IIb.

The Hartree-Fock Approximation:

The Hartree-Fock approximation underlies the most commonly used method in chem-
istry for calculating electron wave functions of atoms and molecules. It is the best approx-
imation to the true wave function where each electron is occupying an orbital, the picture
that most chemists use to rationalize chemistry. The Hartree-Fock approximation is, fur-
thermore, the usual starting point for more accurate calculations that can, in principle,
become exact.
It is most convenient to use ’atomic units’ in calculations of electronic wave functions
for atoms and molecules. The unit chosen for
length is the Bohr radius a0 ,
mass is the electron mass me ,
charge is the electron charge e,
energy is the Hartree = 27.211 eV = 2 EI ,
where EI is the energy of the ground state of the hydrogen atom with respect to separated
electron and proton. In these units, h̄ becomes unity.
The full Hamiltonian for a system of N electrons in the presence of M nuclei with
charge ZA then becomes

N N XN
exact
X X 1
H = h(i) +
i i j>i
rij

where
M
1 X ZA
h(i) ≡ − ∇2i − .
2 riA
A

and riA is the distance between nucleus A and electron i.

Solving the Schrödinger equation with this Hamiltonian is very difficult because the
1/rij terms correlate the distribution of all the electrons. As is frequently done with such
many body problems, we will seek a mean field approximation, where each electron is
treated separately and the effect of all the other electrons is included in an average way.
The way this is done is to carry out a variational calculation where the trial function is of
the form of a single Slater determinant

|ψ0 >= |χ1 χ2 . . . χN > .

That is, the expectation value of the energy

E0 = < H > = < ψ0 |H|ψ0 >

is minimized with respect to the determinantal trial function |ψ0 >. In the process an
optimal single determinant approximation to the stationary state wave function is obtained.
Such a form of the function can, however, only be exact if the Hamiltonian separates into
terms where each one only depend on one of the electron coordinates. Since the true

32
Hamiltonian does contain terms of the form 1/rij , this procedure can never yield an exact
solution. But, it is a useful first approximation.
One can better understand the approximation being made here by thinking of the
resulting eigen function as an exact solution to a different problem, one where the Hamil-
tonian is an approximation to the true Hamiltonian

N 
X 
H app = h(i)+ viHF (i) = H1 + H2 + . . . HN .
i

Here, viHF (i) is an effective potential experienced by the i − th electron due to the presence
of the other electrons, but it cannot depend on the coordinates of the other electrons and
thus represents a spatially averaged interaction. During the variational optimization of the
spin-orbitals an optimal effective interaction viHF (i) is obtained as well as the stationary
state wave functions. Since the approximate Hamiltonian H app separates, its eigenfunc-
tions can indeed be written as a Slater determinant formed from spin-orbitals

|ψ0 >= |χ1 χ2 . . . χN > .

The variational minimization of the energy with respect to arbitrary variations of the
spin-orbitals leads to equations for the spin-orbitals and the optimal, effective potential.
The derivation of these equations, called the Hartree-Fock equations is given below. The
expectation value of the Hamiltonian for a Slater determinant wave function was previously
found to be
N N N
X 1 XX
E0 = [a|h|a] + [aa|bb] − [ab|ba]
a
2 a
b

where the summation indices a and b range over all occupied spin-orbitals. In searching
for the optimal wave function, we must impose the constraint that all the spin-orbitals
remain orthonormal, i.e.
[a|b] − δab = 0

for a = 1, 2, . . . , N and b = 1, 2, . . . , N , a total of N 2 constraints.

The standard method for finding an extremum (minimum or maximum) subject to
a constraint is Lagrange’s method of undetermined multipliers: The constraint equations
are each multiplied by some constant and added to the expression to be optimized. Thus,
we define a new functional L:

N X
X N  
L ≡ E0 − ǫba [a|b] − δab .
a b

When the constraints are satisfied, this new quantity equals the expectation value of the
Hamiltonian, E0 . The unknown constants ǫba are the Lagrange multipliers. The quantity
L (as well as E0 ) is a functional of the spin-orbitals χa , χb , . . . , χN and the problem is

33
to find stationary points of L. That is, given infinitesimal change in the spin-orbitals,
χa → χa + δχa , the change in L, (L → L + δL), should be zero, i.e.:
N X
X N
0 = δL = δE0 − ǫba δ[a|b] .
a=1 b=1

We now evaluate the terms on the right hand side of this expression. By inserting the
new spin-orbitals χa + δχa , etc. into the expression for E0 , and using the fact that the
integration indicated by [ ] is a linear operation, the change in E0 is to first order:
N
X
δE0 = ([δχa |h|χa ] + [χa |h|δχa ])
a=1
N N
1 X X n
+ [δχa χa |χb χb ] + [χa δχa |χb χb ] + [χa χa |δχb χb ] + [χa χa |χb δχb ]
2 a=1
b=1
o
− [δχa χb |χb χa ] − [χa δχb |χb χa ] − [χa χb |δχb χa ] − [χa χb |χb δχa ] .

From the definition of the integrals it is clear that [δχa |h|χa ]∗ = [χa |h|δχa ] and [δχa χa |χb χb ]∗ =
[χa δχa |χb χb ], etc. Furthermore, [δχa χa |χb χb ] = [χb χb |δχa χa ] as can be seen by relabel-
ing the integration variables representing the electron coordinates. The change in E0 can
therefore be rewritten as:
N
X N X
X N
δE0 = [δχa |h|χa ] + [δχa χa |χb χb ] − [δχa χb |χb χa ] + c.c.
a=1 a=1 b=1

The notation c.c. stands for complex conjugate.

Using the factor rule of differentiation on the second term in the expression for δL

δ[a|b] = δ[χa |χb ] = [δχa |χb ] + [χa |δχb ]

gives X X X
ǫba δ[χa |χb ] = ǫba [δχa |χb ] + ǫba [χa |δχb ] .
ab ab ab

Interchanging the summation indices a and b in the second sum on the right hand side
gives: X X X
ǫba δ[χa |χb ] = ǫba [δχa |χb ] + ǫab [χb |δχa ] .
ab ab ab

L is a real quantity and by taking the complex conjugate of the expression defining L, it
can be shown that ǫba = ǫ∗ab , that is the Lagrange multipliers are elements of a Hermitian
matrix. This means the second sum is just the complex conjugate of the first, and we have
X X
ǫba δ[χa |χb ] = ǫba [δχa |χb ] + c.c..
ab ab

34
 Finally, the expression for δL becomes:
N
X X N n
N X o
δL = [δχa |h|χa ] + [δχa χa |χb χb ] − [δχa χb |χb χa ] − ǫba [δχa |χb ] + c.c.
a=1 a=1 b=1

In this expression we have [δχa appearing on the left hand side of each term. We can
symbolically rewrite
N N
!
X X
δL = [δχa |h|χa ] + {χa |χb χb ] − χb |χb χa ] − ǫba |χb ]} + c.c.
a=1 b=1

More explicitly, the expresssion for δL is

N Z
X  N
X 


δL = d~x1 δχ∗a h(1)χa (1) + { Jb (1) − Kb (1) χa (1) − ǫba χb (1)} + c.c.
a=1 b=1

where we have defined two new operators, J and K. The Coulomb operator, Jb , is defined
as Z
1
Jb (1) ≡ d~x2 |χb (2)|2
r12
such that Z 
1
Jb (1)χa (1) = d~x2 χ∗b (2) χb (2) χa (1)
r12
and, in particular we have
Z
d~x1 χ∗a (1)Jb (1)χa (1) = [aa|bb] .

The exchange operator, Kb (1), is defined such that

Z 
∗ 1
Kb (1)χa (1) ≡ d~x2 χb (2) χa (2) χb (1) .
r12

Note how the labels a and b on spin-orbitals for electron 1 get interchanged. In particular,
we have Z
d~x1 χ∗a (1)Kb (1)χa (1) = [ab|ba] .

Note that the exchange operator is a non-local operator in that there does not exist
a simple potential function giving the action of the operator at a point ~x1 . The result of
operating with Kb (1) on χa (1) depends on χa throughout all space (not just at ~x1 ).
Now set δL = 0 to obtain the optimal spin-orbitals. Since δχ∗a is arbitrary, we must
have " #
N
X N
X
h(1) + {Jb (1) − Kb (1)} χa (1) = ǫba χb (1)
b=1 b=1

35
for each spin-orbital χa with a = 1, 2, . . . , N . Defining the Fock operator as

N
X
f (1) ≡ h(1) + {Jb (1) − Kb (1)} ,
b

the solution to the optimization problem, i.e. the optimal spin-orbitals, satisfy

N
X
f χa = ǫba χb .
b=1

This equation can be diagonalized, i.e., we can find a unitary transformation of the
spin-orbitals that diagonalizes the matrix ǫ which has matrix elements ǫba . The Fock
operator is invariant under a unitary transformation, that is a transformation where a new
set of spin-orbitals is defined by taking a linear combinations. The new set of spin-orbitals
is defined as X
χ′a = χb Uba
b

where U † = U −1 such that ǫ̃′ = U † ǫ̃ U is diagonal. Then

f χ′a = ǫ′a χ′a .

This is the Hartree-Fock equation, a one electron equation for the optimal spin-orbitals. It
is non-linear, since the Fock operator, f , itself depends on the spin-orbitals χa .

Occupied and Virtual Orbitals:

From the Hartree-Fock equation we get a set of spin-orbitals (dropping the primes
now):
f χj = ǫj χj j = 1, 2, . . . , ∞.

By solving this equation we can generate an infinit number of spin-orbitals. The Fock op-
erator, f , depends on the N spin-orbitals that have electrons, the occupied orbitals. Those
will be labeled with a, b, c, .... Once the occupied orbitals have been found, the Hartree-
Fock equation becomes an ordinary, linear eigenvalue equation and an infinit number of
spin-orbitals with higher energies can be generated. Those are called virtual orbitals and
will be labeled with r, s, ....

The orbital energy

What is the significance of the orbital energies ǫi ? Left multiplying the Hartree-Fock
equation with < χi | gives

< χi |f |χj >= ǫi < χi |χj >= ǫj δij .

36
Therefore
ǫi = [χi |f |χi ]
N
X
= [χi |h + (Jb − Kb )|χi ]
b
X
= [i|h|i] + [ii|bb] − [ib|bi]
b

where the summation index, b, runs over all occupied spin-orbitals.

The first term < i|h|i > is a one body energy, the electron kinetic energy and the
attractive interaction with the fixed nuclei. The second term, the sum over all occupied
spin-orbitals, is a sum of two body interactions, the Coulomb and exchange interaction
between electron i and the electrons in all occupied spin-orbitals. The total energy of the
system is not just the sum of ǫi for all occupied orbitals, because then the pairwise terms
would be double counted. Recall the expression for E0 :

N N N
X 1 X X X
E0 = [a|h|a] + [aa|bb] − [ab|ba] 6= ǫa .
a
2 a a
b

The factor 1/2 prevents double counting the two electron integrals.
The significance of the ǫi becomes apparent when we add or subtract an electron to
the N electron system. If we assume the spin-orbitals do not change when we, for example,
remove an electron from spin-orbital χc , then the determinant describing the N −1 electron
system is
|N−1 ψc >= |χ1 χ2 . . . χc−1 χc+1 . . . χN >
with energy
N−1
Ec =<N−1 ψc |H|ψcN−1 >
X 1 X X
= [a|h|a] + [aa|bb] − [ab|ba].
2
a6=c a6=c b6=c

The energy required to remove the electron, which is called the ionization potential, is:

IP =N−1 Ec − E0
N N
1 X X 
= −[c|h|c] − [cc|bb] − [cb|bc] + [aa|cc] − [ac|ca] .
2 a
b

We do not need to restrict the summation to exclude c since the [cc|cc] term cancels out.
Using the fact that [ac|ac] = [ca|ca] this can be rewritten as

N
X
IP = −[c|h|c] − [cc|bb] − [cb|bc]
b
= −ǫc .

37
So, the orbital energy is simply the ionization energy.
Similarly, after adding an electron to the N -electron system into a virtual orbital χr ,
the state is
|N+1 ψr >= |χ1 χ2 . . . χN χr >
and the energy is
N+1
Er = <N+1 ψr |H|ψrN+1 > .
The energy difference is called the electron affinity, EA. Assuming the spin-orbitals stay
the same, we have
EA = E0 − N+1Er = − ǫr .

Koopman’s Rule:
The orbital energy ǫi is the ionization potential for removing an electron from an
occupied orbital χi or the electron affinity for adding an electron into virtual orbital χi ,
in either case assuming the spin-orbitals do not change when the number of electrons is
changed. This is a remarkably good approximation due apparently to cancellations of
corrections due to adjustments in the orbitals.

Restricted Hartree-Fock:
For computational purposes, we would like to integrate out the spin functions α and β.
This is particularly simple when we have spatial orbitals that are independent of spin, in
the sense that a given spatial orbital can be used twice, once for spin up and once for spin
down. For example, from a spatial orbital ψa we can generate two orthogonal spin-orbitals
χ1 and χ2 :
χ1 (~x) = ψa (~r)α(ω)
χ2 (~x) = ψa (~r)β(ω) .
Determinants constructed from such spin-orbitals are called restricted determinants.

Transition from Spin Orbitals to Spatial Orbitals:

The restricted determinant can be written as

|ψ > = |χ1 χ2 χ3 . . . χN−1 χN >

= |ψ1 ψ̄1 ψ2 ψ̄2 . . . ψN/2 ψ̄N/2 >

where the ψi denote spatial orbitals occupied by a spin-up electron and ψ̄i denote the same
spatial orbitals occupied by a spin-down electron.
The energy of a determinantal wave function is

N N N
X 1 X X
E =< ψ|H|ψ >= [a|h|a] + [aa|bb] − [ab|ba].
a
2 a
b

38
We will, furthermore, assume here that all the electrons are paired (closed shell). The
wave function then contains N/2 spin orbitals with spin up and N/2 spin orbitals with
spin down, and we can write:

N N/2
X X
χa = (ψa + ψ̄a ).
a a

Any one electron integral involving spin-orbitals

R ∗ with opposite spin vanishes because of
the orthogonality of the spin functions, α β dω = 0. For example,

[ψi |h|ψ̄j ] = [ψ̄i |h|ψj ] = 0 .

R
Since the spin functions are normalized, |α|2 dw = 1, the integration over spin does not
affect the value of non-vanishing matrix elements. We therefore define yet another notation
for matrix elements
(ψi |h|ψj ) ≡ [ψi |h|ψj ] = [ψ̄i |h|ψ̄j ].
The round brackets indicate spatial integration only. Spin has already been integrated out.
Similarly, for two electron integrals:

[ψi ψj |ψk ψℓ ] = [ψi ψj |ψ̄k ψ̄ℓ ]

= [ψ̄i ψ̄j |ψk ψℓ ]
= [ψ̄i ψ̄j |ψ̄k ψ̄ℓ ]
≡ (ψi ψj |ψk ψℓ ).

Any two electron integral with only one bar on either side vanishes, for example:

[ψi ψ̄j |ψk ψl ] = [ψ ψ̄j |ψk ψ̄l ] = 0.

The energy for a single determinant wave function becomes:

N/2
X
E=2 (ψa |h|ψa )
a
N/2 N/2
X X
+ 2(ψa ψa |ψb ψb ) − (ψa ψb |ψb ψa )
a b
X X
=2 (a|h|a) + 2(aa|bb) − (ab|ba)
a ab

with the summation being over the spatial orbitals only. The first type of two electron
integrals, Jij ≡ (ii|jj), is called the Coulomb integral since it represents the classical
Coulomb repulsion between the charge clouds |ψi (~r)|2 and |ψj (~r)|2 . The second type,
Kij ≡ (ij|ji), is called exchange integral and does not have a classical interpretation but

39
arises from the antisymmetrization of the wave function. It results from the exchange
correlation. The energy of two electrons in orbitals ψ1 and ψ2 is

E(↑↓) = h11 + h22 + J12

if their spin is antiparallel, but

E(↑↑) = h11 + h22 + J12 − K12

if their spin is parallel. The energy is lower when the spin is parallel (K12 > 0) because
the antisymmetrization prevents the electrons from being at the same location.

In summary: Given a determinantal wave function, the energy can be obtained in

the following way:
(1) each electron in spatial orbital ψi contributes hii to the energy,
(2) each unique pair of electrons contributes Jij (irrespective of spin),
(3) each pair of electons with parallel spin contributes −Kij .

Restricted Hartree-Fock equation

Using the above expression for the energy, the Hartree-Fock equation becomes:

f (1)ψj (1) = ǫj ψj (1)

where the Fock operator can now be expressed as:

N/2
X
f (1) = h(1) + 2Ja (1) − Ka (1)
a

and the restricted Coulomb and exchange operators are:

Z
1
Ja (1) = d~r2 ψa∗ (~r2 ) ψa (~r2 )
r12

and Z 
1
Ka (1)ψi (1) = d~r2 ψa∗ (~r2 ) ψi (~r2 ) ψa (~r1 ) .
r12
The total energy of the system can be written as:

N/2 N/2 N/2

X X X
E=2 (a|h|a) + 2(aa|bb) − (ab|ba)
a a b
N/2
X X X
=2 haa + 2Jab − Kab
a a b

40
and the orbital energies are:

N/2 N/2
X X
ǫi = (i|h|i) + 2(ii|bb) − (ib|bi) = hii + 2Jib − Kib
b b

All these expresssions are in terms of the spatial orbitals only, there is no explicit reference
to spin.

The Roothaan Equations:

The spatial Hartree-Fock equation:

f (~r1 )ψi (~r1 ) = ǫi ψi (~r1 )

can be solved numerically for atoms. The results of such calculations have been tabulated
by Hermann and Skilman. However, for molecules there is no practical procedure known
for solving the equation directly (recall f depends on the orbitals) and the orbitals ψi are
instead expanded in some known basis functions φµ :

K
X
ψi = Cµi φµ i = 1, 2, . . . , K.
µ

If the number of basis functions is K, we can generate K different orbitals. If the set {φµ }
is complete the results would be the same as a direct numerical solution to the Hartree-
Fock equation. But, for practical reasons the set {φµ } is always finite and therefore not
complete and some error is introduced by expanding ψi . This is called the basis set error.
The problem now is reduced to determining the expansion coefficients Cµi . Inserting
the expansion into the Hartree-Fock equation gives
X X
f (1) Cγi φγ (1) = ǫi Cγi φγ (1).
γ γ

Left multiplying with φ∗µ (1) and integrating gives:

X Z X Z
Cγi d~r1 φ∗µ (1)f (1)φγ (1) = ǫi Cγi d~r1 φ∗µ (1)φγ (1)
γ | {z } γ | {z }
≡ Fµγ ≡ Sµγ
the Fock matrix the overlap matrix
X X
Fµγ Cγi = ǫi Sµγ Cγi
γ γ

F̃ C̃ = S̃ C̃ǭ.

41
This is a matrix representation of the Hartree-Fock equation and is called the Roothaan
equations. The matrices F̃ , S̃ and C̃ are K × K matrices and ǭ is a vector of length K.
The problem is therefore reduced to solving algebraic equations using matrix techniques.
Only if K → ∞ are the Roothan equations equivalent to the Hartree-Fock equation.
The Roothaan equations are non-linear:

F̃(C̃) C̃ = S̃ C̃ǭ.

Since F̃ depends on C̃ and therefore they need to be solved in an iterative fashion. Given
an estimate for C̃n we can find an estimate for F̃(C̃n ) and then solve the equation

F̃(C̃n ) C̃n+1 = S̃ C̃n+1 ǭ

to obtain a new estimate for the C̃ matrix. If C̃n+1 = C̃n to within reasonable tolerance
then self consistency has been achieved and C̃n is the solution. Most workers refer to such
a solution as self-consistent-field (SCF) solution for any finite basis set {φi } and reserve
the term Hartree-Fock limit to the infinite basis solution. The equation is solved at each
step by diagonalizing the overlap matrix S̃, i.e., by finding a unitary transformation X̃
such that
X † SX = 1.
The transformed basis function are
X
φ′µ = Xγµ φγ µ = 1, 2, . . . , K
γ

and form an orthonormal set, i.e.,

Z
d~rφ′∗ ′
µ φγ = δµγ .

Then the equation becomes an ordinary eigenvalue equation:

F ′C ′ = C ′ǫ where F ′ ≡ X † F X and C ′ ≡ X −1 C.

The main computational effort in doing a large SCF calculation lies in the evaluation of
the two-electron integrals. If there are K basis functions then there will be on the order
of K 4 /8 unique two-electron integrals. This can be on the order of millions even for small
basis sets and moderately large molecules. The accuracy and efficiency of the calculation
depends very much on the choice of basis functions, just as any variational calculation
depends strongly on the choice of trial functions.

Basis Set Functions:

Two types of basis functions are frequently used:

42
 ~ A has the form
(1) Slater type functions, which for a 1S orbital centered at R

p ~
φSF ~ A) = ζ/π e−ζ|~r−RA |
1S (ζ, ~
r−R

with ζ a free parameter and

(2) Gaussian type function

 3/2
~ A) = 2α ~ 2
φGF
1S (α, ~
r −R e−α|~r−RA |
π

with α a free parameter.

The Slater type functions have a shape which matches better the shape of typical
orbital functions.In fact, the wave function for the hydrogen atom is a Slater type function
with ζ = 1. More generally, it can be shown that molecular orbitals decay as ψi ∼ e−ar
just like Slater type functions and at the position of nuclei |~r − R ~ A | → 0 there is a cusp
because the potential −e/|r − RA | goes to −∞.
Gaussian type functions have zero slope at |~r − R ~ A | = 0 (i.e., no cusp) and decay
much more rapidly than Slater functions. Since Slater type functions more correctly de-
scribe qualitative features of molecular orbitals than Gaussian functions, fewer Slater type
functions are needed to get comparable results. However, the time it takes to evaluate the
integrals over Slater function is much longer than for Gaussian functions. The two electron
~ A, R
integrals can involve four different centers R ~ B, R
~ C and R ~ D which makes the evaluation
of integrals over Slater functions very time consuming. The product of two Gaussians, on
the other hand, is again a Gaussian

φGF ~ p)
~ B ) = KAB φGF (p, ~r − R
~ A ) φGF (β, ~r − R
1S (α, ~
r−R 1S 1S

where the new Gaussian is centered at

~ ~
~ p = αR A + β R B .
R
α+β

A common practice is to choose basis functions φµ that are constructed from a few Gaus-
sians
XL
CGF ~
φµ (γ, ~r − RA ) = dpµ φGF ~ A)
p (αpµ , ~
r−R
p=1

in such a way as to mimic (in a least squares sense) a Slater function. Those are called con-
tracted Gaussian functions and a standard notation for such basis functions is ST O − N G,
meaning Slater Type Orbital constructed from N Gaussians. A typical value for N is 3,

43
 i.e. three gaussians are used in each orbital.

Figure II.2 Representation of the 1s orbital of the hydrogen atom with a linear combination
of three Gaussians.

In a more flexible basis set called 6-31G, the core electrons are represented by a single
Slater type orbital which is described by six Gaussians (contracted) while valence electrons
are represented by two Slater type orbitals, one described by three Gaussians (contracted)
and the other described by a single Gaussian. When an atom is placed in an external
field, the electron cloud is distorted (polarized). To describe this, it is necessary to include
also excited atomic orbitals, i.e. orbitals which are not occupied in the ground state. In
the 6-31G∗∗ basis set, excited atomic orbitals are included for all atoms (for example d-
orbital functions for O atoms), while in the 6-31G∗ basis set, the excited atomic orbitals
are included for all elements but H atoms. It turns out that H atoms are hard to polarize
so it is often a good approximation to only include polarization of the heavier atoms.

The main computational effort in doing a large SCF calculation lies in the evaluation
of two-electron integrals. If there are K basis functions then there will be on the order of
K 4 two-electron integrals. This can be on the order of millions even for small basis sets
and moderately large molecules. The accuracy and efficiency of the calculation depends
very much on the choice of basis functions, just as any variational calculation depends
strongly on the choice of trial functions.

The results of an ST O − 3G calculation for H2 using restricted Hartree- Fock is

shown in the figure below The limit of large r is not reproduced correctly because H2 does
not dissociate into two closed shell fragments. In restricted Hartree-Fock the dissociation

44
 products incorrectly include H − and H + .

Figure II.3 Energy of a H2 molecule as a function of the distance between the atoms. The
Restricted Hartree-Fock approximation using STO-3G basis set is shown as well
as a more accurate calculation based on the configuration interaction method (see
below).
The Charge Density: In a system with paired electrons, the electron density, i.e., the
probability of finding an electron in a volume element d~r around a point ~r is
N/2
X
ρ(~r)d~r = 2 |ψa (~r)|2 d~r.
a

Because the orbitals are orthogonal, the total charge density is just a sum of charge densities
for each of the accupied orbitals. The integral is
Z N/2 Z N/2
X X
2
d~rρ(~r) = 2 dr|ψa (~r1 )| = 2 1=N
a a

the total number of electrons.

Configuration Interaction:
Recall that the Hartree-Fock solution does not include any correlation in the motion
of electrons with opposite spins because of the approximate treatment of the 1/r12 in-
teraction. However, the ‘exact’ solution, i.e., the solution to the Hamiltonian H exact can

45
be obtained from the orbitals generated in the Hartree-Fock procedure because they from
a complete set. Note that this ‘exact’ solution is still approximate because it involves
the non-relativistic approximation and the Born-Oppenheimer approximation. When K
spatial basis functions are used, 2K spin orbitals are generated in the Hartree-Fock calcu-
lation. The best estimate of the Hartree-Fock ground state is a single Slater determinant
generated from the N spin orbitals with the lowest energy:

|ψ0 >= |χ1 χ2 . . . χa χb . . . χN > .

A singly excited determinant is one with an electron in a virtual orbital, for example χr
rather than χa :

|ψar >= |χ1 χ2 . . . χr χb . . . χN >

and a doubly excited determinant is, similarly:

rs
|ψab >= |χ1 χ2 . . . χr χs . . . χN > .

A total of
 
2K (2K)!
=
N N !(2K − N )!

determinants can be formed. We can consider these determinants as a set of N electron

basis functions which we use to expand the ‘exact’ wave function

X X X X X X
|Φ >= C0 |ψ0 > + Car |ψar > + rs rs
Cab |ψab > + . . . .
r a a b>a r s>r

The first term is the Hartree-Fock approximation. Since each term can be thought of as
a specific configuration the procedure is called configuration interaction (CI). In the limit
of infinit basis functions, K → ∞, the first term |ψ0 > reaches the Hartree-Fock limit
with energy E0 and the set of determinants becomes a complete set so |Φ > becomes the
‘exact’ wave function with energy ǫ0 (not quite exact, really, because it still involves non-
relativistic and Born-Oppenheimer approximations). The correlation energy is defined to
be

Ecorr ≡ ǫ0 − E0 .

2K


For any finite number of basis functions, K, the limit of ‘full CI’ means that all N

46
 determinants are used to in the representation of |Φ >.

Figure II.4 A diagram illustrating how a calculation of the wave function of a many-electron
system converges with respect to the number of basis functions and the number
Slater determinants in the configuration interaction calculation (i.e. the level of
correlation included in the theoretical method. Only when both the number of basis
functions and the number of determinants used in the trial function have reached
a sufficient level is convergence reached in the variational calculation.

47