DFT Density Functional Theory

1. Basic quantum mechanical concepts

The aim of DFT is to calculate the ground state energy of an electronic system. Imagine having a well-
defined collection of atoms (or molecules), one of the most important that we would want to know is
their energy, and, more importantly, how their energy changes if we move the atoms around.

This is where the Schrdinger Equation comes into play. Erwin Schrdinger postulated a wave equation
that describes the time evolution of an quantum-mechanic state in an non-relativistic system. In
classical mechanics the path of an particle or mass is obtained through the Newtonian equations of
motion, in quantum-mechanics however, one obtains a wave-function, and this wave-function
contains all the information we need about the particle (electrons in this thesis). Energy, location and
impulse of the particle are called observables, and each observable can be calculated using the wave
function. For example, if one wants to calculate the probability distribution of an particle, one can take
the square of the absolute value of the wavefunction.

To calculate the energy one needs the Hamilton Operator. The most generic form of the time
dependent Hamilton looks like this:

and the time-independent (the form we are interested in) looks like this:

H Phi = E Phi

Where E is the groundstate energy.

The Hamilton operator depends largely on the form and structure of the system of atoms and
molecules that we are trying to analyze. In the next chapter the time-independent Hamiltonian for a
many-electrons system and the Born-Oppenheimer approximation will be discussed.

1.1 The many body Hamiltionian and the Born-Oppenheimer approximation

The Hamiltonian for a system of interacting nuclei (M) and electrons (N) depends on all the nuclear
coordinates R and all electronic coordinates r.

Where the first two term represent the kinetic energy of every electron and every nucleus in the
system, the next three terms are all representing potential energies. The third describes the potential
energy arising from coulombic electron-electron repulsions, the fourth the attraction between all the
nuclei and electrons and the last term describes the coulomb based nuclei-nuclei repulsion.

Since the mass of an electron is about 1800 times smaller than the mass of a proton or neutron, one
can assume that the fast moving electrons respond much more rapidly to changes in their surroundings
than nuclei can.[Scholl] As a result, one can formulate a Hamiltonian for the electrons only,
compressing the forces originating from the nuclei into an external potential. The electron-only
Hamiltionian (He) has the form:

Or, in more general terms and applied to the Schrdinger equation:

Where V(ri) depends on the type of system (in our case it represents the coulomb interaction between
each electron and the collection of all nuclei [Scholl])and U(ri) represents the interaction between the
electrons themselves.

The wavefunction of the system depends on all coordinates of all nuclei Ri and electrons ri in the
system, so , and this is our main problem. One cannot calculate an individual electron wave
function without considering the individual electron wave functions associated with all the other
electrons, in other words, the Schrdinger Equation is a many-body problem.

There are three coordinates for each molecule and three coordinates for each electron, which means
that the wavefunction of the system is dependent on 3N+3M coordinates, which makes it extremely
difficult to calculate.

The normal approach to solving this equation is to specify the system by choosing an V(r), plug it into
the Schrdinger equation and then solve for the wavefunction. [birds eye]

One of the observables that can be calculated through the wavefunctions is the particle density

There are many powerful methods for solving Schrdingers equation. The diagrammatic perturbation
theory (based on Feynman diagrams and Greens functions) is often used in physics, while in chemistry
Configuration interaction (CI) methods (which are based on systematic expansion in Slater
determinant) are applied more frequently. Those methods work well for small systems, but not for
bigger and more complex ones, at least not efficiently. The computation time needed to do so would
be huge. (birds view seite 6)
2. DFT Density Functional Theory (hauptschlich birds view ab seite 7)

DFT provides a more versatile alternative to the methodes mentioned above. It recognizes that
nonrelativistic Coulomb systems differ only by their potential v(r) and supplies methods for dealing
with the universal operators T and U. It provides a way to break down the many body problem with U
to a single body problem without U.

In the last chapter we introduced the particle density as an observable that can be calculated through
the wavefunctions, in DFT however, the calculations of the wavefunctions and therefore of all other
observables dependend on those are based on the calculation of the electron density n(r).

A lot of what we know today about the electronic, magnetic and structural properties of materials has
been calculated using DFT, it is no surprise that the 1998 nobel price for chemistry was awarded to
Walter Kohn, the founding father of DFT, and John Pople.

So the density-functional works like this:

2.1 Functionals and their derivates

Before going describing in detail the workings of DFT, I would like to introduce and explain functionals.
A functional is a rule for going from a function to a number. The particle number is, for example, a
functional:

As you can see, N is dependent on the particle density, which is a function of the location r. A
functional can also depend on both a function and on a parameter.

This potential is called the Hartree potential, it will be discussed in detail below.

2.3 Hohenberg and Kohn Theorems [Scholl 11] dont forget the proof, refference the original
paper.

DFT rests on two fundamental mathematical theorems. Both were proven by Hohenberg and Kohn, (in
their respective papers (zitat hier)). The first one states that the ground-state wave function from the
Schrdinger equation is a unique functional of the electron density n(r) PHI_0[n(r)]. This means that
there is a one-to-one mapping between the ground-state wave function and the ground-state electron
density, implying that the energy E is a functional of the density n, E[n(r)].

Not only the energy, but also every observable calculated via the ground-state wave function is now a
function of the density. So instead of solving the Schrdinger equation for 3N spatial variables one can
solve it for only three spatial variables and still get useful results.
[jetzt wieder birds eye s.10] The ground-state wave function must not only reproduce the ground state
density, it also must minimize the energy. If n_0 is the ground-state density then this requirement can
be written as:

Where E_v,0 is the ground-state energy in potential v(r). In more general terms, the energy in an
external potential v defined by a density n(r) can be written as:

A consequence of the variation principle is that the wave functions obtained by a density n(r) (which
is different from the ground-state density n_0(r)) are different then the wave functions produced by
n_0(r). The minimal energy E_v[n]obtained from the non-groundstate density must therefore be
higher (or equal to) the ground-state energy E_v[n_0].

This equation is very important for the practical application of DFT that is is called the second
Hohenberger-Kohn theorem.

One can split the total-energy functional in two parts:

The functional F[n] is only determined by the kinetic electron energy T and the potential energy
between electrons U, both are defined as expectation values of the ground-state wave function. If the
ground-state is nondegenerate (in the frame of this thesis is assumed to be the case) it is possible to
calculate Phi_0, the wave function that minimizes F[n] and reproduces n_0(r) without considering the
external potential v(r). The potential energy in a given potential v(r) can be calculated through the
density like this (for comparison, take a look at eq. ..)

As soon as v(r) is known, calculating the potential energy is not very difficult. If v(r) is not constant, one
can calculate the potential Energy through the ground state density. This is easily shown by rewriting
the SE as
2.3.1 Proofs of the Hohenberg Kohn Theorem

There are many proofs of the Hohenberg Kohn Theorems, the simplest one based on the variation
principle (as shown in [zitat paper]). There are also other proofs, like the constrained-search proof of
the Hohenberg-Kohn theorem [paper birdsview seite 11].

2.3.1.1 Outline of the original proof

This proof shows that if one assumes there to two different external potentials based on the same
density leads to an impossible inequality, thus showing that a potential is uniquely determined by its
density.

Assume you have two potentials v_a and v_b (both are one-electron potentials), those are associated
with the hamiltonians H_a and H_b. The variational theorem states that the expectational value of H_a
over the wave function of b must be higher than the ground-state energy of a. So

By adding and subtracting H_b, one gets

The last line can be rewritten as:

One can do this calculation with interchanged variables for a and b, and get the following result in the
same manner:

By adding the last two equations you get

Since the density is assumed to be the same for each potential v, the integrals cancel each other out
and the resulting inequality makes no sense. This implies that one of our premises was wrong, namely
if two different potentials can have the same density.

2.3.1.2 Strong form of the Hohenberg-Kohn Theorem

This proof derived in [26,27,28] is based on the inequality

It shows that

1. A change in potential delta_v must result in a change of the density delta_n, thus showing that
one uniquely determines the other. But this has already been proven in the original proof.
2. It also tells us about the sign of the change. If delta_v is positive, then delta_n has to be
negative. Otherwise the equation above would not hold.

2.3.1.3 Proof based on Katos theorem

Katos theorem ([29,30]) is only valid for coulomb-potentials. It tells us that in such systems the
electron density has cusps at the position of the nuclei. It also satisfies

Where a0 is the bohr-radius, R_k the position of the nuclei and Z_k is the atomic number. All the
information needed to define the Hamiltonian for such a system is contained in the density. For the
total number of particles one can integrate n, the positions of the nuclei R_k can be obtained from the
maxima of n(r), and the atomic number is calculated via eq. ().

2.3.2 A preview of practical DFT and the Thomas-Fermi approximation

Given an external potential v(r) (by defining the system), one needs to calculate the total ground-
state energy E_v[n].

As described in the chapters above, the real ground state density can be found by minimizing the
energy with respect to n. The density is dependent on the structure and geometry of the molecule or
lattice that we are looking at. The distribution and arrangement of atoms for example in a lattice define
n. This distribution depends on the lattice constant a. Furthermore, the potential v(r) also depends on
a, so by changing a and v(r) one can plot the curve for the energy variations Ev,0(a). If one finds a
minimum of the energy, it is easy to derive the electron density for that system. For the kinetic and
interaction energy T and U approximations are needed, those will be provided by the Kohn-Sham
equations below. An older alternative is the Thomas-Fermi approximation.
The functionals U and T have the following form in the T-F approximation:

Where t^hom (n) is the kinetic-energy density of a homogenous interacting system with constant
density n. This type of approximation is an LDA a local density approximation. The density n(r)is not
constant, but one can choose an area so small that the density n(r) and the potential v(r) are
approximately constant. In those areas one can approximate T through the homogenous kinetic-
energy density t^hom(n). Making the cells infinitesimally small and summing over all of them yields
Eq. It is also possible to approximate T^LDA[n] with TsLDA[n]

where tshom is the kinetic-energy density for a non-interacting system.

The problem with the Thomas-Fermi approximation is that the total energy of isolated atoms is higher
than that of a molecule consisting of the same atoms. So, according to this model, all molecules are
unstable, which is not the case.

3. The Kohn-Sham equations

The Kohn-Sham equations use single-particle orbital wave functions to find the minimal energy of a
system. The approach discussed up until now works, but is not very efficient. In the first few chapters
I outlined how to approach the many-body Schrdinger equation using the density as the basis for all
other calculations, but now we must return to the SE, more precisely, to the single-body SE to calculate
the ground state energy.

The kinetic-energy functional T[n] can be decomposed into two parts, one part that represents the
kinetic energy of noninteracting particles of density n (Ts[n]) and into Tc[n] (s stands for single-
particle, c for correlation).

Ts[n] can be expressed using single-particle orbital wave functions of a noninteracting with
density n.

The total kinetic energy is just the sum of the individual energies, since the particles are noninteracting.
Ts depends on , and depends on n.

The exact energy functional is now written as

Where Exc contains the differences T-Ts (=Tc) and U-UH. Exc is yet unknown, we just know that it must
be a functional of the density. Exc is called the exchange-correlation (xc) energy, it can be written as

Where Ex is the exchange energy (due to the pauli-principle) and Ec is the correlation energy (Tc is
then a part of Ec).

Kohn and Sham suggested following scheme to calculate the minimal energy [ref. to paper]. The
minimization of the energy can now be written in terms of functionals

The relation V/n = v(r) is derived by the equation for V mentioned above. UH/n yields the Hartree
potential (ref). For the functional derivative of Exc the variable vxc has been chosen. Now one can set
up the same differential equation for a system on noninteracting particles in a external potential. In
that equation, there will be no U and no Exc

The density that solves this equation is denoted ns(r). Comparing those equations, one notices that the
resulting minimal density n is the same for both if vs is chosen to be vs(r) = v(r) + vH(r)+vXC (r).

So, by choosing this vs(r) the density of the interacting particles is calculated through the eqation for
noninteracting particles. The noninteracting Schrdinger equation then reads

Where the phis are the single-particle orbital wave functions mentioned above. The density is then
calculated via

Where fi is the occupation number for the ith orbital. Those equations are the famous Kohn-Sham
equations. So n depends on Phi, and Phi depends on vs, and vs depends on vH and vxc, and those in
turn depend on n. Its easy to see that this is a non-linear problem. One starts with a proposed n(r),
then calculates the corresponding potential vs(r), and then solve the SE for Phi. After that the density
is calculated with the Phis and the whole process is repeated until the density converges. When the
groundstate density has been calculated, the total energy of the system can be calculated using eq. xx.