Professional Documents
Culture Documents
The total scattering atomic pair distribution function (PDF) method was used in
the experimental work of this thesis to investigate the local structure of the
FeSb2 Iron Antimonide. We have different structural model to represent the
structure of FeSb2 compound where different structural models can be used and
compared, and the model resulting in a better fit is the best model to represent
the data. Where modeling can be done Under different length scales allowing
comparison local and average structural features. From (PDF gui) software we
can compare between the calculated PDF and the measurement PDF and this
agreement between calculated and measurement PDF can be calculated
through the residual function (Rw) and its value determines the goodness of fit,
where lower (Rw) represent better agreement.
Acknowledgments
Dedication
TABLE OF CONTENTS
Abstract ……………………………………………………………........... i
Acknowledgments …….……………………………………………………………. ii
Dedication ....……………….……………………………………………... iii
List of Figures ....……………….………...………………………………….
List of Tables ....……………….……………………….…………………….
Chapter 1. Introduction
1.1 Semiconductor Iron Antimonide, FeSb2
When extending the field of view beyond a single unit cell to consider many
units cells at once, the repetitious arrangement of atoms in a crystal creates the
appearance of planes of atoms separated by empty spaces. Many planes exist
for each space group. A common system for identifying lattice planes is with
Miller indices h, k, and l. In simple terms, the h, k, and l represent the
fractional coordinates where a particular plane intersects the a, b, and c axes of
a unit cell, respectively. This method avoids troublesome coordinates by
working in reciprocal space. For example, if a plane never intersects with one
of the axes, the intersection point in real space would be “infinity.” However,
the Miller index value for this (non-existent) intersection would be zero, as the
reciprocal of infinity approximates as zero. Likewise, a plane which intersects
at a halfway point along a unit cell axis would have a Miller index of 2 for that
direction.
The distance between atoms in a lattice, and thus between atomic planes, is on
the angstrom to nanometer scale. X-rays have wavelengths on the order of an
angstrom, and as a result X-rays in a crystal can be diffracted by lattice planes.
FIG. 1.2 Diagram for the formulation of Bragg’s Law. Bragg’s Law simplifies the
conditions for constructive interference of diffracted X-rays. In this diagram, X-rays are
incident from the upper left and reflected to the upper right. The detector, not shown in
this diagram, would be placed at a distance to the upper right.
When the incident X-ray wavelength is constant, there will be a unique angle θ
where constructive interference can occur for every family of planes with the
same d-spacing. The technique of X-ray diffraction relies on measuring the
angle θ of diffracted X-rays to calculate the d-spacing between lattice planes.
The relationship between lattice spacing d, lattice parameters a, b, and c, and
Miller Indices h, k, and l is known for the 7 lattice systems. Once the d-spacing
is calculated using Bragg’s Law, the corresponding plane can be identified by
its Miller Indices and the lattice parameters can be calculated using equations
such as in Table 1.1
CUBIC 1 h +k +l
2 2 2
=
d2 a2
Tetragonal 1 h 2+ k 2 l 2
= 2 + 2
d2 a c
Hexagonal 2
1 4 h +hk + k l
2 2
(and Trigonal) 2
= 2
+ 2
d 3 a c
Orthorhombic 1 h k l
2 2 2
= + +
d 2 a 2 b2 c 2
Monoclinic 1 h2 k2 l2 2 hlcosβ
2
= 2 2
+ 2
+ 2 2
+ 2
d a sin β b c sin β ac sin β
[
RICLINIC 1
2
h 2 kl k
2
2 hl
2
l 2 hk
= 2 2 + ( cosβcosγ −cosα ) + 2 2 + ( cosαcosγ −cosβ ) + 2 2 + (co
d
2
a sin α bc b sin β ac c sin γ ab
FIG. 1.3. Diffraction image for a powder sample. Powder diffraction is exemplified by
the formation of Debye rings in the diffraction image. Intensities at 2θ must satisfy
Bragg’s Law, while image intensities over the range of φ are due to the random
distribution of particles in a powder.
The intensity of counts in the Debye rings can be integrated as a function
of the angle between diffracted X-rays and the axis of the incident beam (the
semi-vertical angle of the diffraction cones). As shown in Fig. 5, the angle
from the incident beam to where the diffracted X-ray intersects the detector
screen is twice the Bragg angle θ. This is such a common experimental
geometry that the intensity of diffracted X-rays is frequently plotted as a
function of 2θ, where it is understood that 2θ is the angle relative to the
incident beam. An example of Debye rings on an area detector is shown in Fig.
1.4 (a), with the accompanying integrated 2- dimensional intensity plot shown
in Fig. 1.4 (b).
(a) (b)
FIG 1.4 (a) Detector plate image of CeO2 diffraction pattern, and (b) Integrated diffraction
pattern of CeO2. Detector plate images are integrated over φ to produce 2-dimensional
intensity plots as a function of 2θ. CeO2 is used as a standard/calibrant in XRD studies.
2.1 Introduction
2 dsinθ=nλ (2.1)
extracted with the help of crystal lattice and unit cell concepts. Peak positions
contain information about the unit cell lattice parameters (a, b, c, α, β, γ) and
the symmetry space group. Atomic positions (x, y, z) and atomic displacement
parameters (U) within the unit cell are contained in the peak intensity. The
peak width and shape are determined mainly by instrument parameters, crystal
from the diffraction patterns, and these include incoherent scattering such as
air, sample holder, etc., and diffuse scattering which is a result of local
information about the local structure. No clear and sharp Bragg peaks exist for
non-crystalline samples or for local disorders. This is why the long-range lattice
reciprocal space (or momentum space) where the diffracted intensity is measured
wavevectors, kinit and kfinal, respectively, and whose magnitude is given by:
4 π sinθ 2 πn
Q= =
λ d
One of the most powerful methods for investigating and determining the local
structural features of materials, and which has proven its capability on a wide
range of complex materials, is the atomic pair distribution function (PDF)
method. The PDF method will be used in the work presented in this thesis to
investigate the local structure and the differences from the average structure on
a sample of the Iron Antimonide FeSb2. Brief description of this method will be
given in this and the next sections.
One of the most general properties of the PDF method is that no assumption
of perfect lattice periodicity is necessary, which means that it can be applied
are needed in PDF analysis (Dinnebier, et al., 2008). The effect of low and
finite Q-ranges on the PDF will be discussed later. The direction of the vector
Q does not really matter for scatterings from isotropic samples (such as liquids,
glasses, and crystalline powder samples) and the scattering will in general
depend only on the magnitude of Q (Egami and Billinge, 2003; Dinnebier, et al.,
and the discussion and the equations of the PDF described in the next
reach higher Q values and to carry the PDF measurements with good quality
and high accuracy. Modern computer programs also made it easier to analyze
the diffraction experimental data. This is why the PDF method is promising
and fastly spreading and becoming widely used for local structural analysis of
a very wide range of complex materials. A brief historical re- view of the
starting from liquids and amorphous materials since 1930 to recent developments
Bordet, 2015) and the definition is written as (Egami and Billinge, 2003;
in a pair, g(r) is the pair distribution function, ρ(r) is the pair density function,
ρ◦
is the average number density of atoms, and G(r) is called the reduced pair
distribution function. ρ(r), g(r), and G(r) in general contain the same structural
information and are all abbreviated as PDF; they are related to each other
according to equation (2.3) with some differences in their basic properties and
behaviors and in general G(r) is the mostly used function (Egami and Billinge,
2003; Dinnebier, et al., 2008). ρ(r) is given by (Egami and Billinge, 2003):
1
2∑ ∑
ρ ( r ) = ρo g ( r ) = δ (r−r νμ )(2.4)
4 πN r υ μ
where N is the total number of atoms in the sample and rνµ is the separation
between the νth atom and the µth atom. ρ(r) describes the distribution of inter-
atomic distances through the Dirac delta function δ. The sum is over all the
atoms in the sample.
where Rµν is the magnitude of the difference between the positions of the νth
atom and the µth atom and b is the atomic scattering length.
The Normalized Structure function is given by (Dinnebier, et al., 2008;
Egami and Billinge, 2003):
I ( Q ) + ⟨ b ⟩ − ⟨ b2 ⟩
2
S ( Q )= 2
(2.6)
N ⟨b ⟩
S(Q) is then Fourier transformed from reciprocal space to real space to obtain
the PDF G(r) (Egami and Billinge, 2003):
∞ ∞
2 2
G ( r )= ∫ Q [ S ( Q )−1 ] sin ( Qr ) dQ= ∫ F (Q ) sin ( Qr ) dQ( 2.7)
π 0 π 0
F(Q) = Q[S(Q) - 1] is the reduced total scattering structure function. The PDF
is thus based on Fourier transformation of the diffraction data into real space
which
has more advantages in analyzing the diffuse scattering than in reciprocal space
generally because the diffuse scattering appears underneath the Bragg peaks
with relatively low intensity making it difficult to analyze in reciprocal space.
G(r) is the function that is directly related to the Fourier transformation
according to equation (2.7) and this is why it is the mostly used in the
description of the PDF (Egami and Billinge, 2003; Dinnebier, et al., 2008). The
correction, normalization, and Fourier transformation to obtain G(r) are done
using computer programs such as PDFgetX2 and PDFgetX3 for x-rays (Qiu, et
al., 2004(a); Juhas, et al., 2013) and PDFgetN for neutrons (Peterson, et al.,
2000). Detailed descriptions of the PDF experiments and data correction and
normalization are given by (Egami and Billinge, 2003).
1 bν bμ 1
G ( r )= ∑ ∑
r ν μ ⟨ b⟩ 2
δ ( r−r νμ ) −4 πr ρo= R ( r )−4 πr ρo (2.8)
r
b is the atomic scattering length, which in the case of x-rays is equal to the
atomic
form factor f(Q) and ⟨ b ⟩ is the average scattering length of the sample. The sum
b ν bμ
is over all the atoms in the sample. R ( r )=∑ ∑ 2
δ ( r −r νμ ) is called the radial
ν μ ⟨b ⟩
distribution function (RDF) and it is related to g(r), ρ (r), and G(r) through
(Egami and Billinge, 2003):
The RDF gives the number of atoms at a distance r from another atom. It can
be described as follows: if we have an arrangement of atoms, a random atom
can
be chosen as a reference atom (or the atom at the origin) and a spherical shell is
drawn centered at this origin. This shell is expanded to find every other atom in
the sample. Whenever an atom is found situated on the shell at a certain
separation r from the origin, an R(r) peak is assigned at this distance. This is
repeated by making each atom in the sample as the origin and finding its
separation from the other atoms. This description is shown schematically (in
two dimensions) in figure (2.1). The intensity of each peak depends on the
number of atoms which were found on the shell at the distance r and on the
scattering length of each atom in the pair. This is a schematic description of
how to calculate a PDF from the atomic arrangements of a structure. It can be
summarized by the statement given by (Billinge and Kanatzidis, 2004): “The
function G(r) gives information about the number of atoms in a spherical shell
of unit thickness at a distance r from a reference atom. It peaks at characteristic
distances separating pairs of atoms”.
Figure 2.1: A schematic representation of the RDF; peaks of R(r) appears whenever several atoms
are situated on a spherical shell of unit thickness at a specific distance r from the origin (from
Billinge, 2007).
The calculated PDF can be fitted and compared with the measured PDF
through a process called modeling to extract the local structural information of
the sample. The parameters of the model are refined in order to obtain a best
agreement be- tween the calculated and the measured PDFs. The PDF
modeling and the refinable parameters will be described later in this chapter.
for lower values of Qmax. This is why it is necessary to reach enough Q max
values experimentally. An example of a structural information that might be
lost when Qmax is low is the different but close bond lengths Zn - Se and Zn -
Te in the compound ZnSe0.5Te0.5 which was studied using neutron diffraction
(Qiu, et al., 2004); the bond lengths differ by 0.14 Å and the peaks start to
resolve, revealing this difference and the exact bond length for the two pairs, at
larger values of Qmax (> 40Å). The resolution in r is usually written as (Egami
and Billinge, 2003; Petkov, 2012):
2π
δr= (2.10)
Qmax
Figure 2.2: The effect of the value of Qmax on the PDF G(r) (from Bordet,
2015).of the real and sharp atomic pair peaks in the PDF (Egami and Billinge,
2003; Toby and Egami, 1992).
Figure 2.3: The effect of the resolution of the instrument on the PDF of nickel;
(up) better instrumental resolution, (down) less instrumental resolution where
the damping at larger values of r is clear (from Proffen, 2006).
2.3.1.3 The Type of the Radiation Beam Used and Counting Statistics
In general, using neutron beams may have advantages more than using x-rays.
Another consequence caused by using x-rays is the fact that the atomic form
factor f(Q) depends on the atomic number (Z) of the element; atoms with low
Z (such as oxygen) have low f(Q), and this is reflected in the PDF as small
peaks for the atomic pairs that contain low Z atoms. This is shown in figure
(2.5) for the x-ray PDF of LaMnO 3 compared to neutrons for which their
scattering length is large for low Z atoms resulting in more intense peaks; this
is clear in the first peak representing the Mn - O pair.
Figure 2.5: An example of the PDFs of LaMnO3 showing the differences when x-rays and neutrons
are used; the difference coming from the fact that neutrons are more sensitive to low Z atoms
unlike x-rays (from Bordet, 2015).
Atomic displacements are generally results of the thermal motion of the atoms
in the pair around the equilibrium. The motions of two atoms may be of two
types: correlated (in-phase) or un-correlated (out-of-phase) (Jeong, et al.,
1999). The degree of thermal motion, which is measured by the atomic
displacement parameter (U), and the type of the motion produce effects in the
PDF, mainly in the peak width. Larger thermal motion (due to higher
temperature) and uncorrelated motion both lead to an increase in the PDF peak
width and a decrease in its intensity, while less thermal motion and correlated
motion produce narrow sharp and intense peaks. The difference between the
correlated and un-correlated atomic motions is schematically shown in figure
(2.6).
Figure 2.6: (a) The un-correlated atomic motion and (b) the correlated atomic
motion (from Jeong, et al., 1999).
At low values of atomic pair separation r, the atoms tend to move in a correlated
manner due to the low density of atomic pairs and this causes the first PDF peaks
of the first neighboring atoms to be sharper (less width) compared to peaks at
larger r coming from far neighbors; the increased atomic separation affects the
correlated motion where it becomes more un-correlated since the density of atomic
pairs in- creases with r. In other words, the PDF peak width depends on r; it
increases with increasing r.
Structural information about the studied sample are contained in the measured
PDF. Using the PDF method allows for extracting information at different length
scales: short, medium, and long ranges, making it appropriate for both local and
average structure determination. Two types of information can be extracted from
the PDF: direct and indirect information, these are described below.
The measured PDF is obtained after correction and normalization of the diffraction
data as discussed earlier; the measured PDF is thus an absolute function where
some structural information is extracted directly and easily from it. This is why this
information are described as direct information and they are contained in the PDF
peak positions, intensities, and shapes (Egami and Billinge, 2003; Dinnebier, et al.,
2008).
2.4.1.1 Peak Position
Peaks of the PDF appear at distances (r) separating pairs of atoms. Thus, the peak
position yields directly the atomic pair separation or the bond length. Peaks at low
r correspond to atomic pairs of shorter bond lengths and peaks at higher r to atomic
pairs of longer bond lengths. The lowest r value is the distance between an atom
and its first nearest neighboring atoms. No real peak exists for r values lower than
the first neighbor distance. The PDF peaks extend to the largest inter-atomic
distance in the whole sample. Non-crystalline materials have very small number of
broad PDF peaks, due to the very short-range order and the long-range disordered
structure, where they appear at low r values only. On the other hand, crystalline
materials have large number of sharp and well-defined peaks extending to large
values of r because of their long-range order.
The PDF Peak intensity, or more precisely the PDF peak integrated intensity, is
directly related to the number of the atoms at a distance r from another certain
atom; this number is called the coordination number. The integrated intensity is
proportional to the coordination number at r; larger integrated intensity
corresponds to a larger coordination number. This can be directly obtained by
b ν bμ
integrating R ( r )=∑ ∑ 2
δ (r−r νμ) (Egami and Billinge, 2003) which describes the
ν μ ⟨b ⟩
coordination number as the number of atoms situated at a spherical shell centered
at an origin atom as described earlier:
r2
N C =∫ R ( r ) dr (2.11)
r1
where NC is the coordination number, and r 1 and r2 are the integration limits
defining the peak with the coordination number NC.
Most of the structural information about the sample cannot be obtained directly.
These include quantitative determination of the different parameters and the
differences between the local and average structures of the sample which are
usually related to the important properties of the sample. These kinds of
information are obtained through modeling of the PDF and thus are called indirect
information. The modeling of the PDF to obtain this information is discussed in the
next section.
√
N
∑ ω ( r i ) [G obs ( r i ) −Gcalc ( r i ) ]2
i=1
Rω = N
(2.12)
∑ ω( r i)G 2
obs (r i)
i=1
Rw is called the residual function and its value determines the goodness-of-fit;
lower Rw represents better agreement. Gobs and Gcalc are the observed and calculated
PDFs, respectively. w(ri) is the weighting factor. Usually, a value of around 10 %
of Rw represents a good fit for crystalline samples.
Modeling is done using computer programs such as the PDFgui program (Farrow,
et al., 2007) which will be used in this thesis work. The refinement parameters
according to the program PDFgui are discussed below. All the equations used to
account for the different effects are used by the program. Initial values of the
sample dependent and experimental dependent parameters are provided by the user
and are then varied until the best fit is reached.
These include the structural information about the sample which are generally
obtained from conventional crystallographic analysis: the lattice parameters, the
atomic positions, atomic occupancy of each site, and the space group symmetry.
They also include information about the atomic thermal motion and correlated or
un-correlated motion of the atoms. Structural parameters can be refined, and space
group symmetry constraints can be imposed in order not to break the symmetry of
the structure. Since the thermal motion and the un-correlated motion of the atoms
contribute to the PDF peak width as described earlier, they must be taken into
account for the PDF peak width of the calculated PDF. This is done through
multiplying by a Gaussian distribution function determined mainly from the atomic
displacement parameters (uij) which can be varied in the PDFgui program. The
correlated atomic motion at low r must also be considered. The parameter δ in
PDFgui sharpens the PDF peak at low r. The parameters affecting the PDF peak
width are considered through the equation:
'
√
σ ij=σ ij 1−
δ1 δ2 2 2
− 2 +Q broad r ij ¿
r ij r ij
This equation is used by PDFgui. σij is the corrected width σ 'ij is the width due to the
thermal and uncorrelated motion, δ1 and δ2 are the peak sharpening parameters; in
PDFgui the δ1 parameter is used in the high temperature case and δ2 in the low
temperature case, and Qbroad is an experimental dependent parameter contributing to
the peak width and is described below. Accounting for the correlated atomic
motion in PDF is discussed in a study by (Jeong, et al., 1999) on samples of Ni and
InAs. Another sample dependent parameter contained in the PDFgui program is the
phase scale factor. This parameter is used to account for different crystallo- graphic
phases in the sample. Refining this parameter allows for the quantitative
determination of the fraction of each phase in the sample in question.
[ ]
∞ '
1 ' sin Q max (r −r ) sin Qmax (r +r ' ) '
'
G ( r )= ∫
π 0
G (r )
r−r '
−
r +r '
dr (2.14)
sin Qmax r
is called the termination function, and G (r) is the PDF without the effect
r
of finite Qmax. This is to take account for the termination ripples that appear due to
the finite Q-range as mentioned earlier.
PDFgui allows for refinements over different length scales defined through a
minimum and a maximum value of r. This is called the refinement range and the
model is calculated through this range only, and the resulting refined values are
valid for this range.
The Qbroad parameter takes account for the broadening of the PDF peaks at high r
that generally comes from the noise at high Q values. In general, the two
parameters Qdamp and Qbroad have greater effect at higher values of r and since they
account for experimental effects, they can be obtained through modeling the PDF
of a standard sample, such as nickel, and the resulting refined values can then be
used for the sample in question without further refinement.
The data scale factor is refined to account for any errors coming from the
correction and normalization of the diffraction data. By default, the value of the
data scale factor is unity and refined values close to unity indicate, in general, good
correction and normalization of the data (Billinge, 2002).