10 Self-consistent field theory

An important unsolved problem in quantum mechanics is how to deal with indistinguishable, in-
teracting particles - in particular electrons which determine the behaviour of almost every object
in nature. The basic problem is that if particles interact, that interaction must be in the Hamil-
tonian. So until we know where the particles are, we can’t write down the Hamiltonian, but until
we know the Hamiltonian, we can’t tell where the particles are.

10.1 Hartree-Fock theory

The idea is to solve the Schroedinger equation for an electron moving in the potential of the nucleus
and all the other electrons. We start with a guess for the trial electron charge density, solve Z/2
one-particle Schroedinger equations (initially identical) to obtain Z electron wavefunctions. Then
we construct the potential for each wavefunction from that of the nucleus and that of all the other
electrons, symmetrise it, and solve the Z/2 Schroedinger equations again.
Fock improved on Hartree’s method by using the properly antisymmetrised wavefunction (Slater
determinant) instead of simple one-electron wavefunctions. Without this, the exchange interac-
tion is missing. This method is ideal for a computer, because it is easily written as an algorithm.

Guess Wavefunction Calculate Charge Density Calculate Potential

Solve Schroedinger equation

Calculate Charge density

Is charge density
No Same as before?

Yes

STOP

Figure 5: Algorithm for Self-consistent field theory.

Although we are concerned here with atoms, the same methodology is used for molecules or even
solids (with appropriate potential symmetries and boundary conditions). This is a variational
method, so wherever we refer to wavefunctions, we assume that they are expanded in some ap-
propriate basis set.
The full set of equations are
µ ¶ XZ 0 2 Z ∗ 0 0
0 ψj (r )ψi (r )
1 2 X
0 |ψj (r )|
²i ψi (r) = − ∇ + Vion (r) ψi (r) + dr ψ i (r) − δσ σ
i j
dr ψj (r) (3)
2 j |r − r0 | j |r − r0 |

The first term is the kinetic energy and electron-ion potential. The second “Hartree” term, is
the electrostatic potential from the charge distribution of N electrons, including an unphysical
self-interaction of electrons when j = i. The third, “exchange” term, acts only on electrons with
the same spin and comes from the Slater determinant form of the wavefunction.
Physically, the effect of exchange is for like-spin electrons to avoid each other. Each electron is
surrounded by an “exchange hole”: there is one fewer like-spin electrons nearby than the mean-field
would imply. The term i = j neatly cancels out the self interaction of the electron.
10.2 Self-consistent fields

Iterative, self-consistent approaches similar to the Hartree-Fock method can be used to calculate
properties of atoms, solids or molecules. All that changes is Vion .
For non-central potentials appropriate boundary conditions are needed (e.g. periodic in the case
of crystals). One of the main problems now is to select an appropriate basis set for the problem.
Various options exist: Plane waves, atomic orbitals, ‘augmented’ plane waves which wiggle more
near to the nuclei, gaussian or ‘muffin tin’ orbitals which are localised on the nuclei. There is still
a huge amount of research going on in this area.

10.3 Correlation: conditional probability

Hartree-Fock theory does not properly describe correlation. In the Copenhagen Interpretation,
the squared modulus of the wavefunction gives the probability of finding a particle in a given place.
The many-body wave function gives the N-particle distribution function, i.e. |Φ(r1 , ..., rN )|2 is the
probability density that particle 1 is at r1 , ..., and particle N is at rN .
But when trying to work out the interaction between electrons, what we want to know is the
probability of finding an electron at r, given the positions of all the other electrons {ri }. This
implies that the electron behaves quantum mechanically when we evaluate its wavefunction, but
as a classical point particle when it contributes to the potential seen by the other electrons.

10.4 Lattice methods: Variational method by computer

The variational method transposes the problem of solving a differential equation onto the problem
of minimising a function of many variables. It is therefore good for use with computers.
One of the simplest ways of solving for the ground state of a system with a computer is to use
a basis set consisting of the values of |φi defined on a lattice. In 1D such a solution is simply a
histogram where we adjust the wavefunction at each point until the energy of the whole system
is minimised. The kinetic energy (second derivative of the wavefunction) must then be obtained
by some interpolation method. The weights of |φi at each point can be regarded as a basis set. It
is not complete, but it becomes more and more complete as the lattice gets finer.
Another common way of solving the Schroedinger equation numerically is to write the wavefunction
as a Fourier series. X
Φ(b) = ab eib.r
b
where ab are the variational parameters. Using Fourier series (also called plane wave expansions)
has several advantages. Increasing accuracy can be obtained by adding more Fourier components
(because each plane wave is orthogonal to all the others), the value of Φ(r) can be quickly found
by a Fourier transform of a(k) and the kinetic energy has particularly simple form because
R P 2 h̄2 b2
−h̄2 Φ∗ ∇2 Φd3 r b ab 2m
R = P 2
2m Φ∗ Φd3 r b ab

which requires no numerical differentiation if used on a computer. The wavefunctions must be

10.5 Pseudopotentials

A drawback to using plane waves is that electron wavefunctions don’t actually look much like
plane waves, so the basis set is very different from the wavefunctions, and very many Fourier
components are required. One way around this is to use a ‘pseudopotential’ which attempts to
describe the potential due to the nucleus and tightly bound shells of ‘core’ electrons which do not
take part in bonding. In silicon for example the pseudopotential describes the nucleus and the
1s2s2p electrons.
The pseudopotential can be deduced from properties of the perfect atom: Consider:
h̄2 ∇2
V (r)Φ(r) = E + Φ(r)
2m
Where we know atomic properties E and Φ(r), but not V (r), the potential seen by the outer
electrons. We can invert the Schroedinger problem, solving for V (r) to give the exact Φ(r) outside
some core radius r > rc , but smoothing it out for r < rc .
In most applications involving chemical binding, the wavefunction only changes in the region
outside rc . So although the pseudowavefunction is not the correct Kohn-Sham eigenfunction,
changes in its energy due to interaction with other electrons and ions are the same as the change
in the Kohn-Sham eigenfunction.
Choosing rc and inverting the Schroedinger equation is non-unique, but in general:
Pseudopotentials depend on the l quantum number, because they must include the fact that, e.g.
3s must be radially orthogonal to 1s and 2s, while 3d are automatically so because of the angular
dependence. This is called non-locality.
The core charge produced by the pseudo wavefunctions must be the same as that produced by the
atomic wavefunctions. This ensures that the pseudo atom produces the same scattering properties
as the ionic core.
Pseudo-electron eigenvalues must be the same as the valence eigenvalues obtained from the atomic
wavefunctions.
Pseudo wavefunctions must be continuous at the core radius as well as its first and second derivative
and also be non-oscillatory.
If you find it surprising that this works - it is! However tens of thousands of calculations give
energies correct to within a few percent, so the approach seems to accord well with reality.

10.6 k-point sampling

DFT reduces the problem of 1026 interacting electrons to 1026 noninteracting quasiparticles. To
reduce this to a manageable number, we recall that electrons in solids can be labelled by a
wavevector k, and that they form bands in which electrons with similar k have similar energy.
The energy is the integral of these, thus we can obtain a good estimate by sampling states from
an evenly-spaced grid of “k-points”. As this grid becomes finer, so the accuracy of the integral
improves. For most systems a surprisingly small number suffices: tens for insulators and hundreds
for metals.
According to the Bloch theorem, any wavefunction must be written:

Φk = u(r) exp −ik.r

If the wavefunction is expanded in plane waves, then

where k correspond to Bloch waves longer than the unit cell, and b to basis function plane waves
shorter than the cell (i.e. b > k).
10.7 A continuum of quantum states: quantum numbers in a crystal

In a crystal quantum states can be indexed by the Bloch quantum number k. In the LCAO
approximation, there is a state for each possible atomic orbital at each value of k. As the number
of electrons tends to infinity, the allowed k’s form a continuum. The most important application
of quantum mechanics in solid state physics is to understand the relationship between energy and
momentum. A graph of energy vs momentum is called a band structure.
States are occupied from the lowest energy upwards according to the exclusion principle. The set
of momenta which correspond to the maximum allowed energy form a surface in the 3-d space -
the so-called Fermi Surface.
Shown is the valence “band structure” of dhcp potassium calculated using DFT and pseudopo-
tentials: letters are crystallographic notation for values of k (Γ = (0, 0, 0), others are on the edge
of the Brilloiun zone). Note the free electron parabola around Γ, as E = h̄2 k 2 /2m. This structure
has four layers of atoms per unit cell, so on average there are two bands below the Fermi surface at
each k-point (each is spin degenerate). There are lots of bands crossing the Fermi level, showing
that electrons can move from one state to another without requiring energy: potassium is a metal.
Γ-A is quite a short distance in k-space, corresponding to waves along the long direction in the
unit cell: the band structure appears like a parabola “folded back” on itself.

2
E (eV)

-2

-4

Γ M K ΓA HKH LML A

Figure 6: Band structure of potassium, energy scaled so that EF =0. x-axis labels denote a path
through the 3d space of k-vectors.