψi is any of the ground states in the degenerate ground state multiplet.n. i. Theorem 6 The functional FL is convex . So n(r)v(r)dr + FEHK [n] = = = ˆ ˆ ˆ n(r)v(r)dr + TrD[n](T + W ) ˆ ˆ ˆˆ TrD[n]H = inf TrDH ˆ D→n n(r)v(r)dr + FL [n] (183) We can again deﬁne an energy functional Ev [n] = n(r)v(r)dr + FL [n] (184) which by a similar proof as for FLL assumes its minimum at the ground state density corresponding to potential v. The second relation also easily follows. We further have the following relations (185) FL [n] = FLL [n] if n ∈ A and FL [n] < FLL [n] if n ∈ B and n ∈ A (186) The ﬁrst relation follows from the fact that if the density n is a pure state vrepresentable density then the minimizing density matrix for FL is a pure state density matrix. Therefore its expectation value with the Hamiltonian must be larger. which is its convexity.
This is important information which enables us to derive the Gˆteaux diﬀerentiability of the functional FL at the set B of ensemble a vrepresentable densities. Suppose ˆ n is an EVdensity corresponding to potential v of Hamiltonian Hv . suppose that a series of densities nk approaches a given density n in some sense. If n = λ1 n1 +λ2 n2 with λ1 +λ2 = 1 and 0 ≤ λ1 . λ2 ≤ 1 then we have λ1 FL [n1 ] + λ2 FL [n2 ] = + = ≥ λ1 inf λ2 inf ˆ D1 →n1 ˆ ˆ ˆ TrD1 (T + W ) ˆ ˆ ˆ TrD2 (T + W ) ˆ ˆ ˆ ˆ Tr(λ1 D1 + λ2 D2 )(T + W ) (189) ˆ D2 →n2 ˆ ˆ D1 .This is easily shown.n2 ˆ D→n inf ˆ ˆ ˆ inf TrD(T + W ) = FL [n] We therefore now have established that FL is a convex functional on a convex space. where C = u 3 2 FL [n] ≤ FL [nk ] + (193) . An obvious question to ask is whether this functional is continuous. one can prove a weaker statement. ? It turns out that this question is not easily answered. To be more precise.D2 →n1 . Then FL [n] + d3 rn(r)v(r) = = and therefore FL [n] ≤ FL [nk ] + d3 rv(r)(nk (r) − n(r)) (191) ˆ ˆ ˆ ˆ TrD[n]Hv ≤ TrD[nk ]Hv FL [nk ] + d3 rnk (r)v(r) (190) Further. We will discuss this feature of FL in the next section. for instance n − nk 1 → 0 and n − nk 3 → 0 for k → ∞ in L1 ∩ L3 . Having obtained some desirable convexity properties of FL we try to obtain some analytic properties of this functional. Now we 3 3 can split up v ∈ L 2 + L∞ as v = u + w where u ∈ L 2 and w ∈ L∞ and we have  d3 rv(r)(nk (r) − n(r)) ≤ u 1 3 2 n − nk 3 3 + w ∞ n − nk 1 (192) + w ∞ So if n − nk ] then ≤ /C and n − nk ] ≤ /C. However. Does this imply that FL [nk ] − FL [n] → 0. since nk → n in the norms on L1 ∩ L3 for each > 0 there is an integer M such that n − nk 1 ≤ and n − nk 3 ≤ for k > M .
We will need this property in the next section.(193) is called lower semicontinuous. r) ∈ B × RG[n] ≤ r} (195) is closed and convex [11]. The notion of lower semicontinuity is an important property for convex functionals which will allow us to make several other useful statements about other properties of FL . The closedness property means that if a series of points (nk . However. .e. This. i. We will. known as semicontinuity. rk ) in the epigraph of G converges to a point (n.for k suﬃciently large. Since the result is important we present it here in the form of a theorem Theorem 7 Suppose nk and n ∈ S and nk → n for k → ∞ in the norms on L1 ∩ L3 . not need that property in the remainder of this review. the set of points that lie strictly above G. the proof of this is not simple and we will therefore not try to reproduce it here. It turns out that one can prove that FL is lower semicontinuous on all densities in S (see theorem 4. This can also be characterized diﬀerently. Because the limit point is a lower bound.4 of Lieb [1]). however. is that for a lower semicontinuous convex functional G : B → R the following set of points epi(G) = {(n. of course. r) then this point also lies in the epigraph. is an open convex set (this means that for every point in this set there is a neighborhood that contains it). What we have proven is a weaker form of continuity. This also implies that the set of interior points of epi(G). One particular consequence of lower semicontinuity that we will use. This set is called the epigraph of G and is simply the set of points that lie above the graph of G. If we deﬁne inf F [nm ] by inf FL [nm ] = inf{FL [nk ]k ≥ m} then lower semicontinuity implies FL [n] ≤ lim inf FL [nm ] m→∞ (194) Since we required that n is an EVdensity we have proven that FL is lower semicontinuous on the set of EVdensities. We therefore see that if we take → 0 from above then FL [nk ] approaches FL [n] from above and nk → n in the norms on L1 ∩ L3 . does not imply that FL [n] is continuous in n. For that we would have to prove that FL [n] − FL [nk ] ≤ rather than FL [n] − FL [nk ] ≤ . the functional with the property in Eq. One can prove an even stronger theorem in which we only need weak convergence of the series nk . Then FL [n] ≤ lim inf FL [nk ] k→∞ In other words FL is lower semicontinuous on the set S.
then there is a unique tangent line that touches the graph of G in a point (n0 . The discussion is simpliﬁed by the fact that we are dealing with convex functionals. To prove existence of the derivative we use the geometric idea that if a derivative of a functional G[n] in a point n0 exists. Diﬀerentiability of FL In this section we will prove that the Lieb functional is diﬀerentiable on the set of EVdensities and nowhere else. This is the basis of the following deﬁnition.13. The geometric idea behind this deﬁnition is illustrated in Figure 1. The functional derivative at a given EVdensity is equal to −v where v is the external potential that generates the EVdensity at which we take the derivative. G[n0 ]). If G : B → R is a diﬀerentiable and convex functional from a normed linear space B to the real numbers then from the convexity property it follows that for n0 . If for a convex functional G there is a continuous linear functional L : B → R such that G[n1 ] ≥ G[n0 ] + L[n1 − n0 ] (199) then L is called a subgradient or tangent functional in n0 . . To discuss this in more detail we have to deﬁne what we mean with a tangent. n1 ∈ B and 0 ≤ λ ≤ 1 that G[n0 + λ(n1 − n0 )] − G[n0 ] ≤ λ(G[n1 ] − G[n0 ]) From the fact that G is diﬀerentiable we then ﬁnd δG G[n0 + λ(n1 − n0 )] − G[n0 ] = [n1 − n0 ] λ δn (196) G[n1 ] − G[n0 ] ≥ lim Therefore λ→0 (197) G[n1 ] ≥ G[n0 ] + δG [n1 − n0 ] δn (198) Now δG/δn is a continuous linear functional which will be identiﬁed with a tangent.
One can further show that if an inﬁmum exists the the minimizing density matrix satisﬁes the Schr¨dinger o equation [1]. Moreover the tangent functional at an EV density n can be identiﬁed with −v where v is the potential that generates this density.(130) to all potentials in L 2 + L∞ . This is an extension of the previously deﬁned functional in 3 Eq. Let us ﬁrst suppose that n0 is an EVdensity corresponding . In order to show this we ﬁrst deﬁne the energy functional ˆˆ E[v] = inf TrDHv ˆ D (200) where the inﬁmum is searched over all density matrices of the form given in Eq.(181).F[n] F[n0 ]+L[nn0 ] n0 n Figure 1: The convex functional F has a unique tangent functional L in the point n0 For the Lieb functional FL we will now prove the following statement: Theorem 8 The functional FL has a unique tangent functional for every EVdensity and nowhere else.
Then FL [n0 ] + d3 rn0 (r)v(r) = = and therefore FL [n] ≥ FL [n0 ] − Therefore the functional L L[n] = − d3 rv(r)n(r) (203) d3 rv(r)(n(r) − n0 (r)) (202) ˆ ˆ ˆ ˆ TrD[n0 ]Hv ≤ TrD[n]Hv FL [n] + d3 rn(r)v(r) (201) deﬁnes a linear and continuous functional (continuity follows from Eq. Since F is deﬁned on B = L1 ∩ L3 the tangent functional must represent an element in the dual space 3 B ∗ = L 2 + L∞ and can be written as L [n] = − 3 d3 r˜(r)n(r) v (204) with v ∈ L 2 + L∞ and v = v + C.(206): E[˜] = inf {FL [n] + v n∈S ˆ ˆ˜ d3 rn(r)˜(r)} ≥ TrD[n0 ]Hv v (207) Now there are two cases. Since we know that there is ground state density matrix corresponding to n0 we obtain from Eq. Now since we assumed that L was also ˜ ˜ tangent functional we have FL [n] ≥ FL [n0 ] − and hence FL [n] + d3 rn(r)˜(r) ≥ FL [n0 ] + v d3 rn0 (r)˜(r) v (206) d3 r˜(r)(n(r) − n0 (r)) v (205) We take the inﬁmum over all densities in S on the left hand side.ˆ to potential v of Hamiltonian Hv . Suppose L is a diﬀerent tangent functional. Now we have to show its uniqueness.(192) ) and is therefore a tangent functional. either the Hamiltonian with potential v is able to ˜ support a bound ground state or it is not. If it does support a ground state then ˆ ˆ˜ TrD[n0 ]Hv > E[˜] v (208) .
(205) ˜ and therefore also Eq. which we call D[n0 ]. there is a unique tangent functional at every EVdensity and nowhere else. We then again obtain Eq. Since n0 is not an EVdensity ˆ D[n0 ] can not be a ground state density matrix for the Hamiltonian with potential v . Together with ˜ Eq.(200) is attained. Let us further suppose that there is a tangent functional at n0 . Consequently. The constrained search for FL [n0 ] ˆ always has a minimizing density matrix.(207) is true. Therefore there v v is no tangent functional at n0 if n0 is not an EVdensity. With this density matrix then also Eq. As a next step we will show that this implies that the functional FL Gˆteaux a diﬀerentiable at every EVdensity and nowhere else.e. We therefore conclude that the EVdensity n0 has a v v unique tangent functional equal to −v where v is the potential that yields ground state density n0 . With the same arguments as before we ﬁnd that Eq. that for some v Eq. If the v v Hamiltonian with potential v does not support a bound ground state then ˜ there is no normalized density matrix for which the inﬁmum of E[˜] as in v Eq.ˆ since D[n0 ] is not a ground state density matrix for v . We are now ready to introduce the following key theorem which we will use to prove diﬀerentiability of FL at the set of EVdensities: Theorem 9 Suppose F : B → R is a lower semicontinous convex functional which is ﬁnite at a convex subset V ⊂ B of a normed linear space B.(208) must ˜ be true which again leads to the contradiction E[˜] > E[˜]. If F has a unique continuous tangent functional L : B → R at n0 ∈ V then F is Gˆteaux diﬀerentiable at n0 and L = δF/δn.(207) this immediately leads to the contradiction E[˜] > E[˜]. With this deﬁnition the functional FL is a convex lower semicontinuous functional of the whole space L1 ∩ L3 . However. i. there is no unique tangent functional . In Figure 2 we display an example of a functional (in this picture the functional is simply a function) which is not diﬀerentiable at n0 . We have therefore proven our statement. inﬁnite values are welldeﬁned in the theory of convex functionals [11] and they are usually introduced to deal in a simple way with domain questions. For the application of the next theorem it is desirable to extend the domain of FL to all of L1 ∩ L3 .(208) and the same contradiction E[˜] > E[˜].(206) are satisﬁed. Let us now suppose that n0 is not an EVdensity. We follow Lieb [1] and deﬁne FL [n] = ˆ ˆ ˆ inf D→n TrD(T + W ) if n ∈ S ˆ +∞ if n ∈ L1 ∩ L3 and n ∈ S (209) The reader may seem surprised with the appearance of +∞ in this definition. a The fact that uniqueness of a tangent functional implies diﬀerentiability is clear from a geometric point of view.
however. Since V is a convex set we have that if n0 ∈ V and n1 ∈ V then n0 + λ(n1 − n0 ) ∈ V for 0 ≤ λ ≤ 1 and since F is convex we obtain F [n0 + λ(n1 − n0 )] ≤ F [n0 ] + λ(F [n1 ] − F [n0 ]) (210) Furthermore since F has a unique continuous tangent functional at n0 we have F [n0 + λ(n1 − n0 )] ≥ F [n0 ] + λL[n1 − n0 ] (211) .at that point. Since the proof of the previous theorem is important for our discussion and not diﬃcult to follow we will present it here. F[n] L1 L2 n0 n Figure 2: The convex functional F has no unique tangent functional in the point n0 . In the example there are in fact inﬁnitely many tangent functionals at n0 of which there are two drawn in the plot. be careful in drawing conclusions from simple plots and we have to resort to formal proofs. Both L1 and L2 are tangent functionals In the case of inﬁnitely many dimensions one should.
n0 + (n2 − n0 )] = lim F [n0 + (n1 − n0 )] ≥ F [n0 ] + F [n0 . F [n0 ] + F [n0 .( 213) we ﬁnd F [n0 . the function g(λ)/λ has a ﬁnite upper and lower bound and therefore the limit λ → 0 of g(λ)/λ exists. 21]. 1] and hence continuous. r) ∈ B × RL [n] + αr = β} (218) where L : B → R is a continuous linear functional and α and β are real numbers (We can take α = 0 since F [n0 . n1 ] = lim F [n0 + λ (n2 − n0 )] − F [n0 ] = F [n0 . n1 ] is ﬁnite and we will have no vertical line L or vertical hyperplane H). 1]} (217) (216) lies below the graph of functional F .Note that n1 − n0 need not be in V but is always in B since that is a linear space. We therefore ﬁnd F [n1 ] − F [n0 ] ≥ F [n0 . n1 ] is continuous and linear and equal to L. n1 ] ≥ L[n1 − n0 ] (213) where F [n0 + λ(n1 − n0 )] − F [n0 ] (214) λ→0 λ It remains to show that F [n0 . n2 ] λ→0 λ (215) Therefore from Eq. We will just use its consequences. Moreover. tells us that if we have an open convex set A (the interior of epi(F ) in our case) and an aﬃne subspace L (which is the line L in our case) that does not intersect A then there is a hyperplane H that contains L in which the hyperplane H has the form H = {(n. We will not prove the HahnBanach theorem as it is geometrically intuitive and can be found in most textbooks on functional analysis [20. Now since F is convex and lower semicontinuous the set of points above the graph of F (the interior points of epi(F )) form an open convex set which is nonempty since F is ﬁnite on V [11]. n1 ] This means that the straight line L = {(n0 + (n1 − n0 ). . Now a famous theorem of functional analysis. the HahnBanach theorem. Now F has the property that for n2 ∈ V F [n0 . as the equation above shows. n1 ]) ∈ B × R ∈ [0. If we combine the two inequalities we ﬁnd that F [n1 ] − F [n0 ] ≥ F [n0 + λ(n1 − n0 )] − F [n0 ] ≥ L[n1 − n0 ] λ (212) Now since F is convex the function g(λ) = F [n0 + λ(n1 − n0 )] − F [n0 ] is a convex real function on the interval [0.
n1 ]) ∈ H we ﬁnd that (220) L [n1 − n0 ] + αF [n0 . B = L1 ∩ L3 . F [n0 ] + F [n0 . The question therefore is: which densities are ensemble vrepresentable? In this section we will prove a useful result which will enable us to put the KohnSham approach on a rigorous basis. If we take F = FL . Ensemble vrepresentability We have seen that the Lieb functional FL is diﬀerentiable at the set of EVdensities in S and nowhere else. Now we can apply this theorem to the Lieb functional.(213) we then see immediately that −L /α is a continuous tangent functional at n0 which then must be equal to L since that was the only continuous tangent functional at n0 . n1 ] = L[n1 − n0 ] and thus F [n0 + λ(n1 − n0 )] − F [n0 ] δF [n1 − n0 ] = lim = L[n1 − n0 ] λ→0 δn λ (222) (221) We therefore obtained what we wanted to prove. 14. Theorem 11 The set of EVdensities is dense in the set S with respect to the norm on L1 ∩ L3 . Consequently F [n0 . This leaves us with the question which densities in set S actually are EVdensities. V = S and use that FL has a unique tangent functional L = −v at every EVdensity and nowhere else. The coeﬃcient β is then determined by the fact that (n0 . A useful result on this point is obtained in the next section. r) ∈ B × RL [n − n0 ] + α(r − F [n0 ]) = 0} (219) Now from the fact that (n0 + (n1 − n0 ). n1 ] = 0 From this equation and Eq. Moreover the functional derivative at an EVdensity is equal to −v where v is the potential that generates this density.Now we know that our line L is contained in H. . For this reason it is desirable to know a bit more about these densities. F is Gˆteaux diﬀerena tiable at n0 and δF/δn = L. F [n0 ]) ∈ H and we ﬁnd β = L [n0 ]+αF [n0 ] and therefore H = {(n. we obtain: Theorem 10 The functional FL is Gˆteaux diﬀerentiable for every EVa density in the set S and nowhere else.
Let F be a functional F : B → R from a normed function space (a Banach space) B to the real numbers. . In other words.This statement means the following. the set of continuous linear functionals on B. then for every > 0 we can ﬁnd an EVdensity n such that n0 − n 1 ≤ and n0 − n 3 ≤ . From the relation E[v] = inf {FL [n] + n∈S d3 rn(r)v(r)} 3 (224) we obtain immediately that for any n ∈ S and any v ∈ L 2 + L∞ FL [n] ≥ − d3 rn(r)v(r) + E[v] (225) Now the functional on the right hand side of the inequality sign is. for a given v. For every density n0 in the set S there is a series of EVdensities nk such that lim n0 − nk p k→∞ =0 (223) where the subindex for the norm takes the values p = 1 and p = 3.(225) that every v ∈ L 2 + L∞ deﬁnes an FL bounded functional where L is deﬁned as in Eq. In order to establish this result we need to use a theorem due to Bishop and Phelps. A linear functional with this property is called FL bounded.(203). This can also be phrased diﬀerently. For the clarity of the discussion we split the theorem into two parts which both yield interesting results for our functional FL . The inequality sign tells us that this functional lies below the graph of FL [n]. Let us give a general deﬁnition of these linear functionals. Then L ∈ B ∗ is said to be F bounded if there is a constant C such that for all n∈B F [n] ≥ L[n] + C (226) where the constant C may depend on L but not on n. The ﬁrst part gives some insight in the set of potentials V that generate a bound ground state.e. We therefore see 3 from Eq. Let B ∗ be the dual space of B. for every density in the set S there is an EVdensity arbitrarily close to it. The geometric picture of an F bounded functional is displayed in Figure 3. i. Suppose we take an arbitrary density n0 from the set S. a linear functional of n.
Consider the function f (x) = exp (x) on the real axis. Therefore the set of tangent functionals is a subset of the set of FL bounded functionals. . This function is convex and all tangent functions are of the form gα (x) = αx + β where α > 0 and β(α) ≤ 1. This can be illustrated with an example. arbitrary close to the slope of the constant f bounded function g0 .e. but not a tangent. The functional F can take the value +∞ but not everywhere. i. However. Then the continuous tangent functionals to F are B ∗ norm dense in the set of F bounded functionals in B ∗ . The following theorem (due to Bishop and Phelps) ensures that a similar situation occurs for the case of general Banach spaces: Theorem 12 (BishopPhelps I) Let F : B → R be a lower semicontinuous convex functional on a real Banach space B.F[n] L1 L2 n Figure 3: Both L1 and L2 are F bounded by the convex functional F Note that the tangent functionals of FL are special cases of FL bounded functionals. The constant function g0 (x) = β with β ≤ 0 is an f bounded function. we can always ﬁnd a tangent with a slope arbitrary close to zero. they are the FL bounded functionals that also touch the graph of FL .
Let us apply this theorem to our functional FL . The ﬁrst two points of this theorem are equivalent to the previous theorem. We know that every tangent functional of FL can be identiﬁed with −v. i. If we choose vk = v0 + wk where wk describes a series of boxes of increasing size but with decreasing height of the potential wall then we see that we have created a series of potentials in V that approaches v0 to an arbitrary accuracy. They say that any F bounded functional L0 can be approximated to arbitrary accuracy by a tangent functional L . We will not prove this theorem here. Let us now discuss the second part of the BishopPhelps theorem Theorem 13 (BishopPhelps II) Let F : B → R be a lower semicontinuous convex functional on a real Banach space B. Suppose n0 ∈ B with F [n0 ] < ∞ and let L0 ∈ B ∗ be an F bounded functional. L − L0 3. Then for every ≥ 0 there exists n ∈ B and a functional L ∈ B ∗ such that 1. n − n0 B∗ 3 2 ≤ ≤ F [n0 ] − L0 [n0 ] − inf n∈B {F [n] − L0 [n]} 2. until one realizes that one may put the system in big box with a wall of ﬁnite height. The inequality in the third point of this theorem .(24). B For the details of the proof we again refer to [22]. The particles will then spread out over the box in order to minimize the repulsion between them. Now we know 3 that every element −v0 for an arbitrary v0 ∈ L 2 + L∞ corresponds to an FL bounded functional.e. This may seem counterintuitive at ﬁrst sight as one can imagine v0 to be a repulsive potential. Therefore for such a v0 there is a series of vk ∈ V such that lim v0 − vk = 0 (228) k→∞ where the norm is the L +L∞ norm described in Eq. where v is a potential that yields a bound ground state. F [n] ≥ F [n ] + L [n − n ] for all n. v ∈ V. This system will have a bound state if the size of the box is chosen to be big enough. The functional F can take the value +∞ but not everywhere. Therefore every 3 potential v0 ∈ L 2 + L∞ can be approximated to arbitrary accuracy by a potential that yields a bound ground state.This means that if L0 is some F bounded functional then we can ﬁnd a set of tangent functionals Lk such that k→∞ lim L0 − Lk B∗ =0 (227) where the limit is taken in the norm on B ∗ . The proof is clearly described with a geometric interpretation in reference [22].
We take = 1 . Let us now apply this theorem to the functional FL and show that that every density in S is arbitrarily close to an EVdensity. Furthermore.(224) we see that we can write FL [n] = sup 3 v∈L 2 +L∞ {E[v] − d3 rn(r)v(r)} (229) If the supremum is attained for some v then n is an EVdensity.allows us to make statements about distances between elements in B. n0 ∈ S and let −vk correspond to the F bounded functional L0 of the theorem. for any n we have E[vk ] = inf {FL [n] + n∈S d3 rn(r)vk (r)} ≤ FL [n] + d3 rn(r)vk (r) (232) which in combination with the previous inequality yields FL [n] + d3 rn(r)vk (r) ≥ E[vk ] ≥ FL [n0 ] + d3 rn0 (r)vk (r) − 1 k (233) We are now ready to apply the BishopPhelps theorem to the Lieb functional FL . for every integer k and any density n0 ∈ S we can always ﬁnd some vk such that E[vk ] − d3 rn0 (r)vk (r) ≥ = sup 3 v∈L 2 +L∞ {E[v] − 1 k d3 rn0 (r)v(r)} − 1 k FL [n0 ] − (231) where we note that the series vk does not converge to any v if n0 is not an EVdensity. If n is not an EVdensity then the supremum is not attained for any v. In any case. From Eq. which in our case will be densities. We ﬁrst need some preliminaries. According to the theorem we can then ﬁnd a tangent functional −wk (the L of the theorem) such that vk − w k ≤ 1 (234) . Note that the right hand side of this inequality has the geometric meaning of being the diﬀerence of the vertical distance of the functionals L0 and F in n0 and the shortest possible distance between L0 and F . This ˆ follows because then there is a density matrix D[n] that yields density n (see Theorem 5) such that ˆ ˆ TrD[n]Hv = FL [n] + ˆˆ d3 rn(r)v(r) = E[v] = inf TrDHv ˆ D (230) ˆ The density matrix D[n] must therefore be a ground state density matrix.
The BishopPhelps theorem then tells us that nk − n0 p ≤ − FL [n0 ] + n∈S d3 rn0 (r)vk (r) d3 rn(r)vk (r)} (236) inf {FL [n] + where the normindex has the values p = 1 and p = 3. This proves the theorem in the beginning of this section.(233) we see immediately that n∈S inf {FL [n] + d3 rn(r)vk (r)} ≥ FL [n0 ] + nk − n0 ≤ d3 rn0 (r)vk (r) − 1 k (237) and therefore 1 (238) k for p = 1 and p = 3. Since this equation is true for any k we see that any density n0 ∈ S can be approximated to any accuracy by the EVdensity nk in the norms on L1 ∩ L3 .in the norm on L 2 + L∞ . Note that the inﬁmum in this equation can be taken over S rather than L1 ∩ L3 since FL is deﬁned to be +∞ outside S. FL [n] ≥ FL [nk ] − d3 rwk (r)(n(r) − nk (r)) (235) 3 where wk is the potential that generates density nk .e. p 15. In order to put the KohnSham procedure on a rigorous basis we introduce the functional ˆˆ TL [n] = inf TrDT ˆ D→n (239) We see that this is simply the Lieb functional with the twoparticle interaction omitted. All the properties of the functional FL carry directly over to TL . From our previous investigations we know that the tangent −wk touches the graph of FL in an EVdensity nk . The reason is that all these properties were derived on the basis . Then from Eq. Kohn and Sham [8] introduced an auxiliary noninteracting system of particles with the property that it yields the same ground state density as the real interacting system. i. The KohnSham approach and noninteracting vrepresentability We have now come to the discussion of the central equations which form the basis of almost any practical application of density functional theory: the KohnSham equations.
The functional derivative at a noninteracting EVdensity n is given by: δTL = −vs [n](r) δn(r) where the potential vs generates the density n in a noninteracting system. We therefore conclude that TL is a convex lower semicontinuous functional which is diﬀerentiable for any density n that is ensemble vrepresentable for the noninteracting system and nowhere else. still true if we omit ˆ the Coulomb repulsion W . From the last point in this theorem we see that if we want to know if a given density n from the set S can be obtained as an EVdensity of a noninteracting system we may try to calculate the derivative of TL for this density. if we apply the BishopPhelps theorem to the functional TL we obtain the following result: Theorem 15 The set of noninteracting EVdensities is dense in the set S with respect to the norm on L1 ∩ L3 . however. We can in particular choose n to be an interacting EVdensity. If it exists then the derivative yields the potential that we were looking for.e. we can make a number of useful conclusions. This means that for any density n ∈ S there is an noninteracting EVdensity arbitrarily close.ˆ ˆ of the variational principle in which we only required that T + W is an operator that is bounded from below. in . This is. We also know from the BishopPhelps theorem applied to FL that the set of interacting EVdensities is dense in S and therefore for any n ∈ S. n ∈ B and ﬁnd a noninteracting EVdensity arbitrarily close. This is currently not known. TL is Gˆteaux diﬀerentiable at the set of noninteracting EVdensities a and nowhere else. 2. 3. For any n ∈ S there is a minimizing density matrix D[n] with the ˆ ˆ property TL [n] = TrD[n]T . However. Let us collect all the results for TL in a single theorem: Theorem 14 TL is a convex lower semicontinuous functional with the following properties: ˆ 1. First of all. i. Now Kohn and Sham [8] assumed that for any density of an interacting system there is a noninteracting system that has the same density as its ground state. We refer to such densities as noninteracting EVdensities and denote the set of all noninteracting EVdensities by B0 . We can now ask the question whether the sets of interacting and noninteracting EVdensities are equal.
We therefore conclude Theorem 16 The set B0 of noninteracting EVdensities is dense in the set B of interacting EVdensities. however. Then we can construct a noninteracting KohnSham system with ground state wavefunction 1 ΨKS (r1 σ1 . (r) = − δTL (n ) δn(r) i. r2 σ2 ) = √ ϕ(r1 )ϕ(r2 )(α(σ1 )β(σ2 ) − β(σ1 )α(σ2 )) 2 (240) . 2. The density n is a ground state ensemble density of a noninteracting system with potential vs. has not been proven until now. we can ﬁnd an interacting EVdensity arbitrarily close. This theorem tells us that in practice we can always set up a KohnSham scheme. If we combine this result with previous theorems we obtain the following important consequence for the KohnSham scheme: Theorem 17 Suppose n ∈ B is an interacting EVdensity. then we have to show that the potentials vs. We also see that if we want to prove that the sets B and B0 of interacting and noninteracting EVdensities are equal.e. i. One might therefore expect that the sets of interacting and noninteracting EVdensities are in fact equal. This. Let the interacting system have density n. but seems to happen in a rather smooth way. Let us describe a couple of cases for which we know this conjecture to be true. In numerical calculations (see the discussion at the end of this section) one has indeed always succeeded in obtaining a KohnSham potential for given interacting EVdensities obtained from accurate Conﬁguration Interaction (CI) calculations.0 for → 0 in some smooth way. B = B0 . In these calculations convergence to a given KohnSham potential is sometimes diﬃcult to obtain.e. Consider a system of two particles. In order to encourage further work in this ﬁeld we put it here as a conjecture Conjecture 1 The sets B and B0 of interacting and noninteracting EVdensities are equal. n − n p ≤ for p = 1 and p = 3. we can always set up a KohnSham scheme that produces a given interacting EVdensity to arbitrary accuracy. Then for every > 0 there is a noninteracting EVdensity n ∈ B0 such that 1.particular n ∈ B0 . and vice versa. in this theorem approach some potential vs.
This work may be used to prove the general conjecture above if one can prove that the continuum limit can be taken in some smooth way. Let us ﬁnally discuss the exchangecorrelation functional which is the central object in any application of density functional theory. This was proven by Chayes. The lattice system is obtained by discretizing the manybody Hamiltonian on a grid.1 where α and β are the usual spin functions with α( 2 ) = β(− 1 ) = 1 and 2 1 1 α(− 2 ) = β( 2 ) = 0. The derivative of Eq. We deﬁne the exchangecorrelation functional Exc by FL [n] = TL [n] + 1 2 d3 r 1 d3 r 2 n(r1 )n(r2 ) + Exc [n] r1 − r2  (243) Since both TL and FL are deﬁned on the set S the exchangecorrelation functional Exc is also deﬁned on that set.e. Because the density is positive the orbital ϕ = n/2 has no nodes and must be a ground state orbital. This wavefunction has a density n(r) = 2ϕ(r)2 and the KohnSham orbital ϕ satisﬁes − 1 2 2 + vs [n](r) ϕ(r) = ϕ(r) (241) If we now choose ϕ(r) = ϕ(r)/2 then for an interacting EVdensity n the KohnSham Hamiltonian with potential √ 1 2 n √ + (242) vs [n](r) = 2 n has orbital ϕ as an eigenfunction. Since FL and TL are diﬀerentiable respectively on the sets B and B0 and nowhere else. the functional Exc is diﬀerentiable on B ∩ B0 and nowhere else.(243) on that set is given by −v(r) = −vs (r) + d3 r n(r ) δExc + r − r  δn(r) (244) If we deﬁne the exchangecorrelation potential by vxc (r) = δExc δn(r) (245) . Then any intero acting EVdensity on the grid is also a noninteracting EVdensity. which is even a pure state density. we replace the diﬀerential equation by a diﬀerence equation as one might do to solve the Schr¨dinger equation numerically. Chayes and Ruskai [23]. i. Another case where the conjecture is true is for lattice systems. We therefore have constructed a noninteracting system with ground state n.
However. We will therefore in general have q Φi = j=1 αij Dj (252) . The density must now be calculated from a ground state ensemble q ˆ Ds [n] = i=1 µi Φi Φi  (248) of the KohnSham system and is explicitly given as q ˆ n(r) = TrDs [n]ˆ (r) = n i=1 µi Φi ˆ (r)Φi n (249) These equations can be written more explicitely in terms of the KohnSham orbitals deﬁned to be the eigenstates of the single particle equation − 1 2 2 + vs (r) ϕi (r) = i ϕi (r) (250) It is then easily seen that any Slater determinant Φ = ϕi1 . However.then we see that the KohnSham potential can be written as vs (r) = v(r) + d3 r n(r ) + vxc (r) r − r  (246) We see that on the set B ∩ B0 we have obtained the usual KohnSham equations ˆ ˆ ˆ ˆ T + V + VH [n] + Vxc [n] Φi = E0 Φi (247) where i = 1 . . For instance. . .(249) we require that the density is representable by a ground state ensemble of a noninteracting system. q runs over the q degenerate ground states Φi of the KohnSham system. . ϕiN (251) built out of KohnSham orbitals is an eigenstate of the KohnSham system with eigenvalue E0 = i1 + . In Eq. if two Slater determinants have the same energy eigenvalue. . not every eigenstate is necessarily a Slater determinant. we did not require that it must be representable by an ensemble of ground state Slater determinants. + iN . then a linear combination of them also has the same eigenvalue. has been shown by Englisch and Englisch [3] and Lieb [1]. . This leads to a subtle point for KohnSham theory of degenerate states. They constructed an explicit example of a pure state density of a noninteracting system that can not be obtained from a single Slater determinant. That one should be careful at this point.
27]. We further know that every Φi must be a ground state for potential vs . Because both determinants are eigenfunctions of the KohnSham Hamiltonian it is easily seen that these diﬀerent orbitals must have the same KohnSham eigenvalue. This is an indication that the true KohnSham density could be an EVdensity. The matter has been investigated in detail by Schipper et al. We then introduce new orbitals for the highest occupied level: ϕi (r) = ˜ j Uij ϕj (r) (254) Since these orbitals are linear combinations of degenerate orbitals. We can now write the density as n(r) = i <µ ϕi (r)2 + k =µ nk ϕk (r)2 ˜ (255) with occupation numbers nk between 0 and 1. These authors calculated by CImethods accurate charge densities for a couple of molecules.(255) in which also the coeﬃcients nk must be determined. If this were not the case then we could lower the energy by transferring an electron from the highest level to a lower level which would lower the energy and therefore Φi would not be a ground state for potential vs . This also means that if two determinants Di and Dj diﬀer. they are again eigenfunctions of the KohnSham singleparticle Hamiltonian for the highest occupied level. 25]. then this diﬀerence must be due to diﬀerent orbitals in the highest level. This means that we can diagonalize this matrix with a unitary matrix Uij .(250) together with Eq. This eigenvalue is the KohnSham orbital energy of the highest occupied state which we will denote by µ. When do we need fractional occupation numbers in practice? In practice one sometimes ﬁnds that when solving the KohnSham equations with occupations equal to one that the converged solution has a hole below the highest level. This means that all orbitals below the highest occupied level must be occupied.were Dj is a Slater determinant. From these considerations we ﬁnd that the density is of the general form n(r) = i <µ ϕi (r)2 + βkl ϕk (r)ϕ∗ (r) l k = l =µ (253) where βkl is an Hermitian matrix with eigenvalues between 0 and 1. In practice the lowest energy state is then obtained by a procedure called ”evaporation of the hole”. [26. Subsequently they constructed the KohnSham potentials that generate these . This happens for instance for the iron atom within the local density approximation [24]. In this method one increases the occupation of the hole while descreasing the occupation of the highest level until both levels are degenerate [24. We see that for the case of a degenerate ground state we have to solve Eq.
The corresponding bond distances P R(C −C) are displayed in the inset.e. This is then not a proper KohnSham state since it is not the state of lowest energy for the corresponding potential. Other examples are provided by neardegeneracy situations such as in avoided crossings [25] and the H2 + H2 reaction [27]. A recent discussion on KohnSham theory for ensembles can be found in Ref. If one then tries to construct an ensemble solution by the technique of ”evaporation of the hole”. The potentials vxc are the KohnSham potentials P E corresponding to a ground state ensemble. This may be explained by the fact that the wavefunction of the C2 molecule has a strong multideterminantial character. . i. When one tries to construct a KohnSham potential that produces the ground state density of the C2 molecule using occupation numbers equal to one. one does ﬁnd a proper ground state ensemble for the KohnSham system. The potentials vxcS−hole correspond to the exchangecorrelation potentials of the improper KohnSham solutions with a hole.densities. the exchangecorrelation potentials of the C2 molecule for four bonding distances. As an illustration we display in Figure 4. for a pure state. Both vxcS−hole and vxc give an accurate representation of the true density of the molecule. The distortions in vxcS−hole therefore appear to be a price we have to pay for producing the density of such a strongly multideterminantial state with a single Slater determinant. This means that the extension of density functionals to EVdensities is not only of theoretical. P The potential vxcS−hole on the other hand. but also practical interest. i. The potentials are plotted along the C −C bond axis as functions of the distance z from the bond midpoint. [28] which also gives the beryllium series as an explicit example of systems which are ensemble vrepresentable but not pure state vrepresentable. The potential vxc has the features found in many diatomic molecules. but an excited state. With this we mean that the simple HartreeFock approximation has a relatively small coeﬃP cient in the CIexpansion of the wave function. although E E the accuracy attained with vxc is a bit higher. the usual well around the atom.e. The corresponding state is a pure state consisting of a sinE gle Slater determinant. the small intershell peaks and a plateau around the bond midpoint. is heavily distorted as compared E to vxc . A particularly interesting case is the C2 molecule [26]. one ﬁnds that the solution corresponds to a state with a hole below the highest occupied level. From this we can conclude that the ground state density of the C2 molecule is a noninteracting EVdensity but not a noninteracting PSVdensity. The use of an ensemble formulation of KohnSham theory seems therefore especially relevant in cases of strong electron correlations.
but solution potential that an exchangecorrelationhas been obtained. b 2. producing Applying the results of Levy [1] Figure 4: Exchangecorrelation potentials 17. tempt to construct the PS leads to a nonAufbau soludistances to be EVs representable. 37]. c 3.0 a. we knowrate there are many ground . obtained depending on the R(C±C) distance.8 a.u.0 a. (or class b.state density offunction 2 molecule at various that interacting for CH2 for which the wave the C has a relatively weak multideterminantal E character. 3a±d The exchangecorrelation potential mxc along the bond axis of C2 . 1). the possibility has been demonstrated of an essentially accurate ensemble KS interacting densities. PSÀhole mxc (dotted lines) corresponds to the singledeterminant solutions with a hole below the Fermi level. calculation in the densities that are not PSVs representable but that are bonding distances. The yields the ensemble solution and only with the EVs representable. the Fermi level'' has been employed [15.u. prove this not to be solution with accidental degeneracy (EVs representonly an academic possibility. essentially multideterminantal character acting system.. To a. but EVs representability is in fact called for to handle cases with strong electron ability) for a q that is PSV representable for the intercorrelation. mE (solid lines) xc is the potential corresponding to the ensemble solution. ranging The KS solution for the lowest singlet states of CH2 from a singledeterminantal purestate at R(C±C) 1:8 and C2 has been constructed from the ab initio CI q.e. through an ensemble with a weak accidental deobtain the ensemble solution in the cases when an atgeneracy at Re to an ensemble with a strong degeneracy . A variety of KS solutions has been of the interacting groundstate wavefunction. d 4.u. as tion. Our results sity for a single Slater determinant with a hole. Already the accuand Lieb [2] to noninteracting systems.341 Fig. The potentialKSxcS−hole is potential for to extended QZ basis the purestate onedeterminantal v P the set of PS noninteracting ground state densities. a Re . the procedure of ``evaporation of the hole below described in Sect.. reproduces the same dendoes belong to the set of ensemblerepresentable nonWith the example of C2 ..u. i. It is thus certainly possible that a TZ basis potential vxc is the exchangecorrelation given interacting ground state density does not belong a ground state ensemble.
δn(rm ) (259) . We further deﬁned (m) Kxc (n0 . In this approximation the exchangecorrelation functional is taken to be LDA (256) Exc [n] = d3 r xc (n(r)) where xc (n) is the exchangecorrelation energy per volume unit for a homogeneous electron gas with density n. 30] and has the following form Exc [n] = + LDA Exc [n] + d3 rg1 (n(r))( n(r))2 2 d3 rg2 (n(r))( n(r))2 + . r1 . This simple and crude approximation turns out to be surprisingly succesful for realistic and very inhomogeneous systems although one would expect that the LDA would only be accurate for systems with slowly varying densities. We will come back to the question of convergence. The gradient expansion presents an. In this section we will show that the LDA is in fact the ﬁrst term in a systematic expansion of the exchangecorrelation functional in terms of spatial derivatives with respect to the density. . . . The gradient expansion Until now we only considered the formal framework of density functional theory. the theory would be of little use if we would not be able to construct good approximate functionals for the exchangecorrelation energy and exchangecorrelation functional. . δn(rm ) (258) where d3 r = d3 r1 . To derive the gradient expansion we start out by expanding the exact exchangecorrelation energy functional around its homogeneous electron gas value n0 as follows Exc [n] = Exc [n0 ]+ 1 m! m=1 ∞ (m) d3m rKxc (n0 . . Historically the ﬁrst approximation for the exchangecorrelation functional to be used was the local density approximation. rm )δn(r1 ) . However. rm ) = δ m Exc n=n0 δn(r1 ) . . . We therefore treat the inhomogeneous system locally as an electron gas. . .16. provided the series converges. in principle. r1 . . . The LDA therefore seems a suitable starting point for more accurate approximations. . This expansion is commonly referred to as the gradient expansion [5. 29. d3 rm and n(r) = n0 + δn(r). (257) where the functions gi (n) are uniquely determined by the density response functions (of arbitrary order) of the homogeneous electron gas. exact way to construct the exchangecorrelation functional for solids.
rm + a) (m) Kxc (n0 . r2 − r1 ) δn(r1 )δn(r2 ) (266) which leads to an expansion of the form Exc [n] = Exc (n0 ) + 1 2 d3 r1 d3 r2 L(2) (n0 . . rm ) (m) = Kxc (n0 . Since the electron gas has translational. rm − r1 ) (m) = L (n0 . r1 . The function Kxc is given by δExc (1) Kxc (n0 . rm ) = Kxc (n0 . . due to translational invariance. . rm ) (m) Kxc (n0 .(258) does not contribute to Exc [n]. r1 = (m) Kxc (n0 . (267) The function L(2) (often denoted by Kxc in the literature) has been the subject of many investigations [32. . qm ) = d3m rf (r1 . r2 − r1 . If we introduce the Fourier transform of a function f as follows ˜ f (q1 . rm ) = where a is an arbitrary translation vector and where R is an arbitrary (m) rotation matrix. . r2 − r1 . . Let us give some speciﬁc examples. .. does not depend (1) on r1 . . . r1 (m) Kxc (n0 . . . Rrm ) (m) Kxc (n0 . . r1 . . 33]. rm )e−iq1 ·r1 −. m. r2 − r1 )δn(r1 )δn(r2 ) + . rm ) = . r2 ) = δ 2 Exc n=n0 = L(2) (n0 .The ﬁrst term Exc [n0 ] in Eq. We will therefore write this term as Exc (n0 ). . rm − r1 ) (264) where the latter equation deﬁnes the function L(m) as a function of m − 1 (1) variables. Furthermore. −r1 . Let us now go back to the more general case. . (m) (m) Kxc (n0 . . . This function is by now wellknown from extensive investigations of the homogeneous electron gas [31]. r1 ) = (265) n=n0 = vxc (n0 ) δn(r1 ) and we see that this function is simply the exchangecorrelation potential of the electron gas which. i. (m) rotational and inversion symmetry the functions functions Kxc satisfy (m) Kxc (n0 . . If we choose a = −r1 in Eq. The ﬁrst nontrivial term is therefore (2) Kxc (n0 .(258) is the exchangecorrelation energy of a homogeneous system with constant density n0 and is therefore a function of n0 rather than a functional. . . rP (m) ) (263) for an arbitrary permutation P of the indices 1 . . − rm ) (260) (261) (262) . 0. Since δn(r) integrates to zero the term with Kxc in the expansion Eq. . .e. . r1 + a . rP (1) . Rr1 .. . r1 . .−iqm ·rm (268) . . . . r1 . (260) we see that (m) Kxc (n0 . . the function Kxc has full permutational symmetry.
An important special case is obtained for m = 1: ∂vxc ∂n0 = = ∂Kxc (n0 ) = ∂n0 (1) (2) d3 r2 Kxc (n0 . . . q1 . . This can also be written in terms of the functions L(m) as follows Exc [n] = Exc (n0 )+ 1 m! m=1 ∞ d3 qm ˜ (m) d3 q2 . . However. r1 . . q2 . δ˜ (qm ) n n n The gradient expansion is then obtained by expanding L(m) in powers of qi around qi = 0 and subsequently transforming back to real space. . If we do this our ﬁnal expansion will then still depend on our reference density n0 which is undesirable for application of the functional to general systems. qm )δ˜ (q1 ) .. r. . q2 . r2 ) ˜ d3 r2 L(2) (n0 .then ˜ (m) ˜ Kxc (n0 . .(273) can alternatively be derived directly from the properties of the response functions [30] and is related to a generalized form of the wellknown compressibility sumrule of the electron gas. . . d3 qm . r1 . r1 . . L (n0 . r2 − r1 ) = L(2) (q = 0) (274) . . . . . . δ˜ (qm ) (270) (2π)3m xc where d3m q = d3 q1 . . qm ) = (2π)3 δ(q1 + . we will show that we can get rid of this dependence on n0 by doing inﬁnite resummations. r1 . . r. qm ) (2π)3 (2π)3 (271) ×δ˜ (−q2 − . rm ) = ∂n0 (m) (m+1) d3 rKxc (n0 . Nevertheless Eq. + qm )L(m) (n0 . . q1 . . − qm )δ˜ (q2 ) .. . . . qm ) The expansion for Exc then becomes Exc [n] = Exc (n0 )+ 1 m! m=1 ∞ (269) d3m q ˜ (m) n n K (n0 . The key formula that will enable us to do (m) these resummations is the following ﬁrst order change of Kxc : (m) δKxc (r1 . rm ) = (m+1) d3 rKxc (n0 . . . rm )δn(r) (272) If we take δn(r) = δn0 to be a constant change we obtain ∂Kxc (n0 . rm ) (273) Note that by taking δn(r) = δn0 we violate the condition that δn(r) integrates to zero. .
.)(276) in which the Ki are coeﬃcients and Pi are symmetric polynomials. qm ) ∂n0 (275) The importance of this formula will become clear if we work out. q2 . −q2 − . − qm . − qm )2 + q2 + . . .(275) we obtain for the coeﬃcients in the expansion of L(m) the following equations ∂L0 ∂n0 ∂L1 ∂n0 (m) = = L0 L1 (m+1) (n0 ) (n0 ) (281) (282) (m) (m+1) . . . . . . . q1 . . . This is a direct consequence of the permutational symmetry of Kxc . . . .(278) and Eq.˜ We see that the q = 0 value of L(2) can be directly calculated from the knowledge of vxc (n0 ). From the (m) symmetry properties of Kxc one ﬁnds that its Fourier transform has the following expansion in powers of qi : ˜ (m) Kxc (n0 . qm ) = q2 + . . . . qm ) = L0 (n0 ) + L1 (n0 )p(m) (q2 . qm ) = = (−q2 − . it transforms according to a onedimensional representation of the permutation group. . . the lowest order in the gradient expansion. . qm ) + . + q2 m 1 and is invariant under permutation of the indices. For m ≥ 2 relation (273) directly implies the following relation between L(m) and L(m+1) ∂L(m) (n0 . . + qm ) ×(K0 + K1 P1 (q1 . The (m) ﬁrst polynomial P1 (taking into account the δfunction) has the explicit form (m) (277) P1 (q1 . as an explicit example. qm )+ . . . . . From the properties of ˜ (m) Kxc we ﬁnd that for m ≥ 2 the function L(m) has the following expansion L(m) (n0 . . . . . q2 . i. . qm ) = p(m) (q2 . . . . . where the polynomial p(m) is given by p(m) (q2 . More details on the group theoretical treatment of these reponse functions can be found in the work of Svendsen and von Barth [34]. . . + q2 2 m m m (m) (m) (m) (m) (m) (m) (m) (m) (m) (278) 2 i=2 q2 − 2 i i>j≥2 qi · q j (279) It is easily seen that this polynomial has the property p(m+1) (−q2 − . qn ) = (2π)3 δ(q1 + . qm ) (280) If we use this property together with Eq. . q2 . − qm . q2 . . qm ) + K2 P2 (q1 .e. qm ) = L(m+1) (n0 .
(285) as ∞ n(r) this can be rewritten with help of Eq. We obtain LDA Exc [n] = Exc [n] + d3 rL1 (n(r))( n(r))2 + . If we insert the explicit form of L(m) of Eq. . Together with ∂vxc ∂ 2 xc ˜ L(2) (q = 0) = (n0 ) = (n0 ) ∂n0 ∂n2 0 (283) where xc (n0 ) is the exchange energy per volume unit of the electron gas.(278) into Eq.(286) 1 (m) L (n0 )m(m − 1) m! 1 m=2 Since δn(r) = Eq.i. . LDA Exc [n] = d3 r xc (n0 + δn(r)) (2) (288) whereas the third term involves an expansion of the coeﬃcient L1 (n0 + δn(r)) around n0 . .(271) and Fourier transform back to real space we obtain Exc [n] = Exc (n0 )+ 1 (m) L (n0 ) m! 0 m=1 ∞ ∞ d3 r(δn(r))m + d3 r( δn(r))2 (δn(r))m−2 + .(284) and Exc [n] = Exc (n0 )+ 1 ∂ m xc (n0 ) m! ∂nm 0 m=1 ∞ d3 r(δn(r))m + (2) 1 ∂ m−2 L1 (n0 ) (m − 2)! ∂nm−2 0 m=2 d3 r( n(r))2 (δn(r))m−2 + .where m ≥ 2. (2) (289) . .(287) We see that the ﬁrst two terms simply give the expansion of the LDA functional around the constant density n0 . this yields L0 (n0 ) L1 (n0 ) (m) (m) = = ∂ m xc (n0 ) ∂nm 0 ∂ m−2 L1 (n0 ) ∂nm−2 0 (2) (284) (285) It is these two equations that will allow us to eliminate the dependence on n0 in the lowest order gradient expansion. . .e.
However. q) to order q2 . namely perturbation theory starting from the KohnSham Hamiltonian [43. However. The dependence on n0 can also be removed to higher order in powers of q. One ﬁnds [37] that the gradient expansion performs very well for metallic systems. Currently the most fruitful approach to the construction of simple and accurate functionals is the socalled Generalized Gradient Approximation (GGA) [40. This question has been investigated for the gradient expansion of the exchangeenergy functional for which a comparison with exact exchange ˜ energies is possible. 37]. For an explicit example up to order q4 we refer again to Svendsen and von Barth [34]. Further progress along the lines of the gradient expansion may be obtained by study of the socalled gapped electron gas [39]. Let us now address the question of convergence of the gradient expansion. This may not be so surprising if one realizes that the appearance of a gap drastically changes the low qbehavior of (m) the response functions Kxc which determine the gradient coeﬃcients [38]. 41]. This means that the standard gradient expansion can not deal with insulators or ﬁnite systems (which may be modelled as insulators by repeating them periodically with a large lattice constant). For this case the analytic form of L(2) (n0 . 44]. 36] and some higher order gradient coeﬁcients have been determined by Svendsen. 17. but that the quality deteriorates as soon as the system acquires an energy gap. Morever we see that the local density approximation naturally appears as the ﬁrst term in the gradient expansion.and we see that we succeeded in eliminating the dependence on n0 in the gradient expansion. since the approach is not systematic it is diﬃcult to improve the quality of the current GGA functionals. q) is known from the impressive work of Engel and Vosko [35. We also see that the replacement n0 → n(r) requires a resummation over response functions of arbitrary order. For this reason the method is less suited to the calculation of properties of large systems. Furthermore we ﬁnd that the coeﬃcient g1 (n) in our ﬁrst equation (257) is completely ˜ determined by the qexpansion of the function L(2) (n0 . The optimized potential method and the e2 expansion In this section we will describe a second systematic method to construct density functionals. Springer and von Barth [34. In this approach one speciﬁes a form of the paircorrelation function of the manyelectron system and determines the parameters in this function by sumrules and information from the straightforward gradient expansion. the method has the theoretical advantage that it can be used as a . The method is based on traditional perturbation theory and is comparable in computational cost. These functionals have lead to large improvements in molecular binding energies as compared to the local density approximation [42].
The ﬁrst two terms are explicitely given by [43. We consider the Hamiltonian ˆ ˆ ˆ ˆ ˆ ˆ ˆ H = Hs + (H − Hs ) = Hs + (W − VHxc ) where N (290) ˆ W ˆ VHxc = i>j N e2 ri − rj  vH (ri ) + vxc (ri ) (291) = i=1 (292) and vxc the exchangecorrelation potential.e. { k [n]}.i 2 Es − Es. orbitals. . i. and energies. . E (i) [n] = E (i) [{ϕk [n]}. i. We will come back to the point of convergence later. The arbitrariness with . We now note that the energies E (i) are implicit functionals of the density through their dependence on the KohnSham potential. vxc [n]] (297) This follows directly from the HohenbergKohn theorem applied to a noninteracting system. simply a sum over orbital energies and the terms E (i) is the energy to order i in powers of . The perturbation term is simply the diﬀerence between the true and the KohnSham Hamiltonian and we are interested in the case = 1 although this makes expansion in powers of rather doubtful.i (295) (296) where Es. The density n uniquely determines the KohnSham potential vs (up to a constant) and therefore the also the orbitals (up to a phase factor) and eigenvalues (up to constant).i and Φs. We can now do standard perturbation theory and obtain the ground state energy E( ) = Es + E (1) + 2 E (2) + .benchmark to test the quality of diﬀerent approximate density functionals.e. (294) where Es is ground state energy of the auxiliary KohnSham system.i are the excited state energies and wave functions of the KohnSham system. 44] E (1) E (2) = = i=1 where e2 is the square of the electronic charge and vH is the usual Hartree potential n(r ) vH (r) = e2 d3 r (293) r − r  Φ s W ∞ ˆ − VHxc Φs ˆ ˆ ˆ  Φs W − VHxc Φs.
The diﬀerence. This is because the perturbing Hamiltonian contains the exchangecorrelation potential which is unknown from the start and has to be determined selfconsistently from the perturbation series.i (303) The term of order e2 can be written explicitly in terms of the KohnSham orbitals as follows: (1) Exc [n] = − 1 2 N d3 rd3 r k.i 2 Es − Es. is that the orbitals in this expression are KohnSham . the equations can be simpliﬁed if we expand the energies and potentials in powers of the interaction strength e2 and take = 1.l=1 ϕ∗ (r)ϕl (r)ϕ∗ (r )ϕk (r ) k l r − r  (304) We see that this expression has exactly the same form as the usual expression of the exchange energy within the HartreeFock approximation. We therefore have the following series of implications n(r) → vs (r) → {ϕk (r). however. However. This leads to the following set of equations [44] E[n] = + vHxc (r) (i) vxc (r) Ts [n] + e2 2 d3 rn(r)v(r) n(r)n(r ) (i) + e2i Exc r − r  i=1 ∞ d3 rd3 r ∞ (299) (300) (301) = = vH (r) + i=1 (i) e2i vxc (r) δExc δn(r) (i) The e2 and e4 terms in the expansion of the exchangecorrelation energy have the explicit form e2 (1) ˆ e2 Exc [n] = Φs W Φs − 2 and e 4 (2) Exc [n] ∞ d3 rd3 r n(r)n(r ) r − r  (302) = i=1 ˆ ˆ ˆ (1)  Φs W − VH − e2 Vxc Φs.respect to a constant shift and with respect to the phase factor cancels out in the energy expression and therefore the ith order energy becomes a pure density functional. k} → E (i) (298) Note that the perturbation theory that we constructed is not yet in a form that we can use in practical calculations.
r )ϕk (r ) (307) (308) (309) δϕk (r) Gk (r. r )δvs (r ) (312) . The numerical value (1) of Exc [n] and the HartreeFock exchange will therefore in general diﬀer from each other. The change δϕk and δ k of the orbitals and orbital energies of the KohnSham system due to a small change δvs in the potential is given by solution of the equation − This gives δ k 1 2 2 + vs (r) − k δϕk (r) = (δ k − δvs (r))ϕk (r) (306) = = − = d3 rϕ∗ (r)ϕk (r)δvs (r) k d3 r Gk (r.(303) that we need to know this potential in order to calculate the e4 contribution to the exchangecorrelation energy. Let us therefore start by calculating vx . This is a general feature of the present perturbation theory. For this we have to calculate the ﬁrst order change δEx in the exchange functional due to a change δn in the density. are eigenfunctions of a singleparticle Hamiltonian with a local potential. In order to calculate (i) (i−1) Exc we need to calculate vxc ﬁrst. Corresponding to the exchangefunctional there is a local exchange potential deﬁned as vx (r) = δExc δEx (1) = = vxc (r) δn(r) δn(r) (1) (305) We see from Eq.c. r ) l=k From these equations we obtain the following functional derivatives δ k δvs (r) δϕk (r) δvs (r ) N = = (310) (311) The density change δn is given by δn(r) = k=1 ϕ∗ (r)δϕk (r) + c. This is readily done by perturbation theory. r )ϕk (r )δvs (r ) ϕl (r)ϕ∗ (r ) l l− k ϕ∗ (r)ϕk (r) k −Gk (r. Since densities and potentials are in 11correspondence this task amounts to calculating the change in Ex due to a change δvs in the KohnSham potential. in contrast to the HartreeFock orbitals.orbitals which. However. = k d3 r χs (r. because of the similarity to the HartreeFock deﬁnition of exchange we will deﬁne the exchange functional within density (1) functional theory to be Ex [n] = Exc [n].
c.c.c. the exchange potential can be found from simultaneous solution of the equations k ϕk (r) = = − 1 2 2 + v(r) + vH (r) + e2 vx (r) ϕk (r) (316) (317) Λx (r) d3 r χs (r. r )vx (r ) Once we have solved these equations and determined vx we can go on (2) and calculate Exc from Eq.where we deﬁned the static density response function χs of the KohnSham system by χs (r. δϕk (r ) (315) Since both Λx and χs are given as explicit functionals of the KohnSham orbitals and orbital energies.(303). However. except for the term δvx /δvs . r )ϕk (r ) + c. r )vx (r ) (314) where the inhomogeneity Λx is given by N Λx (r) = − d3 r k=1 δEx Gk (r . k k=1 (313) We can now readily derive the following integral equation for vx : Λx (r) = δEx = δvs (r) d3 r δEx δn(r ) = δn(r ) δvs (r) d3 r χs (r. δϕk (r ) δvs (r) (2) (2) + δExc δ k + δ k δvs (r) d3 r δExc δvx (r ) (319) δvx (r ) δvs (r) (2) All terms in this equation are explicitly known. To calculate the e6 contribution to the (2) exchangecorrelation energy we ﬁrst have to evaluate vxc . This potential is the solution of the integral equation Λ(2) (r) = xc (2) δExc = δvs (r) (2) (2) d3 r χs (r. r )vxc (r ) (318) Since Exc is an explicit functional of the orbitals. r ) = δn(r) =− δvs (r ) N ϕ∗ (r)Gk (r. r)ϕk (r) + c. for this term we can ﬁnd an integral equation by diﬀerentiation . orbital energies and vx (2) the inhomogeneity Λxc can be calculated from ∞ Λ(2) (r) xc = k=1 ∞ d3 r d3 r k=1 δExc δϕk (r ) + c.
(320) determines δvx /δvs uniquely. From vxc one could go on along the same (3) (3) lines to determine Exc and subsequently vxc . ˆ E1 [vs ] = Φs [vs ]HΦs [vs ] (322) We may now try to ﬁnd an approximation to the true total energy of the system by choosing a local potential vs that minimizes the energy expression E1 [vs ].c.e. δϕk (r ) δvs (r) (323) If one works out this expression one obtains equations that are identical to Eqns. So we see that in order to deter(2) mine vxc we have to solve two integral equations. r3 ) δvx (r3 ) δvs (r2 ) (320) Since both Λx and χs are explicitly known in terms of orbitals and orbital energies their derivatives with respect to vs are also explicitly known in terms of these quantities (see reference [44] for more details) and therefore Eq. This means that we have to solve the variational equation 0= δE1 = δvs (r) N d3 r k=1 δE1 δϕk (r ) + c.(320) see reference [34])). The determination of these higher order energies and potentials becomes more involved. Since in our procedure we optimized the energy of a Slater determinant wavefunction under the constraint that the orbitals . Because of the 11correspondence between the density n and the potential vs we can also regard E1 as a functional of the potential vs . It is readily seen that one can write the total energy to order e2 . the perturbation series represents an explicit construction procedure for the exact exchangecorrelation energy and potential.(317) with respect to vs : δΛx (r1 ) = δvs (r2 ) d3 r 3 δχs (r1 . i. r3 ) vx (r3 ) + δvs (r2 ) d3 r3 χs (r1 .(316) and (317). which we denote by E1 [n]. Nevertheless. For realistic systems these equations have only be solved approximately (for an explicit solution (2) of Eq. let us brieﬂy review the ﬁrst order equations from a diﬀerent viewpoint. provided that the series converges. as E1 [n] = = Ts [n] + ˆ Φs HΦs d3 rn(r)v(r) + e2 2 d3 rd3 r n(r)n(r ) + e2 Ex [n] r − r  (321) ˆ This is just the expectation value of the true Hamiltonian H of the system with the KohnSham wave function Φs . Before we go on to discuss the convergence properties of this series. These equations were ﬁrst derived by Talman and Shadwick [45].of Eq.
in the Slater determinant come from a local potential, the method is also known as the Optimized Potential Method (OPM). We have therefore obtained the result that the OPM and and the expansion to order e2 are equivalent procedures. The OPM has many similarities to the HartreeFock approach. Within the HartreeFock approximation one minimizes the energy of a Slater determinant wavefunction under the constraint that the orbitals are orthonormal. One then obtains oneparticle equations for the orbitals that contain a nonlocal potential. Within the OPM, on the other hand, one adds the additional requirement that the orbitals must satisfy singleparticle equations with a local potential. Due to this constraint the OPM total energy E1 will in general be higher than the HartreeFock energy EHF , i.e. E1 ≥ EHF . We refer to [46, 47] for an application of the OPM method for molecules. We ﬁnally make some comments on the calculation of functional derivatives in this section. We stress this point since careless use of the chain rule for diﬀerentiation has led to wrong results in the literature [48]. As an example we consider the exchange functional Ex . When we regard this functional as an explicit functional of the orbitals then the functional derivative with respect to ϕk is given by δEx δϕk (r)
N e
=−
d3 r
i=1
ϕ∗ (r)ϕi (r ) ∗ i ϕk (r ) r − r 
(324)
where we used the subindex e to indicate that we regard the functional as an explicit orbital functional. This functional derivative represents the change in Ex due to a change δϕk in orbital ϕk , while keeping all other orbitals ﬁxed. Moreover, we regard ϕk and ϕ∗ as independent variables. If k we regard Ex as a density functional we must require that all orbitals are eigenfunctions of a noninteracting Hamiltonian with a local potential v s . It is clear that we can we can never ﬁnd a change δvs in a local potential that induces a change in only one orbital while keeping the other orbitals ﬁxed. If a potential changes one orbital ϕk to ϕk + δϕk then all other orbitals will change too. The change in Ex regarded as a functional of vs is then given by the functional derivative δEx δvs (r)
N
=
k=1 N
d3 r d3 r
k=1
δEx δϕk (r ) δEx δϕ∗ (r ) k
e
δϕk (r ) δvs (r) δϕ∗ (r ) k δvs (r) (325)
+
e
We see that in this case the explicit orbital derivatives only occur in the sum in which all the orbital changes must be taken into account. We may
also introduce an implicit derivative (δEx /δϕk )i , which can be given the meaning of giving the change in Ex if we know that there is a potential change δvs that changes ϕk to ϕk + δϕk . This implicit orbital derivative, which we denote by subidex i is expressed in terms of the explicit orbital derivatives as follows δEx δϕk (r)
N
=
i i=1 N
d3 r d3 r
i=1
δEx δϕi (r ) δEx δϕ∗ (r ) i
e
δϕi (r ) δϕk (r) δϕ∗ (r ) i δϕk (r) (326)
+
e
The derivatives δϕi /δϕk appear now to take into account that, if orbital ϕk changes to ϕk + δϕk by some potential change δvs , then the other orbitals ϕi change to ϕi + δϕi . By using the implicit orbital derivative we regard Ex as a functional of the potential, or equivalently of the density, and we can therefore use the chain rule δEx δϕk (r) =
i
d3 r
δEx δn(r )
δn(r ) δϕk (r)
=
i
d3 r vx (r )
δn(r ) δϕk (r)
(327)
i
In this equation (δn/δϕk )i is also an implicit derivative given by: δn(r ) δϕk (r)
N
=
i i=1
d3 r
ϕ∗ (r ) i
δϕi (r ) δϕ∗ (r ) + ϕi (r ) i δϕk (r) δϕk (r)
(328)
which should be compared to the explicit derivative δn(r ) δϕk (r) = ϕ∗ (r) k
e
(329)
for k = 1 . . . N . We stress that Eq.(327) is only true when using the implicit derivatives. If in Eq.(327) we replace the implicit derivatives by the explicit ones of Eq.(324) and Eq.(329) we obtain a result that is wrong. As pointed out this is due to the fact that a change in just one orbital can not be induced by a change in a local potential which in density functional theory is in 11correspondence with the density. Most of the confusion can be avoided by regarding all functionals as functionals of the potential v s and by calculating the change in the functionals by means of perturbation theory. In this way one can avoid use of the implicit orbital derivatives as in Eq.(326). Let us ﬁnally come back to the question of convergence of the perturbation series. The perturbation series presented in this section is very similar to MøllerPlesset perturbation theory starting from the HartreeFock approximation. For the MøllerPlesset perturbation theory it is known that it is
in general divergent [49, 50, 51]. However, it is wellknown that when carried out to low orders this perturbation theory gives reasonable answers. The MøllerPlesset perturbation series has therefore all the features of an asymptotic series [52]. Since the perturbation series in this section is very similar to the MøllerPlesset series it will in general also diverge. This has indeed been found in a perturbation theory on the basis of some approximate KohnSham Hamiltonians [53]. Nevertheless, it has been found for the method presented in this section that low orders in perturbation theory give good results [54] and we conclude that our series give at least an asymptotic expansion for the exchangecorrelation energy and potential.
18.
Outlook and conclusions
In this work we have given on overview of the mathematical foundations of stationary density functional theory. We discussed in great detail the question of diﬀerentiability of the functionals and showed that the KohnSham theory can be put on a solid basis for all practical purposes, since the set of noninteracting EVdensities is dense in the set of interacting EVdensities. The question whether these two sets are in fact identical is still an open question. We further discussed two systematic approaches for the construction of the exchangecorrelation functional and potential. What can we say about future developments within density functional theory? There have been many extensions of density functional theory involving spins, relativistic eﬀects, temperature, superconductivity and timedependent phenomena. The last few years we have, for instance, seen many applications of response properties, rather than ground state properties, using timedependent density functional theory. For these extended density functional theories it is, of course, more diﬃcult to provide a rigorous theoretical basis. This is, however, not particular to density functional theory, but applies to any method that deals with manybody systems, especially if one is interested in phenomena like superconductivity or interactions with laser ﬁelds. Nevertheless, also for these complicated cases simple density functionals have provided encouraging results although there is still a clear need for more accurate density functionals. In view of the success of density functional theory for ground state calculations it seems worthwhile to explore these new areas.
19.
Acknowledgments
I would like to thank Evert Jan Baerends, Pieter Schipper and Oleg Gritsenko for providing me with Fig.4.
Rev 56. Berlin (1995) [21] W.Acad. 2974 (1988) [19] M. 136B.J.Stat.Rall Nonlinear functional analysis and applications Academic Press. A1133 (1965) [9] H. 253 (1983) [17] M.References [1] E.Englisch and R.J.Hellmann. Einf¨hrung in die Quantenchemie.M.Stott Phys.Solidi B123. Stuttgart (1979).Rev. B15. New Delhi (1986) [15] R. New Delhi (1973) .Nat. 5.Math. 14 (1964) [7] H.Kohn and L.De Dominicis and P.Jeggle Nichtlineare Funktionalanalysis Teubner. 373 (1984) [4] G.Temam Convex analysis and variational problems NorthHolland Publishing.Rev. Leipzig (1937) and u R. [14] A.H. McGrawHill.de Barra Introduction to Measure Theory Van Nostrand Reinhold.Englisch Phys.USA 76.Englisch Physica 121A.Chem.Levy Phys.Ekeland and R.P.C.Phys. Amsterdam (1976) [12] M.Englisch and R.Sham Phys.1969 (2001) [16] H.Vainberg Variational method and method of monotone operators in the theory of nonlinear equations John Wiley.Werner Funktionalanalysis. New York. 140.van Leeuwen Int.B.Rev. 340 (1939) [8] W.Sci.Lieb Int. New York (1973) [13] M.Kohn Phys.J.Bose Some topics in nonlinear functional analysis Wiley Eastern. A26.Stat. New Delhi (1985)..Martin J. [10] L. 711 (1984) [3] H.J.Aryasetiawan and M. 243 (1983) [2] H. B38.Englisch Phys.Feynman Phys.Phys. 6062 (1979) [20] D.Rev.Quant.Joshi and R. London (1974) [5] P.Mod.K.Rudin Functional Analysis.Hohenberg and W. 1200 (1982) [18] F..Solidi B124. (1971) [11] I. Springer. 24.C.Englisch and R. 864 (1964) [6] C.Levy Proc.H.Siddiqi Functional analysis with applications McGrawHill.
Svendsen and U.R.Savin Int.Averill and G. 8188 (1998) . 329 (1998) [27] P.Lett.Baerends Theor. 5041 (1987) [39] J.Vignale and M.Rev. 38. ed. 16533 (1996) [42] H.P.Engel and S. A51.Chem.H.Baerends J.Phys.Schmider and A. New York (1997) [41] J.[22] R. K. Springer.F. 99.C.Wang and W.Chem.L.Quant.Burke. 497 (1985) [24] F.Gritsenko and E. 7283 (1995) [37] M.Rev.Perdew. 87. 69. 4641 (1996) [26] P.T.V. 21.B.P. Berlin (1996) [30] Y.Dreizler and E.Chem.Kohn Phys.M.Chem. 111. B54. B54.Wang in Electronic Density Functional Theory: Recent Progress and New Directions eds.Rasolt Phys.Rev.Osaka J.B.J.Rey and A. B42. 109.Chayes.V.Acc.Rev. B54.Das.Phys.Perdew and Y.Vosko Adv.Soc.L.Schipper.Glasser Phys.Rev.Japan 36. L.Painter Phys.Phys.Ullrich and W.Svendsen and U. B36.Nalewajski. 17402 (1996) [35] E.Schipper. O.D.Stat.Langreth and S.Chayes and M.S.Springer.W.J. Berlin (1990) [32] D.S.A.Schwarz J.Engel Phys.Wang Phys.Chem. B15.Vosko Phys.J.P.Rev. Plenum.Rev. New Jersey (1979) [23] J.Burke and Y. B51.J.U.Ruskai J.376 (1974) [31] R. 4056 (1999) [28] C.H.Gross Density Functional Theory: An Approach to the ManyBody Problem Springer.von Barth Phys.F.Quant. 093001 (2001) [29] D.T.R.K. G.Israel Convexity in the Theory of Lattice Gases Princeton University Press.Rev.Dobson.E.Chem.Rev.G. 17392 (1996) [38] M. 175 (1990) [33] E. O. 2498 (1992) [25] S. 581 (1998) [40] K.Gritsenko and E.Geldart in Topics in Current Chemistry 180.J. Princeton. 4940 (1990) [36] M.H.R.T.S. P. 105.Phys.W.Becke J.Phys. 31. 1159 (1995) [34] P.von Barth Phys. J.
86.Shadwick Phys.Colle Phys.Bartlett Phys. McGrawHill. Malmqvist Chem.Engel.J. 83.G¨rling Phys.Levy Phys. 5459 (1999) o [48] R.Rev. P. 5455 (1999) [47] A.FaccoBonetti. 112.K.Olsen.Rev.Comp.Jørgensen.Warken Chem.Rev.Dreizler . 9736 (2000) [51] F. 256 (1995) [54] A. 20.Jørgensen.Rev.M.A.Engel and R.Lett.Olsen.Lett.Helgaker and O.Rev. 012503 (1999) ˚ [49] O. A61. A50.Chem. A14. 237.Dreizler J. 31 (1999) [45] J.Phys. Phys.Christiansen. 112. J. 36 (1976) [46] S.Lett. 83.F. S.Phys.Lett.Stillinger J.Talman and W.Nesbet and R.G¨rling and M.Lett.Schmid and R.Ivanov.N.M.Christiansen J. P.Chem. 9711 (2000) [52] C.Rev.A. Singapore (1978) [53] M.Phys.Bender and S.Orszag Advanced Mathematical Methods for Scientists and Engineers.[43] A. 2241 (2001) R. 196 (1994) o [44] E. 369 (1996) [50] J. H.Phys. T.M. E.Koch and P.Hirata and R. 261.H.D.Chem.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.