FYSAST

Examensarbete 15 hp
Juni 2010
Shape functions in calculations
of differential scattering cross-sections
Anders Johansson
Institutionen fr fysik och astronomi
Department of Physics and Astronomy
Abstract
Two new methods for calculating the double dierential scattering cross-
section (DDSCS) in electron energy loss spectroscopy (EELS) have been
developed, allowing for simulations of sample geometries which have been
unavailable to earlier methods of calculation. The new methods concerns
the calculations of the thickness function of the DDSCS. Earlier programs
have used an analytic approximation of a sum over the lattice vectors of
the sample that is valid for samples with parallel entrance and exit-surfaces.
The rst of the new methods carries out the sum explicitly, rst identifying
the unit cells illuminated by the electron beam, which are the ones needed
to be summed over. The second uses an approach with Fourier transforms,
yielding a nal expression containing the shape amplitude, the Fourier trans-
form of the shape function dening the shape of the electron beam inside
the sample. Approximating the shape with a polyhedron, one can quickly
calculate the shape amplitude as sums over its faces and edges.
The rst method gives fast calculations for small samples or beams, when
the number of illuminated unit cells is small. The second is more ecient in
the case of large beams or samples, as the number of faces and edges of the
polyhedron used in the calculation of the shape amplitude does not need to
be increased much for large beams.
A simulation of the DDSCS for magnetite has been performed, yielding
diraction patterns for the L
3
edge of the three Fe atoms in its basis.
Supervisor: Jn Rusz
Department: Department of Physics and Astronomy,
Division of Materials Theory
Examiner: Kjell Pernestl
Populrvetenskaplig
sammanfattning
Fr att analysera material anvnder man ofta olika typer av diraktion, det
vill sga man lter en vg av ngot slag spridas mot materialets atomer.
Denna vg kan vara en elektromagnetisk vg, vanligtvis rntgenstrlning,
eller en elektrons vgfunktion. P andra sidan fngar man upp ett dirak-
tionsmnster, som med rtt tolkningar ger mycket information om vilken
struktur materialet har.
En annan vanlig metod fr att analysera material r elektronenergifr-
lustspektroskopi (EELS). Med den fr man reda p egenskaper hos material,
inte bara genom att studera vilka diraktionsmnster som uppstr, utan ock-
s genom att underska hur mycket energi de elektroner som skickas genom
materialet frlorar i de inelastiska processer som sker dr. D man analyse-
rar energispektrumet frn experimentet kan man f detaljerad information
om vilka atomer som ingr i materialet. En relativt ny metod inom EELS,
som har utvecklats fr att underska materials magnetiska egenskaper, r
EMCD (electron magnetic chiral dichroism). Dr jmfr man spektrum frn
olika delar av diraktionsbilden som representerar olika riktningar i vgvek-
torverfringen, q, skillnaden mellan elektronens vgvektor fre och efter
reaktionen. De mjliga reaktionerna i en magnetiserad atom varierar med q
och genom skillnaden i spektrumen fr olika q kan man utlsa atomspecik
information om materialets magnetisering.
Vid diraktion r det dierentiella tvrsnittet r den storhet som beskri-
ver den utgende strlens intensitet i en given riktning. Kan man berkna
detta kan man frutsga vilket diraktionsmnster som ett givet experiment
kommer att ge. Berkningarna r i de esta fall omjliga att utfra analy-
tiskt och mste gras numeriskt med datorprogram. Tidigare har det bara
funnits program som kunnat berkna tvrsnittet fr vissa srskilda typer av
experiment. Dessa program har jag nu utkat fr att mjliggra simulering-
ar fr en strre klass av experiment. De formler som har anvnts tidigare
krvde att ingngs- och utgngsytorna fr strlen i det material som skul-
le analyseras var parallella och att kristallstrukturen sg ut p ett srskilt
stt. Med de tv nya metoder jag implementerat kan dremot berkningar
utfras fr godtyckliga kristaller.
Fr att prova dessa nya metoder har jag gjort en simulering av tvrsnit-
tet hos magnetit med en viss magnetisering, ngot som inte kunde gras
tidigare.
Contents
1 Introduction 1
2 Theory 2
2.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1.1 Scattering experiments in materials science . . . . . . 2
2.1.2 Electron energy-loss spectroscopy . . . . . . . . . . . . 2
2.1.3 Dichroic spectroscopy . . . . . . . . . . . . . . . . . . 3
2.2 Cross-section calculations . . . . . . . . . . . . . . . . . . . . 4
2.3 Thickness functions . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.1 Analytic expression . . . . . . . . . . . . . . . . . . . . 9
2.3.2 Brute-force summation . . . . . . . . . . . . . . . . . . 10
2.3.3 Shape amplitude calculations of thickness functions . . 10
2.4 Calculating shape amplitudes . . . . . . . . . . . . . . . . . . 12
2.4.1 Calculating the shape amplitude . . . . . . . . . . . . 12
2.4.2 Constructing the polyhedron . . . . . . . . . . . . . . 14
3 Implementation 20
3.1 Earlier versions and a new structure . . . . . . . . . . . . . . 20
3.2 Implementing Brute-force calculations . . . . . . . . . . . . . 20
3.3 Implementing shape amplitude calculations . . . . . . . . . . 20
4 Results and discussion 23
4.1 Comparisons of calculations . . . . . . . . . . . . . . . . . . . 23
4.2 Case study: Magnetite in (111)-direction . . . . . . . . . . . . 24
5 Conclusions 27
Acknowledgements 28
References 29
Nomenclature
DDSCS Double dierential scattering cross section,

2

E
,
see equation (2.1)
EELS Electron energy loss spectroscopy
ELNES Electron energy-loss near edge structures
EMCD Electron energy-loss magnetic chiral dichroism
MDFF Mixed dynamic form factor, see equation (2.18)
Shape amplitude The Fourier transform of a shape function, see equa-
tion (2.42)
Shape function A function having value 1 when inside a body and
0 when outside, see equation (2.40)
TEM Transmission electron microscopy (or microscope)
Thickness function A part of the expression for the DDSCS in dynam-
ical diraction theory, see equation (2.29)
q Wave vector transfer, see equation (2.2)
1 Introduction
This project is about the calculation of dierential scattering cross sections
in the theory of electron spectroscopic experiments. More specically, it will
expand upon the theory of calculating the thickness functions being part of
the expressions for the cross-sections of this theory, as well as expanding
earlier programs for calculating the cross-sections.
Simulations of the cross-sections for various dierent samples and geo-
metries are important to provide guidance to experimentalists doing elec-
tron energy loss spectroscopy (EELS) and more specically EMCD, elec-
tron magnetic chiral dichroism. The simulations give information on how
the diraction patterns should look for specic samples, and also on the
dichroic signals, the dierences in intensities between dierent directions
corresponding to dierent wave-vector transfers, which are important for
EMCD experiments.
The earlier program packages have implemented a method which re-
stricts the calculations to samples with parallel surfaces and with two of the
lattice vectors parallel to these surfaces and the third perpendicular. The
new methods will allow for an arbitrary orientation of sample surfaces and
lattice vectors but be more computationally demanding.
The report is structured into several parts: First comes theory, with a
general introduction to the experiments and theory as well as details regard-
ing dynamical diraction theory and my own derivations aiming at the new
methods for calculating the cross-section. The second part is about imple-
mentation, describing briey what was done in order to make programs with
the new methods. Following this is results and discussion, containing tests
and comparisons of the methods as well as the results from a simulation of
the cross-section of magnetite. Last comes conclusions.
1
2 Theory
2.1 General
In this section a general introduction to scattering experiments and theory
will be given ending in the subject of Electron energy-loss magnetic chiral
dichroism.
2.1.1 Scattering experiments in materials science
Determining the structure and properties of solids have been a major con-
cern in materials science aiming towards a better understanding of material
properties and the ability to control them.
One specic and widely used method of determining crystal structure is
diraction, which can, in a simplied way, be described as directing waves
with a suciently short wavelength of some kind into a crystal and see
what comes out on the other side. What comes out is a so called diraction
pattern resulting from the scattering of the wave on the atoms of the crystal.
The most common diraction experiments use X-rays.
Another way of determining material properties is through spectroscopy.
In spectroscopic experiments one illuminates a sample with waves of varying
wavelength and, looking at the spectrum coming out, one can deduce which
wavelengths have been absorbed and from that obtain information about
the properties of the material.
In scattering experiments, the fundamental quantity of interest to both
theorists and experimentalists is the dierential cross section,

, which
measures in what solid angle a given cross-sectional area of an incident
beam is spread. This determines the observed intensity at various positions
of the detector. The cross section can be calculated from theory. [1]
2.1.2 Electron energy-loss spectroscopy
A specic spectroscopic method for determining material properties is Elec-
tron energy loss spectroscopy (EELS), which deals with the inelastic scat-
tering of an electron beam with on a studied material. These experiments
are often carried out in a transmission electron microscope (TEM) where
the intensity of electrons after leaving the sample is measured as a func-
tion of energy by an electron spectrometer. The inelastic events can be of
many types, as explained in gure 2.1, and sorting those out enables one
to deduce dierent properties of the sample. The excitations to valence
bands of inner shell electrons (atom core-loss) give rise to ionization edges
in the spectrum. Since every element have dierent binding energies for the
2
2.1. General Theory
electrons in the inner shells, examining these edges is a way of determining
what elements constitute the sample. The spectrum near these edges gives
detailed information and is referred to as ELNES (Electron energy-loss near
edge structures) [2]. [3]
R
e
l
a
t
i
v
e
f
r
e
q
u
e
n
c
y
Zero loss peak
Atom core-loss peak
Fine structure
Plasmon resonance
Electron energy loss (eV)
Figure 2.1: Idealized version of an EELS spectrum. The zero-loss peak
corresponds to electrons passing through the material without taking
part in any inelastic events. The plasmon resonance corresponds to
electrons interacting with electrons in the valence band of the material.
The atom core-loss and its ne structure corresponds to events where
an electron of some inner shell of the materials atoms are excited by
the incident electron. (Source: [4])
When doing experiments with an energy resolution the dierential cross
section is replaced by the double dierential scattering cross-section (DDSCS),
in a general form described by

I,F
[I[V [F)[
2
(E E
F
E
I
), (2.1)
where I and F refers to possible initial and nal states of the combined
system of the target and probe electron.
2.1.3 Dichroic spectroscopy
For determining the magnetic properties of materials on the microscopic
scale several techniques have been developed. A material is said to exhibit
3
2.2. Cross-section calculations Theory
dichroism if the absorption of light (or electron waves for that matter) de-
pends on its polarization. This often depends on the magnetisation of the
material and can thus be used as a quantitative measure of magnetisation.
The method of using circularly polarized x-rays in x-ray absorption spec-
troscopy (XAS) to discover dichroism in materials has been around since
1987 [5] and is commonly referred to as XMCD, x-ray magnetic circular di-
chroism. Starting in 2006 a new technique called electron energy-loss mag-
netic chiral dichroism (EMCD) have been developed and used for measuring
dichroism with the ordinary methods of ELNES spectroscopy[6]. Using elec-
trons instead of x-rays allows for a higher spatial resolution. In the theory
of EMCD the polarisation vector of the x-rays is replaced by the wave
vector transfer q, the dierence between nal and initial wave vector of the
incident electron:
q = k
f
k
i
. (2.2)
Instead of having the waves polarized in a specied way, something which
is harder to do with electrons than with x-rays, one can use an unpolarized
beam and place the detector at two dierent positions representing dier-
ent wave vector transfers that are orthogonal and phase shifted by
1
4
of a
wavelength [6]. This corresponds formally to circular polarization. The di-
chroic signal is then obtained as the dierence between the near-edge spectra
for these detector positions, displaying the dierence in excitations allowed
for dierent atomic orbitals in a magnetized sample.
This is just a schematic way of describing the interaction of the electron
beam with the atoms of a sample. In a crystal, the wave functions of the
probe electrons cant be described as single plane waves, which is implied
in the description above, but rather has to be written as a superposition
of Bloch waves, the solutions to the Schrdinger equation for electrons in
a periodic potential. Describing this theory will be the subject of the next
section.
2.2 Cross-section calculations
Here I will briey present the theory of dynamical diraction. It is the theory
used to calculate the DDSCS of ELNES and thus the theory employed in
the programs simulating those experiments.
The DDSCS in ELNES, within the rst-order Born approximation and
for an incident electron represented by a plane wave, can be written as [7]

E
=
4
2
a
2
0
k
f
k
i
S(q, E)
q
4
, (2.3)
with
S(q, E) =

i,f
[i[e
iq r
[f)[
2
(E
f
E
i
E), (2.4)
4
2.2. Cross-section calculations Theory
where a
0
is the Bohr radius, = 1/
_
1 v
2
/c
2
a relativistic factor, (E
i
, [i)),
(E
f
, [f)) the initial and nal energies and electron states of the ionised atom
and E the observed energy loss of the incident electron. S(q, E) is known
as the dynamical form factor (DFF).
As mentioned in the previous section, approximating the probe electron
as one plane wave is in general not applicable for scattering inside a crystal.
The wave functions of the incoming and outgoing electron should instead be
written as superpositions of Bloch waves with dierent wave vectors (indexed
j, l below), in r-representation this is

in
(r) =

jg

(j)
C
(j)
g
e
i(k
(j)
+g)r
, (2.5)

out
(r) =

lh

(l)
D
(l)
h
e
i(k
(l)
+h)r
, (2.6)
where
(j)
,
(l)
are the excitation factors of the Bloch waves labelled j, l and
C
(j)
g
, D
(l)
h
are the so called Bloch coecients. Finding these coecients is
the objective of the theory of dynamical diraction.
To calculate the DDSCS one has to solve the Schrdinger equation in
the crystal potential using the proposed forms of the wave-functions given in
Eq. (2.5) and (2.6). In doing this one has to apply continuity conditions of
the wave-functions (which are modelled as plane waves outside the crystal)
at the crystal surfaces. We have:
k
(j)
=
in
+
(j)
( n
1
), (2.7)
k
(l)
=
out
+
(l)
n
2
, (2.8)
where
(in)
,
(out)
are the respective wave-vectors of the incoming and out-
going electron outside the sample, n
1
, n
2
are the normal vectors of the
crystal faces where the electrons enter and leave the sample as pictured in
Fig. 2.2 and
(j)
,
(l)
are small parameters giving the change in the normal
components (the only ones allowed to change) of the wave vectors.
The details of solving the Schrdinger equations of incoming and outgo-
ing waves are given in ref. [2, 7]. The solution is, for high energies, in the form
of a linear eigenvalue problem with the s as eigenvalues and vectors formed
of the Bloch coecients as eigenvectors. Solving this, in principle innite
dimensional, eigenvalue problem thus gives the s and Bloch coecients.
Dening the planes containing the crystal faces by
n
1
r = t
1
, (2.9)
n
2
r = t
2
, (2.10)
for the face where the beam enters and leaves the sample respectively, and
5
2.2. Cross-section calculations Theory
n
1
n
2
x
y
z
Figure 2.2: The conventions for coordinate systems and surface nor-
mals used throughout.The xyz-coordinates are dened by the lattice
of the crystal. One usually chooses z to be parallel to the third lattice
vector and x parallel to the rst reciprocal lattice vecto. The order of
the lattice vectors are choosen after conventions for diernt lattices.
imposing the continuity conditions of Eq. (2.7) and (2.8) one gets[7]:

(j)
= C
(j)
0

e
i
(j)
t
1
, (2.11)

(l)
= D
(l)
0

e
i
(l)
t
2
, (2.12)
where

here and in the following denotes complex conjugation. We will
however always use an origin in the plane of the entrance face which gives
t
1
= 0 and hence
(j)
= C
(j)
0

.
Adding an absorptive part to the crystal potential and treating it as a
rst order perturbation, the Bloch coecients will not change but the s
obtain an imaginary part:

(j)

(j)
+ i
(j)
, (2.13)
where the real part,
(j)
, will equal the
(j)
of the non-absorptive calculation.
For the outgoing wave we will, for reasons explained further by Nelhiebel
[2], need to do some complex conjugation, which here will be equivalent to
postulating

(l)

(l)
i
(l)
. (2.14)
This follows the convention
(l)
> 0 and the physical requirement that the
outgoing wave is attenuated more if the inelastic event happened further
away from the exit surface.
6
2.2. Cross-section calculations Theory
This allows us to write the wave functions as:

in
(r) =

jg
C
(j)
0

C
(j)
g
e
(i
(j)
+
(j)
) n
1
r
e
i(
in
+g)r
, (2.15)

(out)
(r) =

lh
D
(l)
0

D
(l)
h
e
(i
(l)
+
(l)
)( n
2
rt)
e
i(
out
+h)r
, (2.16)
where t hereafter will be used instead of t
2
. This t is what will be referred
to as the thickness of a sample.
The DDSCS, which was already given in Eq. (2.3), should actually be
written as being proportional to the square of the transition matrix element
[2], i. e.

E
[
out
[f[V [i)[
in
)[
2
. (2.17)
This means it will obtain phase factors composed of the terms in the expan-
sions of the wave functions in Eq. (2.15) and (2.16). In squaring the sums we
will have to sum over each index twice. This means that we must introduce
primed indices. Furthermore we will have interference terms so that the
DFF will be generalized to the MDFF (mixed dynamical form-factor)[8]:
S(q, q

, E) =

i,f
i[e
iq r
[f)f[e
iq

r
[i)(E
f
E
i
E), (2.18)
where the wave vector transfers q, q

are dened as
1
q =
out
+ (
(l)
+ i
(l)
) n
2

in
(
(j)
+ i
(j)
)( n
1
) +h g, (2.19)
q

=
out
+ (
(l

)
+ i
(l

)
) n
2

in
(
(j

)
+ i
(j

)
)( n
1
) +h

. (2.20)
All this will yield the nal DDSCS:

E
=
4
2
a
2
0

out

in
1
N
a

aghg

jlj

X
jlj

ghg

(a)
S
a
(q, q

, E)
q
2
q
2
, (2.21)
where the phase factor X
jlj

ghg

is given by
X
jlj

ghg

(a) = C
(j)
0

C
(j)
g
D
(l)
0
D
(l)
h

C
(j

)
0
C
(j

)
g

D
(l

)
0

D
(l

)
h

. .
=Y
jlj

ghg

e
[i(
(l)

(l

)
)(
(l)
+
(l

)
)]t
e
i(qq

)a
,
(2.22)
and where a denotes the positions of all atoms in the sample. The MDFF can
be calculated in various ways but that will not be covered here and instead
I will refer to [7, 9, 10]. The summation in Eq. (2.21) can be simplied in
dierent manners and that will be the focus of the next section.
1
Here it should be noted that Nelhiebel [2] uses the opposite denition, Q = k
initial

k
nal
, of wave vector transfer.
7
2.3. Thickness functions Theory
2.3 Thickness functions
This section will focus on dierent ways of simplifying the calculation of the
DDSCS of Eq. (2.21). First the thickness function, being a part of the total
DDSCS, will be dened and then three dierent methods of calculating it
will be presented.
First dening

(ll

=
(l)

(l

)
,

(ll

)
+
=
(l)
+
(l

)
,
(2.23)
with similar denitions for j, j

, we can write the DDSCS of Eq. (2.21) in a

perhaps more suggestive form:

E
=
4
2
a
2
0

out

in

ghg

jlj

Y
jlj

ghg

e
(i
(ll

(ll

)
+
)t

1
N
a

a
e
i(qq

)a
S
a
(q, q

, E)
q
2
q
2
.
(2.24)
This can be simplied if we write a as composed of a lattice vector, R, and
a basis vector, u,
a = R+u. (2.25)
The dierences in q and q

comes mostly from the h, h

, g, g

-vectors since
the and are small [7]. This allows us to take the MDFF out of the sum
over j, l, j

, l

. The MDFF does not depend on what unit cell we are in, R,
but only on the basis, u, enabling us to only sum it over the basis vectors.
We can note that the combination hg(h

) = G, a reciprocal lattice
vector.
2
All this enables us to simplify things. We write
1
N
a

a
e
i(qq

)a
=
1
N
R

R
e
[(i
(ll

(ll

)
+
) n
2
+(i
(jj

(jj

)
+
) n
1
]R
e
GR
. .
=1

1
N
u

u
e
i(qq

)u
.
(2.27)
2
Writing things out explicitly we have:
q q

= (
(l)
+ i
(l)
) n
2
(
(j)
+ i
(j)
)( n
1
)
(
(l

)
i
(l

)
) n
2
(
(j

)
i
(j

)
)( n
1
) +G
= (
(ll

+ i
(ll

)
+
) n
2
+ (
(jj

+ i
(jj

)
+
) n
1
+G
(2.26)
8
2.3. Thickness functions Theory
This allows us to write the DDSCS as
4
2
a
2
0

out

in

ghg

1
N
u

u
e
i(qq

)u
S
u
(q, q

, E)
q
2
q
2

jlj

Y
jlj

ghg

T
jlj

l
(t), (2.28)
where T
jlj

l
(t) is the so called thickness function dened by:
T
jlj

l
(t) = e
(i
(ll

(ll

)
+
)t
1
N
R

R
e
[(i
(ll

(ll

)
+
) n
2
+(i
(jj

(jj

)
+
) n
1
]R
. (2.29)
This function can be evaluated by dierent means for dierent sample
geometries. These are explored in the following sections.
2.3.1 Analytic expression
For specic samples the summation in Eq. (2.29) can, with some approx-
imations, be carried out analytically and brought to a closed form. The
conditions are as follows:
The samples surfaces are parallel and orthogonal to the z-axis. The samples
lattice have one lattice vector c along z and the other two orthogonal to z.
This could for instance be a cubic or hexagonal lattice oriented according
to above mentioned conditions and with 001-surfaces. The lattice vector
parallel to z are denoted c and the other two a and b. This means that we
have
n
1
= z n
2
= z. (2.30)
The summation over R can now be split up into three sums over a, b and c
where the sum over a, b will yield the total number of atoms in each xy-plane
of the lattice, denoted N
xy
. The total number of planes (lattice points) in
z-direction is denoted N
z
.
We introduce the notation,

=
(jj

(ll

=
(jj

)
+

(ll

)
+
, (2.31)
enabling us to write T
jlj

l
(t) as
T
jlj

l
(t) =
N
xy
N
a
e
(i
(ll

(ll

)
+
)t
N
z

n=0
e
[(i
(ll

(ll

)
+
)( z)+(i
(jj

(jj

)
+
) z]( z)nc
=
N
xy
N
a
e
(i
(ll

(ll

)
+
)t
N
z

n=0
e
(i

)nc
.
(2.32)
The sum can be simplied further with some approximations. If we assume
the thickness of the sample to be much larger than the lattice constants,
9
2.3. Thickness functions Theory
t c, or equivalently N
z
1, c(i

) to be a small number and

write N
z
N
xy
= N
R
we get (in accordance with Nelhiebel [2])
3
T
jlj

l
(t) = e
[i(
(jj

+
(ll

)(
(jj

)
+

(ll

)
+
)]
t
2

cosh

t
2
sin

t
2
+ i sinh

t
2
cos

t
2
(

+ i

)
t
2
,
(2.33)
where the factors N
R
and 1/N
R
have cancelled. In the case of no absorption
this simplies to
T
jlj

l
(t) = e
i(
(jj

+
(ll

)
t
2
sin

t
2

t
2
. (2.34)
2.3.2 Brute-force summation
For more general lattices, or nonparallel surfaces, one has to evaluate T
jlj

l
(t)
of Eq. (2.29) in other ways. One such way is determining what lattice points
are illuminated by the electron beam and then carry out the summation in
Eq. (2.29) for all those lattice vectors. For reference I state the general
thickness function assuming no absorption:
T
jlj

l
(t) =
e
i
(ll

t
N
R

R
e
i(
(ll

n
2
+
(jj

n
1
)R
. (2.35)
2.3.3 Shape amplitude calculations of thickness functions
A dierent approach to calculating the thickness function is using a shape
function to dene the illuminated volume and after some manipulations
arrive at an expression containing the shape amplitude, i. e. the Fourier
transform of the shape function. This section will explore that possibility.
The formalism below is limited to the case with no absorption
4
.
The summation over lattice vectors of Eq. (2.35) is to be carried out over
all lattice vectors inside a body (t):
T
jlj

l
(t) =
e
i
(ll

t
N
R

R(t)
e
i(
(ll

n
2
+
(jj

n
1
)R
. (2.36)
Dening
p = (
(ll

n
2
+
(jj

n
1
), (2.37)
3
To be explicit: we evaluate the sum as a geometric sum, using N
z
+ 1 N
z
in the
denominator, replacing the resulting e
(

)c
in the numerator with the rst two terms
in the Taylor series and then expand the exponentials in trigonometric and hyperbolic
functions and collect terms.
4
The kernel, e
ipr
, of the Fourier transform used below requires p to be a real vector
10
2.3. Thickness functions Theory
this can be rewritten as:
T
jlj

l
(t) =
e
i
(ll

t
N
R

R(t)
e
ipR
. (2.38)
The sum can be rewritten as an integral with a Dirac-comb:

R(t)
e
ipR
=
_
e
ipr

R(t)
(r R) dr
=
_
e
ipr

R
(r R)(r) dr,
(2.39)
where (r) is the shape function of , dened by:
(r) =
_
1 if r (t),
0 if r / (t).
(2.40)
Equation (2.39) can be interpreted as a Fourier transform using the kernel
e
ipr
. The conventions for the Fourier transform used here will be:

f(p) = T [f] (p) =

_
f(r)e
ipr
dr, (2.41a)
f(r) = T
1
_

f
_
(r) =
1
(2)
3
_

f(p)e
ipr
dp. (2.41b)
The Fourier transform of a product of two functions is, according to the
convolution theorem of Fourier analysis, equal to the convolution of the
Fourier transform of each function. We can hence rewrite the last expression
of Eq. (2.39) as

R(t)
e
ipR
= T
_

R
(r R)
_
(p) T [(r)] (p)
=
1
V
u

G
(p G) (p),
(2.42)
recognising that the Fourier transform of a Dirac comb is a Dirac comb
in reciprocal space divided by the unit cell volume V
u
[11] and denoting
the Fourier transform of the shape function, called the shape amplitude, by
(p). If we note that that N
R
, being the number of unit cells inside (t),
roughly (but close enough for large N
R
) equals V

/V
u
and nally evaluate
the convolution of Eq. (2.42) we can write the thickness function as:
T
jlj

l
(t) =
e
i
(ll

t
V

G
(p G), (2.43)
11
2.4. Calculating shape amplitudes Theory
where the sum over G in practical cases only has to be computed for a few
G within some radius since the shape amplitude decays rapidly with larger
G-vectors.
To use the formula Eq. (2.43) in calculations one needs a convenient way
to calculate the shape amplitude. The shape of the beam can for a general
sample be modelled by a cone cut o by the entrance and exit surfaces as
pictured in Fig. 2.2. How to calculate the shape amplitude of such shapes
will be explored in the next section.
2.4 Calculating shape amplitudes
In this section I present a method for calculating the shape amplitude of
a general cut-o cone, a body which approximately describes the electron
beam through some sample with arbitrarily oriented top and bottom sur-
faces. This shape amplitude can easily be calculated by approximating the
cone with a polyhedron since, as described by Komrska [11], the shape amp-
litude of all polyhedrons can be calculated as a nite sum. The rst of the
following sections will present the formulae for calculating the shape amp-
litude of a general polyhedron while the second will describe how to construct
a suitable polyhedron from the parameters describing the cone.
2.4.1 Calculating the shape amplitude
The trivial case for p = 0 gives
(0) =
_
(r) dr =
_
V

1 dr = V

. (2.44)
The equations below deals with the cases p ,= 0.
As mentioned earlier, the shape amplitude, (p), of any polyhedron
can be expressed analytically as nite sums, allowing for easy computation.
The formalism below follows Komrska [11] but is corrected to correspond
to the convention of the Fourier transform mentioned above in Eq. (2.41).
Following Komrska [11] Eq. (3.3) we have:
(p) =
F

f=1

f
(p), (2.45)
where F is the number of faces of the polyhedron and
f
(p) is the two-
dimensional shape amplitude calculated for each face indexed f.
f
(p), in
turn, is given by:

f
(p) = i
p

N
f
p
2
e
ip
(O
f
)
i
[p
2
(p

N
f
)
2
]
E

e=1

fe
(p), (2.46)
12
2.4. Calculating shape amplitudes Theory
where

N
f
denes the unit outward normal of each face,
(O
f
)
is a vector
lying in each face serving as an origin for that face, E the number of edges
bounding each face and
fe
(p) the contribution of each edge, e, belonging
to the face f. The expression for
fe
(p) is:

fe
(p) = ip n
e
e
ip
(V
e
)
O
f
_
L
e
0
e
ip

t
e
l
dl
=
_

_
ip n
e
e
ip
(V
e
)
O
f
L
e
p

t
e
= 0,
p n
e
p

t
e
(e
ip
(V
e
)
O
f
e
ip
(V
e+1
)
O
f
) p

t
e
,= 0,
(2.47)
where n
e
is the unit outward normal of each edge (lying in the plane of the
face),
(V
e
)
O
f
denes the vertex at the beginning of the edge (relative to
(O
f
)
)
and
(V
e+1
)
O
f
the one at the end, L
e
is the length of the edge and

t
e
is the
unit vector pointing along the edge, i. e. from
(V
e
)
O
f
towards
(V
e+1
)
O
f
.
If p is perpendicular to the face indexed f,
f
(p) has to be calculated
in another way ([11] Eq. 3.11):

f
(p) =
i
p
P
f
e
ipd
f
, if p = p

N
f
. (2.48)
Here P
f
denotes the area of the face and d
f
denotes the orthogonal distance
from the plane the face is part of to the origin.
All these cases can be brought together and rewritten in the slightly
more handy form:
(p) =
_
V

if p = 0,

F
f=1

f
(p) otherwise,
(2.49)
where
f
(p) is given by

f
(p) =
_
_
_

i
p
P
f
e
ipd
f
if p = p

N
f
,
i
p
2
p

N
f
[p
2
(p

N
f
)
2
]

E
e=1

fe
(p) if p ,= p

N
f
,
(2.50)
where

fe
(p) is given by

fe
(p) =
_
_
_
i p n
e
e
ip
(V
e
)
L
e
if p

t
e
= 0,
p n
e
p

t
e
(e
ip
(V
e
)
e
ip
(V
e+1
)
) if p

t
e
,= 0.
(2.51)
Here (e
ip
(V
e
)
e
ip
(V
e+1
)
) could equally well be written
e
ip
(V
e
)
(1 e
ip

t
e
L
e
). (2.52)
This could suit some implementations more.
13
2.4. Calculating shape amplitudes Theory
2.4.2 Constructing the polyhedron
The quantities dening the shape are the following (see also Fig. 2.3):

k The normalised direction of the beam.

r
0
The radius of the beam before entering the sample.
The spreading angle of the beam once inside the sample.
u
0
The point where the centre of the beam enters the sample. This is can
mostly be set to zero, using the beam impact point as the origin. The
origin will, as noted after Eq. (2.9), always be placed in the upper
plane.
n
1
The upper surface normal.
n
2
The lower surface normal.
u
0

k
r
0

n
1
n
2
Figure 2.3: The parameters dening the cut-o cone. The vector u
0
dening the central impact point of the beam is drawn from some
origin in the upper plane.
The rst task is tracing out the circular cross-section of the beam outside
the sample. One can start by nding any normal vector not parallel to

k.
Then x
k
as depicted in Fig. 2.4 can be constructed as
x
k
=

k , (2.53)
allowing us to dene
y
k
=

k x
k
, (2.54)
14
2.4. Calculating shape amplitudes Theory
yielding a right handed coordinate system with basis x
k
, y
k
,

k where x
k
, y
k
is a basis for the cross-sectional plane of the cylinder representing the beam.
n
1
n
2
x
y
z

k
x
k
y
k
r
n
Figure 2.4: The basis of the cross-sectional plane of the cylinder rep-
resenting the beam outside the sample, x
k
, y
k
. The r
n
-vector is one
of the evenly distributed vectors which projected on the upper plane
will give the upper vertices of the polyhedron.
Now a set of N points evenly distributed around the circular cross-section
of the beam can be constructed as
r
n
= r
0
_
cos
_
n
N
2
_
x
k
+ sin
_
n
N
2
_
y
k
_
, (2.55)
where n runs from 1 to N.
Constructing the vertices of the polyhedron is now a task of projecting
these points rstly onto the upper surface along

k and secondly from there
to the lower surface with the correct slope away from

k.
The general problem of projecting a vector, r, unto a plane, along
some unit vector, m can be solved in the following way: The projected
vector r
m
is
r
m
= r + m m, (2.56)
where m is the unknown distance between the plane and r along m. The
plane itself is dened by its normal vector, n and the normal distance to
the origin, t, i. e.
n = t (2.57)
denes all points . Since r
m
we have the following equation for
determining m:
(r + m m) n = t, (2.58)
15
2.4. Calculating shape amplitudes Theory
yielding
m =
t r n
m n
. (2.59)
Returning to the initial problem of projecting the circle outlining the
beam onto the rst surface, hereafter denoted , the vertices of the poly-
hedron lying in can be written:

(V
n
)
= r
n

r
n
n
1

k n
1

k, (2.60)
where, as mentioned earlier, the origin is in , meaning t = 0.
r
n

l
n

(V
n
)

(V
n
)

n
1
n
2
Figure 2.5: The vertices of the polyhedron lying in the upper and
lower planes. The ones with the same index are connected via an edge
along

l
n
. The upper plane is denoted and the lower .
To get the vertices lying in the lower plane, , we have to construct
vectors

l
n
pointing out from

k with the correct angle.

l
n
is

l
n
= cos()

k + sin()
r
n
r
n
, (2.61)
as can be seen from Fig. 2.6. The vertices in the lower plane can now be
written

(V
n
)
=
(V
n
)
+ l
n

l
n
, (2.62)
where

l
n
is dened as in Eq. (2.61) and l
n
is
l
n
=
t
(V
n
)
n
2

l
n
n
2
, (2.63)
as follows from Eq. (2.59) with t the normal distance from u
0
to .
16
2.4. Calculating shape amplitudes Theory
r
n
r
n

l
n

l
n
l
n

(V
n
)

(V
n
)

(V
n
)

l
n
Figure 2.6: The vector

l
n
pointing from upper to lower vertex can be
calculated as shown in this gure. The length along the nth edge is
also shown. The inset shows the vectors in the grey plane of the large
picture in more detail.
Knowing that the edges of the polyhedron are those connecting
(V
n
)
and
(V
n
)
as well as those connecting
(,V
n
)
and
(,V
n+1
)
the polyhed-
ron is completely specied.
Further specications of the polyhedron
To use the polyhedron dened above for calculating the shape amplitude as
in section 2.4.1 the directions and normals of all edges and faces as well as
the lengths of all edges and areas of and distances to all faces have to be
found. How to nd these is explored here.
Using the conventions depicted in Fig. 2.7 all the necessary vectors and
lengths for each face are obtained by specifying:

t
(n)
1
=

(V
n+1
)

(V
n
)

(V
n+1
)

(V
n
)

L
(n)
1
=

(V
n+1
)

(V
n
)

, (2.64a)

t
(n)
2
=

l
n
(of Eq. (2.61)) L
(n)
2
= l
n
(of Eq. (2.63)), (2.64b)

t
(n)
3
=

(V
n+1
)

(V
n
)

(V
n+1
)

(V
n
)

L
(n)
3
=

(V
n+1
)

(V
n
)

, (2.64c)

N
(n)
=

t
(n)
1


t
(n)
2

t
(n)
1


t
(n)
2

, (2.64d)
n
(n)
1
=
n
1


t
(n)
1

n
1


t
(n)
1

n
(n)
2
=

t
(n)
1


N
(n)

t
(n)
1


N
(n)

, (2.64e)
17
2.4. Calculating shape amplitudes Theory
n
1
n
2
L
(n)
1
L
(n)
2
L
(n)
3
n
(n)
1
n
(n)
2
n
(n)
3
n
(n)
4
n
(n)
5
n
(n)
6

t
(n)
1

t
(n)
2

t
(n)
3

N
(n)

(V
n
)

(V
n
)

(V
n+1
)

(V
n+1
)
Figure 2.7: The conventions used for all normal vectors, edge vectors
and edge lengths dened by Eq. (2.64).
n
(n)
3
=

t
(n)
3
n
2

t
(n)
3
n
2

n
(n)
4
=

N
(n)

t
(n)
3

N
(n)

t
(n)
3

, (2.64f)
n
(n)
5
=

t
(n)
2


N
(n1)

t
(n)
2


N
(n1)

n
(n)
6
=

N
(n)

t
(n)
2

N
(n)

t
(n)
2

, (2.64g)
for all n. For the calculation of the shape amplitude the correct vectors
have to be picked out and sometimes reversed to keep the correct orientation
around the faces.
The area of a convex polygon (which all faces will be) can be calculated
by choosing one vertex and drawing the vectors pointing to all other vertices,
calculating the sum of the absolute value of the cross-product of all pairs
and dividing by 2 (i. e. dividing the polygon into triangles and using A

=
ab sin
2
).
The distance d
f
from the origin to each plane can be calculated as d
f
=
r
f


N
f
where r
f
is any vector in the plane, e. g. one of the vertices.
To calculate the volume of a general polyhedron one can utilise the local
topology formula stated by Franklin [12]:
V =
1
6

nij

t
(n)
i

(V
n
)
n
(n)
ij

(V
n
)

N
(n)
ij

(V
n
)
, (2.65)
where the sum is taken over all quadruples (
(V
n
)
,

t
(n)
i
, n
(n)
ij
,

N
(n)
ij
) with
(V
n
)
denoting the vertices,

t
(n)
i
the unit vectors along all edges i radiating from
18
2.4. Calculating shape amplitudes Theory
x
1
x
2
x
3
x
4
x
5
Figure 2.8: The area of any convex polygon can be calculated by the
method which in this picture gives: A = [x
1
x
2
+x
2
x
3
+x
3

x
4
+x
4
x
5
[/2
the vertex n, n
(n)
ij
the unit normal of the edge i pointing into the adjacent
face j and

N
(n)
ij
the unit normal of the face j pointing into the polyhedron.
These vectors are not exactly the ones dened by the conventions of Eq.
(2.64) but are nevertheless easy to retrieve from those vectors, if care is
taken in introducing minus signs where it is appropriate.
19
3 Implementation
In this section I will present the steps needed to implement the ways of
calculating the thickness function described above into an existing program
package.
3.1 Earlier versions and a new structure
This project has been built upon programs developed by Rusz et al. [7]. The
earlier version of the program, implemented in FORTRAN90, contained one
part for solving the problems of dynamical diraction, called dyndif, and
another separate part for calculating the MDFF, given the necessary data
(pairs of q, q

-vectors) from dyndif. The MDFF can be calculated with

various methods [7, 9, 10]. The version that have been used in the testing of
this project is mdff_dip using a dipole approximation and sum rules [9] to
calculate the MDFF in a simple way. In the earlier versions, the analytical
thickness function, Eq. (2.33), was implemented in dyndif and the program
ow was as described in Fig. 3.1.
Upon extending the earlier program with dierent ways of calculating
the thickness function I choose to split the thickness function into a separate
program, taking as input from dyndif the coecients Y
jlj

ghg

of Eq. (2.28)
paired with all

[
+
(ll

)|(jj

)
and outputting the products Y
jlj

ghg

T
jlj

l
(t)
to be used in the nal summation done in dyndif. This ow of the pro-
grams is depicted in Fig. 3.2 When doing calculations of the cross-sections
of materials one now has the choice between three dierent methods of cal-
culating the thickness function: Analytic, Brute-force (called Flood), and
Shape amplitude (called Shape).
3.2 Implementing Brute-force calculations
The brute-force summation was implemented with the help of a program
developed earlier by J. Rusz. This program, utilising the ood-ll algorithm
for determining what lattice vectors are inside the beam, was incorporated
into thick_func. A list of illuminated unit cells is generated and used for
the explicit summation performed according to Eq. (2.29) (or Eq. (2.35) in
the case of no absorption).
3.3 Implementing shape amplitude calculations
The implementation of the shape amplitude calculation of the thickness
function was performed along the lines presented in section 2.3.3 and 2.4.
20
3.3. Implementing shape amplitude calculations Implementation
dyndif
mdff
dyndif (with analytic
thickness function)
Figure 3.1: Program ow
before implementation of
dierent thickness func-
tions.
dyndif
mdff thick_func
dyndif
Figure 3.2: Program ow
after the implementation
of dierent thickness
functions.
Figure 3.3: An example of illuminated unit cells, here displayed as
single spheres, identied by the ood-ll algorithm.
21
3.3. Implementing shape amplitude calculations Implementation
The polyhedron approximating the beam was constructed as described in
section 2.4.2 along with all edge vectors, normal vectors, surface areas etc. A
sphere of a few reciprocal lattice vectors is also calculated for the summation
in Eq. (2.43), where only a few vectors around 0 will give non-negligible
contributions. Then, using the data from dyndif, the thickness function
is calculated as in Eq. (2.43) with the expressions for the shape amplitude
given in Eq. (2.49) and the following equations.
22
4 Results and discussion
First, this section will present the results of some comparisons done between
the old and new ways of calculating the DDSCS. Second, a simulation of
the DDSCS of magnetite in orientation with zone axis (111), a system which
could not be simulated earlier, will be described.
4.1 Comparisons of calculations
To see whether the new implementations of Flood and Shape were correct
they were tested against the old implementation of the analytic calculation
and each other. All these tests were done using a Fe
3
O
4
FCC (magnetite)
crystal. Figure 4.1 shows a comparison of the Analytic and Flood-ll cal-
culation with an increasing tilt of the upper surface. There is a perfect
match for a tilt angle of 0

but the dierence increases with larger angles,

as the approximations for the analytic formula become less valid. A similar
0
5
10
15
20
25
30
35
0 10 20 30 40 50 60 70 80 90 100
D
D
S
C
S
(
a
r
b
.
u
n
i
t
s
)
t/nm
0

10

15

20

25

30

Analytic
Flood-ll
Figure 4.1: Comparison of DDSCS between analytic calculation and
ood-ll calculation with increased tilt of the upper surface. The lines
for each tilt angle are shifted upwards, hence the scale on the y-axis
should not be interpreted as an absolute scale.
comparison between Flood and Shape, which should give the same results,
does show a large dierence with the current implementation of the shape
amplitude method. There are probably still some bugs to be sorted out in
the program. However, for samples with parallel surfaces orthogonal to z,
23
4.2. Case study: Magnetite in (111)-direction Results and discussion
all methods seem to work as indicated by Fig. 4.2, which shows a perfect
correspondence for dierent beam directions for all three methods.
0
5
10
15
20
25
30
35
0 10 20 30 40 50 60 70 80 90 100
D
D
S
C
S
(
a
r
b
.
u
n
i
t
s
)
t/nm
0

12

16

20

Analytic
Flood
Shape
Figure 4.2: Comparison of all three methods for a sample satisfying
the conditions for the analytic approximations to be valid. The plots
are for some dierent tilts of the incoming electron beam relative to
the sample surface and the values are both shifted and scaled with a
factor 10 in all cases but the 0

-beam.
The problems with Shape seems to lie in how small but non-zero p-
vectors are handled and this seems to be a larger problem for tilted surfaces.
Figure 4.3 shows plots of some tests in calculating the shape amplitude with
small p-vectors. The values around the origin get very high, although they
should not exceed 1. At p = 0 the shape function is explicitly set to 1.
4.2 Case study: Magnetite in (111)-direction
To test the new programs on a system that could not be simulated with the
earlier versions, a simulation of the DDSCS of magnetite was done. The
crystal structure of magnetite is shown in Fig. 4.4. The incident beam as
well as the magnetisation were set to the (111)-direction. Then calculations
of the DDSCS were performed for the L
3
edge for all three Fe atoms in
the basis, separating out the real and imaginary parts of the MDFF before
doing the last summation in dyndif. Calculations were made over a grid of
directions around the transmitted beam with (220) along the x-axis. This
gave the diraction patterns in Fig. 4.5. One can clearly see the pendellsung
oscillations, the variations in the intensity with thickness, and some small
dierences for the dierent atoms.
24
4.2. Case study: Magnetite in (111)-direction Results and discussion
100
0
100
200
300
400
0.2 0.1 0 0.1 0.2
I
m

(
p
)
/
V

|p| = p
x
100
0
100
200
300
400
500
0.2 0.1 0 0.1 0.2
I
m

(
p
)
/
V

|p| = p
y
100
0
100
200
300
400
500
600
700
0.2 0.1 0 0.1 0.2
I
m

(
p
)
/
V

|p| = p
z
0.5
0
0.5
1
1.5
2
2.5
3
0.2 0.1 0 0.1 0.2
R
e

(
p
)
/
V

|p| = p
x
0.5
0
0.5
1
1.5
2
2.5
3
0.2 0.1 0 0.1 0.2
R
e

(
p
)
/
V

|p| = p
y
0.5
0
0.5
1
1.5
2
2.5
3
0.2 0.1 0 0.1 0.2
R
e

(
p
)
/
V

|p| = p
z
Figure 4.3: The real and imaginary parts of the shape amplitude cal-
culated for some small p-vectors in x, y and z direction. This was
done with a beam in z-direction and a slightly tilted exit surface. The
sample thickness was set to 50 nm and the surface was tilted with 6

along x.
Figure 4.4: Magnetite structure, two unit cells shown. The red atoms
are Fe and the brown O.
25
4.2. Case study: Magnetite in (111)-direction Results and discussion
1.0
0.5
0.0
0.5
1.0
4 nm 14 nm 24 nm 34 nm 44 nm
1.0
0.5
0.0
0.5
1.0
1.0
0.5
0.0
0.5
1.0
1.00.5 0.0 0.5 1.0 1.00.5 0.0 0.5 1.0 1.00.5 0.0 0.5 1.0 1.00.5 0.0 0.5 1.0 1.00.5 0.0 0.5 1.0
1.0
0.5
0.0
0.5
1.0
54 nm 64 nm 74 nm 84 nm 94 nm
1.0
0.5
0.0
0.5
1.0
1.0
0.5
0.0
0.5
1.0
1.00.5 0.0 0.5 1.0 1.00.5 0.0 0.5 1.0 1.00.5 0.0 0.5 1.0 1.00.5 0.0 0.5 1.0 1.00.5 0.0 0.5 1.0
Figure 4.5: DDSCS (diraction patterns) for magnetite around the
transmitted beam with (220) along the x-axis. The rows correspond
to the three dierent Fe atoms in the basis and the columns to dierent
thicknesses. The tick labels are multiples of the Bragg vector G
220
and
the intensities of the maps are relative to the maximum and minimum
DDSCS in the calculation.
26
5 Conclusions
The earlier software package dyndif used to calculate the DDSCS in ELNES
has been extended with new methods of calculating the thickness function.
This allows for simulations with arbitrarily oriented sample-surfaces and
lattice vectors. One method, called Flood since it uses an implementation
of the ood-ll algorithm, has been implemented identifying the illuminated
unit cells of the sample and then explicitly summing over them. This is very
fast for small samples or beams, where the number of illuminated cells is
small.
Another method, called Shape calculates the thickness function by using
the shape amplitude of the body representing the beam in the sample. This
method scales very well since the number of summations depends on the
number of faces used in the polyhedron representing the beam, a number
that does not have to be increased much as the beam or the sample is
enlarged.
The DDSCS of magnetite with magnetisation along (111)-direction and
an incoming beam in (111)-direction have been calculated for a grid of dir-
ections around the transmitted beam, yielding the rst (111) zone axis sim-
ulation of the energy-ltered diraction pattoern of this system and being
the rst in a now much wider class of possible simulations to be used in
guiding the experiments in e. g. EMCD.
27
Acknowledgements
Thanks to my supervisor Jn Rusz, who has provided excellent guidance,
encouragement and proofreading. Thanks also to Stina and my other friends
for encouragement and some proofreading.
28
References
[1] David B. Williams and C. Barry Carter. Transmission Electron Mi-
croscopy: A Textbook for Materials Science. 2nd ed. Springer, 2009.
[2] Michael Nelhiebel. Eects of crystal orientation and interferometry in
electron energy loss spectroscopy. PhD thesis. cole Centrale Paris,
1999.
[3] R F Egerton. Electron energy-loss spectroscopy in the TEM. Reports
on Progress in Physics 72.1 (Jan. 2009), p. 016502. issn: 0034-4885.
doi: 10.1088/0034-4885/72/1/016502.
[4] Wikipedia User HatnCoat. EELS_Idealised (graphic). Licensed un-
der the Creative Commons Attribution-ShareAlike 3.0 License. 2009.
url: http://en.wikipedia.org/wiki/File:EELS\_Idealised.svg
(visited on 13/05/2010).
[5] G. Schtz et al. Absorption of circularly polarized x rays in iron.
Physical Review Letters 58.7 (Feb. 1987), pp. 737740. issn: 0031-
9007. doi: 10.1103/PhysRevLett.58.737.
[6] P. Schattschneider et al. Detection of magnetic circular dichroism us-
ing a transmission electron microscope. Nature 441.7092 (May 2006),
pp. 4868. issn: 1476-4687. doi: 10.1038/nature04778.
[7] Jan Rusz et al. First-principles theory of chiral dichroism in electron
microscopy applied to 3d ferromagnets. Physical Review B 75.21 (Apr.
2007), pp. 214425214434. issn: 1098-0121. doi: 10.1103/PhysRevB.
75.214425.
[8] H. Kohl and H. Rose. Theory of Image Formation by Inelastically
Scattered Electrons in the Electron Microscope. Advances in Elec-
tronics and Electron Physics 65 (1985), pp. 173227.
[9] Jn Rusz et al. Sum rules for electron energy loss near edge spectra.
Physical Review B 76.6 (June 2007), p. 4. issn: 1098-0121. doi: 10.
1103/PhysRevB.76.060408.
[10] L Calmels and J Rusz. Momentum-resolved EELS and EMCD spec-
tra from the atomic multiplet theory: Application to magnetite. Ul-
tramicroscopy In print (Apr. 2010). issn: 1879-2723. doi: 10.1016/j.
ultramic.2010.04.015.
[11] Ji Komrska. Algebraic expressions of shape amplitudes of polygons
and polyhedra. Optik 80.4 (1988), pp. 171183.
[12] W. Randolph Franklin. Local Topological Properties of Polyhedra.
Unpublished talk. 1992. url: http : / / wrfranklin . org / wiki /
Research/localtopo.pdf (visited on 12/05/2010).
29