Special Course in Theoretical Physics

PHYS-E0546: DFT for Practitioners

Aalto University, September 2019

Tutorial 3: Periodic systems

Solid bulk and surfaces

Manuscript for Exercise Problems

Prepared by Patrick Rinke, Xi Chen and Ondrej Krejci

Department of Applied Physics
Aalto University, 2019
Introduction

The aim of this tutorial is to familiarize you with the basic concepts of periodic density functional
theory (DFT) calculations and the related settings necessary to run FHI-aims. Part I teaches you
about the basic concepts of periodic systems. Part II introduces the basic concepts of surface
calculations.

Part I: Basic properties of solids and convergence tests

Problem I: Basic calculations of bulk structures: cohesive energy curve

Problem II: Energy convergence tests

Problem III: The cohesive energy curves for bulk Si

Problem IV: The equation of state for bulk Si

Problem V: Unit cell relaxation

Problem VI: Electronic band structure and density of states

Part II: Basic surface calculations

Problem VII: Basic surface slab calculations

Problem VIII: Relaxing surface structures

Appendix I: Structural information

Appendix II: High symmetry k-points

Appendix III: Visualisation with VESTA

For every exercise, we also provide solutions and sample input files. As always, they can be found
in the $TUTORIALS/ directory. However, we strongly recommend to use the provided input files
only in case of time shortage. You will maximize your learning by trying to generate the input
files on your own. In case you get stuck with a particular problem, do not hesitate to ask one of
the tutors. An executable of FHI-aims, as well as species files and utility scripts, are provided in
the $COURSE/CODE/ directory.

Take your time to read the tasks carefully before proceeding to prepare the calculations. Each
subtask starts with a short summary (gray box) and gives details and hints afterwards. Also, feel
free to consult the supplementary information presented in the Appendices.

It is good practice to use a separate directory for every single run of FHI-aims in
order to preserve the exact input files along with the output files.

Additional tools and programs

Visualization tools:
To visualize structures, vibrational modes, charge density plots, etc., several programs including
molden, vmd, and jmol are installed on your workstations.

Utility scripts:
DFT output files are complex, so we use utility scripts to extract important data. In the
$COURSE/CODE/utilities directory you will find tools like aims-SCF_convergence.awk and
get_relaxation_info.pl to monitor the simulation, as well as create_xyz_movie.pl to track

2
the geometry optimisation. In this tutorial, additional scripts can be found in
$TUTORIALS/Tutorial_3/utilities.

To visualize periodic structures, we recommend VESTA. A brief tutorial is provided in Ap-

pendix III. To open a geometry.in file in your current directory, type :
vesta ‘pwd ‘/ geometry . in &

(Please note that the “‘” character is a backtick). To get periodic images, choose “Objects” →
“Boundary...” in the top panel. Then set the range of fractional coordinates that you want to
see and press “Apply”. A range bigger than 1 results in a periodic image that is bigger than the
specified unit cell. Note: if you see only the word "VESTA" on your screen, the file load didn’t
work and you should load the coordinates manually.

To visualize a structure given in geometry.in with Jmol (Fig. 2), please type:
jmol geometry . in &

Make sure the lattice vectors in your geometry.in are specified before your atomic positions to
visualize the geometry properly with Jmol. To get periodic images, click with the right mouse
button inside the Jmol drawing area and choose “Symmetry” → “Reload {444 666 1}.

3
Part I: Basic properties of solids and convergence tests

In the first part of this tutorial, we will work on different structural phases of bulk silicon. The
correct description of their pressure dependence by Yin and Cohen [1] is one of the early success
stories of computational materials science. In this exercise, we show how to calculate basic proper-
ties of solids like lattice constants, cohesive energies, band structures, and density of states. Please
use the basic settings given in Fig. 1 as default for part I of this tutorial.

# Physical settings
xc pw - lda
spin none
relativistic atomic_zora scalar

# SCF settings
sc_accuracy_rho 1E -4
sc_accuracy_eev 1E -2
sc_accuracy_etot 1E -5
sc_iter_limit 40

# k - grid settings ( to be adjusted )

k_grid nkx nky nkz

Figure 1: Default physical and computational settings for control.in for part I
.

The Perdew-Wang LDA (xc pw-lda) exchange-correlation functional will be used for all calcu-
lations. Silicon is known to be non-magnetic, so no explicit spin treatment is needed. The
“relativistic atomic_zora scalar” setting is not strictly necessary for silicon. The nuclear
charge of silicon (Z = 14) is still small enough to allow for a non-relativistic treatment. But as
the correction is computationally inexpensive, it does not hurt to use it (just be sure to never
compare total energies from different relativistic settings). The k_grid setting will be discussed in
the actual exercises. Please use the default “light” species settings for Si in:
$SPECIES/light/14_Si_default.

Problem I: Basic calculations of bulk structures

Here, we learn how to set basic parameters in control.in for periodic calculations. Next, we
construct periodic geometries in the FHI-aims geometry input format (geometry.in) and visualize
them. Finally, we compare total energies of different Si bulk geometries.

Educational Objectives
• Learn how to set periodic boundary conditions (PBC) in calculations.
• Visualise and observe periodic unit cells.

• Learn to define the k-space sampling in a calculation.

• Compare energies of different lattice structures as a function of lattice constant.

The control.in for periodic calculations looks much the same as for the cluster calculations
(molecules). There is one important difference: a k-grid for the Brillouin zone integrations must

4
be specified. The number of grid points determines the sampling density of the first Brillouin zone
in 3D reciprocal space, needed to describe the variation of electronic bands E(k) with wave vector
k in periodic systems. The 3D Brillouin zone is defined by reciprocal lattice vectors computed
from the real-space lattice vectors in the geometry.in. The order in which k-space grid settings
are written is consistent with the ordering of the reciprocal lattice vectors, which follows from the
order of real-space lattice vectors.

To set up a periodic structure in FHI-aims, all three lattice vectors as well as the atomic positions
in the unit cell must be specified. The lattice vectors are denoted by the keyword lattice_vector.
There are two ways to specify the atomic positions. You can specify absolute Cartesian positions
with the keyword atom. Alternatively, you can specify the atomic positions in the basis of the
lattice vectors, the fractional (or commonly, direct) coordinates, with the keyword atom_frac.
The fractional coordinates si are dimensionless and the coefficients for the linear combination of
the lattice vectors ai . Written out as a formula, this linear combination reads as follows

R = s1 · a1 + s2 · a2 + s3 · a3 , (1)

where R is the Cartesian position of the specified atom. For example, fcc Si with a lattice constant
a = 4 Å is defined by
lattice_vector 0.0 2.0 2.0
lattice_vector 2.0 0.0 2.0
lattice_vector 2.0 2.0 0.0
atom_frac 0.0 0.0 0.0 Si

The simplest way to check the geometry.in file is to visualize the corresponding geometry. This
should always be done before any calculation, to verify that the structure is plausible (and that
no atoms are extra or missing).

Tasks:

1. In the control.in file, set a 3x3x3 k-space grid effectively making nkx=nky=nxz=3. You will
see that for a small unit cell this k-point density is never enough, but we can try to find out
what happens.

2. Construct geometry.in files for the Si fcc, bcc, and diamond structures. Use the approximate
lattice constants a of 3.8 Å for fcc, 3.1 Å for bcc, and 5.4 Å for the diamond structure. A full
set of lattice vectors and atomic positions of primitive unit cells (the unit cell with the
smallest volume, containing only one point of the crystal lattice) for fcc, bcc, and diamond
can be found in Appendix I. Be sure to translate the given atomic positions (which are in
Cartesian coordinates) to the right value for either the atom or the atom_frac keyword.

Figure 2: Single unit cell (left) and periodic images (right) of diamond Si in Jmol.

5
 3. Visualize geometry.in files using VESTA (or Jmol). Can you spot how the primitive cell
motif is repeated throughout space?

4. Make dedicated folders for each Si crystal structure and copy the input files there. Also copy
over the provided run_diamond.py script to calculate total energies of the diamond Si phase
as a function of lattice constant a.

5. Start with the diamond phase calculation and edit the script run_diamond.py to insert the
range of unit cells to be computed: choose 7 different values of a in steps 0.1 Å around
the lattice constant given above. Use the command python run_diamond.py to start the
calculation.

6. Copy the run_diamond.py script to the remaining directories and carefully modify the ge-
ometry sections to adjust the lattice vectors for other primitive cells. Insert new unit cell
ranges as appropriate and complete the calculations.

7. Use the postprocess.py script to retrieve the total energies which will be written to the file
energies.dat, along with the lattice constants. For each phase of Si, modify the script to
input the range of lattice constants you had tried and run the script.

8. Compute the total energy per atom, not per unit cell: this is important for the diamond
structure (in the case of fcc and bcc, each unit cell contains one atom). You can do so
by re-running the postprocess.py script after modifying line 44 of the script to read:
“energy = float(linesplit[5])/2.”.

9. Plot the total energy per atom for all structures as a function of the lattice constant. At
which lattice parameters do your curves reach the minimum? How does this compare to the
experimental lattice constants?

10. What is the most stable bulk phase of Si according to your results?

You should see that, with the current computational settings, the diamond Si phase is unfavorable
compared to the other two phases by about 0.1 eV. However, the experimentally most stable phase
is the diamond structure. We will show in the next two problems that the coarse 3×3×3 k-grid is
the reason for this disagreement.

Problem II: Energy convergence tests

The results of unconverged calculations can range from inaccurate to unphysical, as seen in the
previous exercise. Now we explicitly check total energy convergence with respect to the k-grid and
to the basis set. Each phase needs to be checked separately, but in the interests of time, choose
one Si phase to proceed with.

Educational Objectives

• Develop a feeling for the k-point sensitivity.

• Deepen the understanding of convergence.

Subproblem 1: Convergence with k-grid

1. For your chosen phase, employ the input files from the previous exercise. Prepare control.in
files with k-grids of 8×8×8, 12×12×12, and 16×16×16.

6
 2. For each k-grid setting, compute the total energy curve as before (repeat steps 5-8 from the
previous exercise).

3. Plot the total energies versus the lattice constant for the different k-point settings. Add the
3×3×3 results too.

4. Which k-grid should be used to achieve convergence within 10 meV?

5. How does the computation time increase with the number of k-points? Hint, you can find
the total run time in the aims.out output file.

6. Which k-grid offers the best compromise between accuracy and runtime?

Subproblem 2: Convergence with basis set size

To change the basis size settings, take a look at the species-dependent settings in control.in.
There, you will find a line starting with “# "First tier" - ...”. Each line after this defines a
basis function which is added to the minimal basis. Right now, there is one additional function
for each valence function (s and p) as well as a d function to allow the atoms to polarize. This is
what we call minimal+spd in the context of this tutorial. In quantum chemistry and in particular
the Gaussian community, this type of basis set is often called “double zeta (ζ) plus polarization”
(DZP).

1. Use your control.in file from the previous problem with the 12×12×12 k-point grid. Cal-
culate the total energies for your phase of Si as a function of the lattice constant for the
minimal and the tier1 basis sets.

2. Plot the total energies curves obtained for the three different basis sets and write down the
values of the lattice constants that minimise the energy. Also, consider the values of the
lowest energy reached in each curve, and compare the total time of the calculations.

3. What do you observe? How do the curves change? How does this compare to the k-point
convergence?

Optional subproblem 3: Effect of Gaussian broadening of the Kohn-Sham occupation

numbers

FHI-aims always outputs three different energies. While these energies are all the same for systems
with a band gap, they differ for metallic systems with finite Gaussian broadening. The “Total
energy uncorrected” gives the value of the Kohn-Sham energy functional for the final self-
consistent electronic structure. However, due to the Gaussian broadening the self-consistency pro-
cedure does not minimize this total energy but a “free energy” specified to the right of “Electronic
free energy”. From these two numbers, FHI-aims extrapolates back to the total energy without
broadening and writes the resulting number to the right of “Total energy corrected”. For true
metals, it is generally best to make use of the correction. For finite systems and in particular for
isolated atoms, however, the back-extrapolation is unphysical and should not be used.

Tasks:

1. Use the control.in file and scripts for the fcc Si system, with the 12×12×12 k-point grid
setting and the minimal+spd basis. You can explicitly set the Gaussian broadening to σ =
0.1 eV by specifying
occupation_type gaussian 0.1

in control.in.

7
 2. Prepare a plot with the corrected, uncorrected total energies, and the electronic “free ener-
gies” for a broadening of σ = 0.1 eV and the default value of σ = 0.01 eV from the previous
calculations.

3. How does the broadening affect the results? Which broadening is safe to use?

In the calculations that follow, we will use the default value of σ = 0.01 eV. We will stick to the
corrected total energy for the periodic systems in this part of the tutorial as it is the most accurate
value for metals and makes no difference for semiconductors.

Problem III: The cohesive energy curves for bulk Si

Having established well-converged computational settings, we now revisit the phase stability of
bulk silicon (in practice, one should always check convergence first). We will learn how to compute
the basic cohesive properties and study the pressure dependence of phase stability. To proceed, we
need to calculate the cohesive energy curve from the E(a) curve computed in previous exercises.

Educational Objectives

• Learn to compute the cohesive properties of a solid.

The cohesive energy (Ecoh ) of a crystal is the energy per atom needed to separate it into its
constituent neutral atoms. It is computed per simple cubic unit cell defined by the lattice constant
a to compare cohesion of different phases on an equal footing. Ecoh is defined as
 
Ebulk − N Eatom Ebulk
Ecoh = − =− − Eatom , (2)
N N

where Ebulk is the bulk total energy per unit cell and N the number of atoms in the unit
cell. Eatom is the energy of the isolated atom. The number of atoms in the simple cubic unit cell
defined by the lattice constant a is different for each phase of Si, as illustrated in Fig. 6, Appendix I.

Tasks:

1. Use the converged settings (k-grid 12×12×12, minimal+spd basis, and σ = 0.01 eV occupa-
tion broadening) to compute the total energy of fcc, bcc, and diamond Si as a function of
lattice constant a. Which phase of Si is the most stable?

2. Compute the energy of a single Si atom following what you learned in Tutorial 1 (make sure
you use different settings in the input file, and note the correct energy in the output file).

3. Determine the number of atoms N per cubic unit cell defined by the lattice constant a for
fcc, bcc, and diamond Si crystal phases.

4. For each phase of Si, use the script convert-coh.awk to compute the cohesive energies
cohesive.dat from energies.dat:
python convert-coh.py --Natoms ..--Eatom .. -i energies.dat -o cohesive.dat
where the “..” need to be replaced by the corresponding values of Nsc and Eatom , respectively.

5. Plot the cohesive energies of all three phases on one graph with the atomic volume (not
lattice constant a) on the x axis.

6. What do you observe?

8
Problem IV: The equation of state for bulk Si

Educational Objectives
• Learn to analyze the equation of state for a solid.

The equilibrium lattice constant a0 is an important quantity which we can calculate from our data.
In principle, this can be done with a quadratic ansatz for E(a) or E(V ). Here, we will discuss
and use a thermodynamically motivated and more accurate fitting function, the Birch-Murnaghan
equation of state [2, 3]. The energy per atom (E = −Ecoh ) is expressed as a function of the atomic
volume (V = Vatom )
B0
" #
B0 V (V0 /V ) 0 B0 V 0
E(V ) = E0 + +1 − 0 . (3)
B00 B0 − 1
0 B0 − 1
The fitting parameters V0 and E0 are the equilibrium atomic volume and atomic energy, respec-
tively, B0 the bulk modulus and B00 its derivative with respect to pressure. Equation (3) can be
derived by assuming a constant pressure derivative B00 .

The fitting script murn.py is part of the FHI-aims distribution. If you have prepared the files
cohesive.dat using the provided gawk script, you can simply use this script with
murn . py cohesive . dat -o fit . dat

The program then outputs the parameters V0 , E0 , B0 , and B00 for the given data set as output. As
a quick plausibility check of the fit, you can use the option -p to see a plot. The script performs no
unit conversions, so the bulk modulus B0 is given in units of eV/Å3 because the cohesive energies
and atomic volumes were provided in eV and Å3 , respectively (1 eV/Å3 is 160.21765 GPa). √ The
optimal lattice constant can be calculated from the equilibrium atomic volume V0 by a0 = 3 Nsc V0
with Nsc the number of atoms in the cubic unit cell.

Tasks:

1. For each phase of Si, apply the script murn.py to perform the Birch-Murnaghan fit of the
cohesive energy curves cohesive.dat.

2. Take note of the output parameters from the script, and use them to compute the equilibrium
lattice constant a0 in Å, the bulk modulus B0 in GPa, and the cohesive energy Ecoh in eV.

3. Compare the calculated quantities for the diamond phase with the experimental values of
a = 5.430 Å, B0 = 98.8 GPa, and Ecoh = 4.63 eV [4]. Exact agreement with experimental data
is not expected, since xc functionals are approximated. Based on your observations, what
can you tell about the tendencies of the pw-lda functional - does it overbind or underbind?

4. Plot the resulting fitted curves saved in fit.dat into one figure. You should get something
similar to Fig. 1 in the paper by Yin and Cohen [1].

5. Bonus: Given this data, can you estimate at what pressure a phase transition would occur?
Recall the Maxwell construction (see below).

By exposing the crystal to the right pressure, one can enforce many different atomic volumes
smaller than the equilibrium ones. In principle, the most stable phase for a given atomic volume
V can simply be deduced from the curve with the lowest E(V ). The corresponding pressure can
be calculated from the slope of the curve by the simple thermodynamic relation p = −∂E/∂V .

However, in the regime of about 13 Å3 < Vatom < 18 Å3 coexistence of a diamond phase at ∼ 18 Å3
and a bcc phase at ∼ 13 Å3 is favorable (see Fig. 3). The fraction of atoms in the two phases
then determines the average atomic volume. The resulting average atomic energy is a straight line

9
 phase 2

Cohesive energy
phase 1
dE

dV

Volume

Figure 3: Maxwell construction

between the corresponding points which is below both the lines of diamond and bcc Si. This is
called the Maxwell construction. From the slope of this line (a common tangent), we can deduce
a transition pressure of about 14 GPa at which diamond and bcc Si could coexist according to our
calculations. This is more than 105 times the ambient pressure of about 100 kPa. Note that there
are additional phases for silicon which have not been calculated here. In fact, the Si β-tin phase is
more stable than the bcc phase at high pressures. For a more thorough discussion, refer to [1].

Problem V: Unit cell relaxation

In the previous tutorial you learned how to optimize the positions of atoms in molecules. To obtain
optimized periodic structures, not only the atomic positions must be optimized but also the lattice
vectors. Here, we perform a structure relaxation from bcc to diamond Si.

When performing a local structure optimization, we do not go uphill in the energy landscape or
cross energy barriers. The path from the starting structure to the desired local minimum has no
barriers. This means that we have no assurance that we have found the global minimum, which
may lie over an energy barrier from our starting geometry (to cross energy barriers, one can do,
for example, variable cell-shape molecular dynamics, but this is out of the scope of this tutorial).

Since we want to end up in a diamond structure, we have to start with at least 2 atoms: it is
crucial that the number of atoms in the initial and final unit cells is the same. Our starting unit
cell is a cubic bcc structure consisting of 2 atoms. Of course, we cannot start with the ideal cubic
bcc unit cell because we will be stuck in the local minimum of this high symmetry structure. Even
if we distort the starting geometry significantly (atomic positions & lattice vectors), we can easily
end up in an unwanted local minimum. Here we provide a suitable starting geometry, feel free to
try other ones when you are finished with this one.

Tasks:

1. Prepare input files from the previous exercise (k-grid 12×12×12, minimum+spd basis set)
for unit cell relaxation by adding the following keywords to control.in:
relax_geometry trm 1E -2
relax_unit_cell full

The latter keyword enables full optimization of the lattice vectors (their lengths and the

10
 angles between them). The convergence criterion will be the same as the one specified for
the atoms.

2. In the convergence section of the input file, add the line:

sc _a cc ur ac y_ fo rc es 1E -3

The computed forces will change during the SCF cycle as the electronic structure evolves,
so the convergence of forces within the SCF cycle also needs to be considered for accurate
relaxation calculations. Tolerance of 10−3 eV/Å is sufficient given the desired relaxation
accuracy above.

3. Prepare the initial geometry for structure relaxation as follows:

lattice_vector 3.1 0.4 0.4
lattice_vector 0.45 3.1 0.4
lattice_vector 0.45 0.45 3.1
atom_frac 0.0 0.0 0.0 Si
atom_frac 0.3 0.3 0.3 Si

4. Run the calculation and use the get_relaxation_info.pl script on the output file to check
the calculation has finished correctly. Take a closer look at the output file to see that not
only the atomic forces are calculated but also a quantity called stress tensor. The stress
tensor is a measure of the forces acting on the unit cell itself.

5. To visualise the geometry optimisation, apply the create_geometry_zip.pl to your output

file. Unzip the resulting file geometries.zip to see the individual geometries or use Jmol
by typing the command “jmol -s geometries.spt”.

6. Analyze the final structure (angles between lattice vectors, their length, and atomic positions)
and compare it to the equilibrium values of diamond Si obtained from the Birch-Murnaghan
fit in the previous problem.

Problem VI: Electronic band structure and density of states

The electronic band structure describes the dispersion relation of electrons inside a solid. The
band structure gives information about ranges of energy that electrons are allowed to have or not.
Permitted ranges are called bands, forbidden ones band gap. Many properties of a solid can be
deduced from its band structure, e.g., if it is a metal, semiconductor, or insulator.

The density of states (DOS) is one of the basic concepts in solid state physics. Particularly, the
DOS around the Fermi level is of interest as it will give you information about the properties of
your system, for example, whether it is conducting, semiconducting, or insulating. Many material
properties depend on the DOS, for instance the conductivity. The number of states n within a
given energy interval (0 − ∆) <  < (0 + ∆) per unit volume Vcell is given by
Z 0 +∆
n= dg() (4)
0 −∆

where g() is the density of states (DOS). In a free atom or an isolated molecule, the DOS consists
of a series of discrete energy levels (δ peaks) and can be written as
X
g() = δ(i − ). (5)
i

In a periodic system, the single particle energies become k dependent and the DOS continuous.

11
The number of states per energy is averaged over k
1 X
Z
g() = d3 k δ(i,k − ). (6)
VBZ i BZ

In order to calculate the density of states numerically, we have to replace the integral over the
Brillouin zone (BZ) in Eq. (6) by a sum over k-points. In the case of infinite k-points, this
replacement is exact. To compensate the deficiency of a finite grid, we broaden the δ(k,i − )
distribution by a Gaussian function with an broadening factor σ,
" 2 #
1 1 XX 1  − k,i

g() = √ exp − . (7)
2πσ nk i 2 σ
k

Tasks:

1. Use your control.in and geometry.in for the equilibrium lattice parameter for diamond Si
from Problem IV (Murnaghan fit) to prepare input files for electronic structure calculation.

2. In the control.in file, add the keywords:

output dos -18. 0. 200 0.1

The first two values define the energy window of interest, the first value is the lower energy
bound and the second value is the upper energy bound. The third value is an integer spec-
ifying at how many energy data points the DOS is evaluated, and the last value gives the
Gaussian broadening σ. All energies (bounds and broadening) are given in eV.

3. For a well-resolved DOS function, a dense k-grid to determine the sum over k-points in
Eq. (7) is necessary. After self-convergence is reached, the DOS can be computed using an
auxiliary k-grid that is made denser by factors n1 , n2 , n3 , respectively. To switch on this
function, specify factors n1 to n3 in the control.in with the keyword:
dos _kgrid _facto rs 5 5 5

The main calculation will be executed using the k1 × k2 × k3 , but the density of states is
calculated on a denser grid after the SCF cycle. The dimensions of the new k-point grid are
k1 × n1 , k2 × n2 , k3 × n3 .

4. To calculate the band structure, specify the path along high-symmetry points in reciprocal
space: L → Γ → X → W → K in control.in file:
# diamond band structure:
output band 0.5 0.5 0.5 0.0 0.0 0.0 21 L Gamma
output band 0.0 0.0 0.0 0.0 0.5 0.5 21 Gamma X
output band ...
Please refer to Appendix II for the location of the other high symmetry points in the Brillouin
zone. In each line, the first three numbers are the coordinates of the starting point in units
of the reciprocal lattice vectors. The next three numbers specify the end point. The band
structure is then calculated along the path connecting these two points.

5. Run the calculation. Densities of states will be output into separate files: KS_DOS_total_raw.dat,
where the zero of energy has not been altered, and KS_DOS_total.dat, where the zero of
energy has been shifted to the computed Fermi level. Plot the last file to visualize the density
of states of silicon in the diamond phase. How does the LDA bandgap compare with the
experimental band gap of silicon of 1.17 eV [4]?

6. To visualise the computed band structure, use the script aimsplot.py without any arguments
in the calculation directory. Note that the DOS plotted along the band structure reflects the
density of the electronic bands at any energy.

12
Part II: Basic surface calculations

Figure 4: The hydrogen-saturated ideal 2×1 Si(001) surface. The cyan (light) atoms correspond to the
bulk Si atoms, the blue (dark) atoms are the top Si atoms. The white atoms on the bottom
layer are hydrogen atoms.

In this exercise we focus on the clean Si(001) surface, technologically the most relevant Si surface.
This surface terminates by unsaturated (dangling) silicon bonds, so the surface atoms re-arrange
into a more energetically favourable surface reconstruction. The nature of the surface reconstruc-
tion was unclear for many years. After long discussions and different models used to understand
the differences between various experimental (LEED, STM) and theoretical approaches, it was
shown by direct evidence in STM by Wolkow [5] that the main surface feature at low temperatures
is the asymmetric Si dimer in a 2×1 reconstructed surface unit cell.

We will continue with the same level of accuracy, so you can take your control.in and only adjust
the k-grid settings as shown below. For "saturating" the bottom layer of the slab, we additionally
need the light species defaults for hydrogen, so add this to the control.in file. As shown in Fig. 4,
each silicon atom has to be saturated by two hydrogen atoms.

# Physical settings
xc pw - lda
spin none
relativistic atomic_zora scalar

# SCF settings
sc_accuracy_rho 1E -4
sc_accuracy_eev 1E -2
sc_accuracy_etot 1E -5
sc_iter_limit 40

# k - grid settings
k_grid 12 24 1

Figure 5: Physical and computational settings for control.in for part III.

13
Problem VII: Basic surface slab calculations

Educational Objectives

• Learn to build and visualize a surface structure.

• Compute and analyse surface band structures.

Building surface slab models is not difficult, but one must proceed with care. The geometry of
four layers of Si(001)-(1×1) in diamond structure is provided in the Appendix I. By convention
the surface is rotated with respect to the bulk structure by 45° around the z axis, so that the z
axis is perpendicular to the surface plane. Adding the next four layers in thickness would involve a
repetition of the first four layers, with coordinates shifted by a in z direction. To expand this 1×1
unti cell into the (2×1) reconstructed surface model, one must add a repetition of all the atomic
√
coordinates, shifted by a/ 2 in x direction and double the corresponding lattice vector. This is
handily done in a spreadsheet or with scripts: here we provide script write-geom.py for your
convenience.

To determine a sufficient amount of vacuum between slabs and reduce interactions between periodic
images, you would need to run several calculations with different vacuum thickness (typically, at
least 10Å of vacuum is needed ). In FHI-aims vacuum is cheap, you can easily use a vacuum of
Lvac = 30 Å or more without noticeable performance impact.

For periodic calculations with a slab, the k-point grid needs to be modified to request only one
k-point in the z direction (there should be no interaction between different periodic images of the
slab). Also, in a 2×1 unit cell, the lattice vector ax is twice as large as ay , so the reciprocal vector
rx is twice smaller than ry and needs only half the number of k-points.

Tasks

1. Use the python scripts write-geom.py to build the 2x1 ideal, hydrogen-saturated, 4 and 8
layer surface slabs. Edit the scripts to insert the optimized lattice constant obtained from the
Murnaghan fit and look at the difference between the scripts. Leave the value Lvac = 30 Å
for vacuum thickness. The scrips needs no input file: ./write-geom_4L.py > geometry.in

2. Visualise the 2x1 slabs with Jmol or VESTA, and note its orientation with respect to the
coordinate axes. How many atoms are there in each layer of the unit cell?

3. In the control.in files, adjust the k-point settings for the surface using the precomputed
converged setting of 12 × 24 × 1.

4. Switch on the DOS calculation by adding the following lines:

output dos -18. 0. 200 0.1
dos _kgrid _facto rs 5 5 1

5. For band structure calculations in the 2D reciprocal space of the surface (see Appendix II )
following the path Γ̄ → J¯ → K̄, introduce the following settings:
# Si 2x1 surface band structure:
output band 0.0 0.0 0.0 0.5 0.0 0.0 21 Gamma J
output band 0.5 0.0 0.0 0.5 0.5 0.0 21 J K

6. Perform single-point calculations for the four layer and eight layer surface slabs. Use the
aimsplot.py script (type aimsplot.py) to visualise and compare the band structures be-
tween the two slabs.

14
When plotting the band structures as before, note the differences between the four- and the eight-
layer slabs. While the bands within the fundamental band gap (which arise from the surface atoms)
hardly change with the number of layers, the bands in the valence and the conduction bands get
about twice as dense. These bands are bulk-like; the more layers you add, the more of these bands
appear.

Problem VIII: Relaxing surface structures

Educational Objectives
• Familiarize yourself with atomic relaxation at surfaces.

Relaxing surface slabs is conceptually different than relaxing bulk structures. To mimic realistic
surfaces, lower layers of the slab are assigned bulk-like character and kept in fixed position, while
the upper layers are relaxed. In FHI-aims, particular atoms can be constrained by inserting the
following line right under the atom coordinates.
c o n s t r a i n _ r e l a x a t i o n . true .

An example excerpt from geometry.in reads like this:

atom -1.2063524529754976 0. 00 00 00 00 000 00 00 0 -0.8530200000000001 H
c o n s t r a i n _ r e l a x a t i o n . true .
atom 1. 206 35 24 52 97 54 97 6 0. 00 00 00 00 00 000 00 0 -0.8530200000000001 H
c o n s t r a i n _ r e l a x a t i o n . true .
atom ...

Tasks

1. Start with the input files from a previous calculation, using the four layer Si slab structure.
Fix the bottom two layers (one H and one Si layer) and leave the top three (all Si) layers to
relax.

2. Adjust the control.in file for structural relaxation by inserting the following keywords:
sc _a cc ur ac y_ fo rc es 1E -3
relax_geometry trm 1E -2

The settings above with relax the atoms inside the unit cell without changing unit cell
parameters. These are fixed to allow lower slab layers to mimic ideal bulk structure.

3. In structural relaxations, it is a useful trick to depart from perfectly symmetric coordiantes in

the initial geometry. Introduce a slight perturbation (about 0.5 Å) in one or more coordinates
of the two Si atoms in the surface layer of geometry.in. This ensures that the initial geometry
is different from the symmetry-induced saddle point of the ideal geometry. A possible choice
for the top silicon atoms is:
atom 2.4 14 84 51 63 45 31 70 7 0. 00 00 00 000 00 00 00 0 3.262 Si
atom 5.2 44 53 54 90 35 95 12 1 0. 00 00 00 000 00 00 00 0 4.062 Si

Here, the two top Si atoms have been moved in the x-direction by ±0.5 Å, and one Si atom
has been pushed down by 0.8 Å.

4. Perform the calculation (it should take about 5min on 4CPUs). Check that it completed
correctly with the get_relaxation_info.pl script. To visualise the geometry optimisa-
tion, apply the create_geometry_zip.pl to your output file. Unzip the resulting file

15
 geometries.zip to see the individual geometries or use Jmol by typing the command
“jmol -s geometries.spt”. Has the surface changed its morphology?

5. Apply scripts to compute the band structure of the relaxed surface structure. Compare the
bands with the data from the previous, ideal surface calculation.

6. What do you observe?

16
Appendix I: Structural information

Figure 6: Cubic (black) and primitive (grey) unit cell for bcc, fcc and diamond Si (from left to right).

Crystal structures atomic coordinates lattice vectors

0 a/2 a/2

fcc 0 0 0 a/2 0 a/2

a/2 a/2 0
0 a/2 a/2
0 0 0
diamond a/4 a/4 a/4
a/2 0 a/2
a/2 a/2 0
−a/2 a/2 a/2

bcc 0 0 0 a/2 −a/2 a/2

a/2 a/2 −a/2
0 a/2 a/2
zincblende with A 0 0 0 a/2 0 a/2
atom species A and B B a/4 a/4 a/4
a/2 a/2 0

Table 1: Solids: Atomic coordinates and lattice vectors for different crystal structures.

diamond(001)
atomic coordinates lattice vectors
0 0 0 √
√ a/ 2 0 0
a/2 2 0 a/4 √
√
a/2 2
√
a/2 2 a/2
0 a/ 2 0
√ 0 0 L
0 a/2 2 3a/4

Table 2: Surfaces: The atomic coordinates of an ideal diamond(001) surface. Note: a is the bulk lattice
constant and L is the total slab thickness (L = a + Lvac with the vacuum size Lvac ).

17
Appendix II: High symmetry k-points

fcc x1 x2 x3
fcc(001) x1 x2 x3
L 0.5 0.5 0.5
Γ 0 0 0 Γ̄ 0.0 0.0 0.0
X 0 0.5 0.5 J¯ 0.5 0.0 0.0
W 0.25 0.5 0.75 K̄ 0.5 0.5 0.0
K 0.375 0.375 0.75

Table 3: High symmetry points for fcc/diamond bulk and (001)surface structures given in units of the
three reciprocal lattice vectors (~k = x1~b1 + x2~b2 + x3~b3 ). In the case of fcc/diamond bulk, the
reciprocal lattice vectors form a bcc structure and the corresponding lattice vectors can be found
in Tab. 1.

Figure 7: Brillouin zone and high symmetry points for fcc/diamond structure. The three coordinate axes
(~kx , ~ky , and ~kz ) form a Cartesian coordinate system – not to be confused with the reciprocal
lattice vectors of the fcc/diamond structure.

18
Appendix III: Visualisation with VESTA

Visualizing your results from density functional calculations is useful and important for a couple
of reasons. First, you can visualize the structure as it was read by FHI-aims and check its
accuracy. This is a fool-proof way to check that the geometry handled by the code is actually the
one you want. Second, you can view ‘volumetric data’ superimposed on the structure of whatever
system you are dealing with. Volumetric data can be any kind of data written to a real-space grid
including charge densities, wave functions, potentials, etc. Visualization of volumetric data can
help you extract meaningful physics from your calculation. This document is a quick-start guide
to visualizing output from FHI-aims with VESTA.

VESTA is a public, easy-to-use application for visualization of atomic structures and volumetric
data. You can find it here: http://jp-minerals.org/vesta/en/

The complete manual for VESTA is here:

http://www.geocities.jp/kmo_mma/crystal/download/VESTA_Manual.pdf

Visualizing crystal structures

It is difficult to understand what a bulk, three-dimensional crystal structure looks like just by
reading atomic coordinates. Luckily, VESTA will read the geometry.in file needed for FHI-aims
and visualize the crystal for you. Type vesta to launch.

Figure 8

Now navigate to File and open your geometry.in file for FHI-aims. It should show the crystal
structure you are interested in. An example geometry.in file for wurtzite ZnO and its appearance
in VESTA are shown below and in Fig. 8.

lattice_vector 3.27534473 -0.07200997 -0.00749503

lattice_vector -1.57091511 2.87405204 -0.05082762
lattice_vector 0.01006313 0.10435241 5.29125719
atom 0.04290507 1.88121444 -0.06093020 Zn
atom 1.66090983 0.94898484 2.59987036 Zn
atom 0.04188438 1.91975639 1.94486746 O
atom 1.66718882 0.99019168 4.60509159 O

You can rotate the structure just by clicking and dragging the mouse. If you lose track of what

19
direction you are looking at, click the ‘a,’ ‘b,’ or ‘c’ buttons on the top left. These will automatically
orient the camera with the first, second, or third crystallographic direction. Notice the three
crystallographic directions are given by the axes in the bottom left corner. If you click on ‘Tools’
at the top left of Fig. 12, you can select the ‘Animation’ rotation mode. With this selected, the
crystal will slowly rotate in the direction you choose.

Fig. 8 looks ok, but it can be improved a lot. When looking at a periodic structure, it is helpful
to increase the number of unit cells displayed so that you can see how different cells meet at their
boundary. Click on Objects → Boundary to get the menu shown in Fig. 9. The boundaries of the
displayed structure are set by ‘xmin’ and ‘xmax’ in reduced coordinates (and likewise for y and z).
Setting the boundaries to 0 and 1 shows one cell. Try setting all of the maximum boundaries to 2.

Figure 9

That looks better, but it is still difficult to see how the material is bonded. To fix that, click
on Edit → Bonds. You should see the menu shown in Fig. 10. You have a few choices on how
VESTA will search for bonds. The simplest, catch-all method is to click on ‘New’, choose ‘Search
molecules,’ check ’Do not search atoms beyond the boundary,’ and then set a maximum bond
length. This choice will draw a bond for any two atoms that are less than the maximum bond
length away from each other. You need to determine a reasonable maximum bond length for your
structure. By searching atoms bonded only to ‘A1’ and/or ‘A2,’ you can search for bonds only
involving those one or two chemical species. Now that the bonds are drawn, it is much easier to
visualize the crystal and how the atoms are arranged when you rotate the crystal. For reference,
a single unit cell is still outlined in black by VESTA.

Visualising cube files

Navigate to File and open your .cube file from FHI-aims. It should show the atomic structure
and some colored surfaces, something like Fig. 11. The yellow surfaces are called isosurfaces −
surfaces of constant value. In the case of plotting ρ or ρi , they are surfaces of constant charge
density. These isosurfaces are analogous to reading a topographic map in 2D. Instead of lines at
constant value in the 2D case, we now visualize surfaces. Try clicking the different options under
‘Style’ on the left to change the appearance of the structure (Fig. 12). You can turn isosurfaces on
or off by clicking the ‘Show isosurfaces’ box. And you can change the isosurface style by choosing
a different option under ‘Style’ (the second ‘Style’ panel). The ‘Show sections’ box is probably also
checked. Sections are cuts through the charge density on the boundary of your plotting box. They
appear as contour plots on the sides of the plotting box. These can be useful for periodic systems,

20
 Figure 10

but for molecular systems, the density near the edges of your plotting box is so low, there is not
much meaningful information there.

Figure 11

The tools on the left side of the screen (very left side of Fig. 12) allow you to measure bond
lengths, angles, etc. with your mouse. For example, click the one with the little arrow between two
atoms, then click on two atoms in the structure. In the output dialogue on the bottom, VESTA
should tell you the distance between those two atoms. The next tool down the list will allow you
to measure bond angles. Click it, then click on three atoms in the structure. The angle formed by
these three should be printed in the output dialogue.

If you navigate to Objects → Properties → Isosurfaces, you should see the ‘Properties’ menu
shown in Fig. 12. Here, you can change the isosurface value by clicking on the number underneath
the ‘level’ label. You can also change the color or add multiple isosurfaces by selecting ‘New.’ Try

21
selecting a reasonable isosurface value that allows you to see the character of the charge density.
Selecting isosurface values is subjective and can be tricky, as different values can show very different
shapes. One possibility is to select isosurface values that are a certain percentage of the maximum
value on the grid, which is conveniently reported by VESTA. Always selecting the same percentage
of the maximum offers some plotting consistency between different molecules.

Figure 12

Under the ‘General’ tab (Fig. 13), you can change the overall appearance of the viewer. You can
turn on/off the outline of the plotting box, remove the coordinate axes, or adjust the material
appearance. Under ‘Atoms,’ you can change the color of each species, the size of each atom, etc.
There are lots of customizable aspects to create exactly the look you want. Explore all the options!

Images can be exported by going to File → Export Raster Image or File → Export Vector
Image. Either of these will save a snapshot of the viewing window. Be sure to zoom in and orient
the molecule correctly for the best quality image. That’s it! Now you can visualize your results
from FHI-aims.

VESTA also has a 2D display mode. Sometimes, making a 2D contour plot on a slice through a
certain plane looks better than an isosurface in 3D. To start the 2D mode, click on Utilities →
2D Data Display. You should see the menu shown in Fig. 14.

First, you must select which plane you want to make your 2D slice through. Click on ‘Slice’ to
get the second menu shown in Fig. 14. Set a plane based on its (hkl) index. Don’t worry about
setting the distance from the origin - you will be able to visualize data in all planes parallel to the
one set by (hkl). The default setting is (hkl) = (001). If your molecule is planar and oriented in
the xy-plane, this is a good choice. Click ‘Ok.’

Now you need to set the saturation levels for the color scale. The maximum value will be the grid

22
 Figure 13

value that appears red, and the minimum value will appear blue. These do not have to be 100%
and 0%. Usually, the maximum and minimum values on the grid are very far apart, with most
of the grid points taking a value much closer to the minimum than the maximum. If you simply
set the maximum and minimum to 100% and 0%, you probably won’t be able to see much. The
majority of the data points will be in a very narrow color band, and you won’t be able to see much
detail at all. Next to ‘Max,’ set the saturation level to something around 3%. Then set the ‘Min’
level to 0%. This will probably take some tuning to get a nice image − you can reset these later.
Now click on the blue plane with an arrow going through it and scroll with the wheel on the mouse.
You should see a new image with every wheel click as you scroll through different planes parallel
to the one you set, something like Fig. 15. Each plot is a slice through the density parallel to the
(hkl) plane that you set. Try setting different saturation levels and a different (hkl) plane. Again,
you can export an image by navigating to File → Export Raster Image.

Now check the ‘Bird’s-eye view’ box. As long as the blue plane with the arrow going through
it is checked, you can still scroll through different plane levels. Now, though, click and drag the
plotting area. You should be able to rotate the plotting plane, and the plot now has depth. The
colors correspond to the height of the features out of the plotting plane. To see more dramatic
features, you might want to increase the maximum saturation level. Now you can generate a
surface corresponding to the density through a single plane. 2D displays can be easier to look at
than 3D isosurfaces, especially for planar molecules.

23
Figure 14

Figure 15

24
References

[1] M. T. Yin and Marvin L. Cohen. “Microscopic Theory of the Phase Transformation and
Lattice Dynamics of Si”. In: Phys. Rev. Lett. 45 (12 Sept. 1980), pp. 1004–1007. doi: 10.1103/
PhysRevLett.45.1004. url: http://link.aps.org/doi/10.1103/PhysRevLett.45.1004.
[2] F. D. Murnaghan. “The compressibility of media under extreme pressures”. In: Proc. Natl.
Acad. Sci. 30 (9 Sept. 1944), pp. 244–247. doi: 10 . 1073 / pnas . 30 . 9 . 244. url: http :
//www.pnas.org/content/30/9/244.short.
[3] Francis Birch. “Finite Elastic Strain of Cubic Crystals”. In: Phys. Rev. 71 (11 June 1947),
pp. 809–824. doi: 10.1103/PhysRev.71.809. url: http://link.aps.org/doi/10.1103/
PhysRev.71.809.
[4] Charles Kittel. Introduction to Solid State Physics. 6th. John Wiley & Sons Inc, Dec. 1986.
isbn: 0471874744.
[5] Robert A. Wolkow. “Direct observation of an increase in buckled dimers on Si(001) at low tem-
perature”. In: Phys. Rev. Lett. 68 (17 Apr. 1992), pp. 2636–2639. doi: 10.1103/PhysRevLett.
68.2636. url: http://link.aps.org/doi/10.1103/PhysRevLett.68.2636.

25