Robert E. Dinnebier, Andreas Leineweber, John S.O.

Evans
Rietveld Refinement
Also of interest
Highlights in Applied Mineralogy.
Soraya Heuss-Aßbichler, Georg Amthauer, Melanie John (Eds.), 
ISBN ----, e-ISBN (PDF) ----,
e-ISBN (EPUB) ----

Structures on Different Time Scales

Theo Woike, Dominik Schaniel (Eds.), 
ISBN ----, e-ISBN (PDF) ----,
e-ISBN (EPUB) ----

Multi-Component Crystals. Synthesis, Concepts, Function

Edward R. T. Tiekink, Julio Zukerman-Schpector (Eds.), 
ISBN ----, e-ISBN (PDF) ----,
e-ISBN (EPUB) ----

Superconductors at the Nanoscale.

From Basic Research to Applications
Roger Wördenweber, Victor Moshchalkov,
Simon Bending, Francesco Tafuri (Eds.), 
ISBN ----, e-ISBN (PDF) ----,
e-ISBN (EPUB) ----

Electrochemical Storage Materials.

From Crystallography to Manufacturing Technology
Dirk C. Meyer, Tilmann Leisegang, Matthias Zschorner,
Hartmut Stöcker (Eds.), 
ISBN ----, e-ISBN (PDF) ----,
e-ISBN (EPUB) ----

X-ray Studies on Electrochemical Systems.

Synchrotron Methods for Energy Materials
Artur Braun, 
ISBN ----, e-ISBN (PDF) ----,
e-ISBN (EPUB) ----
Robert E. Dinnebier, Andreas Leineweber,
John S.O. Evans

Rietveld
Refinement

Practical Powder Diffraction Pattern Analysis using

TOPAS
Physics and Astronomy Classification Scheme 2010
Primary: 61.05.-a, 61.05.cp, 61.05.fn; Secondary: 61.72.-y, 61.72.Dd, 61.72.Nn

Authors
Prof. Robert E. Dinnebier
Max Planck Institute for
Solid State Research
Heisenbergstr. 1
70569 Stuttgart, Germany
[email protected]

Prof. Andreas Leineweber

TU Bergakademie Freiberg
Institute of Materials Science
Gustav-Zeuner-Str. 5
09599 Freiberg, Germany
[email protected]

Prof. John S.O. Evans

Durham University
Department of Chemistry
South Road
Durham DH1 3LE, UK
[email protected]

ISBN 978-3-11-045621-9
e-ISBN (PDF) 978-3-11-046138-1
e-ISBN (EPUB) 978-3-11-046140-4

Library of Congress Control Number: 2018954391

Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed
bibliographic data are available on the Internet at http://dnb.dnb.de.

© 2019 Walter de Gruyter GmbH, Berlin/Boston

Typesetting: Integra Software Services Pvt. Ltd.
Printing and binding: CPI books GmbH, Leck.
Cover image: Robert E. Dinnebier

www.degruyter.com
Preface
A detailed insight into atomic-level structure is of crucial importance in understanding,
designing and optimizing structural and functional materials. The technique of choice for
structure determination is usually single crystal diffraction. Unfortunately, single crystals
of suitable size and shape are simply not available for many materials, and one has to rely
on powder diffraction methods. These are often viewed as information poor, but actually
contain a wealth of information which is not available in a routine single crystal study.
Some of this is summarized schematically in Figure 1: Bragg peak intensities give detailed
information about a sample’s phase make-up and the composition and structure of
phases; peak shapes give information about its microstructure; and the diffuse intensity
between the Bragg peaks gives information on various aspects of local structure or
disorder. The technique can also be more easily adapted to perform in situ or operando
studies than single crystal methods, allowing us to follow materials in action. The
information content of a powder pattern is huge, but sophisticated analysis methods
are often needed to harvest the maximum amount of information from the data
(Dinnebier & Billinge, 2008). This book discusses how this can be done.
Although the powder diffraction method was developed as early as 1916 by Peter
Debye and Paul Scherrer (1916), for the first 50 years its use in crystallography was
almost exclusively limited to qualitative and semiquantitative phase analysis. In materi-
als science residual stress, texture and microstructure were also heavily investigated.
What held the field back, especially in the area of determination of the atomic crystal
structure, was the often-cited problem of powder diffraction: the accidental and sys-
tematic peak overlap caused by a projection of three-dimensional reciprocal space on to
the one-dimensional 2θ axis of a powder pattern. This leads to a strongly reduced
information content compared to a single crystal data set. The introduction of what
has become known as the Rietveld method marked a turning point (Rietveld, 1967, 1969).
Hugo Rietveld created a program to sidestep the overlap problem by modeling the whole
powder pattern with a set of parameters that can be refined by minimizing the difference
between the calculated and the measured powder pattern. One then finds that there
is sufficient information in the one-dimensional data set to reconstruct the three-
dimensional structure. The early development of different methods and Rietveld codes
has been discussed by various authors (Young 1993, van Laar 2018). A great feature of the
Rietveld method is the possibility to separate the main constituents of a powder pattern
(background, lattice, crystal structure and microstructure – see Figure 1), allowing
different aspects of a material’s properties to be probed separately. We cover the funda-
mentals of powder diffraction in Chapter 1 and describe the Rietveld whole pattern fitting
method in Chapter 2.
In its original implementation, the Rietveld method was used to refine a structural
model from neutron diffraction data of single-phase samples. The method has been
significantly extended over the last 50 years, and a variety of different whole-pattern
fitting techniques are now used. One set is the LeBail (1988) and Pawley (1981) methods

https://doi.org/10.1515/9783110461381-201
VI Preface

Powder diffraction pattern

Background Reflections

Position Intensity Profile

Sample Scattering from
(FWHM, peak shape)
sample holder, air,
etc.
Inelastic scattering: Instrument Sample
Compton scattering function broadening

Elastic diffuse scattering: Lattice parameters, Crystal structure: Real structure:

Space group:
Local structure Macrostrain Atomic positions Microstrain
Amorphous fraction Qualitative phase Magnetic moments Domain size
Lattice dynamics analysis Temperature factor Stacking disorder
Occupancy
Texture
Quantitative
phase analysis

Figure 1: Schematic view of the different parts of a powder pattern and their physical meaning
(adapted from Dinnebier & Billinge (2008) with permission from The Royal Society of Chemistry).

that are structure-independent whole powder pattern fitting approaches for deriving
reflection intensities for structure determination, testing lattice parameters and space
groups and for deriving microstructural parameters without the need for a crystal
structure. The uses of these and related methods are discussed in Chapter 3.
In the earliest Rietveld codes, the shapes of peaks in a powder pattern were fitted
with analytical functions such as Gaussian, Lorentzian or pseudo-Voigt functions, and
the 2θ-dependence of peak widths was described using simple empirical functions.
With increasing computational power it has become possible to develop more sophis-
ticated approaches in which the contributions to the peak shape from the instrument
and sample are treated separately giving access to exquisite detail about the micro-
structure of a material (size, strain and so on). We explore various approaches for this
in Chapter 4. Recently, faster and more sophisticated algorithms and much faster
computers have allowed Rietveld refinement using large supercells. This allows the
description of complex stacking faults as structural contributions to peak shapes
during Rietveld refinement (Coelho et al., 2016), and is discussed in Chapter 10.
The extension of the Rietveld method to X-ray powder diffraction data was
published in 1977 (Malmros & Thomas, 1977) and has had enormous impact in
many different areas. Most obviously, it allows the structural analysis of samples
using laboratory-derived data. It has also opened up the possibility of routine quanti-
tative analysis of multiphase mixtures (including amorphous content) by the Rietveld
method. This is now a major application of the Rietveld method in industry (Bish
 Preface VII

& Howard, 1988). We discuss the quantitative phase analysis of crystalline and amor-
phous samples in Chapter 5.
Chapter 6 introduces some of the ideas that are needed to refine particularly
complex structures, where the information content in the powder diffraction data
alone may not be sufficient. In these cases, extra information (“chemical knowl-
edge”) can be included in the form of constraints, restraints and rigid bodies. These
can help stabilize refinements, and even small protein structures can now be refined
using synchrotron powder diffraction data (Von Dreele, 1999). Similar information is
also needed if one tries to solve structures ab initio using the Rietveld method coupled
with simulated annealing (Newsam et al., 1992). We discuss some of the tricks needed
to do this in Chapter 7, but refer the reader to more specialized texts (David et al.,
2006) for discussion of other structure solution methods.
Chapter 8 discusses another approach that can help reduce the number of
parameters needed in Rietveld refinement of samples that undergo symmetry-low-
ering phase transitions – so-called symmetry-adapted distortion mode refinements.
The method can allow efficient analysis of otherwise complex problems and can help
rapidly identify the key structural changes occurring at the transition. The symmetry
ideas underlying this approach can be reapplied to a number of related problems such
as understanding spin arrangements in magnetically ordered materials, which are
covered in Chapter 9. We discuss here how the interaction of neutrons with magnetic
moments allows the determination of magnetic structure and some of the different
approaches to magnetic Rietveld refinement.
Turning to the bottom left corner of Figure 1, elastic diffuse scattering is mea-
sured during any powder diffraction experiment, but is usually ignored during
Rietveld refinement and treated as part of the general “background” of the pattern.
It is, however, possible to extract information about the local structure of a material
from this scattering. With modern X-ray and neutron sources, the powder pattern can
be measured over a large range of reciprocal space with good counting statistics. One
can then take the Fourier transform of the normalized pattern to produce the real
space pair distribution function (PDF) in the so-called total scattering approach. In
Chapter 11, we introduce the real-space Rietveld-type approaches to analyzing such
data and modeling local structure.
Finally, with modern instrumentation it is now possible to collect a vast number of
powder diffraction patterns of a sample as a function of external variables with a time
resolution in the second or even sub-second regime. This allows the detailed in situ or
operando study of many important processes. In Chapter 12, we describe some of the ways
to simulate multiple powder patterns to help plan such experiments. We also discuss how
to automate data analysis to allow fitting of large numbers of patterns using either a
sequential approach, where each data set is analyzed independently, or a parametric
approach, where parameters are refined across multiple data sets (Stinton & Evans,
2007). There are many as yet unexplored applications of these methods.
VIII Preface

Over 25 years have passed since the last textbook exclusively devoted to Rietveld
refinement (Young, 1993) was published, and there have been significant developments
since that time. We therefore decided to write a text that combines the fundamentals
of the method with practical details of its implementation. Several powerful public
domain Rietveld codes have been developed such as GSAS (Larson & Von Dreele,
1986), FULLPROF (Rodriguez-Carvajal, 1993), RIETAN (Izumi, 1989) and JANA (Petricek
et al., 2014), to name just a few. When preparing this book, we decided to focus on the
academic version of the program TOPAS (Coelho, 2018). The reasons are manifold. First
of all, TOPAS is extremely fast and extremely robust, making it a great tool for beginners.
It also tackles the different types of refinement problems (X-ray, neutron, fixed wave-
length, time of flight, energy dispersive, structure solution, distortion modes, magnetic
refinement, PDF analysis, multiple data sets and so on) that we think are most important.
Our number one reason, however, is related to the unique scripting language of TOPAS,
which allows experienced users to implement new developments in Rietveld refinement
themselves. One example of this is the symmetry mode approach described in Chapter 8,
which can be programmed without changing the TOPAS code (Campbell et al., 2007).
The macro language of TOPAS also allows the implementation of new ideas by the user
that can be rapidly shared with the entire user community (Scardi & Dinnebier, 2010). In
this spirit we provide TOPAS code/macros for many of the corrections and algorithms
described throughout the book. Many of these are available online at http://topas.dur.ac.
uk/topaswiki/doku.php?id=book, so the reader can test the ideas. Most of the examples
will work with any version of TOPAS, but some of the more advanced topics may only
work with version 6 and above (Chapters 9–11, some of Chapter 12). Finally, the provision
of standard macros via the topas.inc file acts as a wonderful “dictionary” detailing the
mathematics behind many TOPAS commands.
It should be noted that we expect the reader to be familiar with the basics of
crystallography and diffraction that are covered in many excellent introductory texts
(e.g., Giacovazzo et al., 2011; Pecharsky & Zavalij, 2009; Gilmore et al. 2019). We have
also kept the mathematical detail in the body of the text at the minimum level, which is
necessary for understanding. Where more explanation is required we refer the reader to a
stand-alone Chapter 13 that covers most of the higher mathematics needed to understand
Rietveld refinement.
We would like to acknowledge Prof. Jörg Ihringer who gave permission for using his
excellent script about diffraction for Chapter 13. Special thanks goes to Dr. Sebastian
Bette for writing a major fraction of the chapter on “stacking faults.” We would also like
to thank Prof. Branton Campbell and Dr Phil Chater for educating some of us through
on-going collaborations in the areas of Chapters 8, 9 and 11.

Robert Dinnebier
Andreas Leineweber
John Evans
 Preface IX

The book is dedicated to Alan Coelho (Brisbane) and Arnt Kern (Karlsruhe) for developing
the unique TOPAS program and making it available to the powder diffraction community.

References
Bish, D., Howard, S.A. (1988): Quantitative phase analysis using the Rietveld method. J. Appl. Cryst.
21, 86–91.
Campbell, B. J., Evans, J.S.O., Perselli, F., Stokes, H.T. (2007): Rietveld refinement of structural distortion-
mode amplitudes. IUCr commission on crystallographic computing newsletter No. 8, 81–95.
Coelho, A.A. (2018): TOPAS and TOPAS-Academic: an optimization program integrating computer
algebra and crystallographic objects written in C++. J. Appl. Cryst. 51,210–218.
Coelho, A.A., Evans, J.S.O., Lewis, J.W. (2016): Averaging the intensity of many-layered structures for
accurate stacking-fault analysis using Rietveld refinement. J. Appl. Cryst. 49, 1740–1749.
Debye, P., Scherrer, P. (1916): Interferenzen an regellos orientierten Teilchen im Röntgenlicht.
Nachrichten Kgl. Gesell. Wiss. Göttingen I. 1–15; II. 16–26.
David, W.I.F., Shankland, K., McCusker L.B., Baerlocher, Ch. (2006): Structure determination from
powder diffraction data, IUCr Monographs on Crystallography, Oxford University Press, 337 pages.
Dinnebier, R.E., Billinge, S.J.L. (eds.) (2008): Powder diffraction – theory and practice, The Royal
Society of Chemistry (RCS), Cambridge UK, 574 pages.
Giacovazzo, C., Monaco, H.L., Artioli, G., Viterbo, D., Milanesio, M., Gilli, G., Gilli, P., Zanotti,
G., Ferraris, G. (2011): Fundamentals of Crystallography (International Union of Crystallography
Monographs on Crystallography), 3rd edition, Oxford University Press, 872 pages.
Gilmore, C.J., Kaduk, J.A., Schenk H. (eds.): International tables for crystallography, volume H,
powder diffraction. Wiley, New York (US), to be published in 2019.
Izumi, F. (1989): RIETAN: a software package for the Rietveld analysis of X-ray and neutron diffraction
patterns. The Rigaku J. Vol. 6 10–20.
Larson, A.C., Von Dreele, R.B. (1986): GSAS – general structure analysis system, Report LAUR
86–748, Los Alamos National Laboratory, New Mexico, 13 Pages.
Le Bail, A., Duroy, H., Fourquet, J.L. (1988): Ab-initio structure determination of LiSbWO6 by X-ray
powder diffraction. Mat. Res. Bull. 23, 447–452.
Malmros, G., Thomas, J.O. (1977): Least squares structure refinement based on profile analysis of
powder film intensity data measured on an automatic microdensitometer. J. Appl. Cryst. 10, 7–11.
Newsam, J.M., Deem, M.W., Freeman, C.M. (1992): Accuracy in powder diffraction II, NIST Special
Publ. No. 846, 80–91.
Pawley, G. S. (1981): Unit-cell refinement from powder diffraction scans. J. Appl. Cryst. 14, 357–361.
Pecharsky, V., Zavalij, P. (2009): Fundamentals of powder diffraction and structural characterization
of materials, 2nd edition, Springer US, 744 pages.
Petricek, V., Dusek, M., Palatinus, L. (2014): Crystallographic computing system JANA2006: general
features. Z. Kristallogr. 229, 345–352.
Rietveld, H.M. (1967): Line profiles of neutron powder-diffraction peaks for structure refinement. Acta
Cryst. 22, 151–152.
Rietveld, H.M. (1969): A profile refinement method for nuclear and magnetic structures. J. Appl. Cryst. 2,
65–71.
Rodriguez-Carvajal, J. (1993): Recent advances in magnetic structure determination by neutron
powder diffraction. Physica B 192, 55–69.
Scardi, P., Dinnebier, R.E. (eds.) (2010): Extending the reach of powder diffraction modelling by user
defined macros, Materials Science Forum 561, Trans Tech Publications Ltd., 219 pages.
Stinton, G.W., Evans, J.S.O. (2007): Parametric Rietveld refinement. J. Appl. Cryst. 40, 87–95.
X Preface

Van Laar, B., Schenk, H. (2018): The development of powder profile refinement at the Reactor Centre
Netherlands at Petten. Acta Cryst. A74, 88–92.
Von Dreele, R.B. (1999): Combined Rietveld and stereochemical restraint refinement of a protein
crystal structure. J Appl. Cryst. 32, 1084–1089.
Young, R.A. (ed.) (1993): The Rietveld method, Oxford University Press, New York (US), Oxford, 298
pages.
Contents
Preface V

1 The powder diffraction method 1

1.1 Diffraction by crystallites 1
1.2 The Bragg equation 5
1.3 The Debye equation 10
1.4 Information loss in a powder pattern 13
References 14

2 The Rietveld method 16

2.1 The peak position 21
2.1.1 Corrections to the peak position 22
2.2 The intensity of a Bragg reflection 27
2.2.1 The atomic form factor 27
2.2.2 The coherent atomic scattering length 30
2.2.3 Displacement parameter 31
2.2.4 The structure factor 33
2.3 Intensity correction factors 36
2.3.1 Multiplicity 36
2.3.2 Lorentz–polarization factor 37
2.3.3 Absorption correction 41
2.3.4 Surface roughness 46
2.3.5 Extinction 47
2.3.6 Overspill effect 48
2.3.7 Preferred orientation 50
2.4 The scale factor 52
2.5 The peak profile 53
2.5.1 The box function 55
2.5.2 Gaussian distribution 56
2.5.3 Cauchy (Lorentz) distribution 58
2.5.4 The Voigt distribution 59
2.5.5 The TCHZ pseudo-Voigt function 62
2.5.6 The circles function 62
2.5.7 The exponential function 63
2.5.8 The 1/X-function 67
2.5.9 The emission profile 68
2.6 The background 71
2.7 The mathematical procedure 74
2.8 Agreement factors 77
XII Contents

2.8.1 Weighting schemes 78

2.9 Data collection and refinement strategy 80
2.10 Example of a Rietveld refinement 83
References 86

3 Structure independent fitting 89

3.1 Introduction 89
3.2 Constrained single peak fitting 90
3.3 The Le Bail method 91
3.4 The Pawley method 95
References 96

4 Peak shapes: Instrument  microstructure 97

4.1 Introduction 97
4.2 Determination of the instrumental resolution function (IRF) 98
4.2.1 Bragg–Brentano geometry (Cu-Kα1,2 doublet): FP approach 99
4.2.2 Bragg–Brentano geometry (Cu-Kα1,2 doublet): extended TCHZ
approach 106
4.2.3 Bragg–Brentano geometry with Cu-Kα1 radiation 107
4.3 Treatment of sample-dependent line broadening 109
4.3.1 Size broadening 111
4.3.2 Microstrain broadening 116
4.4 Neutron diffraction data with focus on the TOF method 125
References 129

5 Quantitative phase analysis 131

5.1 Introduction 131
5.2 External standard method 132
5.3 Rietveld method 133
5.4 Internal standard method 134
5.5 PONKCS method 134
5.6 Important correction factors 135
5.7 Elemental composition 136
5.8 Practical application 137
5.8.1 Determination of the IRF and background 137
5.8.2 Standard-less Rietveld QPA 138
5.8.3 External standard method for determining unknown or amorphous
content 140
5.8.4 Internal standard for determining unknown or amorphous
content 146
 Contents XIII

5.8.5 Application of the PONKCS method 149

References 149

6 Restraints, constraints and rigid bodies 151

6.1 Constraints in TOPAS 153
6.2 Restraints & penalties in TOPAS 154
6.2.1 Distance, angle and flatten restraints 157
6.2.2 Electrostatic potentials 160
6.3 Rigid bodies 161
6.3.1 Definition of a rigid body 162
6.3.2 Cartesian coordinates 162
6.3.3 Fractional coordinates 168
6.3.4 Spherical coordinates 168
6.3.5 Using internal rotations and translations to create a rigid body 169
6.3.6 Z-matrix 171
6.3.7 External degrees of freedom of a rigid body 174
6.3.8 Finding a starting orientation 176
6.3.9 Rotation of rigid bodies around an arbitrary axis 177
6.3.10 TLS matrices 177
6.3.11 Example of a Rietveld refinement using rigid bodies 179
References 182

7 Solving crystal structures using the Rietveld method 183

7.1 Introduction 183
7.2 Local and global minima 183
7.3 The method of simulated annealing 184
7.4 The simulated annealing algorithm in TOPAS 187
7.4.1 Example of a structure determination with simulated
annealing 190
References 194

8 Symmetry mode refinements 196

8.1 Introduction 196
8.2 Symmetry lowering phase transitions 196
8.3 Symmetry adapted distortion modes 199
8.3.1 Overview 199
8.3.2 Definitions and terminology 200
8.3.3 Symmetry mode labels and TOPAS implementation 201
8.3.4 Symmetry mode description of WO3 203
8.4 Examples of symmetry mode refinements 205
8.4.1 WO3 850 °C P4/ncc and 25 °C P21/n Rietveld refinements 205
XIV Contents

8.4.2 Using a genetic algorithm (GA) to identify symmetry 208

8.4.3 Using inclusion runs to identify symmetry 209
8.4.4 Exhaustive searching using a subgroup tree 210
8.4.5 Other order parameters – occupancies, rotational modes and
magnetism 210
8.5 Example of a Rietveld refinement using symmetry modes 210
References 216

9 Magnetic refinements 218

9.1 Introduction 218
9.2 Magnetic diffraction 218
9.3 Magnetic symmetry 220
9.4 Ambiguities in magnetic structures 227
9.5 LaMnO3 TOPAS worked example 228
9.5.1 Single phase nuclear and magnetic Shubnikov
approach 229
9.5.2 Separate nuclear and magnetic phases 230
9.5.3 Symmetry modes approach 232
9.5.4 Symmetry modes – determining the Shubnikov group 234
References 235

10 Stacking disorder 237

10.1 Introduction 237
10.2 Description of stacking orders 237
10.3 Types of stacking faults 239
10.4 Modeling the influence of stacking faults on diffraction 242
10.5 TOPAS implementation 246
10.6 Determining fault probabilities by Rietveld refinement 250
References 252

11 Total scattering methods 253

11.1 Introduction 253
11.2 Total scattering and atomic pair distribution function (PDF)
analysis 254
11.3 Parameters influencing the PDF 260
11.3.1 Peak position in a PDF 260
11.3.2 Peak area of a PDF 260
11.3.3 r-space resolution of a PDF 262
11.3.4 Processing artefacts 263
11.4 Example: Ni standard 264
 Contents XV

11.5 Calculation of a PDF 265

11.6 Joint refinements | Bragg + PDF 266
References 268

12 Multiple data sets 269

12.1 Simulation of angle dispersive powder patterns 269
12.1.1 Simulation of a single powder pattern 269
12.1.2 Simulation of a series of powder patterns 271
12.1.3 Scripting and the command line mode 272
12.2 Sequential Rietveld refinement 274
12.2.1 Introduction 274
12.2.2 Sequential refinement 274
12.2.3 Example of a sequential Rietveld refinement 277
12.3 Parametric Rietveld refinement 281
12.3.1 Parametric quantitative refinement 282
12.3.2 Non-crystallographic parameters 293
12.3.3 Plotting tricks 294
References 295

13 Appendix: Mathematical basics 296

13.1 Complex numbers 296
13.2 Power series 297
13.3 The Euler formula 299
13.4 Summing pointers 299
13.5 Fourier transformation 300
13.6 Convolution 301
13.7 Correlation function 303
13.8 The delta (δ) function 303
13.9 Matrix algebra 305
13.10 Crystallographic computing 307
13.11 Basis transformations 310
13.12 Rotations in the Cartesian system 311
13.13 Eulerian angles 312
13.14 Spherical coordinates 312
13.15 Triaxial ellipsoid 313
13.16 Symmetry operator 314
13.17 Probability density functions 315
13.17.1 Gaussian distribution 315
13.17.2 Lorentz distribution 316
13.17.3 The Voigt distribution 317
13.17.4 The pseudo-Voigt distribution 318
XVI Contents

13.17.5 More width parameters 319

13.17.6 Error function 320
13.18 The method of least squares 321
13.19 Spherical harmonics 325
References 327

Index 328
1 The powder diffraction method

Powder diffraction is one of the most powerful techniques for studying the atomic
structure of real materials. In this chapter we assume the reader has a knowledge of
basic crystallography and refer them to one of the excellent texts on the topic for
more advanced information (e.g., Giacovazzo et al., 2011; Egami & Billinge, 2012).
We will restrict ourselves to giving sufficient information that we can discuss
diffraction phenomena that are particularly important for powder diffraction. We
describe the basic effects of diffraction by crystalline and noncrystalline samples.
We also discuss the origins of peak broadening in powder diffraction patterns and
quantify how peak overlap leads to information loss relative to single crystal
experiments. We refer readers to specialist texts (e.g., Pecharsky & Zavalij, 2009;
Dinnebier & Billinge, 2008) for detailed information on experimental aspects of
powder diffraction.

1.1 Diffraction by crystallites

When radiation with a wavelength comparable to interatomic distances is incident on

a crystallite, two types of coherent elastic scattering1 occur: Bragg scattering that is
restricted to the lattice points in reciprocal space and elastic diffuse scattering that
can be found over wide regions of reciprocal space if a sample lacks translational
order. In this section we discuss the basic concepts for understanding scattering by
ordered crystallites.
An ideal crystal is an infinite three-dimensional periodic arrangement of structural
motifs. These motifs can consist of any number of atoms. If the motifs are replaced by
points, the points form a three-dimensional lattice in real space. The empty lattice can
be represented by a three-dimensional sum of equidistant δ-functions (see also Chapter
13 – Appendix: Mathematical Basics) called the lattice function g:

X
Na X
Nb
X
Nc
gðxÞ = δðx − a νa Þ δðx − b νb Þ δðx − c νc Þ, (1:1)
νa = − N a νb = − Nb νc = − N c

with the origin in the center of a lattice with basis vectors a, b and c and a number of
unit cells in the direction of each lattice vector of 2Na , 2Nb , 2Nc . For an ideal crystal
the number of unit cells is infinite. In Fourier (= reciprocal/diffraction) space, the

1 Coherent means a set phase relationship between the scattered radiation from two objects such that
interference occurs giving structural information. Elastic means no change in wavelength between
incident and scattered beams.

https://doi.org/10.1515/9783110461381-001
2 1 The powder diffraction method

amplitude scattered by the δ-functions at the lattice points in a direction represented

by scattering vector s is:

X
Na X
Nb
X
Nc
GðsÞ = e2πiðνa a  s + νb b  s + νc c  sÞ . (1:2)
νa = − Na νb = − Nb νc = − Nc

The scattering vector can be expressed in reciprocal coordinates:

a  s = h1 ; b  s = h2 ; c  s = h3 , (1:3)

which results in the so-called Laue function which when squared gives the distribu-
tion of diffracted intensity in reciprocal space:

sin2 ðπNa h1 Þ sin2 ðπNb h2 Þ sin2 ðπNc h3 Þ

GðsÞ2 = Gðh1 , h2 , h3 Þ2 = . (1:4)
sin2 ðπah1 Þ sin2 ðπbh2 Þ sin2 ðπch3 Þ

The intensity of GðsÞ2 is highest in directions where the phases of all summed
scattered waves differ by multiples of 2π. GðsÞ2 is plotted in two dimensions in
Figure 1.1, and we see that the maxima occur at integer coordinates of the scattering

G(s)2 G(s)2

2 2
1 1
–2 0 h –2 0 h
–1 –1 2 –1 –1 2
0 0
1 –2 h1 1 –2
h1 2 2

Motif b Lattice = b Crystal structure

a a

Figure 1.1: Top: The G(s)2 function of eq. (1.4) plotted for Na = Nb = 3 and Na = Nb = 10; note how the
peaks sharpen to become more like δ-functions as N increases. Bottom: Mathematical convolution of
a structural motif (consisting of two atoms) and a three-dimensional lattice (consisting of mathe-
matical points) to give a crystal structure.
 1.1 Diffraction by crystallites 3

vectors. Geometrically, these are at the lattice points of the reciprocal lattice.2 This
can be expressed mathematically as:

a  s = h; b  s = k; c  s = l, (1:5)

where h, k and l are integers. These are the three well-known Laue equations. They
tell us how the direction of the scattered beam is related to the real or reciprocal
lattices. Dividing the integer triple (h, k, l) by the largest common integer leads to the
coprime Miller indices.
We can also see from Figure 1.1 that GðsÞ2 becomes increasingly sharp at the
reciprocal lattice points as the number of unit cells increases.
The motifs associated with the crystal lattice can be represented by a
distribution of density (electron or nuclear density) %cell . X-rays interact with
the electrons while neutrons interact with the nuclei (apart from magnetic
scattering, see Chapter 9). The content of each unit cell can be expressed as
the sum of individual objects (atoms). If symmetry elements in addition to the
identity element (one-fold axis) exist, only the atoms of the asymmetric unit
must be given, as the remaining atoms in the unit cell are created by the
symmetry operators, while the entire crystal is built by the translational sym-
metry of the lattice.
Each object is represented by the convolution of its density and a δ-function
defining its position xj in the unit cell:
X
n  
%cell ðxÞ = %j ðxÞ  δ x − xj . (1:6)
j=1

In Fourier space the amplitude scattered by the unit cell at scattering vector s is:

n Z Z Z Z∞ Z∞ Z∞
∞ ∞ ∞
X  
Fcell ðsÞ = %j ðxÞe2πi sx
dx δ x − xj e2πi sx dx. (1:7)
j=1
−∞ −∞ −∞ −∞ −∞ −∞

The Fourier transform of a single atom depends on its form and scattering power and
is called the atomic form factor in the case of X-rays and the scattering length in the
case of neutrons:
Z∞ Z∞ Z∞
fj ðsÞ = %j ðxÞe2πi sx dx (1:8)
−∞ −∞ −∞

and is usually assumed to show a spherically symmetric distribution of scattering

power, implying that fj ðsÞ = fj ðsÞ. The structure factor then greatly simplifies to:

2 See Section 13.10 for a definition of the reciprocal lattice.

4 1 The powder diffraction method

X
n
Fcell ðsÞ = fj ðsÞ e2πi sxj . (1:9)
j=0

Finally, a crystal structure can be viewed as a convolution of the structural motif

(content of one unit cell) and the lattice (Figure 1.1) such that:

%crystal ðxÞ = %cell ðxÞ  gðxÞ (1:10)

represents the density distribution of the crystal. In Fourier space the amplitude
scattered by the crystal with scattering vector s reads:

Fcrystal ðsÞ = Fcell ðsÞ  GðsÞ. (1:11)

We therefore find that a single crystal gives rise to scattering in certain discrete
directions determined by the cell dimensions (through GðsÞ) with intensities deter-
mined by the cell contents (through Fcell ðsÞ). These scattered or diffracted beams
would appear as sharp spots on an X-ray film or a two-dimensional detector.
If we now consider a microcrystalline powder, there will be a large number of
crystallites in different orientations. Ideally, the individual crystallite size will be of the
order of 1 μm or less and all orientations will be equally probable. A key property of any
Fourier transformation is that the intensity distribution of the diffracted image is invar-
iant against translation of the object, which means that all diffracting crystallites in a

a*
re d ≥ λ/2
300 301 302 sphe
ng
200 201 202 iti
Lim

100 101
bea

102
ary

000 001 002

c*
Prim

here
sp Origin
ms
bea
ld

racted
Ewa

Diff
100

I
101

301

d*

Figure 1.2: Left: Schematic illustration of how in-plane rotation of the projection of the reciprocal
a*c*-plane around its origin produces the one-dimensional powder pattern. Right: Illustration of the
region of reciprocal space that is accessible in a powder measurement (outer circle). The smaller
circle represents the Ewald sphere (adapted from Dinnebier & Billinge (2008) with permission from
The Royal Society of Chemistry).
 1.2 The Bragg equation 5

Figure 1.3: Comparison between the diffracted beams originating from a single crystal (top) and a powder
(bottom). For the latter, some Debye–Scherrer cones are drawn (adapted from Dinnebier & Billinge (2008)
with permission from The Royal Society of Chemistry).

powder sample, which have identical orientation, contribute to the same diffraction
spots. Due to the different orientations of the crystallites, a single diffraction spot
becomes smeared out on the surface of a sphere with a radius given by the length of
the reciprocal lattice vector d*. Therefore, the orientation of the vector d* (= s) gets lost,
or, in other words, the three-dimensional reciprocal space is projected onto the d*-axis
(Figure 1.2, left).
The circular intersection of the smeared reciprocal lattice with the Ewald sphere
(Figure 1.2, right) results in the diffracted X-rays of the reflection hkl forming coaxial
cones, the so-called Debye–Scherrer cones (Figure 1.3). The smearing of reciprocal space
makes the measurement easier but results in a loss of information. Reflections from
lattice planes whose vectors lie in different directions but have the same d-spacing will
overlap and cannot be resolved or separated in the measurement. Some of these over-
laps are dictated by symmetry (systematic overlaps) and others are accidental.
Systematic overlap is not a serious issue for symmetry equivalent reflections (e.g., the
six Bragg peaks ð100Þ, (100), ð010Þ, ð010Þ, ð001Þ, ð001Þ of a cubic sample) since all
will have the same intensity and multiplicity (6). The problem is more serious for
nonsymmetry-related peaks (e.g., (300) and (221) of any cubic material, or (210) and
(120) of a cubic material with m3 Laue symmetry) or accidental overlap.
The two most important formulas for understanding powder diffraction, Bragg’s
law (W. L. Bragg, 1912) and the Debye equation (Debye, 1915), are derived below.

1.2 The Bragg equation

The most famous way of understanding the peak positions in a powder diffraction
pattern is via the Bragg equation (W. L. Bragg, 1912) that describes the principle of X-ray
diffraction as what appears to be a reflection of X-rays by sets of parallel lattice planes,
6 1 The powder diffraction method

Figure 1.4: Schematic drawing of a set of parallel

lattice planes (111) passing through all points of the
x cubic lattice.

(130)
k
s = k–k0

b*
2θ
a* k0 k
θ

M
θ ε Q θ d
P α
ε0 N
r Figure 1.5: Schematic illustration of the geometry
required to derive Bragg’s law.

characterized by the Miller indices hkl. All planes in a set are identical and separated by
a distance dhkl (Figure 1.4).3 Although this isn’t a good picture of what is happening
physically, it can be related to the Ewald approach given above using Figure 1.5.
The different path lengths of X-rays/neutrons scattered from atoms in different
planes lead to a phase shift Δ, which can immediately be deduced from Figure 1.5
using standard trigonometric relationships4:

3 Note that X-rays and neutrons are scattered by the electrons and/or nuclei of atoms. The view of
diffraction as reflection by lattice planes that may or may not contain atoms is a convenient
mathematical description but doesn’t represent the true physical process occurring.
4 cosðα + θÞ = cos α cos θ − sin α sin θ; cosðα − θÞ = cos α cos θ + sin α sin θ
 1.2 The Bragg equation 7

Δ ¼ PN þ NQ
¼ ðMN Þ cos ε0 þ ðMN Þ cos ε ¼ ðMN Þ½ cosðα þ θÞ þ cosðα þ θÞ (1:12)
¼ ðMN Þ½2 sin α sin θ

with

d = ðMN Þ sin α (1:13)

For constructive interference, Δ must be a multiple n of the wavelength λ, which

immediately leads to the Bragg equation:

nλ = 2d sin θ. (1:14)

For convenience, d is usually divided by n leading to:

d
λ = 2 sin θ = 2dhkl sin θ. (1:15)
n
Alternatively, Bragg’s law can be written in vector notation (Figure 1.5). The scattering
vector s is always perpendicular to the scattering plane, and is the difference between
the wave vectors of the incoming and outgoing beams given by k0 and k, respectively:

s = k − k0 . (1:16)

Setting the magnitude of k0 and k to 1/λ (the radius of the Ewald sphere) leads to the
Bragg equation in terms of the magnitude of the scattering vector s:

s = d*hkl . (1:17)

Bragg’s law results in sharp spots of high intensity that emerge from the crystal-
lite in specific directions given by the Bragg equation, and there is a one-to-one
correspondence between these Bragg spots and each set of crystallographic planes.
Each Bragg spot is therefore labeled with the same set of Miller indices, hkl, as the set
of planes that gave rise to it. The Debye–Scherrer cone created by the orientational
averaging becomes a sharp peak in a one-dimensional powder pattern.
Peaks in powder patterns of real samples will invariably be broadened. We can
understand one source of broadening by calculating the total derivative of the Bragg
equation with the d-spacing as the subject. Applying the chain rule, this is:

∂d ∂d nλ cos θ n
dd = dθ + dλ = dθ + dλ, (1:18)
∂θ ∂λ 2 sin θ sin θ 2 sin θ
which simplifies to:

dd dθ dλ
=− + . (1:19)
d tan θ λ
The dimensionless quantity dd/d can, for example, be interpreted as a microscopic
strain. Any variation of the d-spacing within a crystallite will then give rise to a strain
peak broadening dθ as discussed in Chapter 4.
8 1 The powder diffraction method

Finite crystallite domain size is a second source of broadening. If we assume an

infinite stack of lattice planes, the Bragg equation gives the position of δ-function
Bragg peaks. If the angle between the incoming beam and the lattice plane θ is
different by an amount ε from the Bragg condition, there will always be a lattice
plane inside the crystallite for which the accumulated extra path length produces a
phase shift of ðn + 1=2Þλ causing destructive interference (Figure 1.6). For a thick
crystal this will occur for an arbitrarily small ε, which explains why Bragg reflec-
tions are sharp.

)λ
+½
0
=

2Δ

(n
3Δ
Δ

θ+ε θ+ε =
Δ
0
1
dhkl
2
2θ
3

p–1
p

Figure 1.6: Path length difference of the scattered ray versus the depth of the lattice plane in the
crystal (adapted from Dinnebier & Billinge (2008) with permission from The Royal Society of
Chemistry).

The situation changes if the size of the crystallites is finite. In this case, for small ε the
plane for which Δ = ðn + 1=2Þλ holds may not be present in the crystal, thus leading to
an intensity distribution over some small angular range in θ. This is called size
broadening and is described by the Scherrer equation (Klug and Alexander, 1974).
We can derive the Scherrer equation approximately by considering the crystallite in
Figure 1.6, which will have a thickness in the direction perpendicular to p+1 lattice
planes of separation dhkl (Figure 1.6) of:

Lhkl = pdhkl . (1:20)

The additional beam path Δ between consecutive lattice planes at the angle
θ + ε is5:

5 Note that for small θ, sin θ  θ and cos θ  1.

 1.2 The Bragg equation 9

Δ ¼ 2d sinðθ þ εÞ ¼ 2dðsin θ cos ε þ cos θ sin εÞ

(1:21)
¼ nλ cos ε þ 2d cos θ sin ε  nλ þ 2d ε cos θ:

The phase difference is then:

Δ 4πεd cos θ
’ = 2π = 2πn + . (1:22)
λ λ
The phase difference ’L between the top and the bottom layer p is then:

4πεd cos θ 4πεLhkl cos θ

’L = p = . (1:23)
λ λ
If a is the amplitude of a wave diffracted at a single lattice plane, then according to
Klug and Alexander (1974), the resulting amplitude of all diffracted waves is:

sinð’L =2Þ
A=a p , (1:24)
ð’L =2Þ

with the maximum amplitude A0 = a p at ε = 0. Half maximum intensity is then:

A2 1 sin ð’L =2Þ

2

2
= = , (1:25)
A0 2 ð’L =2Þ2

which is satisfied when:

’L
= 1.4. (1:26)
2
From this we can approximate the misalignment angle at half maximum intensity
using eq. (1.23):

2.8λ
ε1=2 = . (1:27)
4πLhkl cos θ
The measured angular width fwhmhkl between the two points of half maximum
intensity on a 2θ scale leads to the Scherrer equation:

0.89λ
fwhmhkl = 4ε1=2 = , (1:28)
Lhkl cos θ
which gives a measure of the peak width in radians due to the finite particle
size. The prefactor depends on the shape of the grains (e.g., it is 0.89 for
perfect spheres and 0.94 for cubic-shaped grains) but is always close to unity.
Note that this equation is not valid for crystallites that are too large or too
small. For very large crystallites the peak width is governed by the coherence
of the incident beam and not by particle size. For nanoscale crystallites,
Bragg’s law fails and needs to be replaced by the Debye equation discussed
in the next section.
10 1 The powder diffraction method

1.3 The Debye equation

In general, decreasing the size of a crystal leads to an increase in the width of its
Bragg peaks as we have seen in the previous section. When the size of the crystallite
becomes very small, the width of the Bragg peaks is so large that they merge and
overlap, and it no longer makes sense to use δ-function Bragg peaks as the starting
point for the analysis. Even though the coherent diffraction is diffuse in nature and
distributed throughout reciprocal space at this point, it still contains useful structural
information.
Under the assumption that only elastic scattering occurs and each photon is
scattered only once, the structure amplitude for an ensemble of n atoms at fixed
positions without periodicity is given by6:

X
n
AðsÞ = fj ðsÞe2πi sxj (1:29)
j=1

with the conjugate complex:

X
n
A* ðsÞ = fj ðsÞe − 2πi sxj . (1:30)
j=1

The intensity can be calculated as:

X
n X
n
I ðsÞ = jAðsÞj2 = AðsÞA* ðsÞ = fj ðsÞfk ðsÞe − 2πi srjk , (1:31)
j=1 k=1

where rjk = xk − xj is the vector between the atoms j and k. For a simple diatomic
molecule (Figure 1.7), this would be7:

I ðsÞ = f1 ðsÞ2 + f2 ðsÞ2 + f1 ðsÞf2 ðsÞe − 2πisr12 + f1 ðsÞf2 ðsÞe2πisr12

= f1 ðsÞ2 + f2 ðsÞ2 + 2f1 ðsÞf2 ðsÞcosð2πs  r12 Þ (1:32)

Eq. (1.32) simplifies further for a homonuclear diatomic molecule (e.g., oxygen O2)
with f1 ðsÞ = f2 ðsÞ to:

I ðsÞ = 2f1 ðsÞ2 ð1 + cosð2πs  r12 ÞÞ. (1:33)

For O2 gas, scattering from an individual molecule will be coherent but that
from different molecules will be incoherent. In this case we will see a sum of

6 For simplicity, only the real part of the form factor is taken into account. See Section 2.2.1 for more
detail.
 
7 Using cos ’ = 21 ei’ + e − i’ .
 1.3 The Debye equation 11

scattering from all the molecules present, which will have every orientation
relative to the beam with equal probability. It is therefore necessary to take an
orientational average of the scattering leading to spherically averaged inten-
sity. The result is the equation known as the Debye equation, which is derived
below (Debye, 1915).
Analogous to the first step of eq. (1.32), eq. (1.31) can be rewritten as:

X
n X
n X
n
I ðsÞ = fj ðsÞ2 + fj ðsÞfk ðsÞe2πisrjk . (1:34)
k=1 j = 1 k = 1, k ≠ j

The double sum in eq. (1.34) contains nðn − 1Þ addends, which can be grouped
into 21 nðn − 1Þ pairs of addends describing the same atom–atom pair with the
distance vectors rjk and rkj = − rjk . Hence the corresponding pairs of addends
are complex conjugates. Decomposition of each addend according to eq. (13.1)
(see Chapter 13) leads to cancellation of the imaginary terms. Thus, eq. (1.34)
can be rewritten as:
X
n X
n X
n  
I ðsÞ = fj ðsÞ2 + fj ðsÞfk ðsÞ cos 2πs  rjk (1:35)
k=1 j = 1 k = 1, k ≠ j

corresponding to the second step in eq. (1.32). In order to determine the direction
average of eq. (1.35), let’s consider an arbitrary addend of the double sum, and
assume that φ is the angle between the scattering vector s and the direction vector
rjk between two atoms j and k (Figure 1.7):

s  rjk = s rjk cos ’. (1:36)

The direction average of the cosine of that particular addend then becomes the
integral over the surface of the sphere of radius rjk (Figure 1.8) divided by the area
2
4πrjk of that sphere:

k0
s = k–k0

2θ
s

k0
k

2
r12 φ r12
ε
1
Figure 1.7: Scattering by an object consisting of
ε0
two scatterers 1 and 2, separated by the vector r12.
12 1 The powder diffraction method

s
r.dφ

φ r.sinφ.dω
r

Figure 1.8: Definition of a volume element in spherical

coordinates.

Z
  1  
cos 2πs  rjk ¼ 2
cos 2πs rjk cos ’ dS: (1:37)
4πrjk
surface of sphere

As illustrated in Figure 1.8, the area element dS in eq. (1.37) is calculated by:
2
dS = rjk sin ’d’dω. (1:38)

Using appropriate limits of integration, this leads to:

Zπ Z2π
  1  2
cos 2πs  rjk = 2
cos 2πs rjk cos ’ rjk sin ’dωd’. (1:39)
4πrjk
0 0

The integration over ω simply yields 2π. The second integration over ’ can be
accomplished by substitution using z = cos ’, d’ = − dz= sin ’ and adaption of the
integration boundaries to give:
 
  sin 2πs rjk
cos 2πs  rjk = . (1:40)
2πs rjk

The direction average of eq. (1.35) is then obtained by using eq. (1.40), leading to the
Debye equation:
 
Xn X n X n
sin 2πs rjk
I ð sÞ = fj ðsÞ2 + fj ðsÞfk ðsÞ . (1:41)
k=1 j = 1 k = 1, k ≠ j
2πs rjk

The intensity must be multiplied by the number of particles N (molecules, nanocrys-

talline grains and so on) that contribute to the scattering. An example of using the
Debye equation to describe the scattering of CCl4 gas is shown in Figure 1.9. For
clusters of atoms such as small nanoparticles that are intermediate in size between a
diatomic molecule and a small chunk of crystal, the Debye equation is exact and may
be used to calculate the scattering intensity. We find that as the length scale of order
 1.4 Information loss in a powder pattern 13

Intensity [a.u]

1200

1000 CCl4 regular form factor

CCl4 constant form factor
800

600

400

200
2 sin (θ)
0.5 1.0 1.5 2.0 2.5 3.0 λ

Figure 1.9: Qualitative intensity distribution from elastic diffuse scattering (Debye equation) of CCl4
gas taking the form factors of carbon and chlorine into account (lower blue line) and with a constant
form factor (upper orange line).

increases, the diffracted intensity gradually evolves from having broad features as a
function of s = 2sinθ=λ (as in Figure 1.9) to the sharp peaks we associate with micro-
crystalline powders.

1.4 Information loss in a powder pattern

The loss of information due to peak overlap in a powder pattern can be quantified.
According to David et al. (2006), the average number of reflections within a shell of
width Δd* at a radius d* = 1=d (=s) in reciprocal space can be approximated for a
triclinic material with unit cell volume V by:
  2
ΔN d* e 2πVd* Δd* . (1:42)

The value steadily increases with a constant Δd* . If the expression is converted to
equidistant intervals in °2θ using the Bragg equation (eq. (1.15)), a maximum occurs
as illustrated in Figure 1.10 (left), which shows the average number of reflections in
0.1° 2θ intervals as a function of the scattering angle for Cu-Kα1 radiation for triclinic
unit cell volumes of 1000 and 1500 Å3.
All peaks with a separation that is larger than the experimentally resolvable
separation can be counted as distinguishable reflections. Choosing the resolvable
separation as a constant strain variation with the Bragg peak resolution Δd=d, the
number of distinguishable reflections for a triclinic system can be calculated as
(David & Shankland, 2008):
14 1 The powder diffraction method

Δ = 0.1° 2θ (Cu-Kα1 )
16 5000
Na4P4O8S4(H2O)6 V = 1000 Å3 Cu-Kα1, V = 1000 Å3
V = 1000 Å3 Δd/d = 10–3

Number of independent reflections

V = 1500 Å3 Δd/d = 10–4
4000
12 Single crystal
Number of reflections

3000
8
2000

4
1000

0 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
2θ/° 2θ/°

Figure 1.10: Left: Number of Bragg reflections within 0.1° 2θ intervals for Cu-Kα1 radiation for different
triclinic unit cell volumes over the entire diffraction range. The bar chart contains binned data from a
simulated powder pattern of Na4P4O8S4(H2O)6 that has a unit cell volume of 1000.04 Å3 (a = 8.815 Å,
b = 9.313 Å, c = 14.259 Å, α = 78.33°, β = 95.07°, γ = 119.26° (Ilyukhin et al., 1982)). Right: Number of
distinguishable Bragg reflections for a triclinic unit cell volume of 1000 Å3 for different high-
resolution powder and single crystal experiments over the entire diffraction range.

3
  1 − expð − 2πV Δd
d dmax Þ
*
Nind d*max e . (1:43)
3 Δd
d

Figure 1.10 (right) shows the effect of increasing resolution on the number of distin-
guishable reflections for a unit cell volume of 1000 Å3 as a function of the scattering
angle for Cu-Kα1 radiation. Even for the highest instrument resolution of Δd=d = 10−4,
as typical for third generation synchrotron sources, the number of reflections that can
be resolved (even assuming no sample broadening) quickly falls below what is
possible in single crystal diffraction as 2θ increases.

References

Bragg, W.L. (1912): The diffraction of short electromagnetic waves by a crystal. Proc. Camb. Phys. Soc.
17, 43–57.
David, W.I F.,Shankland, K. (2008): Structure determination from powder diffraction data. Acta Cryt.
A64, 52–64.
David, W.I.F., Shankland, K., McCusker L.B., Baerlocher, Ch. (2006): Structure determination from
powder diffraction data, IUCr Monographs on Crystallography, Oxford University Press, New
York (US), 337 pages.
Debye, P. (1915): Zerstreuung von Röntgenstrahlen. Annalen der Physik, 351, 809–823.
 References 15

Dinnebier, R.E., Billinge, S.J.L. (eds) (2008): Powder diffraction – theory and practice, The Royal
Society of Chemistry (RCS), 574 pages.
Egami, T., Billinge, S. (2012): Underneath the Bragg peaks - structural analysis of complex materials,
Volume 16, 2nd edition, Pergamon, 422 pages.
Giacovazzo, C., Monaco, H.L., Artioli, G., Viterbo, D., Milanesio, M. Gilli, G., Gilli, P., Zanotti, G.,
Ferraris, G. (2011): Fundamentals of crystallography (International Union of Crystallography
Monographs on Crystallography), 3rd edition, Oxford University Press, New York (US), 872
pages.
Ilyukhin, V.V., Kalinin, V.P., Kuvshinova, T.B., Tananaev, I.V. (1982): The crystal structure of
Na4P4O8S4(H2O)6. Dokl. Akad. Nauk SSSR 266, 1387–1391.
Klug, H.P., Alexander, L.E. (1974): X-ray diffraction procedures, 2nd edition, John Wiley & Sons, New
York, 966 pages.
Pecharsky, V., Zavalij, P. (2009): Fundamentals of powder diffraction and structural characterization
of materials, 2nd Edition, Springer US, 744 pages.
2 The Rietveld method

As we discussed at the end of Chapter 1, one of the key features of powder diffraction
is the accidental and/or systematic overlap of some Bragg reflections. The degree of
overlap of individual reflections depends on their separation in Δd/d and their peak
widths. Ultimately, the intensity information content in the powder pattern will be
determined by these factors (Figure 2.1).
A powder pattern consists of a set of N consecutively measured intensities yobs, i ,
where the running index i 2 ½1, .., N represents the measured position Xobs, i as the
diffraction angle 2θ (°), energy E (keV) or time-of-flight TOF (ms).1 The steps do not
have to be equidistant.
Most frequently, a powder pattern is measured at constant wavelength as a
function of the diffraction angle 2θ with equidistant angular steps of width Δ2θ.
In such a case, the index i represents the angular position in the powder pattern
according to 2θi = 2θstart + ði − 1ÞΔ2θ, with the starting angle 2θstart .
In the case of angular dispersive constant wavelength X-ray or neutron data
(Figure 2.2), the relationship between the measured angular space and the d-spacing
scale follows directly from the Bragg equation2:


λ
2θ=rad = 2arcsin (2:1)
2d

For energy dispersive X-ray data (Figure 2.4), the relationship between the measured
energy scale and the d-spacing scale is given by:

6.2
E=keV =   (2:2)
d= Å  sin θfixed

with θfixed the fixed detector angle used for data collection. The conversion factor
depends on the chosen units.
For TOF data (Figure 2.5), the relationship between the measured TOF scale and
the d-spacing scale is given by:

TOF = t0 + t1 d + t2 d2 (2:3)

where t0 , t1 and t2 are diffractometer constants characteristic of a given detector bank

on a TOF powder diffractometer. TOF is typically given in milli- or microseconds.
The fundamental idea behind the Rietveld method is simple: Instead of analyzing
the integrated peak intensities from a powder pattern in a single crystal-like fashion,
the entire information content of a powder pattern (Figure 2.2) available in step-
scanned intensity data is fitted with a model whose parameters are refined using a

1 TOPAS uses X for all the scales. We adopt this practice where needed.
2 Note the need in TOPAS to multiply by 360=2π to convert to degrees.

https://doi.org/10.1515/9783110461381-002
 2 The Rietveld method 17

40
35
30
Sqrt (rel. intensity)

25
20
15
10
5
0
–5
–10
67 68 69
2θ/°

Figure 2.1: Zoomed-in part of the powder pattern of quartz showing the measured pattern (black
circles), three single peaks (red lines) fitted to the individual Bragg reflections exhibiting different
degrees of overlap, and the difference curve below. The square root of the intensity is displayed.

least squares procedure to optimize the fit. This procedure intrinsically accounts for
peak overlap. According to the method of least squares, the squared sum of differ-
ences between the N observed yobs, i and calculated ycalc, i step-scanned intensities is
subjected to minimization (Figures 2.2–2.5):
X 
wi ðyobs, i − ycalc, i Þ2 ! Min (2:4)
i

The weight wi is usually derived from the variance of yobs, i as 1=σ2 ðyobs, i Þ while all
covariances between different yobs, i are assumed to be zero.
The calculated intensity ycalc, i is expressed by combinations of mostly nonlinear
analytic or nonanalytic functions as:
0 1
X X 2 
ycalc, i = @Sp Fcalc, s, p  Φs, p, i Corrs, p, i A + Bkgi (2:5)
p sðpÞ

The outer sum runs over all crystalline phases p with Bragg peaks in the powder
pattern, while the inner sum runs over all Bragg reflections s = ðhklÞ of a phase
p, which contribute to the position i in the powder pattern.3 A scaling factor Sp,
which is proportional to the weight fraction of phase p, is applied to the
reflection intensities of each phase. Corrs, p, i represents the product of various
 2
correction factors that need to be applied to the reflection intensities Fcalc, s, p 
that may depend on the diffraction geometry and/or individual reflection
indices. The contributions to the factor Corr are discussed in more detail in
Section 2.3. The value of the profile function Φs, p, i is given for the profile point i

3 The components of s, the Laue indices h, k and l, refer to the basis vectors a*, b* and c* of the
reciprocal lattice.
18 2 The Rietveld method

34000
32000
30000
28000
26000
24000
22000
Rel. intensity

20000
18000
16000
14000
12000
10000
8000
6000
4000
2000
0
–2000
20 30 40 50 60 70 80 90 100 110 120 130 140 150
2θ/°
180
170
160
150
140
130
Sqrt (rel. intensity)

120
110
100
90
80
70
60
50
40
30
20
10
0
20 30 40 50 60 70 80 90 100 110 120 130 140 150
2θ/°
10
9
8
7
In (rel. intensity)

6
5
4
3
2
1
0
20 30 40 50 60 70 80 90 100 110 120 130 140 150
2θ/°

Figure 2.2: Typical single phase Rietveld plot showing the observed angular dispersive powder
pattern (black circles) of quartz, the calculated powder pattern (red line), the difference curve (grey
line) and the markers of the reflection positions below. Top: linear intensity; middle: square root of
the intensity; bottom: logarithmic intensity.
Note: For better visibility of lower intensity reflections, the square root representation of the
measured intensity is typically used throughout the book.
 2 The Rietveld method 19

180
170
160
150
140
130
Sqrt (rel. intensity)

120
110
100
90
80
70
60
50
40
30
20
10
0
5 4 3 2 1
d/Å

Figure 2.3: Rietveld plot of Figure 2.2 converted to d-spacing (in Å units) with the square root of the
intensity.

90

80

70

60
Sqrt (rel. intensity)

50

40

30

20

10

30 35 40 45 50 55 60 65 70 75 80
E/keV

Figure 2.4: Typical Rietveld plot showing the energy dispersive powder pattern of silicon recorded
with θfixed = 7.77°.

relative to the position s = jsj = 2sinθ=λ of the Bragg reflection s. Therefore, the
peak profile depends only on the peak position given by the scalar s, and not on
hkl (i.e., s). This restriction is lifted in the case of anisotropic line broadening,
where an explicit hkl dependent Φs, p, i is considered (see Chapter 4). The
observed background coming from thermal diffuse scattering, incoherent
20 2 The Rietveld method

12
11
10
Sqrt (rel. intensity)

9
8
7
6
5
4
3
2
1
2000 4000 6000 8000 10000 12000 14000 16000 18000
TOF/μs

Figure 2.5: Rietveld fit to TOF powder pattern of CeO2.

scattering, inelastic scattering, sample environment and so on at position i in

the powder pattern is denoted as Bkgi .
This approach requires modeling of the entire powder pattern. To simplify and
understand this complex task, the information content of the powder pattern can be
divided into several parts, allowing the separation of groups of parameters with
respect to their origin:
– Peak position s that is geometrically determined by the crystallographic lattice,
space-group symmetry and instrumental factors.
 2
– Integrated peak intensity Fcalc, s, p  Corrs, p that is determined by the time- and
space-averaged crystal structure and geometrical contributions. Corrs, p is the
value of Corr at the peak position s for phase p.
– Peak profile Φs, p, i that is determined by the instrument profile and microstruc-
tural parameters of the sample.
– Background Bkgi .
Each part has contributions from both the sample and the instrument. Since
Rietveld refinement requires starting values of all parameters within a (relatively
narrow) range of convergence, it can be useful to consider different aspects of the
pattern separately according to different empirical, phenomenological or physical
models.
In the following sections, the factors affecting these different parameter groups are
described. The main focus will be on angular dispersive powder diffraction as this is the
standard method in most laboratories. Where useful, the corresponding code in the
TOPAS scripting (INP) language is given between border lines.4 Table 2.1 summarizes

4 In general, we quote fewer significant figures than would be carried within the TOPAS script for
clarity of presentation.
 2.1 The peak position 21

Table 2.1: Selection of important instrumental and sample contributions affecting position (as a shift)
and/or intensity (as a correction factor) and/or shape (as a convolution) of the Bragg reflections.

Correction Position Intensity Profile Section

(shift) (factor) (convolu-
tion)

Instrument
Zero shift x ..
Specimen displacement x ..
Equatorial divergence (fixed slit) x ..
Equatorial divergence (variable slit) x x .., ..
Size of source in the equatorial plane x ..
Specimen tilt x ..
Receiving slit length in the axial plane x ..
Receiving slit width in the equatorial plane x ..
Emission profile x x ..
Tube tails x .., ..
Axial divergence (prim., sec. Soller etc.) x x ..
Lorentz-polarization x ..
Sample
Linear absorption/transparency x x x .., .., ..
Surface roughness x ..
(An)isotropic microstrain x ..–..
(An)isotropic crystallite size x ..–..
Preferred orientation x ..
Extinction x ..
Overspill x ..
Multiplicity x ..
Displacement parameter x ..

common aberrations/contributions affecting position and/or intensity and/or shape

(as a convolution) of the Bragg reflections in a Rietveld refinement.

2.1 The peak position

For an angular dispersive powder pattern, the observed scattering angle 2θs of a
Bragg reflection s = ðhklÞ can be calculated from the corresponding d-spacing by
the Bragg equation corrected by aberrations Δ2θcorr due to misalignment of the
diffractometer or the sample, or due to transparency, axial divergence (or similar)
effects:


λ 1
2θs = 2arcsin + Δ2θcorr . (2:6)
2 ds
22 2 The Rietveld method

Given a set of lattice parameters (a, b, c, α, β, γ) or their reciprocal counterparts

(a*, b*, c*, α*, β*, γ*) and the unit cell volume V, the positions for all possible
reflections s can be calculated according to:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u h2 b2 c2 sin2 α þ k2 a2 c2 sin2 β þ l2 a2 b2 sin2 γþ
1 1uu
¼ 2hkabc2 ðcosαcosβ  cosγÞ
ds V t (2:7)
þ2kla2 bcðcosβcosγ  cosαÞþ2hlab2 cðcosαcosγ  cosβÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ h2 a2 þ k2 b2 þ l2 c2 þ 2hka b cos γ þ 2hla c cos β þ 2klb c cos α

for the triclinic case. The equation simplifies considerably with increasing lattice
symmetry (Giacovazzo et al., 2011); for example, for orthorhombic, tetragonal and
cubic systems to:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 h2 k2 l2
= + + . (2:8)
ds a2 b2 c2

2.1.1 Corrections to the peak position

The position of the Bragg reflection can be affected in a linear or nonlinear manner by
a series of contributions coming from the sample and from the instrument. For
constant wavelength data, the absolute error in the interplanar spacing Δd as a
function of the measured diffraction angle can be easily calculated by evaluating
the exact differential of the Bragg equation (eq. (1.15)) (Chapter 1):


λ cos θ
dd = dθ + 2d cos θdλ. (2:9)
2 sin2 θ

Neglecting the error in the wavelength,5 eq. (2.9) immediately shows the strong
nonlinear increase of Δd at low diffraction angles even for small constant errors in
2θ (Figure. 2.6).
The simplest angular correction is a constant shift (“zero error”), which in TOPAS
is defined as6:

prm c 0.0 min -0.1 max 0.1

th2_offset = c;

5 This would not be a valid assumption with, for example, wavelength (energy) dispersive data.
6 We’ll see later that the parameter (prm) c is allowed to refine in this example. A fixed parameter
would be preceded by ! giving !c. In the early code snippets of the book we will often omit the ! of fixed
parameter names for ease of reading.
 2.1 The peak position 23

0.1
Δ2θ = 0.01°
Δ2θ = 0.02°
0.08 Δ2θ = 0.10°

0.06
Δd/Å

0.04

0.02

0
0 20 40 60 80
2θ/°

Figure 2.6: Absolute error in the interplanar spacing Δd versus diffraction angle 2θ for different
constant errors in the measured diffraction angle using monochromatic Cu-Kα1 radiation.

or as predefined macro:

Zero_Error(@, 0)

For TOF data, any constant shift is included in the parameter t0 of the transforma-
tion function between the TOF scale and d-spacing. The relevant TOPAS code is:

prm !t0 -16.50

prm !t1 48262.60
prm !t2 -8.35
pk_xo = t0 + t1 D_spacing + t2 D_spacing^2;

or as predefined macro:

TOF_x_axis_calibration( !t0, -16.50, !t1, 48262.60, !t2, -8.35)

t1 is the most important factor and is related to the source to detector flight path. t0 is
usually a small contribution from electronic factors and t2 describes minor d2-dependent
aberrations. Typically, all three parameters are refined from a reference pattern and are
fixed in later refinement.
A constant shift in energy dispersive data is also included in the transfor-
mation of the measured energy scale into a d-spacing scale. The corresponding
TOPAS code is:
24 2 The Rietveld method

prm zero -0.16068

prm !detector_angle_in_radians = 7.77 Deg_on_2;
prm wavelength = 2 D_spacing Sin(detector_angle_in_radians);
prm energy_in_eV = 10^5 / (8.065541 wavelength);
pk_xo = 10^-3 energy_in_eV + zero;

A common nonlinear correction for angular dispersive Bragg–Brentano geometry is

the cos θ-dependent peak shift caused by a flat sample whose surface deviates from
the focusing circle (Figure 2.7). This is the so-called height error c in mm:


180 cos θ
Δ2θcorr = = − 2 c (2:10)
π RDS

0.15

0.1
Displacement error/°2θ

0.05 RDS = 173.5 mm

–200 μm displacement
–100 μm displacement
0
100 μm displacement
200 μm displacement
focussing circle Δ2θ –0.05

–0.1

–0.15
sample
0 20 40 60 80 100 120 140 160
2θ/°

Figure 2.7: Angular shift due to specimen displacement in Bragg–Brentano geometry. Schematic
drawing (left) and dependence on diffraction angle for different displacements (right).

with RDS the distance between sample and detector in mm. In TOPAS language7:

prm c 0.0 min -0.1 max 0.1

th2_offset = -2 Rad c Cos(Th) / Rs;

7 Note that the reserved parameter name Th will take all the different values of s for each reflection
s = (hkl) of a phase occurring in the diffraction data. In most equations, however, s will be simply
written as θ. Similar things are valid for the parameter D_spacing.
 2.1 The peak position 25

The corresponding predefined macro in TOPAS is:

Specimen_Displacement(@, 0.0)

Note that due to the nature of the cosine function, there is a high correlation between
the constant zero shift and specimen height displacement, particularly if the powder
pattern covers only a small angular range. It’s also worth noting that sample height
corrections are particularly problematic when indexing powder data. Equation. (2.10)
shows that even a 15 μm sample height error causes sufficient peaks shifts (e.g., 0.01°
2θ at 10° 2θ for Cu-Kα radiation with a diffractometer of radius 173 mm) that indexing
could be difficult.
A displacement of a capillary away from the center of the goniometer in Debye–
Scherrer geometry also causes a nonlinear shift in the angular position. A suitable
correction function is:



dL dV
Δ2θcorr = arcsin sinð2θÞ − arcsin cosð2θÞ (2:11)
RDS RDS

with the displacement of the capillary in the direction of the beam dL and perpendi-
cular to the beam dV (Gozzo et al., 2010). Figure 2.8 shows that these effects are much
less significant than in Bragg–Brentano geometry.

RDS = 173.5 mm
0.008 DL = DV = 0.1 mm
DL = DV = 0.5 mm
Displacement error/°2θ

0.004

–0.004

–0.008

0 20 40 60 80 100 120 140 160

2θ/°

Figure 2.8: Angular shift as a function of diffraction angle caused by the displacement of a capillary
with respect to the center of the goniometer.
26 2 The Rietveld method

In TOPAS this correction can be realized by:

prm dv 0
prm dl 0
th2_offset =(Rad ArcSin(dv Cos(2 Th))/Rs))+(Rad ArcSin(dl Sin(2 Th))/Rs));

Another type of peak shift for capillary samples in angular dispersive powder diffrac-
tion experiments is caused by θ-dependent absorption. An empirical expression was
given by Sabine (1988) (Figure 2.9):

Δ2θcorr = ¼ 2AθB ð90  θÞC

A ¼ 0:000033μeff R
 2
(2:12)
B ¼ 1:168  0:22μeff R þ 0:0168 μeff R
 2
C ¼ 1:155 þ 0:2054μeff R  0:0224 μeff R

0.1

0.08
Displacement error/°2θ

0.06
μeff R = 0.05
μeff R = 0.20
0.04

0.02
Figure 2.9: Peak shift of capillary samples in
angular dispersive powder diffraction
0
experiments caused by θ-dependent
0 20 40 60 80 100 120 140 160 absorption as a function of diffraction angle
2θ/° for different values of μeffR.

with the effective (taking packing effects into account) linear absorption of the
sample μeff in mm−1 and the radius R of the capillary in mm. This correction can be
applied in TOPAS as:

prm c 0.15 ‘ = μeff R

th2_offset = 0.000033 c (Th Rad)^(1.168 - 0.22 c + 0.0168 c^2)
(90 - Th Rad)^(1.155 + 0.2054 c - 0.0224 c^2);

or by using the predefined macro:

 2.2 The intensity of a Bragg reflection 27

Cylindrical_2Th_Correction(@, 0.15)

Other factors, which affect the position of Bragg reflections are axial divergence of the
incident beam and transparency/absorption. The appropriate corrections are nor-
mally built into the peak shape functions. Due to high correlation, it is important to
keep the number of correction factors in a refinement low.

2.2 The intensity of a Bragg reflection

The (integrated) intensity of a Bragg reflection is proportional to the squared complex
structure factor that itself is the vector sum of complex atomic form factors (for
X-rays) or coherent atomic scattering lengths (for neutrons) weighted by additional
complex phase factors. The following sections discuss factors that influence peak
intensities.

2.2.1 The atomic form factor

The atomic form factor describes the scattering power of an atom or ion as a function
of the scattering vector length s. In the case of X-rays, the form factor fj depends
strongly on s with a marked decrease at higher values. Note that most explicit
parameterizations of fj and others are formulated as a function of ~s = s=2 = sin θ=λ
and not of s.
The value at ~s = 0 is normalized to the number of electrons of the scatterer (atom
or ion). The form factor consists of a wavelength independent (normal scattering) and
a complex wavelength-dependent part (anomalous scattering):
pffiffiffiffiffiffiffi
fj ð~sÞ = fj0 ð~sÞ + Δfj′ ðλÞ + − 1Δfj′i ðλÞ. (2:13)

The real part of the anomalous scattering factor has a phase shift of 180° with respect
to the normal scattering factor, thus directly reducing the scattering power, while the
complex part has a phase shift of 90° (Figure 2.10).
Anomalous scattering effects are often disregarded for simplicity, but become
extremely important if the wavelength used is in the vicinity of an absorption edge of
an atomic species in the sample. For a strong scatterer, the change in scattering
power can amount to the equivalent of several electrons and so-called anomalous
dispersion measurements can be used to give extra element-specific information on
structures. By default, TOPAS uses dispersion coefficients Δfj′ ðλÞ and Δfj′i ðλÞ (from
http://henke.lbl.gov/optical_constants/) that cover the energy range from 10 eV to 30
keV (Figure 2.11).
28 2 The Rietveld method

Im

fʹi f0
f

fʹ
φ(s) Figure 2.10: Vector (pointer) representation of the
Re complex atomic scattering factor f with normal scattering
(f0) and real (f') and complex (f'i) parts of anomalous
scattering.

80 pb.nff_f1
75 pb.nff_f11
70
65
60
55
50
Electrons

45
40
35
30
25
20
15
10
5
0
2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 22000 24000 26000 28000 30000
Energy/keV

Figure 2.11: TOPAS screenshot showing anomalous


scattering factors Z − Δfj′ ðλÞ (nff_f1) and Δfj′i ðλÞ
(nff_f11) for Pb as a function of energy (λ=A = 12.3985=ðE=keVÞ). Data were taken from http://henke.
lbl.gov/optical_constants/.

The functional dependence of the form factors for all common atoms and ions has
been parameterized by an empirical linear combination of four Gaussian functions:

X
4 j
− b ~s2
fj0 ð~sÞ = cj0 + ajk e k (2:14)
k=1

with the nine parameters a1, a2, a3, a4, b1, b2, b3, b4 and c0 tabulated, for example, in
Tables for X-Ray Crystallography (1995) Vol C. The obtained form factors are valid in
the range 0 ≤ ~s ≤ 2, which is sufficient for most cases (Figure 2.12).
For more precise approximations covering the range 0 ≤ ~s ≤ 6 (Waasmaier &
Kirfel, 1995), the normal form factor is approximated by an empirical linear combi-
nation of five Gaussian functions, leading to 11 parameters. This is the default in
TOPAS (Figure 2.13):
 2.2 The intensity of a Bragg reflection 29

80 Cu-Kα1 Mo-Kα1 Pb
Pb4+
70 O
Atomic form factor/electrons

60

50

40

30

20
Figure 2.12: Normal atomic form factors
10 of Pb, Pb4+ and O calculated using the nine-
parameter approximation function as a
0
function of ð~sÞ. The theoretical maximum
0 0.4 0.8 1.2 1.6 2 ranges (0 ≤ 2θ ≤ 180°) for Cu-Kα1 and Mo-Kα1
sin θ/λ radiations are shaded.

f [electrons]

80

Atomic form factors

60

40

20

sin(θ)
1 2 3 4 5 6 λ

Figure 2.13: Atomic form factors of all atoms with atomic numbers ranging from 1 (H) to 92 (U)
calculated using the 11-parameter approximation function as a function of ð~sÞ.

X
5 j
− b ~s2
fj0 ð~sÞ = cj0 + ajk e k . (2:15)
k=1

TOPAS allows manual input and also refinement of fj0 ð~sÞ, Δfj′ ðλÞ and Δfj′i ðλÞ, but high
correlations exist between Δfj′ ðλÞ, Δfj′i ðλÞ, scale factor and displacement parameters
30 2 The Rietveld method

(see Section 2.2.3 for latter). The syntax uses the f0_f1_f11_atom keyword (f0 ≡ fj0 ð~sÞ,
f1 ≡ Δfj′ ðλÞ and f11 ≡ Δfj′i ðλÞ). Defaults are used when either f1 or f11 are not defined.
The following example defines f1 and f11 for Pb2+, S and O2− and the refinement flag
(@) is set for Pb2+:8

' Default values from NFF files

load f0_f1_f11_atom f1 f11
{
Pb+2 @ -3.7275704 @ 8.93529065
S 0.335135942 0.550512703
O-2 0.0523209357 0.0337069703
}

In the following example, the a1 parameter of f0 and f11 of Pb are refined:

' Values for Pb from atmscat.cpp

' Pb+2 27.392647 16.496822 19.984501 6.813923 5.233910 4.065623 1.058874 0.106305
6.708123 24.395554 1.058874

prm a1 25 min -50 max 50

load f0_f1_f11_atom f0 f11 {
Pb+2
= a1 Exp(1.058874 (-0.25) / D_spacing^2) +
16.496822 Exp(0.106305 (-0.25) / D_spacing^2) +
19.984501 Exp(6.708123 (-0.25) / D_spacing^2) +
6.813923 Exp(24.395554 (-0.25) / D_spacing^2) +
5.233910 Exp(1.058874 (-0.25) / D_spacing^2) +
4.065623; ' this is f0 for Pb
@ 5 ' this is f11 for Pb
}

2.2.2 The coherent atomic scattering length

For neutron diffraction data, Δfj′ ðλÞ = Δfj′i ðλÞ = 0, and fj0 ð~sÞ is replaced by the
bound coherent scattering length, b, which is independent of the scattering
angle (Figure 2.14). Neutron scattering length data in TOPAS are from Sears
(1992) and stored in the file NEUTSCAT.CPP.

8 The “load { }” keyword in TOPAS is used to simplify user input. It allows the loading of keywords of
the same type by typing the keywords once.
 2.2 The intensity of a Bragg reflection 31

1.4
Neutron scattering length 1012bcoh/cm

1.2
1
0.8
0.6
0.4 Pb
0.2 O
H
0 D
–0.2
–0.4

0 0.4 0.8 1.2 1.6 2 Figure 2.14: Neutron coherent scattering

sin θ/λ lengths for Pb, O, H and D.

2.2.3 Displacement parameter

At any temperature, atoms vibrate about their equilibrium position. Moreover, static
local atomic displacements may exist in disordered structures like solid solutions.
The corresponding displacements lead to a decrease in peak intensities, which can be
described by multiplying the atomic form factor with a correction factor. One can
distinguish between isotropic and anisotropic displacements. For the isotropic case
the displacement factor (Debye–Waller factor) for the entire crystal structure, groups
of atoms or an individual atom is defined as:
2
t = e − B~s , (2:16)

1
Cu-Kα1 Mo-Kα1 B = 0.5 Å2
B = 1.0 Å2
0.8 B = 2.0 Å2
B = 5.0 Å2
Rel. displacement factor

0.6

0.4

0.2

0 Figure 2.15: Intensity reduction t as a function

0 0.4 0.8 1.2 1.6 2 of ~s for a series of normalized displacement
sin θ/λ parameters B.
32 2 The Rietveld method

where B is the isotropic displacement parameter (usually, and also in TOPAS,

in Å2). t is shown in Figure 2.15 and used in the structure factor calculation of
eq. (2.21).
The physical meaning of B is given by:

B = 8π2 u2 , (2:17)

where u2 is the mean square deviation from the equilibrium position of the atom or
atomic group. A range of 0.1 Å2 ≤ B ≤ 1.5 Å2 is considered normal for inorganic
compounds, while for coordination compounds B ≤ 3 Å2 is usually acceptable.
Larger values usually indicate errors or severe disorder in the crystal structure.
Negative values often indicate systematic errors in the intensities due to, for example,
absorption or surface roughness or misassignment of an atom type. TOPAS uses B as
the isotropic displacement parameter.
In order to determine individual displacement parameters with good precision
from powder data, a large range of ~s must be covered. In addition, the ~s-dependence
of the intensity reduction (Figure 2.15) is similar to that of many other correction
factors that are often poorly treated. For this reason, the displacement factor is
frequently (harshly!) referred to as a “physical trash collector.”
With high-quality powder diffraction data measured over an extended ~s range, it
may be possible to refine anisotropic displacement parameters for the strong scat-
terers in the crystal structure.9 The anisotropic displacement parameter can be
geometrically described as a three-axis ellipsoid and can be described in terms of a
second-rank tensor symmetric about its principal diagonal:
0 1
u11 u12 u13
u ¼ @ u12 u22 u23 A; (2:18)
u13 u23 u33

which leads to a direction s = (hkl) dependent displacement factor of:

t = e − 2π ðu11 h a + u22 k b + u33 l c + 2u12 hka b + 2u13 hla c + 2u23 klb c Þ .

2 2 *2 2 *2 2 *2 * * * * * *
(2:19)

In order to prevent physically meaningless results, the uij matrix must be kept
positive definite, which can be achieved with the following boundary conditions:

uii > 0
uii  ujj > uij 2 (2:20)
u11  u22  u33 þ u212  u213  u223 > u11  u223 þ u222  u213 þ u233  u212 .

9 Especially for neutron data where the scattering factor doesn’t fall off with ~s.
 2.2 The intensity of a Bragg reflection 33

In TOPAS, the site-dependent macro “ADPs_Keep_PD” does this job.10 TOPAS uses
uij as anisotropic displacement parameters, which are activated by the “adps”
keyword:

site Cl1 x @ 0.03947` y @ 0.27566` z @ 0.25331` occ Cl 1 adps

that leads to (in the order u11 , u22 , u33 , u12 , u13 , u23 Þ:

site Cl1 x @ 0.03947` y @ 0.27566` z @ 0.25331` occ Cl 1

ADPs { @ 0.04367` @ 0.06117` @ 0.08074` @ 0.00592` @ 0.01280` @ 0.01262` }

TOPAS automatically places appropriate symmetry restrictions on individual uij

components.
Higher levels of complexity using an anharmonic approximation of the atomic
displacement parameters are not of relevance for powder diffraction, except in very
special cases (e.g., Wahlberg et al., 2016).

2.2.4 The structure factor

Ignoring anomalous scattering, the structure factor of a Bragg reflection is defined as

a complex sum over all atoms j in the unit cell (Figure 2.16):
X 
F ðsÞ = tj fj ðsÞe2πisxj (2:21)
j

with the positional vector xj of an atom j in the unit cell defined by the fractional
crystal coordinates:
0 1
x
xj ¼ @ y A. (2:22)
z

Equation (2.21) contains the displacement factor tj for every atom j. This factor is
omitted in the following for brevity.
Using the Euler identity, the real and complex parts of the structure factor can be
separated (Figure 2.16):
X
 

pffiffiffiffiffiffiffi X


 

F ðsÞ = fj ðsÞcos 2πs  xj + − 1 fj ðsÞsin 2πs  xj = AðsÞ + iBðsÞ. (2:23)
j j

10 This behavior can be turned off using the keyword adp_no_limits.

34 2 The Rietveld method

Im Im
f3
F(s)
f2
f2

φ(s)3 f1 f1
f3 φ(s)
φ(s)1 φ(s)2 Re Re

Figure 2.16: Graphical representation of the complex structure factor as vector sum (right) of the
individual form factors (left).

If anomalous scattering is taken into account, the structure factor amplitude

becomes:
X 0 0
  
FðsÞ ¼ fj0 ðsÞ þ Δfj ðλÞ þ i Δfj i ðλÞ  cos 2πs  xj
j
X 0 0
   (2:24)
þi fj0 ðsÞ þ Δfj ðλÞ þ i Δfj i ðλÞ  sin 2πs  xj :
j

After separating the real and the imaginary parts, this turns into:

8 P 0
   9
>
>
< j fj0 ðsÞ þ Δfj ðλÞ  cos 2πs  xj > >
=
FðsÞ ¼    
>  P Δf 0 i ðλÞ  sin2πs  x 
> >
>
: j j ;
j
8 P 0
   9
>
>
< j fj0 ðsÞ þ Δfj ðλÞ sin 2πs  xj > >
=
(2:25)
þi    
> þ P Δf 0 i ðλÞ  cos2πs  x  >
> >
: j j ;
j

¼ fA01 ðsÞ  B11 ðsÞg þ ifB01 ðsÞ þ A11 ðsÞg ¼ AðsÞ þ i BðsÞ.

The intensity of a Bragg reflection s is proportional to the structure factor multiplied

by its conjugate complex, which is equivalent to the squared absolute value of the
structure factor amplitude jF ðsÞj:

I ðsÞ ∝ F ðsÞ F * ðsÞ = jF ðsÞj2 . (2:26)

For practical purposes it is easier to separate the real and imaginary parts of the
structure factor (Figure 2.17) leading to:
 2.2 The intensity of a Bragg reflection 35

jF ðsÞj2 ¼ AðsÞ2 þ BðsÞ2

¼ ½jAðsÞj þ ijBðsÞj½jAðsÞj  ijBðsÞj
¼ A01 ðsÞ2 þ B01 ðsÞ2 þ A11 ðsÞ2 þ B11 ðsÞ2 þ 2B01 ðsÞA11 ðsÞ  2A01 ðsÞB11 ðsÞ:
(2:27)

The phase angle can be directly deduced from Figure 2.17 as:


jBðsÞj
’ðsÞ = arctan . (2:28)
jAðsÞj

Im

F(s)

B(s)
φ(s)
A(s) Re

–B(s)
Figure 2.17: Vector (pointer) representation of the structure
F*(s)
factor and its conjugate complex.

Due to the squaring of the structure factor amplitude, the relative contribution to the
diffraction pattern of light elements in the presence of heavy elements is low if X-rays
are used. As an example, the squared scattering power (squared form factors) of Pb
and O differ by a factor of more than 100.
TOPAS gives direct access to all the quantities in the structure factor equa-
tion (eq. (2.27)). Even though it is not needed for regular work, this can some-
times be useful. As a simple example, the notation of atomic sites for structure
factor calculation in a structure (str) phase of TOPAS is shown for LaB6. Please
note that the special positions should be given as equations.11 The keywords
num_posns, beq and occ define the number of unique equivalent positions
generated from the space group, the isotropic displacement parameter B and
the site occupancy factor, respectively:

site La num_posns 1 x =0; : 0 y =0; : 0 z =0; : 0 occ La 1 beq @ 0.38934

site B num_posns 6 x @ 0.19986 y =1/2; : 0.5 z =1/2; : 0.5 occ B 1 beq @ 0.23502

11 Particularly for recurring numbers like 1/3 (use =1/3; not 0.33333).
36 2 The Rietveld method

The reflection intensities and structure factors can be written to the file “reflec-
tions.txt” with12:

phase_out “reflections.txt” append

load out_record out_fmt out_eqn
{
"%4.0f" = H; "%4.0f" = K; "%4.0f" = L; " %3.0f" = M;
"%12.2f" = I_no_scale_pks; ' I without any scale_pks corrections applied
"%12.2f" = Iobs_no_scale_pks; ' Iobs from Rietveld decomposition formula
"%10.2f o" = Iobs_no_scale_pks_err;
" %11.5f" = A01; " %11.5f" = A11; " %11.5f" = B01; " %11.5f\n" = B11;
}

2.3 Intensity correction factors

To calculate the integrated reflection intensities observed experimentally in a powder

diffraction pattern, a series of correction factors have to be applied to the squared
structure factors, which depend on the scattering vector s or its length s. A list of the
most common correction factors is given by the product:

CorrðsÞ = MðsÞLPðsÞAðsÞPOðsÞEðsÞ . . . , (2:29)

which includes the multiplicity MðsÞ of a reflection given by the lattice symmetry, an
absorption correction AðsÞ, the (solely geometrical) Lorentz–polarization factor
LPðsÞ, a preferred orientation correction POðsÞ and a correction for primary extinction
EðsÞ, which is only relevant for highly crystalline materials. The absolute values of
the individual correction factors are not important, as any constant gets absorbed in
the scale factor.13

2.3.1 Multiplicity

Due to the overlap of Friedel pairs, the observed intensity is always doubled corre-
sponding to a minimum value of two for the reflection multiplicity for all crystal
systems. In addition, for symmetries higher than triclinic, symmetry-equivalent
reflections of identical d-spacing have identical intensity and overlap completely.
The total number of these reflections is called multiplicity and lies between 2 and

12 Expressions like “ %11.5f\n” are formatting commands, for example writing real numbers ( f ) 11
characters wide with five decimal places followed by a new line (\n).
13 Note that all corrections are applied to the calculated intensities in order to match the observed
intensities.
 2.3 Intensity correction factors 37

48. TOPAS takes care of the multiplicity automatically (Figure 2.18). The reserved
keyword “M” is used to store all reflection multiplicities of a Pawley/LeBail (hkl_Is)
(see Chapter 3) or Rietveld (str) phase. An example of how to save the reflection
parameters including the multiplicity of a hkl_Is phase to the file “reflections.txt” is:

phase_out reflections.txt load out_record out_fmt out_eqn

{
"%3.0f" = H; " %3.0f" = K; " %3.0f" = L; " %3.0f" = M;
" %11.5f" = D_spacing;
" %11.5f" = 2 Rad Th;
" %11.5f\n" = I_no_scale_pks;
}

60
Cu-Kα1 Mo-Kα1
54
48
42
Multiplicity of LaB6

36
30
24
18
12
6
0
Figure 2.18: Multiplicities (6, 8, 12, 24, 48)
0 0.4 0.8 1.2 1.6 2  as a
of the Bragg reflections of LaB6 (Pm3mÞ
sin θ/λ function of ~
s.

2.3.2 Lorentz–polarization factor

The Lorentz and the polarization factors are purely geometric factors. The Lorentz
factor has several contributions. One takes into account the relative time that a
reciprocal lattice point moving with angular velocity ω spends passing through the
finite thickness of the Ewald sphere. According to Figure 2.19, the component of the
linear velocity v of a reciprocal lattice point along the radius of the Ewald sphere is:

2 sin θ 1
v = ωd* cos θ = ω cos θ ∝ sin θ cos θ, L ∝ . (2:30)
λ v
38 2 The Rietveld method

Detector slit

ωd*
θ
s = d* Sample
k
θ 2θ
Ewald sphere θ
k0 De
ω ω by
e–
Sc
he
rre
rc
on
es

Figure 2.19: Schematic drawing of the geometrical origin of the Lorentz factor. Left: Reciprocal lattice
points rotating with different speed through the Ewald sphere. Right: Effect of the opening angle of
the Debye–Scherrer cones on the intensity distribution.

In the case of powder diffraction, an additional geometrical factor occurs, which

normalizes the different radii of the Debye–Scherrer rings (Figure 2.19, right). The
fraction of the diffraction cone that intersects the detector is highest at low angles and
at very high angles (backscattering). The factor is proportional to 1/sinθ. The typical
form of the Lorentz factor for powders is then (Figure 2.20, left):
1 1
L= ∝ (2:31)
cos θsin2 θ sin θ sinð2θÞ

50 10000

40 8000
TOF Lorentz factor
Lorentz factor

30 6000

20 4000

10 2000

0 0
0 20 40 60 80 100 120 140 160 10 8 6 4 2 0
2θ/° d/Å

Figure 2.20: Lorentz factor for angle dispersive data as a function of the scattering angle (left) and for
TOF data as a function of d-spacing (right).
 2.3 Intensity correction factors 39

where any constant factor gets absorbed by the overall scale factor. In the case of
constant wavelength neutron data or fully polarized synchrotron radiation, the
macro Lorentz_Factor can be used to describe these effects.
For neutron TOF data, the Lorentz factor calculates to (Figure 2.20, right):

LTOF = d4 . (2:32)

A0 ⊥
A⊥

A0
A

Figure 2.21: Schematic drawing of the effect of polarization for an unpolarized primary beam.

The polarization factor originates from partial polarization of the scattered electro-
magnetic wave. For the case depicted in Figure 2.21 with:

A0? ¼ A0jj ; I0 ¼ A20? þ A20jj ¼ 2A20? ;

(2:33)
Ajj ¼ A0jj ; A? ¼ A0? cos 2θ; I ¼ A2? þ A2jj ¼ A20? cos2 2θ þ A20? ;

the intensity ratio between the diffracted and the primary beam follows:

I 1 + cos2 2θ
P= = . (2:34)
I0 2
This equation is valid for unpolarized radiation from a laboratory X-ray tube. When a
primary or secondary beam monochromator is present, a more general equation is
used:

1 − cos2 2θ  cos2 2θm

P= (2:35)
2
where 2θm is the Bragg angle of the reflection from the monochromator (Table 2.2).
For unpolarized radiation 2θm can be set to 0° (e.g., X-ray diffractometers without any
monochromator), for fully polarized radiation 2θm can be set to 90° (e.g., synchrotron
radiation or constant wavelength neutron diffraction) (Figure 2.22). In reality, syn-
chrotron radiation is 95–97% polarized. In order to account for fractional polarization
of the beam, a factor K can be introduced with K = 0.5 for circularly polarized X-rays
(i.e., laboratory X-ray tubes), K = 0 for fully polarized X-rays (ideal synchrotron
source) and K ~ 0.05 for a “real” synchrotron source:
40 2 The Rietveld method

Table 2.2: Bragg angle of different combinations of monochromators

and radiation for angle-dispersive laboratory powder diffractometers.

Monochromator Radiation Bragg angle 2θm /°

None Any 
Ge () Cu-Kα .
Ge () Mo-Kα .
Ge () Ag-Kα .
Ge () Mo-Kα .
SiO () Cu-Kα .
Graphite Cu-Kα .

0.5

0.45
Polarization factor

0.4

0.35
Unpolarized,
no monochrom
X-rays,
0.3
Ge(111) monochrom
Neutrons, synchrotron
0.25
0 20 40 60 80 100 120 140 160 180
2θ/°

Figure 2.22: Polarization factor for unpolarized, partly polarized and polarized radiation as a function
of diffraction angle.

ð1 − K + K  cos2 2θ  cos2 2θm Þ

P= . (2:36)
2
For practical reasons, the Lorentz and the polarization factors are usually combined
in a single Lorentz–polarization factor (LP factor) (Figure 2.23, left). The effect of the
LP factor on the intensities of an X-ray powder pattern is enormous (Figure 2.23,
right).
The corresponding correction macros in TOPAS are called LP_Factor(2θm ) for
laboratory X-ray data and LP_Factor_Synchrotron(1 – K, 2θm ) for a “real” synchrotron.
Although possible, one should not attempt to refine the K factor and/or the Bragg
angle of the monochromator. To apply a general LP factor for angular dispersive data
as defined above in TOPAS, the following code can be used:
 2.3 Intensity correction factors 41

100
80 Without LP factor

Rel. intensity
60
50
Unpolarized 40
Monochr. (27.3° 2θ)
Lorentz polarization factor

40 20
0
30 100
With LP factor
80

Rel. intensity
20 60
40
10
20
0 0
0 20 40 60 80 100 120140160 20 40 60 80 100 120 140
2θ/°
2θ/°

Figure 2.23: Effect of the Lorentz–polarization (LP) factor for angular dispersive data as a function of
Bragg angle. Left: LP factor for unpolarized beam (solid line) and partly polarized beam by a Ge (111)
primary beam monochromator with Cu-Kα1 radiation. Right: Simulated powder pattern of LaB6 for
Ag-Kα1 radiation with (bottom) and without (top) LP factor.

prm k 0.5 min 0.0 max 1.0

prm mono 0 min 0.0 max 90
scale_pks = (1/( Sin(Th)^2 Cos(Th))) (1 - k + k Cos(mono) Deg)^2 Cos(2 Th)^2) / 2;

For TOF neutron data, the following code is appropriate:

scale_pks = D_spacing^4;

Note that the keyword scale_pks is used for applying intensity corrections to phase
peaks, while scale_phase_X scales calculated patterns point by point. Multiple defi-
nitions are allowed and each is applied to the peaks and pattern, respectively.

2.3.3 Absorption correction

For accurate powder diffraction work it is important to consider the effects of absorp-
tion on experimental intensities. For simple transmission through a solid material, the
transmitted intensity I with respect to the initial intensity I0 depends on the thickness x
of the material and its linear absorption coefficient μ (Figure 2.24):

I = I0 e − μx . (2:37)
42 2 The Rietveld method

1
μ = 5.0 [mm–1]
μ = 2.0 [mm–1]
0.8 μ = 1.0 [mm–1]
Absorption correction factor

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
Thickness/mm

Figure 2.24: Intensity correction factor required for solid samples in transmission geometry with
different absorption coefficients.

The appropriate absorption correction factor (transmission factor) for calculated

intensities therefore is A = I=I0 . Absorption depends strongly on the energy (wave-
length) of the radiation used and also changes rapidly close to the absorption edges
discussed in Section 2.2.1. To minimize absorption effects it is therefore important to
match the experimental wavelength to the system being studied.
One important consideration in Bragg–Brentano geometry is the requirement that
the sample is “infinitely thick”, meaning that a negligible fraction of the beam passes
straight through the sample. One way to estimate the minimum sample thickness, tmin,
required to meet this criterion is to calculate the sample depth required for the incident
beam to be reduced to 1/1000th of its initial intensity. We set14 tmin = 2x sin θmax and
I ðsurfaceÞ=I ðtminÞ = 1000 in eq. (2.37), which simplifies to give:

3.45 sin θmax

tmin ffi . (2:38)
μ

For Ni powder tmin ≈ 0.013 cm and for a typical organic sample tmin ≈ 1.3 cm at θ = 90
using Cu-Kα radiation. Note that it is practically unlikely that an organic material will
meet the infinite thickness criterion.
In a Rietveld analysis, the effects of absorption on peak intensities must be taken
into account. The corrections differ for different experimental geometries. Many of
these have been reviewed and a collection of TOPAS macros given (Rowles &

14 The factor of 2 arises as both the incoming and the outgoing beams must be taken into account.
 2.3 Intensity correction factors 43

Buckley, 2017). To calculate the reduction of the diffracted intensity, one must take
the total path l of the incident and the diffracted beams in the sample into account
and integration must be performed over the entire volume V of the sample that
contributes to scattering. Instead of the linear absorption coefficient μ, an effective
linear absorption coefficient μeff should be used to account for the lower packing
density of a loose powder.

α β
t' ts
dt' 2θ

Figure 2.25: Schematic drawing of asymmetric flat plate reflection geometry.

For flat plate asymmetric reflection (Figure 2.25), where a (largely) parallel beam15
is incident on the specimen surface at a fixed angle, the irradiated volume
contributing to the diffracted intensity can be calculated (Egami & Billinge,
2003) by:
ðts  
Area 1 +
− μeff t′ sinα 1
sinð2θ − αÞ
V= e dt′
sinα
0

 − 1(  )
Area sinα 1 +
− μeff ts sinα 1
sinð2θ − αÞ
= 1+ 1−e , (2:39)
μ sinð2θ − αÞ

where Area is the area of the incident X-ray beam on the specimen surface, ts is the
specimen thickness, α is the angle between the incident beam and the specimen
surface and 2θ − α ð = βÞ is the angle between the diffracted beam and the specimen
surface. In the case of an “infinitely thick” sample (exponential term in eq. (2.39) is
less than 0.001), the equation simplifies to:

−1
Area sinα
VAsym = 1+ . (2:40)
μeff sinð2θ − αÞ

For symmetric reflection α = β = θ (Bragg–Brentano geometry), this volume further

simplifies into a constant:

15 The tube to specimen distance in most experiments is sufficiently large that we can consider the
beam to be parallel in these calculations.
44 2 The Rietveld method

Area
VBragg = . (2:41)
2μeff

Consequently, for Bragg–Brentano geometry with an infinitely thick sample, the

diffracting volume is constant, and no correction to the intensity is necessary.
For asymmetric reflection geometries a correction will be needed to account for
the changing volume of diffracting material:

− 1
VAsym sinα
A= =2 1+ (2:42)
VBragg sinð2θ − αÞ

that, for samples fulfilling the criterion of infinite thickness, is independent of the
absorption coefficient.
In the case when the specimen does not fulfill the criterion of “infinite thick-
ness,” the correction factor becomes:

− 1 (  )
1 +
− μeff ts sinα 1
sinα sinð2θ − αÞ
A=2 1+ 1−e . (2:43)
sinð2θ − αÞ

Figure 2.26 plots this correction factor for a specimen with different angles between
the incident beam and specimen surface as a function of diffraction angle. For the
case of a 10° incident angle, the effect of different values of the specimen thickness is
shown.

A
2.0

1.5

1.0 α = 2°, ts = 100 μm

α = 10°, ts = 100 μm
α = 10°, ts = 50 μm
0.5
α = 10°, ts = 25 μm
α = 20°, ts = 25 μm
2θ/°
0 50 100 150

Figure 2.26: Intensity enhancement for asymmetric reflection geometry of specimen with different
angle between incident beam and specimen surface. For a 10° angle, the effect of different values for
the specimen thickness is shown using a linear absorption coefficient of 100 cm−1.

Equation (2.43) simplifies significantly for symmetric Bragg reflection of noninfi-

nitely thick samples to (Figure 2.27):
 2.3 Intensity correction factors 45

1.0

0.8

0.6

0.4 ts = 100 μm
ts = 50 μm
0.2 ts = 25 μm

2θ/°
0 50 100 150

Figure 2.27: Intensity enhancement for symmetric reflection geometry (Bragg–Brentano) showing
the effect of different values for the specimen thickness using a linear absorption coefficient of
100 cm−1.

ts 2
A = 1 − e − μeff sinθ (2:44)

In TOPAS scripting language this can be applied by:

prm uT 1.0 min 0.0001 max 12

scale_pks = 1-exp(-uT 2 / Sin(Th));

For cylindrical samples (Debye–Scherrer geometry), the beam must pass through
the entire capillary at low angles. A reasonable approximation for an absorption
correction factor has been given by Sabine et al., 1988:

A = AL cos2 θ + AB sin2 θ (2:45)

where AL and AB are the absorption factors at the Laue condition (θ = 0 ) and the
Bragg condition (θ = 90 ), respectively.
The absorption factors depend on μeff R with R the cylinder (capillary) radius,
and are calculated using modified Bessel and Struve functions. A typical absorption
correction factor for μeff R = 1.0 or 2.5 is shown in Figure 2.28. The formula gives
satisfactory results for μeff R < 10.
In TOPAS scripting language such a cylindrical absorption correction can be
realized by:

prm uR 1.0 min 0.0001 max 12

scale_pks = AL_Cyl_Corr(@, uR)) Cos(Th)^2 + AB_Cyl_Corr(@, uR) Sin(Th)^2;
46 2 The Rietveld method

0.3
Normalized cylindrical absorption factor

0.2 μR = 1.0
μR = 2.5

0.1

0
0 20 40 60 80 100 120 140 160
2θ/°

Figure 2.28: Absorption correction factor for cylindrical samples as a function of scattering angle for
μeff R = 1.0 and 2.5.

Alternatively a predefined macro with μeff R as argument can be used:

Cylindrical_I_Correction(@, 1.0)

A more rigorous treatment using radial symmetry for the calculation of cylindrical
absorption coefficients taking the capillary loading into account was published by
Kalifah (2015).

2.3.4 Surface roughness

If the packing density in Bragg–Brentano geometry varies with depth, thus creating a
“rough surface,” the so-called porosity effect reduces the intensity at low Bragg
angles. This is also a kind of absorption effect. The two most common corrections
are those by Pitschke et al. (1993) (Figure 2.29, left):

1 − a1 ð1=sinθ − a2 =sin2 θÞ
A= (2:46)
1 − a1 ð1 − a2 Þ

and by Suortti (1972) (Figure 2.29, right):

a1 + ð1 − a1 Þe − a2 =sinθ
A= (2:47)
a1 + ð1 − a1 Þe − a2
 2.3 Intensity correction factors 47

1 1

Suortti surface roughness

Pitschke correction factor

a1 = a2 = 0.01
a1 = a2 = 0.05
0.8 0.8
a1 = a2 = 0.01
a1 = a2 = 0.05

0.6 0.6

0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160

2θ/° 2θ/°

Figure 2.29: Correction factor for porosity effect in Bragg–Brentano geometry according to Pitschke
et al. (1993) (left) and Suortti (1972) (right) as a function of diffraction angle.

where a1 and a2 are refinable parameters. The latter function works better on low-
angle reflections. In TOPAS scripting language the surface roughness correction
according to Pitschke can be realized by:

prm a1c min .0001 max 1

prm a2 min .0001 max 1
scale_pks = (1 - a1 (1 / Sin(Th) – a2) / Sin(Th)^2)) / (1 – a1 + a1 a2);

and according to Suortti by:

prm a1c min .0001 max 2

prm a2 min .0001 max 2
scale_pks = (a1 + (1 - a1) Exp( -a2 /Sin(Th))) / (a1 + (1 - a1) Exp( -a2));

The two corresponding macros in TOPAS are:

Surface_Roughness_Pitschke_et_al(@, a1v, @, a2v)

Surface_Roughness_Suortti(@, a1v, @, a2v)

2.3.5 Extinction

In rare cases, for nearly perfectly crystalline materials, primary extinction effects
(within the same crystallite) can occur for powders. The observed intensity for strong
48 2 The Rietveld method

reflections at low Bragg angle will decrease. The interested reader is referred to
Sabine (1985) and Sabine et al. (1988).

2.3.6 Overspill effect

In many diffraction geometries it is important that the incident beam remains smaller
than the sample area at all angles in order to ensure the constant illumination volume
condition (in the case of an infinitely thick specimen). This is particularly important
in Bragg–Brentano geometry. Nevertheless, at low angles it is common for the
irradiated area to become greater than the area covered by the sample on the sample
holder. This “overspilling” reduces the intensities up to the diffraction angle at which
the two areas are identical (Figure 2.30).

Divergence slit
Focus d

S
L
φ
R
θ –φ/2
θ

l1 l2
L

Figure 2.30: Irradiated length on the surface of a flat plate sample in Bragg–Brentano geometry for a
divergent beam (left) and a perfectly parallel beam (right).

For divergent beam Bragg–Brentano geometries with a tube opening angle φ, which
is determined by the divergence slit, the irradiated length calculates to:

R sin ’2 R sin ’2 R ’½rad

L = l1 + l2 =  ’
 +  
’ ffi (2:48)
sin θ + 2 sin θ − 2 sin θ

with the goniometer radius R (Figure 2.30, left) (Fischer, 1996, Krüger & Fischer,
2004, Pecharsky & Zavalij, 2003). In the case of small divergence the beam can be
 2.3 Intensity correction factors 49

regarded as quasi-parallel and the term R ’½rad corresponds to the thickness of the
beam d (Figure 2.30, right).
An intensity correction factor as a function of the diffraction angle can thus be
calculated for a sample length S (Figure 2.31):

S  ’
Ov = for 0 ≤ 2θ½rad ≤ 2arcsin R . (2:49)
LD S
In TOPAS this simple overspill correction for (quasi) parallel beam geometry can be
realized by:16

prm !len 3 min 0.0001 ‘length of sample

prm !wid 1 min 0.0001 ‘beam width
scale_pks = if ( Sin(Th) < wid/len, len/(wid/sin(Th), 1);

25 1
R = 285 mm
φ = 1°
φ = 1/4°
Overspill correction factor

0.8
Total irradiated length/mm

20

15 0.6

10 0.4
R = 285 mm, S = 10 mm
φ = 1/2°
φ = 1/4°
5 0.2

0 0
0 20 40 60 80 0 20 40 60 80
2θ/° 2θ/°

Figure 2.31: Left: Irradiated length on the surface of a flat plate sample in Bragg–Brentano geometry
with a divergent beam for different opening angles φ. Right: Corresponding intensity correction
function for the overspill effect for a sample length of 10 mm.

Alternatively, automatic variable divergence slits can be used during the measure-
ment, which have a small opening at low 2θ and then widen as a function of 2θ. In
TOPAS the correction of the peak intensity can be accomplished by scaling peaks by a
θ-dependent sine function:

16 The Divergence_Sample_Length macro in TOPAS accommodates both overspill and the flat sur-
face aberration. In other words the intensity, peak shift and peak shape is accounted for.
50 2 The Rietveld method

scale_pks = Sin(Th);

The corresponding TOPAS macro is:

Variable_Divergence_Intensity

We recommend not using variable slits in Rietveld work as any imprecision in slit
opening will influence intensities, and there is a progressive deterioration of the para-
focussing condition causing the resolution to continuously decrease with increasing
2θ. Moreover, the peak asymmetry changes (for a correction function see Section 2.5.8).

2.3.7 Preferred orientation

The idea of powder diffraction is based on the perfect randomness of the orientations
of the crystallites as pointed out as early as in 1917 by Hull. Experimentally, this is
only easily realized in the case of spherical crystallites. If needle or plate-like crystal-
lites are prepared in flat plate sample holders for reflection geometry or between foils
in transmission geometry, the crystallites tend to align themselves in one or more
preferred orientation(s). If the corresponding lattice planes are in reflection condi-
tion, their intensities are strongly increased (Figure 2.32). A detailed introduction into
the topic is given by Pecharsky and Zavalij (2009).

Figure 2.32: Schematic drawing of the effect of preferred orientation (in two dimensions).
Left: Randomly oriented crystallites with a proportional number of crystallites in reflection condition.
Right: preferentially oriented crystallites with a disproportional high number of crystallites in
reflection condition.

In one simple approach where there is a single preferred orientation direction, the
angle between the reciprocal lattice vector s of each Bragg reflection and the
specific reciprocal lattice vector spref of the preferred orientation is calculated
using the scalar product:
 2.3 Intensity correction factors 51

spref  s
cosωs =   (2:50)
spref jsj

A correction factor can be calculated by the March–Dollase function (March, 1932)

according to which:

1X N   −3=2
Ts = τ2 cos2 ωis + τ − 1 sin2 ωis , (2:51)
N i=1

where the sum runs over all N symmetry equivalent reciprocal lattice points and τ is
the refined preferred orientation parameter, which is defined as the ratio between the
correction factors for Bragg peaks perpendicular and parallel to the direction of the
preferred orientation.
In TOPAS the code for a single direction of preferred orientation (here 100) is:

str_hkl_angle ang1 1 0 0
prm tau min 0.0001 max 2
scale_pks = Multiplicities_Sum(((tau^2 Cos(ang1)^2 + Sin(ang1)^2 / tau)^(-1.5)));

Alternatively, the following macros for one or two preferred orientation directions
can be used:

PO(@, c, ang, hkl)

PO_Two_Directions(c1, v1, ang1, hkl1, c2, v2, ang2, hkl2, w1c, w1v)

In a more general approach, a symmetry-adapted spherical harmonic expansion17 for

fixed sample orientation can be used:
!
X
L
4π X l
Ts = 1 + C k ðsÞ ,
m m
(2:52)
l=2
2l + 1 m =− l l l

where Clm and klm ðsÞ are the harmonic coefficients and factors (Järvinen, 1993),
respectively, and L is the order of the spherical harmonic. Due to the inversion
symmetry present in any powder pattern, only even orders need to be taken into
account. L in TOPAS can be 2, 4, 6 or 8. A parameter, J, quantifying the magnitude of
the preferred orientation can be calculated with:

17 See Chapter 13 (Mathematical basics).

52 2 The Rietveld method

!
X
L
1 X l  
J=1+  C m 2 (2:53)
l=2
2l + 1 m = − l l

with J ≥ 1 (unity in case of random orientation).

In TOPAS, spherical harmonics functions to scale intensities can be applied using:

spherical_harmonics_hkl sh
sh_order 8
scale_pks = sh;

or alternatively by the predefined macro:

PO_Spherical_Harmonics(sh, order)
normals_plot = sh;normals_plot_min_d .25

where sh is the parameter and order the order of the spherical harmonics function. A
three-dimensional plot of the spherical harmonics function can be performed by
using the keyword normals_plot (Figure 2.33).

Figure 2.33: Three-dimensional second-order

a b
spherical harmonic representation of the preferred
orientation correction of a 2H graphite sample
measured in flat plate mode, showing strong
preferred orientation along the c-direction.

Preferred orientation should be distinguished from graininess, where a small number of

large oriented crystallites will lead to incorrectly measured intensities. There is no
simple correction for graininess, except for collecting data on a better-prepared sample.

2.4 The scale factor

The scale factor is a linear, phase-specific factor that absorbs the constants of all
intensity correction factors. It is specific for a particular instrumental configuration
 2.5 The peak profile 53

and depends on the incident intensity and measuring time. In the case of multi-
phase Rietveld refinement using Bragg–Brentano geometry, the scale factor can be
used for full standardless quantitative phase analysis, based on the following
equation:

Sp ðZMV Þp μ*m
Xp = (2:54)
K
where Xp is the relative weight fraction of the phase p in a mixture of several crystal-
line phases, Z is the number of formula units of phase p in the unit cell, M is the
molecular mass of the formula unit of phase p, V is the volume of the unit cell in Å3 of
phase p and Sp is the scale factor of phase p. K is a scaling factor, which depends on
the instrumental conditions and is independent of sample and phase related para-
meters. The mass absorption coefficient of the entire sample is μ*m . Quantitative
Rietveld refinement can be performed without knowledge of K and μ*m , since the
instrumental conditions and the absorption coefficient enter the equation as con-
stants and are identical for all phases. Therefore, in the case of a multiphase mixture,
the scale factor is directly related to the weight fraction Xα of the phase α and can be
used for quantitative phase analysis according to:

Sα ρ
Xα = P  α  (2:55)
p Sp ρp

with the density of a single phase ρα (in g/cm3), which can easily be calculated
according to:

Zα Mα  1.66055
ρα = . (2:56)
Vα
Correction factors like the Brindley correction for spherical particles (Brindley, 1945)
are often applied for mixtures with very different absorptions. They should be used
with extreme caution. A more detailed analysis on the scale factor and its importance
for quantitative phase analysis is given in Chapter 5.

2.5 The peak profile

In general, the profile ΦðXÞ of a Bragg reflection centered at the peak position x0 can
be regarded as a mathematical convolution18 of contributions from the instrument,
the so-called instrumental resolution function IRF ð XÞ, and from the microstructure
MSð XÞ of the sample (Klug & Alexander, 1974):

18 See Chapter 13 (Mathematical basics).

54 2 The Rietveld method

Φð X Þ = ðIRF  MSÞð XÞ (2:57)

Here X = x – x0, where x0 is the observed peak position on the scale (2θ, TOF, E)
in which the data are recorded. The profile function is therefore described relative
to the peak center x0(s) (i.e., what we called 2θs in Section 2.1; more details in
Chapter 4) for each phase p. In the present section, the indices s and p are omitted
for simplicity.
The instrumental resolution function is split into contributions coming from the
finite width of the X-ray source (X-ray tube or synchrotron), the so-called emission
profile (EPÞ; and a series of horizontal and vertical instrumental aberrations of the
diffractometer. These include the angular acceptance function of the Soller slit(s)
controlling the axial beam divergence, the angular acceptance function of the plug-in
slit controlling the equatorial beam divergence, the angular acceptance function of
the receiving slit and so on. For linear position sensitive detector (PSD) systems, the
receiving slit aberration is replaced by functions describing the defocusing due to
asymmetric diffraction, the parallax error and the point spread function of the
detector.
The microstructure contribution of the sample contains contributions from
effects like domain size, isotropic and/or anisotropic microstrain, dislocations, fault-
ing and so on.
In the following section, several mathematical functions that are frequently used
to approximate the effect of aberrations from the instrument and from the sample to
the profile of a Bragg reflection are discussed in more detail. As most people use
constant wavelength X-ray or neutron data, the following examples are given on a
2θ-scale and we explicitly write Φð XÞ = f ð2θ − 2θ0 Þ.
All functions need to be normalized to unity:
+ð∞

Φð XÞdX = 1 (2:58)
−∞

in order not to alter the integrated intensity of the Bragg reflections. However,
reflection-independent deviations from an integral of unity will only affect the
value of the scale factor.
The different mathematical functions that contribute to the peak profile can be
either of phenomenological or physical nature. When the parameters of these func-
tions are directly related to geometrical properties of the diffraction experiments,
they are called fundamental parameters (FP) and the convolution procedure the
fundamental parameter approach. The convolution in TOPAS is either done numeri-
cally in direct space or by the product of the Fourier transforms in reciprocal space. In
the following sections, a list of common functions is explained. A more detailed
analysis on the influence of microstructural properties on the peak profile is given in
Chapter 4 (Peak shapes: Instrument  microstructure).
 2.5 The peak profile 55

2.5.1 The box function

There are several aberrations that are commonly described by a box function. These
include the size of the source in the equatorial plane, thickness of sample surface as
projected onto the equatorial plane, width of the receiving slit in the equatorial plane,
width of strips in position sensitive strip detectors, and so on. A box function of width
a is defined as:

A for  a2 < ð XÞ < a2

boxð XÞ ¼ (2:59)
0 forð XÞ ≤  a2 and ð XÞ ≥ a2

with the normalization A=1/a (Figure 2.34). The Fourier transform of a box function
with the reciprocal variable h is calculated as:

sinðπhaÞ
BOXðhÞ = Aa (2:60)
πha

A = 1, a = 1
A 1

0.8

0.6
Box(h)
box(x)

0.4

0.2

–0.2

–a/2 +a/2 –8 –4 0 4 8
x h

Figure 2.34: The box function (left) and its Fourier transform (right).

The following TOPAS code defines a box function (called hat in TOPAS) mimicking
the transmittance of a rectangular slit with a width c of 0.1 mm:

prm !c 0.1 min 0.000001

hat = Rad c / Rs;

with the sample to detector distance Rs. The box function is then automatically
convoluted with all Bragg reflections. The predefined equivalent macro is:
56 2 The Rietveld method

Slit_Width( , 0.1)

A delta function δ can be simulated by using a very small slit width of, for example,
10−5.

2.5.2 Gaussian distribution

The normal (or Gaussian) distribution is a very common continuous probability

distribution (Figure 2.35). Physical quantities that are expected to be the sum of
many independent processes (such as measurement errors) often have distributions
that are nearly normal. The expression for a normalized Gaussian distribution in
terms of its full width at half maximum fwhm is:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2
2 lnð2Þ=π − 4lnð2Þ fwhmX
gaussð XÞ = e (2:61)
fwhm

1 1

0.8 0.8
Rel. amplitude

Rel. amplitude

0.6 0.6
fwhm
0.4 0.4

0.2 0.2

0 0
–0.2 –0.1 0 0.1 0.2 –20 –10 0 10 20
X h

Figure 2.35: The Gaussian function (left) and its Fourier transform (right).

The Fourier transform of a Gaussian function is itself a Gaussian function:

π2 fwhm2 2
GAUSSðhÞ = e − 4ln2
h
(2:62)

In TOPAS, a convolution of a Gaussian function with a constant fwhm of 0.1° 2θ into

all Bragg reflections of the powder pattern can be achieved by:
 2.5 The peak profile 57

prm fwhm 0.1

user_defined_convolution = (2 Sqrt(Ln(2)/Pi)/fwhm) Exp(-4 Ln(2)(X/fwhm)^2);

where X is the measured x-axis (usually 2θ). Since the Gaussian function in TOPAS is
predefined, a simpler expression can be used:

prm c 0.1 min 0.00001

gauss_fwhm = c;

1
0.1 tan(θ)
0.1/cos(θ)
0.8

0.6
fwhm/°2θ

0.4

0.2

0
0 40 80 120 160
2θ/°

Figure 2.36: Angular dependence of the fwhm broadening according to microstrain (tan θ depen-
dence) and crystallite size (1=cos θ dependence).

The dependence of peak widths on θ can be introduced in various ways depending on

the origin of the broadening (Figure 2.36). In the case of microstrain the dependence
is usually on tan θ:

prm c 0.1 min 0.00001

gauss_fwhm = c Tan(Th);

while for crystallite size it is 1=cos θ:

prm c 0.1 min 0.00001

gauss_fwhm = c / Cos(Th);
58 2 The Rietveld method

In order to get physically meaningful values for crystallite size and microstrain, the
parameter c must be appropriately scaled as explained in Chapter 4.
Alternatively, a Fourier transform (FT) of a response function (here a Gaussian
function) can be convoluted on to peaks using a Fast Fourier Transform (FFT):

prm !fwhm 0.1 min 0.00001

ft_conv = Exp(-(Pi FT_K fwhm)^2 / (4 Ln(2)));
ft_min = 1e-8;
ft_x_axis_range = 40 fwhm;

FT_K is a reserved parameter name and it returns the Fourier transform variable
divided by the X-axis range of the peak with extension ft_x_axis_range. ft_min defines
the smallest value to which the transform is calculated.

2.5.3 Cauchy (Lorentz) distribution

The normalized Cauchy or Lorentz distribution (Figure 2.37) is defined as:

2π=fwhm
lorentzð XÞ =  2 (2:63)
X
1 + 4 fwhm

1 1

0.8 0.8
Rel. amplitude

Rel. amplitude

0.6 0.6
fwhm
0.4 0.4

0.2 0.2

0 0
–0.4 –0.2 0 0.2 0.4 –4 –2 0 2 4
X h

Figure 2.37: The Lorentzian function (left) and the real part of its Fourier transform (right).

The real part of the Fourier transform of a Lorentzian is:

 2.5 The peak profile 59

LORENTZ ðhÞ = e − 2π fwhm jhj . (2:64)

The Lorentzian function is commonly used to describe the emission profile from an
X-ray tube, as well as crystallite size and strain effects from the sample. Also, in a
perfect infinite crystal, Bragg peaks are not δ-functions, but finite Lorentzians with
the fwhm being the Darwin width.
By analogy with the Gaussian function, convolution of a Lorentzian function
with a constant fwhm of 0.1° 2θ into all Bragg reflections of the powder pattern can be
achieved in TOPAS by:

prm fwhm 0.1

user_defined_convolution = (2 Sqrt(Ln(2)/Pi)/fwhm) Exp(-4 Ln(2)(X/fwhm)^2);

or

prm c 0.1 min 0.00001

lor_fwhm = c;

Microstrain Lorentzian broadening can be described with:

prm c 0.1 min 0.00001

lor_fwhm = c Tan(Th);

and crystallite size with :

prm c min 0.00001

lor_fwhm = c /Cos(Th);

Again, in order to get physically meaningful values for crystallite size and micro-
strain, the parameter c must be appropriately interpreted.

2.5.4 The Voigt distribution

The Voigt distribution (Figure 2.38) can be regarded as the convolution of a Gaussian
and a Lorentzian:

voigtð XÞ = gaussð XÞ  lorentzð XÞ. (2:65)

Convolution of a Voigt function with a constant fwhm of 0.1° 2θ for the Gaussian and
for the Lorentzian part into all Bragg reflections of the powder pattern can be
achieved by:
60 2 The Rietveld method

0.8
Rel. amplitude

0.6
fwhm
0.4

0.2

0
–4 –2 0 2 4
X

Figure 2.38: Peak profile of the Voigt function with a Lorentzian and a Gaussian fwhm of 0.5 each.

prm lfwhm 0.1 min 0.00001

prm gfwhm 0.1 min 0.00001
lor_fwhm = lfwhm;
gauss_fwhm = gfwhm;

The assumption that microstrain and crystallite (domain) size generally contain
Gaussian as well as Lorentzian contributions leads to the so-called double-Voigt
Approach (Balzar, 1999), where the Gaussian and Lorentzian components of two
Voigt functions are refined to represent domain size (csg, csl) and microstrain (sg, sl),
respectively. Although this is a simplification, as no physically based models are
used, its application is quite common:

prm csg 200

prm csl 200
prm sg 0.1
prm sl 0.1
gauss_fwhm = 0.1 Rad Lam / (Cos(Th) csg);
lor_fwhm = 0.1 Rad Lam / (Cos(Th) csl);
gauss_fwhm = sg Tan(Th);
lor_fwhm = sl Tan(Th);

Physically, more meaningful values for average crystallite size and microstrain can
be deduced by calling the following TOPAS code with the refined parameters from
above:
 2.5 The peak profile 61

prm !k = 1
prm !kf = 0.89
prm lvol 0
prm lvolf 0
prm e0 0
LVol_FWHM_CS_G_L(k, lvol, kf, lvolf,, csg,, csl)
e0_from_Strain(e0,, sg,, sl)

where lvol is the integral breadth based volume-weighted column height with shape
factor k (here fixed to 1), lvolf is the fwhm based volume-weighted column height with
shape factor kf (defaults to 0.89) and e0 as the resulting normalized strain.
Voigt (or pseudo-Voigt) functions are also useful for describing the sample
contribution to the peak shape in TOF experiments. The fwhm needs to be converted
to TOF space:

prm lor 0.5 min 0 max 1.0

prm fwhm 0.1
peak_type pv pv_lor lor pv_fwhm = fwhm Constant(t1 0.00001) D_spacing;

where pv_lor and pv_fwhm are the Lorentzian fraction and the fwhm of the peak
profile, respectively. Note that t1 is the linear calibration parameter from the
TOF_x_axis_calibration macro (see Section 2.2.1).19
The corresponding predefined macro in TOPAS is:

TOF_PV(@, 283.50627`, @, 0.0248895257, t1)

Lorentzian and Gaussian size broadening can be similarly convoluted as shown

earlier in this chapter:

prm cl 10 min 3 max 10000

prm cg 10 min 3 max 10000
lor_fwhm = Constant(t1 .1) D_spacing^2 / cl;
gauss_fwhm = Constant(t1 .1) D_spacing^2 / cg;

The same can be achieved by the two predefined macros:

TOF_CS_L(@, 10, t1)

TOF_CS_G(@, 10, t1)

19 The expression Constant(var) assigns a fixed value to the quantity var at the start of refinement. It
is not updated if var changes.
62 2 The Rietveld method

2.5.5 The TCHZ pseudo-Voigt function

Another symmetric peak profile function often used for angular dispersive data is the
modified Thompson–Cox–Hastings pseudo-Voigt “TCHZ” (Thompson et al., 1987,
Young, 1993) that is defined in Chapter 13 (Mathematical basics). The Gaussian
fwhmG and Lorentzian fwhmL are defined as:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z
fwhmG = Utan2 θ + Vtanθ + W +
cos2 θ
Y
fwhmL = Xtanθ + (2:66)
cosθ
where U, V, W, X, Y and Z are refineable parameters. The overall fwhm can be
calculated using eq. (13.89). It should be apparent that U and X are related to
microstrain while Z and Y are related to domain size. The shape of the TCHZ
pseudo-Voigt function looks practically identical to that of the Voigt function
(Figure 2.38). One advantage of the TCHZ pseudo-Voigt function, in contrast to
the Voigt function, is the ability to easily report the fwhm of the fitted Bragg
reflections. The predefined macro for the TCHZ pseudo-Voigt function in TOPAS
is:

TCHZ_Peak_Type(pku, 0.0107, pkv, -0.0011, pkw, 0.0005, !pkz, 0, pkx, 0.0418, !pky, 0)
‘U, V, W, Z, X, Y

In this example, U, V, W and X are refined, while Y and Z are set to zero (no size
contribution).

2.5.6 The circles function

A simple approximate function useful for modeling the asymmetry of a Bragg reflec-
tion is the so called circles function with the curvature εm as an adjustable parameter
(Figure 2.39, left):
sffiffiffiffiffiffiffiffiffi
X m 
circlesð XÞ = 1 −   for 0 ≤ X ≤ Xm (2:67)
X

One of the main applications for this function is the phenomenological model-
ing of the peak asymmetry caused by axial divergence which is mainly due
to the increasing curvature of the Debye–Scherrer rings at very low and extre-
mely high angles that are cut by (typically) rectangular receiving slits of
finite width (Cheary & Coelho, 1998). In order to model axial divergence with
the circles function, a tan(2θ)-dependence is usually used as given below in
 2.5 The peak profile 63

0
c = 20
400 c = 100

–1 200
Rel. factor

Rel. factor
0

–2
Xm = 0.2 –200
Xm = 0.5
Xm = 1.0
–400
–3
–0.8 –0.4 0 0.4 0.8 0 40 80 120 160
x 2θ/°

Figure 2.39: Circle function different curvature εm (left) and dependence of the parameter εm on
diffraction angle as typical for axial divergence (right).

TOPAS notation. Please note that the asymmetry is reversed above 90° 2θ (Figure
2.39, right)20:

prm c 12 min 0 max 20

circles_conv = -.5 Rad ( c / Rs)^2 / Tan(2 Th);

The corresponding predefined TOPAS macro is:

Simple_Axial_Model(@, 12)

2.5.7 The exponential function

Asymmetry of Bragg reflections can also be described by an exponential asymmetry

decay function of the type21:

X ln 0.001
exp conv const = eXm for 0 ≤ X ≤ Xm (2:68)

20 In reality, the asymmetry due to axial divergence has its minimum at around 120° 2θ (Cheary &
Coelho, 1998).
21 ln(0.001) is a scaling factor.
64 2 The Rietveld method

that is convoluted into the peak profile. This function can be useful for describing, for
example, the highly asymmetric instrumental peak shape of TOF data or the effects of
transparency on the peak shape in Bragg–Brentano geometry.

f(
x)
X
α 2θ–α 2θ–α
O P
t' x2 ts
x1
h 2θ
M

Figure 2.40: Schematic drawing of the aberration due to the transparency effect for (a)symmetric flat
plate reflection geometry (in degrees).

In the following, we derive the peak profile aberration due to the transparency effect
following the treatment given by Masson et al. (1996) and Rowles and Buckley (2017).
Figure 2.40 is a schematic drawing of the aberration due to the transparency effect for
(a)symmetric flat plate reflection geometry of an “infinitely thick” sample. The
intensity diffracted at point M is reduced by absorption with respect to the intensity
diffracted at point O according to eq. (2.37) where the additional pathway x traveled
by the X-ray in the sample is:

t′ t′
x = x 1 + x2 = + (2:69)
sinα sinð2θ − αÞ

with:

h
t′ = x1 sin α = sin α (2:70)
sinð2θÞ

that leads to:


h sin α
x= 1+ . (2:71)
sinð2θÞ sinð2θ − αÞ

The angular variable X (in degrees) is here defined by:

h 180
X = 2θ′ − 2θ  (2:72)
Rs π

with the angle 2θ′ where diffraction is observed at the detector and the specimen–
detector distance Rs with Rs h. The intensity ratio (eq. (2.37)) thus becomes:
 2.5 The peak profile 65


 
I Rs sin α π
= exp − μ 1+ . (2:73)
I0 sinð2θÞ sinð2θ − αÞ 180

After introducing the substitution:

X
− μx = (2:74)
δ
with:

 − 1
μRs sin α 180
δ= 1+ (2:75)
sinð2θÞ sinð2θ − αÞ π

the normalized absorption profile change induced by the transparency effect is then
given by:
(
1
exp Xδ X ≤ 0
f ðXÞ ¼ δ (2:76)
0 X>0

Figure 2.41 shows several normalized absorption profiles for different diffraction
angles and linear absorption coefficients due to the transparency effect for an infinitely
thick specimen in asymmetric flat plate reflection geometry (α = 10°). Symmetric flat
plate reflection geometry simply leads to a less pronounced asymmetry of the absorp-
tion profile.

Figure 2.41: Normalized absorption profiles for different diffraction angles and linear absorption
coefficients describing the transparency effect for infinitely thick specimen in asymmetric flat plate
reflection geometry with an angle between incoming X-rays and specimen normal of 10°.

For thin samples, where the X-ray beam does not get fully absorbed within the
sample, the absorption profile will be truncated at Xmin , which can be calculated
according to Rowles and Buckley (2017) by:
66 2 The Rietveld method

tS sinð2θÞ 180
Xmin = − . (2:77)
RS sinðαÞ π

The dependence of Xmin on the scattering angle for different thicknesses of the sample
is shown in Figure 2.42.

Xmin/°
2θ/°
50 100 150

–0.005

–0.010 25 μm
50 μm
100 μm
–0.015

Figure 2.42: Angular dependence of the Xmin cut-off parameter due to transparency for different
thickness of specimen in asymmetric flat plate reflection geometry with an angle between incoming
X-rays and specimen normal of 10°.

In TOPAS language, the transparancy effect can be described as follows by the

exponential convolution function (exp_conv_const):

prm c 1 min 3 max 500

exp_conv_const = Ln(0.001) Sin(2 Th) / ( c Rs 0.003490658 );
' Note: 0.003490658 = 2 Deg / 10 used for scaling purpose

Several predefined macros exist for transparency correction:

‘scaling factor c, sample thickness d in mm.

Absorption_With_Sample_Thickness_mm_Shape(@, c, @, d)
Absorption_With_Sample_Thickness_mm_Intensity(@, c, @, d)
Absorption_With_Sample_Thickness_mm_Shape_Intensity(@, c, @, d)

The significant instrumental asymmetry of TOF peaks (Figure 2.43) can also be
satisfactorily modeled by convoluting one or more exponential functions into the
peak profile. To do this in TOPAS, the usual transformations to TOF space must be
performed as:
 2.5 The peak profile 67

prm a0 774.0 min = Max(Val .3, 1e-6); max = 2 Val + 1;

prm a1 0 min = Max(Val .3, 1e-6); max = 2 Val + 1;
prm wexp 1
prm lr 1 ‘ can be +1 or -1
exp_conv_const = lr Constant(t1) / ( a0 + a1 / D_spacing^wexp );

7
6.5
6
5.5
Sqrt (rel. intensity)

5
4.5
4
3.5
3
2.5
2
1.5
6600 6700 6800 6900 7000 7100 7200 7300 7400 7500 7600 7700 7800
TOF/μs

Figure 2.43: Asymmetric peak profile of TOF data which was modeled by a pseudo-Voigt sample profile
convoluted with two exponential functions one for the left and one for the right side of the reflection.

The predefined macro in TOPAS is:

TOF_Exponential(a0, 774.0,,0, 1, t1, +)

2.5.8 The 1/X function

Another way to describe asymmetry of a Bragg reflection is using a 1/X decay function
of the type:
4
one on x = pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi for x = 0 to Xmin (2:78)
jX Xmin j

where the parameter Xmin specifies the relative extension of the function on the X axis
and can be either positive or negative. It can, for example, be used to describe the
effect of divergence slits (either fixed or variable) on the peak profile for angular
dispersive data.
In TOPAS, the corresponding function is called one_on_x_conv. The angular
dependence of the 1/X convolution function for a fixed divergence slit can be defined
as (Figure 2.44, right):
68 2 The Rietveld method

50 0
Xmin = 1.0
Xmin = 0.5
40 Xmin = 0.2 –0.4

30 –0.8
Rel. factor

c=1

Xm
c=2
20 –1.2 c=3

10 –1.6

0 –2
0 0.2 0.4 0.60 0.80 1 0 40 80 120 160
X 2θ/°

Figure 2.44: 1/X correction function (left) and angular dependence of the correction function for
modeling the peak asymmetry due to divergence with a fixed divergence slit.

prm c 1 min 0.0001 max 6

one_on_x_conv = -c^2 Deg_on_2 / Tan(Th);

or alternatively with the predefined macro:

Divergence(@, v)

The angular dependence for a variable divergence slit is applied with:

prm c min 0.0001 max 60

one_on_x_conv = -c^2 Sin(2 Th) Rad/(4 Rs^2);

or alternatively with the predefined macro:

Variable_Divergence_Shape(@, v)

In both cases, a single refinable parameter is used.

2.5.9 The emission profile

The emission profile depends strongly on the X-ray radiation source used. For
laboratory X-rays it is usually described as a set of one to five Voigt profiles,
representing the distribution of wavelengths, depending on the degree of
 2.5 The peak profile 69

100 Cu-Kα1,2 Lab

Cu-Kα1 Lab
Synchroton
80
√(rel. intensity)

60

40

20

0
67.2 67.4 67.6 67.8 68
2θ/°

Figure 2.45: Emission profiles for angle dispersive experiments with a wavelength of λ = 1.54059 Å for
a lab source with Ni-filter (dotted), a lab source with primary beam monochromator (dashed) and a
synchrotron source (solid) in the angular range between 67.2 and 68.2° 2θ.

monochromatization (Figure 2.45). In addition, tube tails or absorption edges may be

present, which need to be modeled separately. A description of the emission profile is
the first part of any peak shape description.
In the following text, the parameter names in TOPAS are given in parenthesis.
The Voigt lines in the emission profile are defined by wavelength (lo), their relative
area (la), their Gaussian (lg) and Lorentzian (lh) half widths and the limiting ratio
between the maximum and the minimum of a calculated reflection (ymin_on_ymax),
which determines the x-axis extent over which an emission line is calculated. The
area under the emission profile is determined by the sum of the area parameters of
the individual lines, which can be scaled to unity. d-spacings are calculated from the
lo value of the emission profile line with the largest la value (Lam). The following
TOPAS script shows the empirical emission profile of Cu-Kα radiation consisting of
five Lorentzian lines according to Mendenhall et al. (2017):

' official NIST copper kalpha spectrum from JPhysB 2017

lam
ymin_on_ymax 0.0001
la sat 0.05026 lo 1.534753 lh 3.6854
la 3.91 lo 1.5405925 lg 0 lh 0.436
la 0.474 lo 1.5410769 lg 0 lh 0.558
la 1.53 lo 1.5443873 lg 0 lh 0.487
la 0.754 lo 1.5446782 lg 0 lh 0.630
70 2 The Rietveld method

For angular dispersive neutron and synchrotron data, a δ-function can be simulated
by a Lorentzian profile with a very small half width of, for example, 0.00001 mÅ:

Lam
ymin_on_ymax 0.0001
la 1 lo 0.2078 lh 0.000001

For TOF and energy dispersive data, the following macro can be used.

TOF_LAM(0.0001)

where the argument simply defines ymin_on_ymax.

For laboratory instruments, the finite width of the X-ray source is modeled by a
simple hat-function. For accurate line profile analysis it may be necessary to modify
the simple hat function model to accommodate the so-called tube tails effect
(Bergmann et al., 2000) (Figure 2.46). If these “tube tails” are present, the source
width aberration function can be better approximated by the combination of a sharp
and a broad hat function. The parameters introduced to describe tube tails are the
tube filament width [mm], effective width of tube tails in the equatorial plane
perpendicular to the X-ray beam in negative and positive z-direction [mm], and the
fractional height of the tube tails relative to the main beam:

Tube_Tails(, 0.04,, -1.208450694,, 1.529020892,, 0.00120369033)

53.97369
120
110
100
Sqrt (rel. intensity)

90
80
70
60
50
40
30
20
10
54
2θ/°

Figure 2.46: TOPAS screen shot showing the fit of a single Bragg reflection of LaB6 measured with
Cu-Kα radiation where the apparent tube tails are fitted using the Tube_tails macro.

Metal Kß filters for angle dispersive laboratory data introduce an absorption edge
between the Kα and Kβ wavelengths, dependent on the wavelength/filter material and
 2.6 The background 71

90
80
70
Sqrt (rel.intensity)

60
50
40
30
20
10

10 15 20 25 30 35 40 45 50 55 60
2θ/°

Figure 2.47: Laboratory angle dispersive X-ray powder diffraction pattern of corundum measured
with Cu radiation and a Ni-foil Kß filter.

its thickness (Figure 2.47). With position sensitive detectors (PSDs), absorption edges are
often visible due to the high signal to noise of the data. To account for this effect, the
Absorption_Edge_Correction macro can be used in TOPAS. The parameters are:
Maximum wavelength (max_lam) for extending the source emission profiles to include
transmitted Kß and Bremsstrahlung; position of the absorption edge of the filter material
(edge in Å); remnant Bremsstrahlung curvature (a_white, b_white), width of the absorp-
tion edge (a_erf); height of the absorption edge (cedge_extra):

Absorption_Edge_Correction(
1.75, ' max_lam
@, 1.48668, ' edge
@, 0.00121, ' a_white
@, 64.51902, ' b_white
@, 648.8600, ' a_erf
@, 0.00504 ' edge extra
)

2.6 The background

The observed background at position i in the powder pattern Bkgi can be modeled by
an analytical or empirical function or it can be manually defined. Sources for the
background come from the instrument and the sample, such as disorder, thermal
diffuse scattering, incoherent scattering, inelastic scattering and so on.
It is common practice to fit the background with high-order (typically 5–15)
orthogonal Chebyshev polynomials of the first kind. The higher the order, the higher
are the correlations between background coefficients and between background
 72 2 The Rietveld method

coefficients and the intensity of overlapping reflections at higher scattering angle.

Chebyshev polynomials of first kind are defined by a recursive relation:

T0 ðxÞ = 1

T1 ðxÞ = x

Tn + 1 ðxÞ = 2xTn ðxÞ − Tn − 1 ðxÞ (2:79)

where the x-axis is normalized between −1 and 1, which is done for an equidistant
2θ-axis according to:
 
2ð2θi Þ − 2θfinal + 2θstart
xi = (2:80)
2θfinal − 2θstart

The background values are then calculated as:

X
n
Bkgi = ck Tk ðxi Þ (2:81)
k=0

In TOPAS, the predefined keyword bkg followed by coefficients of the Chebyshev

polynomial can be used (Figure 2.48):

bkg @ 0 0 0 0 0 …

80

Powder pattern (silica)

Chebyshev polynomial 5th order
Chebyshev polynomial 15th order
√(rel. intensity)

60 Chebyshev polynomial 45th order

40

20

40 80 120
2θ/°

Figure 2.48: Fitting the powder pattern of amorphous silica by the background function in TOPAS
using Chebyshev polynomials of different order.
 2.6 The background 73

Quite often a steep increase of the background is observed at low scattering angle,
particularly if position sensitive detectors with large opening angles are used. Adding
a 1/2θi term to the Chebyshev polynomial often leads to a reasonable fit:

c Xn
Bkgi = + ck Tk ðxi Þ (2:82)
2θi k = 0

which in TOPAS is done by the predefined macro:

One_on_X(@, c)

A background can also be defined manually by connecting specified points by straight

line segments. In TOPAS, the following list of predefined macros can be used22:

Bkg_Straight_Line_First(0, x0, y0)

Bkg_Straight_Line(0, 1, x1, y1)
…
Bkg_Straight_Line(n-1, n, xn, yn)

The presence of humps in the background due to scattering by amorphous materials

(e.g., a glass capillary) can either be modeled by introducing additional artificial
reflections that are broadened using a small “crystallite size” or by several (more or
less phenomenological) “Debye-like” functions to describe short range order effects
(see Section 1.3):
sinðQi rÞ
Bkgi = w (2:83)
Qi
with:
sinθi
Qi = 2πsi = 4π (2:84)
λ
and a refinable weight w and correlation shell radius r. The built-in TOPAS macro
“Bkg_Diffuse(b, w, bb, r)” can be used, which can be applied in multiple ways.
Another possible background functions is a cosine Fourier series:
X
n
Bkgi = ck cosðk2θi Þ (2:85)
k=0

which in TOPAS can be realized by23:

22 Note that the first background point needs to be explicitly defined. X0 and Xn refer to the first and
last position in the powder pattern, usually 2θstart and 2θend for angle dispersive diffraction data.
23 The keyword fit_obj fits a user defined function to the observed data (usually a powder pattern).
74 2 The Rietveld method

prm bk1 1
prm bk2 1
…
prm bk12 1
fit_obj = bk1 + bk2 Cos(X Deg) + bk3 Cos(X Deg * 2) + … + bk12 Cos(X Deg * 11);

Finally, it is possible to record an experimental background of, for example, an empty

capillary and store it as an xy ASCII file. This can then be included during the Rietveld
fit with the command:

user_y back_expt experimental_background.xy ‘name then filename

prm back_scale 1.82237`_0.00310
fit_obj = back_scale * back_expt;
Plot_Fit_Obj(back_expt) ‘show on screen

2.7 The mathematical procedure

In the following, a brief description of the mathematical background of Rietveld
analysis is given. More detailed information is given (e.g.) in a book chapter by Robert
Von Dreele (2008). The quantity to be minimized (also called the objective function)
in Rietveld analysis can be formally written as:
X
N  2
S¼ wi yobs;i  ycalc;i ðpÞ (2:86)
i¼1

where yobs, i and ycalc, i are the observed and calculated intensities, respectively, at
point i in the powder pattern of N data points.
The parameter vector of size P corresponding to the number of independent
parameters can be written as:
0 1
p1
B .. C
p ¼ @ . A. (2:87)
pP

Since ycalc, i ðpÞ is a nonlinear function, it must be approximated by a Taylor series

which is usually terminated after the first term:

X
P
∂ycalc;i ðpo Þ  
ycalc;i ðpÞ ffi ycalc;i ðpo Þ þ pj  pj;0 (2:88)
j¼1
∂pj

around the initial estimates of the parameters:

 2.7 The mathematical procedure 75

01
p1;0
B C
po ¼ @ ... A: (2:89)
pP;0

The vector of the parameter shift is:

Δp = p − po . (2:90)

The objective function can then be written as:

!!2
X
N X
P
∂ycalc;i ðpo Þ
S¼ wi yobs;i  ycalc;i ðpo Þ þ Δpj . (2:91)
i¼1 j¼1
∂pj

To find the minimum of the objective function we need the first derivative with
respect to the refined parameters, and we introduce subscript k to avoid confusion:
!!
∂S XN XP
∂ycalc;i ðpo Þ ∂ycalc;i ðpo Þ
¼ 2 wi yobs;i  ycalc;i ðpo Þ þ Δpj
∂pk i¼1 j¼1
∂pj ∂pk
!
XN   ∂ycalc;i ðpo Þ X P
∂ycalc;i ðpo Þ ∂ycalc;i ðpo Þ
¼ 2 wi yobs;i  ycalc;i ðpo Þ  Δpj .
i¼1
∂pk j¼1
∂pj ∂pk
(2:92)
At the minimum, the first derivative must be zero:
∂S
¼ 0; (2:93)
∂pk
from which it follows that:
X
N X
P
∂ycalc;i ðpo Þ ∂ycalc;i ðpo Þ X
N   ∂ycalc;i ðpo Þ
wi Δpj ¼ wi yobs;i  ycalc;i ðpo Þ . (2:94)
i¼1 j¼1
∂pj ∂pk i¼1
∂pk

Changing the summations on the left side leads to:

P X
X N
∂ycalc;i ðpo Þ ∂ycalc;i ðpo Þ XN   ∂ycalc;i ðpo Þ
wi Δpj ¼ wi yobs;i  ycalc;i ðpo Þ . (2:95)
j¼1 i¼1
∂pj ∂pk i¼1
∂pk

This is equivalent to a linear set of equations in Δp:

AΔp ¼ Y (2:96)

with the components of the P × P matrix A (each k corresponds to a matrix row and
each j corresponds to a column) given by:

X
N
∂ycalc;i ðpo Þ ∂ycalc;i ðpo Þ
Akj ¼ Ajk ¼ wi (2:97)
i¼1
∂pj ∂pk
76 2 The Rietveld method

and the P components of the vector Y by:

X
N   ∂ycalc;i ðpo Þ
Yp ¼ wi yobs;i  ycalc;i ðpo Þ : (2:98)
i¼1
∂pk

The equations in Δp of eq. (2.96) are solved for every iteration of refinement. The
quantities Δp correspond to the changes in the parameters p that should minimise eq.
(2.86). Unfortunately, due to the Taylor series approximation, the computed shifts Δp
don’t directly lead to a fully minimized solution but to a hopefully better
approximation.24
The default algorithm in TOPAS for solving the system of linear equations is the
Newton–Raphson nonlinear least squares method with the Marquardt method (1963)
included for stability. The Marquardt (1963) method applies a scaling factor to the
diagonal elements of the A matrix when the solution to the normal equations fails to
reduce χ2 :

Aii, new = Aii ð1 + ηÞ (2:99)

where η is the Marquardt constant. After applying the Marquardt constant the normal
equations are solved again and χ2 recalculated; this scaling process is repeated until
χ2 reduces. The Marquardt constant η is automatically determined each iteration.
This determination is based on the actual change in χ2 and the expected change in χ2
and the expected change in χ2 . A bound constrained conjugate gradient (BCCG)
method (Coelho, 2005) incorporating min/max parameter limits is used for solving
the normal equations. Min/max limits are dynamically recalculated during the solu-
tion process.
TOPAS allows many options over the algorithms that are used for solving the
normal equations. Some of them are listed below:

no_normal_equations ‘ Prevents the use of the Marquart method

approximate_A ‘ Approximate the A matrix without the need for calculating
‘ the A matrix dot products. Based on the BFGS method (Broyden, 1970;
‘ Fletcher, 1970; Goldfarb, 1970; Shanno, 1970)
use_LU ‘ LU-decomposition is used instead of BCCG
line_min ‘ Steepest decent method
use_extrapolation ‘ Parabolic extrapolation of parameters as a function of iteration
A_matrix_memory_allowed_in_Mbytes
A_matrix_elements_tolerance

24 Least squares can get stuck in a local rather than global minimum if initial parameter approxima-
tions are poor. Chapter 7 discusses global optimization protocols for avoiding this.
 2.8 Agreement factors 77

2.8 Agreement factors

Many different statistical agreement (R-) factors have been proposed for judging the
quality of a Rietveld refinement. The most common one is the so-called profile R-factor,
which is a measure of the difference between the observed and the calculated profile:
PN
i = 1 jyobs, i − ycalc, i ðpÞj
Rp = PN (2:100)
i = 1 yobs, i

This simple sum of all differences relative to the sum of all observed values has
several problems. First, it tends to overemphasize the strong reflections and it doesn’t
take experimental uncertainties into account. Both problems are overcome by apply-
ing a weighting scheme, where every data point gets a weight wi (see below):
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN ffi
2
i=1 w i ð yobs, i − y calc, i ð p ÞÞ
Rwp = PN 2
(2:101)
i = 1 wi yobs, i

This R-factor is directly related to the Rietveld objective function of eq. (2.86).
The next problem is related to the influence of the background. If the peak to
background ratio is low, the profile R-value can be dominated by the well-fitted back-
ground points and relatively insensitive to the structural model. To avoid this problem, it
is useful to subtract the background from the observed step scan intensities in the
denominator:
PN
′ i = 1 jyobs, i − ycalc, i ðpÞj
Rp = P N
(2:102)
i = 1 jyobs, i − Bkgi j

and
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN ffi
2
w ð y − y ð p ÞÞ
R′wp = i=1 i obs, i calc, i
PN 2
(2:103)
i = 1 wi ðyobs, i − Bkgi Þ

Despite these corrections, profile R-values of different refinements can only be com-
pared for identical statistical conditions. The so-called expected R-factor, which is
mainly determined by counting statistics, gives a measure of the best possible fit:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
N −P
Rexp = PN (2:104)
2
i = 1 wi yobs, i

and:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
N −P
R′exp = PN 2
(2:105)
i = 1 wi ðyobs, i − Bkgi Þ

with the number of data points N and the number of parameters P. On an absolute
basis, the ratio χ between the weighted profile R-value and the expected R-value (also
called goodness of fit, GOF) is a good measure on the quality of the Rietveld refinement:
78 2 The Rietveld method

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN 2
Rwp i = 1 wi ðyobs, i − ycalc, i ðpÞÞ
χ= = (2:106)
Rexp N −P

A χ between 1 and 1.5 is considered good. For comparison with single crystal data, the
Bragg-R-value can be used that is based on integrated reflection intensities rather
than step scan intensities:
PK
k = 1 jIobs, k − Icalc, k j
RBragg = PK (2:107)
k = 1 Iobs, k

Iobs, k and Icalc, k are the “observed” and calculated intensities of the kth reflection out
of K reflections. Rietveld RBragg -values are often lower than those one would expect
in single crystal experiments and should therefore be interpreted with caution. The
reason for this is that for overlapping reflections, the intensity is apportioned to
individual hkl-reflections according to the ratio of the calculated intensities, aver-
aging out misfits of individual reflection intensities. This leads to a biased or overly
optimistic assessment of the Bragg-R-value.
The so called Durbin–Watson statistic (Durbin & Watson, 1971; Hill & Flack, 1987):
PN
i = 1 ðΔyi − Δyi − 1 Þ
d= PN 2
(2:108)
i = 1 ðΔyi Þ

with Δyi = yobs, i − ycalc, i measures serial correlations between adjacent data points in
the difference curve. For a good refinement in which the difference plot is random, a
value of 2.0 is expected. Correlated errors lead to significantly lower values.
All agreement factors can be accessed in TOPAS by keywords at the overall level
or for a specific Pawley, LeBail or Rietveld phase. Some TOPAS keywords related to
agreement factors are listed below:

r_wp 15.6061677 r_exp 12.0417707 r_p 10.0071346

r_wp_dash 17.0540303 r_p_dash 11.1899544 r_exp_dash 13.1589462
gof 1.29600274
r_bragg 1.66505799
weighted_Durbin_Watson 1.46695946

2.8.1 Weighting schemes

Depending on the type of observed data (powder diffraction, pair distribution func-
tion, single crystal and so on), different weights should be applied to the individual
measured intensities. In the ideal situation, the data format (XYE format) contains the
associated errors σðyobs, i Þ in the observed intensity at position i in the data set. The
default weighting for powder diffraction data in this case is the reciprocal value of the
variance of the observed step scan intensities:
 2.8 Agreement factors 79

1
wi = (2:109)
σðyobs, i Þ2

For Poisson-type counting statistics (as found for scintillation counters),

σðyobs, i Þ = yobs, i 1=2 such that the weight can be calculated as:
1
wi = , yobs, i ≥ 1 (2:110)
yobs, i

In general, TOPAS can apply weights as a function of the position Xi , the observed
intensity yobs, i , the calculated intensity ycalc, i and the standard deviation of the
observed intensity σðyobs, i Þ. Some examples of different weighting schemes in TOPAS
are:

weighting = If(SigmaYobs < 1, 1, 1/SigmaYobs^2);

weighting = 1 / Max(Yobs, 1);
weighting = If(Yobs <= 1, 1, 1 / Yobs);
weighting = ( Abs(Yobs-Ycalc) / Abs(Yobs+Ycalc) +1) / Sin(X Deg / 2);

The weighting is usually calculated at the start of each refinement cycle. In cases
where the weight is a function of ycalc, i , a flag recal_weighting_on_iter can be used to
recalculate the weighting at the start of refinement iterations.
Figure 2.49 demonstrates the effect of weighting on a refinement, where the
weighted difference curve shows that the weak peaks at higher diffraction angle
are just as important as the strong ones that are usually found at low angles. The
Rietveld plot in Figure 2.49 also shows the cumulative χ2 function, that is the

120
110
Rel. intensity

Difference curve
100
90 0
Sqrt (rel.intensity)

80
70
60
50 Weighted difference curve
Rel. intensity

40
30
0
20
10
0
20 40 60 80 100
20 40 60 80 100 120
2θ/°
2θ/°

Figure 2.49: Left: Rietveld plot of LaB6 measured with Mo-Kα1 radiation. The gray line is the so-called
cumulative χ2. Right: Unweighted (top) and weighted (bottom) difference curve of the Rietveld
refinement normalized to the sum of the absolute differences.
80 2 The Rietveld method

weighted sum of the squares of the difference between observed and calculated
powder diffraction patterns up to that point in the diffraction pattern. This visualizes
the impact of a Bragg peak or a region of the diffraction pattern on the overall fit to the
data (David, 2004).
One problem with many Rietveld fits is the presence of unknown crystalline
phases that are not included in the refinement. In such a case, a so-called robust
refinement scheme can be implemented using an iterative reweighting of the data at
each step of the refinement using, for example, the expression (Stone et al., 2009):

− 2ln½pðMjD, IÞ
wi = (2:111)
ðycalc, i − yobs, i Þ2

with pðMjD, IÞ representing the probability of the model, M, given the data, D, and
any other available information I. According to Stone et al., such a function can be
replaced by a polynomial and a logarithmic function and a better structural model is
obtained for the phase of interest. An appropriate TOPAS macro can be found in their
paper and in TOPAS.INC.

2.9 Data collection and refinement strategy

In 1999, a paper called “Rietveld refinement guidelines” was published by McCusker
et al. describing good laboratory practice starting from the measurement of a powder
pattern to the final Rietveld refinement. These guidelines are still valid and a highly
recommended source of information. A more recent practical guide should appear
soon (Madsen and Kern, 2019).
Every home-lab Rietveld refinement begins with the measurement of a powder
pattern, most commonly in either Debye–Scherrer or Bragg–Brentano geometry.
Some general rules apply for both geometries:
– The diffractometer must be careful aligned allowing for good reproducibility of
measurements.
– Instrumental aberrations related to misalignment should be avoided.
– The instrumental resolution function (IRF) should be accurately measured and
regularly monitored.
– The sample should be carefully ground to a grain size of 1–5 μm. Overgrinding
destroys crystallinity, leads to peak broadening and should be avoided.
Subsequent annealing might remove such grinding-induced broadening.
– At least 5 data points over the fwhm should be measured.
– The measured range should include all peaks at high d-spacing (low diffraction
angle) and extend until the peaks die off and become indistinguishable from the
background.
– Good counting statistics are essential. Variable counting time, where increasing
time is spent at higher diffraction angles, should be considered.
 2.9 Data collection and refinement strategy 81

Some recommendations for measurements in Bragg–Brentano geometry:

– Rotating sample holders should be used.
– Preferred orientation should be avoided as far as possible (no pressing of the
sample in the sample holder; back or side loading; spray drying etc.).
– Transparency effects on the peak shape can be minimized for very low absorbing
materials by using thin slurries of material on single crystal sample holders; the
trade-off might be the need for additional corrections of the integrated intensity.
– The contribution of the background from the sample holder should be reduced by
using low background single crystal holders (e.g., Si(911) cuts).
– Surface roughness effects can be avoided by either using slurries or by wiping off
excessive powder in the cavity of the sample holder with a glass slide.
– The constant volume condition should be ensured at all times thus avoiding
overspill effects by using a bigger sample area/smaller slits. The use of variable
divergence slits is discouraged. If variable divergence slits are used, proper calcu-
lation of the estimated standard deviations and proper scaling must be ensured.

Some recommendations for measurements in Debye–Scherrer geometry:

– The capillary must be carefully aligned, rotating and not wobbling.
– Absorption should be minimized (ideally μR ≲ 1) using either an appropriate
wavelength, a thin capillary, a sample diluted with a low absorbing amorphous
material (e.g., cork or powdered glass), or by mounting on the surface of a thin
glass rod.
– The illuminated area of the capillary should be evenly filled with powder.
– The material of the glass capillary should be low absorbing glass (e.g., lithium-
borate glass).
– Care must be taken when measuring an empty capillary for background deter-
mination, as a capillary filled with sample might have a much stronger absorp-
tion leading to a nonlinearly lowered background.

At the infancy of the Rietveld method, a proper least squares refinement strategy was
crucial as the radius of convergence for individual parameters was low and refining
too many parameters at the same time or in the wrong order led to divergence or
program crashes. This situation changed with TOPAS which is a very stable least
squares program practically allowing thousands of parameters to be refined simulta-
neously without being particularly sensitive to the order of parameter turn on.
Nevertheless, we can suggest a typical “cautious” recipe for releasing and fixing
sets of parameters which is likely to maximize the success rate of complicated
Rietveld refinements (Figure 2.50). Some of the key-points are:
– The refinement of the background and the lattice parameters and the peak profile
can be separated from the refinement of the crystal structure by starting with a
LeBail/Pawley fit, before switching to Rietveld analysis.
82 2 The Rietveld method

Fundamental parameters y n Estimated fundamental

or IRF
TCHPV from reference parameters Fixed Lorentzian
known
material crystal. size (200 nm)

LP factor LP factor
Lattice parameters, Lattice parameters, background
background FP, in particular axial divergence

LeBail/Pawwley
Sample height (only Bragg–Brentano)
and/or
zero shift

Double-Voigt for
crystallite size and microstrain

Spherical harmonics y Anisotropic

or peak
Stephens model broadening ?

Lattice parameter, background

FP, peak profile
Atomic coordinates, scale factor

Scale factor
Intensity corrections (absorption/surface roughness, etc.)

Rietveld
Overall isotropic displacement parameter
Atomic coordinates

Background
FP, peak profile

Spherical harmonics y Preferred

or orientation ?
March–Dollase model
n

Isotropic displacement parameters

Figure 2.50: Flow chart of a typical Rietveld refinement. Parameters in red are refined, those in blue
are fixed (TOPAS notation).
 2.10 Example of a Rietveld refinement 83

– The LeBail/Pawley fit will give a good indication of the best Rwp achievable for a
given data set.
– If possible, the IRF should be included from the beginning. This has two advantages.
Firstly, the IRF provides excellent starting parameters for the profile, secondly its
use allows the proper separation between instrumental and sample contributions.
– After switching to Rietveld refinement, the background parameters should
be refined again to account for possible correlations (in particular at high
angles) between the background parameters and peak intensities from a
previous Pawley/LeBail refinement.
– At the end of a Rietveld refinement, all refined parameters should be released
simultaneously to ensure proper statistics.
– Models and their associated uncertainties should be critically assessed for their
reliability and uniqueness.

2.10 Example of a Rietveld refinement

Now it is time to give an example of a full Rietveld refinement. We have selected the
laboratory powder pattern of the room temperature phase of the double salt
Mg(H2O)6RbBr3 for this. Atoms lie on general and special positions and we will use
fractional coordinates to describe the crystal structure. Later in the book, this exam-
ple will be reused to discuss alternative ways of describing a crystal structure, namely
rigid bodies (Chapter 6) and symmetry (distortion) modes (Chapter 8).
The room temperature crystal structure of Mg(H2O)6RbBr3 has monoclinic C2=c
symmetry (Dinnebier et al., 2008). Its structure is characterized by a three-dimen-
sional network of corner-sharing RbBr6 octahedra that contains one Mg(OH2)6
octahedron in the center of each void (Figure 2.51).

Br
Mg
Rb
O

Figure 2.51: Projection of the crystal structure of Mg

b (H2O)6RbBr3 as viewed along the c-axis showing
green RbBr6 and brown Mg(OH2)6 octahedra (from
a Dinnebier et al., 2008).

High-resolution laboratory X-ray powder diffraction data were recorded using a

Debye–Scherrer geometry Bruker D8 diffractometer, equipped with a Våntec-1
84 2 The Rietveld method

position-sensitive detector and Cu-Kα1 radiation (1.540596 Å) obtained from a primary

Ge(111) Johansson monochromator. A small amount of cubic RbBr is present in the
powder pattern and is included as a second phase in the Rietveld refinement.
The different parts of the TOPAS INP file25 are shown below. TOPAS INP files have a
treelike structure. At the top level we need to describe the data set using the xdd keyword:

xdd "RbBrMgBr2_6H2O_295K.raw" ‘ Powder pattern in Bruker raw format

‘do_errors ‘ flag to calc. standard uncertainties
bkg @ 0 0 0 0 0 0 ‘ 6th-order background polynomial
One_on_X(@, 6842.94765) ‘ 1/X background term
start_X 10 ‘ Start 2θ

The next section describes information about the instrument and is discussed in
detail in Chapter 4:

lam ‘ Wavelength section

ymin_on_ymax 0.0001 ‘ Extent of peak width
la 1 lo 1.540596 lh 0.401844 ‘ rel.int.,λ, Lorentzian width of emission

Zero_Error(@, 0.00289) ‘ Refined zero shift

LP_Factor( 27.3) ‘ Lorentz-polarisation factor for Ge(111)
Rp 217.5 ‘ Primary radius of the diffractometer
Rs 217.5 ‘ Secondary radius of the diffractometer
axial_conv ‘ Axial convolution section
filament_length 8 ‘ Length of the filament
sample_length 8 ‘ Length of the sample
receiving_slit_length 8 ‘ Length of the receiving slit
secondary_soller_angle @ 2.7 ‘ Angle of the secondary Soller slit
axial_n_beta 20 ‘ Number of rays in the axial plane used
to describe the axial aberration

Slit_Width( 0.1) ‘ Width of a strip of the PSD

The next section describes intensity and positional corrections due to absorption of
the capillary sample:

prm muR 0.027 min 0.01 max 1.0 ‘ Refined μeffR for absorption calculations
Cylindrical_I_Correction(muR) ‘ Intensity correction
Cylindrical_2Th_Correction(muR) ‘ Positional correction

25 We refer to the Technical Reference for full TOPAS syntax. Briefly, “@” flags an unnamed
parameter to refine; if a parameter has a name (e.g., muR) it is refined, unless its name is prefixed
by the “!” symbol, in which case it is fixed. A ’ character flags a commentThe jEdit editor (community.
dur.ac.uk/john.evans/topas_academic/jedit_main.htm) is one useful tool for editing these files in
that it can be easily configured to color-code TOPAS syntax.
 2.10 Example of a Rietveld refinement 85

The next tree level includes a description of each phase (flagged by str) that
contributes to the xdd. We start with the rubidium bromide impurity phase with fixed
coordinates:

str ‘ Tree level for the RbBr impurity phase

phase_name RbBr ‘ Phase name for the GUI
Strain_G(!stg, 0.17437) ‘ Fixed Gaussian microstrain broadening
r_bragg 2.478 ‘ Bragg-R-factor
MVW( 661.485, 327.427538, 3.498) ‘ Mol. mass of cell contents, cell volume, wt%
scale @ 3.560044e-005 ‘ Scale factor
space_group Fm-3m ‘ Hermann Maguin space group symbol
Cubic( 6.89242) ‘ Fixed cubic lattice parameter
‘ Atomic site, label, Wyckoff mult., fract. coord., occupancy, displ. param.
site Rb2 num_posns 4 x=0; y=0; z=0; occ Rb+1 1 beq !B1 2
site Br3 num_posns 4 x=1/2; y=1/2; z=1/2; occ Br-1 1 beq =B1;

It is recommended to write all special positions as equations using integer numbers

or fractions to avoid incorrect multiplicities due to round-off errors (e.g., = 1/3;,
instead of 0.3333). In order to check the correctness of the multiplicity of the site,
the keyword num_posn 0 should be included at each site. The value of the multi-
plicity will be automatically updated after the first refinement cycle. An overall fixed
(!) displacement factor is defined as parameter “B1” and used for all atoms in the list
using the equation “beq = B1;”.
The next section describes the main Mg(H2O)6RbBr3 phase. A total of 13 structural
coordinates can be refined (H atoms are omitted), the remaining ones are fixed by
symmetry:

str ‘ Mg(H2O)6RbBr3 phase

phase_name Mg(H2O)6RbBr3
CS_L(@, 800) ‘ Lorentzian size broadening
CS_G(@, 854) ‘ Gaussian size broadening
Strain_G(@, 0.123)
r_bragg 2.910
MVW( 1781.925, 1311.269, 96.459)
scale @ 8.69364709e-005
space_group C12/c1
a @ 9.641313
b @ 9.865348
c @ 13.786177
be @ 90.08773
site Br1 num_posns 4 x=1/2; y=0; z=1/2; occ Br-1 1 beq B2 2
site Br2 num_posns 8 x @ 0.25283 y @ 0.74052 z @ 0.74863 occ Br-1 1 beq=B2;
site Mg1 num_posns 4 x=1/2; y=1/2; z=1/2; occ Mg+2 1 beq=B2;
site Rb1 num_posns 4 x=1/2; y @ -0.00277 z=3/4; occ Rb+1 1 beq=B2;
site O1 num_posns 8 x @ 0.40784 y @ 0.68248 z @ 0.53974 occ O-2 1 beq=B2;
site O2 num_posns 8 x @ 0.31906 y @ 0.39673 z @ 0.54930 occ O-2 1 beq=B2;
site O3 num_posns 8 x @ 0.41407 y @ 0.51740 z @ 0.36089 occ O-2 1 beq=B2;
86 2 The Rietveld method

The graphical result of the Rietveld refinement is shown in Figure 2.52. The fit
gives satisfactory agreement factors (Rwp = 5.42 %, Rwpʹ = 8.46 %, GOF = 1.48,
RBragg = 2.91%). Most importantly, the visual fit between calculated and observed
patterns is excellent, and all features in the observed pattern are described well.
The small misfit due to an unaccounted impurity phase at 20° 2θ shows up as a step
in the cumulative χ2.

240
Mg(H2O)6RbBr3 96.46 %
220 RbBr 3.54 %
200
180
160
Sqrt (rel. intensity)

140
120
100
80
60
40
20
0

10 20 30 40 50 60 70 80 90 100
2θ/°

Figure 2.52: Rietveld plot of Mg(H2O)6RbBr3 using unconstrained refinement of fractional coordinates.
To emphasize weaker features, the square root of the intensity is displayed. A small amount of
RbBr is included as second phase. As in most figures in the book, experimental data are shown in blue,
the calculated pattern in red and the difference (obs − calc) in gray. Vertical tick marks at the bottom of
the plot show positions of allowed hkl reflections. The gray line shows the cumulative χ2.

References
Balzar, D. (1999): Voigt-function model in diffraction line-broadening analysis in Defect and microstruc-
ture analysis from diffraction, Edited by R.L. Snyder, H.J. Bunge, International Union of
Crystallography Monographs on Crystallography, Oxford University Press, New York (US), 785
pages.
Bergmann, J., Kleeberg, R., Haase, A., Breidenstein, B. (2000): Advanced fundamental parameters
model for improved profile analysis. Mat. Sci. Forum 347–349, 303–308.
Brindley, G.W. (1945): The effect of grain or particle size on X-ray reflections from mixed powders and
alloys, considered in relation to the quantitative determination of crystalline substances by
X-ray methods. Phil. Mag. Ser. 36, 347–369.
Broyden, C.G. (1970): The convergence of a class of double-rank minimization algorithms. J. Inst.
Maths. Applics. 6, 76–90.
Cheary, R.W., Coelho, A.A. (1992): A fundamental parameters approach to X-ray line-profile fitting. J.
Appl.Cryst. 25 (2), 109–121.
 References 87

Cheary, R.W., Coelho, A.A. (1998): Axial divergence in a conventional X-ray powder diffractometer. I.
theoretical foundations. J. Appl. Cryst. 31, 851–861.
Coelho, A.A. (2005): A bound constrained conjugate gradient solution method as applied to
crystallographic refinement problems. J. Appl. Cryst. 38, 455–461.
David, W.I.F. (2004): Powder diffraction: least-squares and beyond. J Res Natl Inst Stand Technol.
109, 107–123.
Debye, P. (1915): Zerstreuung von Röntgenstrahlung. Annalen der Physik, 351, 809–823.
Dinnebier, R.E., Billinge, S.J.L. (eds.) (2008): Powder diffraction: theory and practice. RSC publication,
Cambridge UK, 574 pages.
Dinnebier, R.E., Liebold-Ribeiro, Y., Jansen, M. (2008): The low and high temperature crystal struc-
tures of [Mg(H2O6)]XBr3 double salts (X = Rb, Cs). Z. Anorg. Allg. Chem. 634, 1857–1862.
Durbin, J., Watson, G.S. (1971): Testing for serial correlation in least square regression. III. Biometrika
58, 1–19.
Egami, T., Billinge, S. (2012): Underneath the Bragg peaks - structural analysis of complex materials,
Volume 16, 2nd edition, Pergamon, Oxford (UK), 422 pages.
Fischer, R.X. (1996): Divergence slit corrections for Bragg–Brentano diffractometers with rectangular
sample surface. Powder Diffraction 11, 17–21.
Fletcher, R. (1970): A new approach to variable metric algorithms. Comput. J. 13, 317–322.
Giacovazzo, C., Monaco, H.L., Artioli, G., Viterbo, D., Milanesio, M., Gilli, G., Gilli, P., Zanotti, G.,
Ferraris, G. (2011): Fundamentals of crystallography (International Union of Crystallography
monographs on crystallography). 3rd edition, Oxford University Press, New York (US), 872
pages.
Goldfarb, D. (1970): A family of variable metric updates derived by variational means. Math. Comput.
24, 23–26.
Gozzo, F., Cervellino, A., Leoni, L., Scardi, P., Bergamaschi, A., Schmitt, B. (2010): Instrumental
profile of MYTHEN detector in Debye-Scherrer geometry. Z. Kristallogr 225, 616–624.
Hill, R.J., Flack, H.D. (1987): The use of the Durbin-Watson d statistic in Rietveld analysis. J. Appl.
Cryst. 20, 356–361.
Hinrichsen, B., Dinnebier, R.E., Jansen,. M (2008): Two-dimensional diffraction using area detectors,
in Powder diffraction – theory and practice, Edited by R.E. Dinnebier and S.L.J. Billinge. RSC
Publishing, Cambridge, 414–438 Pages.
Hull, A.W. (1917): A new method of X-Ray crystal analysis. Phys. Rev. 10, 661–697.
Khalifah, P. (2015): Use of radial symmetry for the calculation of cylindrical absorption coefficients
and optimal capillary loadings. J. Appl. Cryst. 48, 149–158.
Klug, H.P., Alexander, L.E. (1974): X-ray diffraction procedures for polycrystalline and amorphous
materials, 2nd edition, John Wiley and Sons, New York, 966 pages.
Krüger, H., Fischer, R.X. (2004): Divergence-slit intensity corrections for Bragg–Brentano
diffractometers with circular sample surfaces and known beam intensity distribution. J. Appl.
Cryst. 37, 472–476.
Le Bail, A., Duroy, H., Fourquet, J.L. (1988): Ab-initio structure determination of LiSbWO6 by X-ray
powder diffraction. Mat. Res. Bull. 23, 447–452.
Levenberg, K. (1944): A method for the solution of certain problems in least squares. Quart. Appl.
Math. 2, 164–168.
Madsen, I.C., and Scarlett, N.V.Y. (2008): Quantitative phase analysis, in Powder diffraction: theory and
practice, Edited by. R.E. Dinnebier and S.L.J. Billinge, RSC Publishing, Cambridge UK, 298–331.
Madsen, I.C., Kern, A. (2019): Guidelines for collecting high quality powder diffraction data on a
laboratory instrument.
Malmros, G., Thomas, J.O. (1977): Least squares structure refinement based on profile analysis of powder
film intensity data measured on an automatic microdensitometer. J. Appl. Cryst. 10,7–11.
88 2 The Rietveld method

Marquardt, D. (1943): An Algorithm for least-squares estimation of nonlinear parameters. J. Appl.

Math. 11, 431–441.
Mendenhall, M.H., Henins, A., Hudson, L.T., Szabo, C.I., Windover, D., Cline, J.P. (2017): High-precision
measurement of the X-ray Cu Kα spectrum. J Phys. B Mol. Opt. Phys. 50, 115004 18pp.
McCusker, L.B., Von Dreele, R.B., Cox, D.E., Louer, D., Scardi, P. (1999): Rietveld refinement guide-
lines. J. Appl. Cryst. 32, 36–50.
Pawley, G.S. (1981): Unit-cell refinement from powder diffraction scans. J. Appl. Cryst. 14, 357–361.
Pecharsky, V., Zavalij, P. (2009): Fundamentals of powder diffraction and structural characterization
of materials, 2nd Edition, Springer US, Softcover, 744 pages.
Pitschke, W., Mattern, N., Hermann, H. (1993): Incorporation of microabsorption corrections in
Rietveld analysis. Powder Diffraction 8, 223–228.
Press, W.H., Flannery, B.P., Teukolsky, S.A., Vetterling, W.T. (1986): Numerical recipes, Cambridge
University Press, Cambridge (UK), 819 pages.
Rietveld, H.M. (1967): Line profiles of neutron powder-diffraction peaks for structure refinement. Acta
Cryst. 22, 151–152.
Rietveld, H.M. (1969): A profile refinement method for nuclear and magnetic structures. J. Appl. Cryst.
2, 65–71.
Rowles, M. R., Buckley, C.E. (2017): Aberration corrections for non-Bragg–Brentano diffraction
geometries. J. Appl. Cryst. 50, 240–251.
Sabine, T.M. (1985): Extinction in polycrystalline materials. Aust. J. Phys. 38, 507–518.
Sabine, T.M., Hunter, B.A., Sabine, W.R., Ball, C.J. (1998): Analytical expressions for the transmission
factor and peak shift in absorbing cylindrical specimens. J. Appl. Cryst. 31, 47–51.
Sabine, T.M., Von Dreele, R.B., Jorgensen, J.E. (1988): Extinction in time-of-flight neutron powder
diffractometry. Acta Cryst. A 44, 374–379.
Sears, V. (1992): Neutron scattering lengths and cross section. Neutron News 3, 26–37.
Shanno, D.F. (1970): Conditioning of quasi-Newton methods for function minimization. Math.
Comput. 24, 647–656.
Suortti, P. (1972): Effects of porosity and surface roughness on the X-ray intensity reflected from a
powder specimen. J. Appl. Cryst. 5, 325–331.
Stinton, G.W., Evans, J.S.O. (2007): Parametric Rietveld refinement. J. Appl. Cryst. 40, 87–95.
Stone, K.H., Lapidus, S.H., Stephens, P.W. (2009): Implementation and use of robust refinement on
powder diffraction in the presence of impurities. J. Appl. Cryst. 42, 385–391.
Thompson, P., Cox, D.E., Hastings, J.B. (1987): Rietveld refinement of Debye-Scherrer synchrotron
X-ray data from Al2O3. J. Appl. Cryst. 20, 79–83.
TOPAS version 6 (2017), Bruker-AXS, Karsruhe, Germany.
Von Dreele, R. B. (1999): Combined Rietveld and stereochemical restraint refinement of a protein
crystal structure. J. Appl. Cryst. 32, 1084–1089.
Von Dreele, R.B. (2008): Rietveld refinement in powder diffraction: theory and practice, Edited by
R.E. Dinnebier and S.L.J. Billinge, RSC Publishing, Cambridge, 266–281 Pages.
Waasmaier, D., Kirfel, A. (1995): New analytical scattering-factor functions for free atoms and ions
Acta Cryst. A 51, 416–431.
Wahlberg, N., Bindzus, N., Bjerg, L., Becker, J., Dippel, A.C., Iversen, B.B. (2016): Synchrotron powder
diffraction of silicon: high-quality structure factors and electron density. Acta Cryst. A 72, 28–35.
Wilson, A.J.C. (ed.) (1995): International tables for crystallography, Volume C: mathematical, physical
and chemical tables. Corrected reprint of 1st edition. Kluwer Academic Publishers, Dordrecht,
The Netherlands.
Young, R.A. (ed.) (1993): Introduction to the Rietveld method, The Rietveld Method, IUCr Book Series,
Oxford University Press, New York (US), 287 pages.
3 Structure independent fitting

3.1 Introduction
Before starting a Rietveld refinement, it is very helpful to have good starting para-
meters for peak positions, peak profile (microstructure) and background. If these
parameters are known with reasonable accuracy, one can focus on just the structural
aspects of the model in the early stages of a Rietveld refinement. The best way to
achieve this is to perform a structure-independent whole powder pattern fitting
(WPPF) beforehand. The predetermined parameters can then be transferred directly
to Rietveld refinement.
Structure-independent WPPF also provides a set of peak intensities, uncertain-
ties and the correlations between them for overlapping reflections. These are often
needed for different structure solution methods such as direct methods, simulated
annealing (Coelho, 2000; David et al., 2006) or charge flipping (Oszlányi & Süto,
2004). The better the peak shape is described, the more accurately partly overlapping
reflections can be separated.
In general, three main WPPF methods are available that differ in the amount of
constraints and how the intensities are obtained.
The simplest method is unconstrained single peak fitting, where a set of para-
meters is refined to fit each peak separately. Although this might lead to an
excellent fit, the refined parameters can be completely meaningless. Therefore a
number of constraints need to be introduced. First, the instrumental resolution
function should be used as a basis for all peak shapes. Second, a limited number of
overall background parameters should be included in the fitting process. In addi-
tion, since shape parameters like asymmetry or strain and crystallite size are
usually smooth functions of the scattering angle, it might be useful to constrain
them. Single peak fitting is used when no lattice parameters are known. The
obtained peak positions are usually of higher quality than typical peak search
methods employing derivatives, and can be used, for example, for indexing the
powder pattern.
If the lattice parameters are known, they should be included (with a suitable
zero shift or sample-displacement function) in the WPPF, thus constraining the
peak positions. This drastically reduces the number of parameters needed. The two
common approaches for this are the Pawley (Pawley, 1981) and the Le Bail (Le Bail
et al., 1988; Le Bail, 2005) methods; they differ in how integrated reflection inten-
sities are extracted. If the space group is unknown, the refinement is performed in a
space group without extinctions, for example, for the primitive orthorhombic lattice
this would be extinction group P— (P222, Pmmm, Pmm2, Pm2m, P2mm). The most
probable space groups can then be found either manually by taking all peaks close
to zero intensity as extinct, or more systematically by fixing all parameters and

https://doi.org/10.1515/9783110461381-003
90 3 Structure independent fitting

refining in all possible space groups one-by-one. The space groups leading to the
best Rwp agreement factors, having a minimum number of unfitted lines (by visual
inspection of the difference curve) and a minimum number of predicted reflections
with zero intensity are the most probable ones. It’s also possible to apply more
statistically sophisticated methods for space group choice (Markvardsen et al.,
2001). Programs such as ExtSym will automatically read TOPAS output files to
achieve this.
One of the problems in WPPF is the possible correlation between background
parameters and peak intensities at high diffraction angle where strong peak overlap
occurs. It is crucial to keep the number of background parameters low and to check
the oscillations of the background function visually.
Following WPPF, switching to Rietveld refinement or simulated annealing is
straightforward. One first fixes all predetermined parameters, adds a crystal
structure (or part of it), refines the scale factor and successively introduces
structural parameters. Toward the end of a Rietveld refinement it is good
practice to free the previously fixed parameters. This is particularly important
for the background that may be poorly described in WPPF due to peak overlap
at higher diffraction angles.

3.2 Constrained single peak fitting

The simplest method for WPPF is to fit all peaks that can be distinguished by eye
individually but with an overall background function and constrained strain and
crystallite size parameters (e.g., using the double-Voigt approach).
From a practical viewpoint, the peak positions should be fixed for the first
iteration while the intensities are refined freely. The starting value for the intensities
should be set to a small number to ensure convergence. After the first iteration, the
peak positions can be released for free refinement. Quite frequently, not all over-
lapping peaks have been identified. Visual inspection of the difference curve some-
times shows unaccounted intensity that belongs to missing peaks. These peaks
should be included for the next iteration.
In the following example K(C5H5) (KCp) was measured at a synchrotron (Figure
3.1). We will use a simple delta-type function to describe instrumental broadening, as
the peak shape is predominantly determined by the sample:

xdd "Kcp.raw"
bkg @ 362.60 -47.51 -57.54 -33.07 64.24 2.02 -9.88 -1.07 1.77 -4.17 0.04
start_X 6
finish_X 31.3
LP_Factor(90)
convolution_step 2
Rs 200.5
Simple_Axial_Model(6.1)
 3.3 The Le Bail method 91

lam
ymin_on_ymax 0.001
la 1 lo 1.14937 lh 1e-006
xo_Is
xo @ 9.118041043
peak_type fp
LVol_FWHM_CS_G_L(1, 135.973, 0.89, 129.658,csg, 146.917,csl, 9918.224)
e0_from_Strain( 0.0011,,,stl, 0.5116)
I @ 1.130485033
xo_Is
xo @ 12.56787108
peak_type fp
LVol_FWHM_CS_G_L(1, 135.9734, 0.89, 129.658,(csg), 146.917,(csl), 9918.224)
e0_from_Strain( 0.0011,,,(stl), 0.5116)
I @ 0.09691388313_LIMIT_MIN_1e-015
…
xo_Is
xo @ 30.71553312
peak_type fp
LVol_FWHM_CS_G_L(1, 135.973, 0.89, 129.658,(csg), 146.917,(csl), 9918.224)
e0_from_Strain( 0.0011,,,(stl), 0.5116)
I @ 4.676600833

3.3 The Le Bail method

In the Le Bail WPPF method (Le Bail et al., 1988; Le Bail, 2005), all parameters except
for structural parameters (atomic positions, occupancies, displacement factors) are
subjected to least squares refinement in a process analogous to regular Rietveld
refinement.
The Le Bail WPPF method iterates the Rietveld formula and thus requires only a
slight modification of the Rietveld code. Recall that the Rietveld formula (eq. (2.5)) of
a single phase for calculating the step-scan-intensity of step i is:
X 
ycalc, i = S jFcalc, s j2  Φs, i  Corrs, i + Bkgi . (3:1)
s

Since Fcalc, s j2 cannot be calculated from a crystal structure, all “calculated” peak
intensities are initially set to an arbitrary value, for example:

jFcalc, s j2 = 1.0. (3:2)

These are then entered in the Rietveld decomposition formula as “calculated” struc-
ture factors as if they had been derived from a structural model.
The Rietveld refinement then determines a set of new “calculated” structure
factors jFcalc, s j2new from the decomposition formula according to:
 92
3 Structure independent fitting

Figure 3.1: Constrained single peak WPPF of KCp. Top: Entire powder pattern; bottom: Zoomed region.
 3.3 The Le Bail method 93

P 
S i yobs, i jFcalc, s j2  Φs, i  Corrs, i
jFcalc, s j2new = (3:3)
ycalc, i

This effectively scales jFcalc, s j2 by the ratio between observed and calculated inten-
sities. These are then used as “calculated” structure factors in eq. (3.3) for the next
iteration until convergence is reached:

jFcalc, s j2 = jFcalc, s j2new (3:4)

Hence, the intensities of the individual peaks are not treated as least squares para-
meters and are not directly refined.
If identical peak intensities are used as starting values, the intensities of (almost)
fully overlapping reflections tend to be equipartitioned after the refinement con-
verges. Negative intensities are not possible.
Alternatively, if part of the crystal structure is known (e.g., the position of heavy
atoms), the intensities from a preliminary Rietveld refinement can be used as starting
intensities for a Le Bail WPPF. Intensity ratios of overlapping reflections are then
closer to their “true” values often leading to better results if direct methods are
subsequently used for structure completion.
The example of a typical Le Bail WPPF of KCp is given below. The small aniso-
tropic peak width due to microstrain is not accounted for (Figure 3.2):

xdd "Kcp.raw"
bkg @ 349.97 -67.75 -59.31 -14.93 64.72 -17.97 -15.28 2.47 4.96 0.119 5.10
start_X 6
finish_X 34
LP_Factor( 90)
Zero_Error(@, -0.002447297771)
Rs 200.5
Simple_Axial_Model(@, 6.022114618)
lam
ymin_on_ymax 0.001
la 1 lo 1.14937 lh 1e-006
hkl_Is
phase_name "KCP hkl_Phase"
lebail 1
TCHZ_Peak_Type(@, 1, @, -0.3174, @, 0.020,, 0,@, 0.4765,, 0)
Tetragonal(@ 9.9695,@ 10.4995)
space_group p-421c
hkl_m_d_th2 1 0 1 8 7.22960567 9.11856079 I 1.110176902
hkl_m_d_th2 1 1 0 4 7.04950809 9.3520298 I 0.008123926706
…
hkl_m_d_th2 5 1 0 8 1.95518172 34.186676 I 0.2618402688
 94
3 Structure independent fitting

Figure 3.2: Le Bail WPPF of KCp. Top: entire powder pattern; bottom: zoomed region.
 3.4 The Pawley method 95

3.4 The Pawley method

In the Pawley WPPF method (Pawley, 1981), the unit cell parameters, background
parameters, zero point errors, peak shape parameters and all reflection intensities are
subjected to nonlinear least squares refinement. In principle, the following system of
equations must be solved:
P
ycalc, 0 = ðIs  Φs, 0 Þ + Bkg1
Ps
ycalc, 1 = ðIs  Φs, 1 Þ + Bkg2
(3:5)
... s
P
ycalc, N − 1 = ðIs  Φs, N − 1 Þ + Bkgn
s

where ycalc, i is the calculated step scan intensity at position i in the powder pattern, Is is
the intensity of reflection s, Φs, i is the value of the normalized peak shape function of
reflection s at position i 2 ½0...N − 1 in the powder pattern. The reflection positions are
determined by the unit cell parameters and the zero point error. The background point
Bkgi is either predetermined or modeled by a polynomial or similar function (see
Chapter 2). In terms of a least squares minimization procedure, this requires typically
a (10+n) × (10+n) square matrix, where 10 is the number of background and lattice
parameters refined and n is the number of symmetry-independent reflections generated
for the 2θ range covered by the data. The correlation between the peak intensities
increases with increasing overlap (up to 100% for peaks with identical d-spacing within
the limits of resolution of the data). Since there is nothing in the least squares procedure
that forces the peaks to have positive intensity, it could be that for two overlapping
peaks one peak shows a negative intensity while the other has an intensity higher than
the sum of the two peaks. To reduce this problem, Pawley introduced restraints into the
least squares procedure such that as the difference between the calculated 2θ values of
two adjacent peaks approaches zero, the intensity of the two peaks is set equal with a
weight dependent on their separation. Alternatively, refinement can be based on |F|
instead of I. This forces the peak intensities to have positive values only.
TOPAS performs a Pawley fit for a hkl_Is phase if the keyword lebail 1 is not defined.
TOPAS is quite sophisticated in taking specific care in avoiding unstable (and ulti-
mately singular) least squares matrices. As a result, Le Bail and Pawley WPPF fits in
TOPAS lead to identical results in terms of agreement factors. A Pawley fit typically
requires fewer least squares cycles for convergence and is slightly faster. With the
Pawley method the refined intensities and their uncertainties can be directly used for
structure determination by charge flipping (Oszlányi & Süto, 2004). The keyword:

Out_for_cf(file)

outputs the intensities and their correlations from a Pawley refinement for use in
charge flipping.
96 3 Structure independent fitting

References
Coelho, A.A. (2000): Whole-profile structure solution from powder diffraction data using simulated
annealing. J. Appl. Cryst. 33, 899–908.
David, W.I.F., Shankland, K., van de Streek, J., Pidcock, E., Motherwell, W.D.S., Cole, J.C. (2006):
DASH: a program for crystal structure determination from powder diffraction data. J. Appl. Cryst.
39, 910–915.
Dinnebier, R.E., Billinge, S.J.L. (eds) (2008): Powder diffraction: theory and practice, The Royal
Society of Chemistry (RCS), 574 pages.
Le Bail, A., Duroy, H., Fourquet, J.L. (1988): Ab-initio structure determination of LiSbWO6 by X-ray
powder diffraction. Mat. Res. Bull. 23, 447–452.
Le Bail, A., (2005): Whole powder pattern decomposition methods and applications: a retrospection.
Powder Diffraction 20, 316–326.
Markvardsen, A.J., David, W.I.F., Johnston, J.C., Shankland, K. (2001): A probabilistic approach to
space-group determination from powder diffraction data. Acta Cryst. A 57, 47–54.
Oszlányi, G., Süto, A. (2004): Ab initio structure solution by charge flipping. Acta Cryst. A 60, 134–
141.
Pawley, G.S. (1981): Unit-cell refinement from powder diffraction scans. J. Appl. Cryst. 14, 357–361.
4 Peak shapes: Instrument
microstructure

4.1 Introduction
A diffraction peak measured on a scale x, which is most commonly x = 2θ (angular
dispersive data), can be understood as a convolution of several different contribu-
tions. For a diffraction peak located at position x0 the convolution is best described
on the scale X = x − x0 . The two most fundamental contributions are the instrumental
contribution, IRF(X) (Instrumental Resolution Function) and the sample contribution
MS(X) (from MicroStructure). MS(X) is also called structural line broadening. The
overall, peak profile Φ(X) of a particular reflection can be described as a convolution
of these two contributions:

Φð XÞ = ðIRF
MSÞð XÞ. (4:1)

As we will show later, both MS and IRF can be regarded as convolutions of several
subcontributions. As TOPAS is able to handle convolutions of arbitrary functions, it
offers great opportunities to model peak profiles. In particular, it is possible:
– to optimize line-broadening models to improve the quality of Rietveld fits.
– to develop sophisticated models to extract microstructural information from the
sample-dependent line broadening.

TOPAS can also handle different scales, like time-of-flight (TOF) or energy (E).
Moreover, other scales are important for theoretical considerations (like strain).
Generally, the parameters characterizing a peak, like its position x0 and its width
parameters δ (e.g., fwhm) will depend on the scale these parameters are referring to.
In order to show how widths on different scales are related, let’s consider another
scale y and let:

y = yð xÞ (4:2)

be a bijective (invertible) function of x. If the position of a given reflection hkl is x0 ,

the position on the y scale is y0 as calculated by eq. (4.2). If a small shift x − x0 with
respect to a position x0 is considered, on the y scale this shift will be:

dy
y − y0 = ðx − x0 Þ: (4:3)
dxx0

The shift can correspond to an actual shift of the peak position due to an effect like
strain. Alternatively, the (positive) peak width δ can be related to the absolute value
of such a shift. The peak width parameter fwhm on the two different scales is then:




dy

fwhmy =



fwhmx , (4:4)
dx x0

https://doi.org/10.1515/9783110461381-004
98 4 Peak shapes: Instrument
microstructure

where the subscript gives the scale. In some cases we will also need to model a hkl-or
other dependence to peak width, and we will flag this using a superscript: fwhmhkl 2θ .
In simple cases, where peak broadening is symmetric and isotropic, it is possible
to directly use Φ(X) to describe peak shapes. However, to model more complex
effects, such as asymmetry or for a quantitative interpretation of the structural line
broadening in terms of, for example, size and microstrain parameters; the IRF hkl and
MShkl terms must be considered separately.
Instrumental line broadening is experimentally assessable using a standard
material with no or only negligible structural line broadening. From its diffraction
pattern, the IRF is assessed either by fitting and/or by considering the diffraction
geometry (see Section 4.2). The IRF will nearly always be isotropic, that
is, IRF hkl = IRF x0 .1 The parameters describing the IRF are then fixed when evaluating
diffraction data recorded from a different material under the same measurement
conditions as those applied for the instrumental standard. The additional broadening
MShkl in eq. (4.1) is then modeled by one or more suitable sample contributions with
refined parameters.
Note the following:
(1) The intrinsic properties of an instrumental standard may affect the measured
instrumental profile. One property is absorption that can affect the diffracting
volume of the specimen and its peak profile (see Chapter 2). Hence, the X-ray
absorption coefficients of the instrumental standard and the material to be
investigated should be matched if possible.
(2) The diffraction pattern from an instrumental standard only provides direct
information about IRF x0 around the positions where diffraction peaks are avail-
able. An appropriate functional description of the x0 -dependence of IRF x0 will
allow determination of the instrumental resolution around arbitrary x0 values,
but extrapolation to values outside the range of measured peaks can be proble-
matic. Physically realistic descriptions of the IRF will behave better upon extra-
polation than purely empirical descriptions.

4.2 Determination of the instrumental resolution function (IRF )

All powder diffraction patterns, even those obtained at the highest resolution synchro-
tron source, have IRFx0 contributions originating from the instrument. It is, therefore
important to know this function in detail. This not only helps monitoring changes of

1 This excludes use of narrowly textured or single crystalline materials for determining the instru-
mental resolution of a powder diffractometer, because certain contributions to the instrumental
broadening like divergence are affected by a narrow texture (see Section 4.2).
 4.2 Determination of the instrumental resolution function (IRF ) 99

the diffractometer over time, but also allows explicit determination of the sample-
dependent line broadening contributions. The latter is usually dominated by isotropic
and/or anisotropic microstrain and domain size. In the following, the most common
case of angle-dispersive diffraction data is considered, that is, x = 2θ. The x = TOF scale
will be considered in Section 4.4.
The first step in the determination of the IRF consists of measuring a high-quality
powder pattern of a line profile standard like NIST SRM 660a LaB6 (current batch is
660c) over the relevant range of diffraction angles under identical experimental
conditions to the samples under investigation. The standard is expected to contain
only negligible sample contributions, meaning “high crystallinity,” no microstrain
broadening and a domain size large enough (>500 nm) not to cause any size broad-
ening but small enough to ensure reasonable particle statistics. Misalignment of the
diffractometer and the sample should be avoided.
In general, the convolution approach by Klug & Alexander (1974) is used to build
up the IRF starting from the emission profile. In the so-called fundamental parameters
(FP) approach (Cheary & Coelho, 1992), the IRF is built up essentially from first
principles by convoluting the contributions due to relevant instrumental aberrations.
Ideally one only uses measurable physical quantities like slit widths, slit lengths,
Soller slit opening angles and so on without refinement. In reality, minor refinement
might be necessary since the FP process involves some approximations for computa-
tional speed and simplicity.
Alternatively, well-established phenomenological line shape functions, like the
extended Thompson Cox Hastings (TCHZ) Pseudo-Voigt function (Thompson et al.,
1987; Young, 1993) can be used to describe the IRF (see Section 2.5.5). The treatment,
however, is usually supplemented with models considering divergence-related asym-
metry of the diffraction peaks.
In the following, a stepwise recipe is given to determine the IRF of different
angular dispersive laboratory powder diffractometers using the LaB6 line profile
standard (NIST SRM 660a LaB6). LaB6 crystallizes in space group Pm
3m with a lattice
parameter of a = 4.155 Å. The lanthanum and boron atoms are located at (0, 0, 0) and
(~0.2, ½, ½) respectively.

4.2.1 Bragg–Brentano geometry (Cu-Kα1,2 doublet): FP approach

In this section LaB6 data from a Bragg–Brenatano diffractometer URD6 (Präzisionsme-

chanik Freiberg, Germany) are analyzed with the FP approach. The diffractometer has a
radius of 250 mm and is equipped with a Cu tube. The primary beam path contained a
divergence slit (giving 0.46° divergence) and Soller slits (length 25 mm, plate spacing 0.5
mm). The secondary beam path contained a receiving slit of 0.52 mm and a curved
graphite monochromator. The sample was a thin layer of LaB6 powder on a “zero-
background” Si plate with its (510) plane parallel to the surface. Figure 4.1 illustrates
100 4 Peak shapes: Instrument
microstructure

1 3 2
3
4 4

2θ
6

Figure 4.1: Schematic of a Bragg–Brentano diffractometer and the main components which affect
the IRF. The tube is located near (1) and the detector near (5). The anode material in the tube
determines the emission profile seen by the sample and also gives rise to the tube tails. The
receiving slit (2) adds an additional contribution. When a position-sensitive detector (PSD) is used
at this position, the simplest treatment is to consider a receiving slit with a width corresponding to
the PSD channel width. The axial divergence (perpendicular to the plane of the paper) and its
contribution to the IRF is determined by the length of the tube focus (1), and the acceptance angles
of the Soller slits (3). The equatorial divergence (within the paper plane) is determined by the size
of the divergence slits (4). Here the primary and secondary diffractometer radii (1)–(6) and (6)–(2)
are identical. The large circle is the (para) focusing circle with a 2θ dependent radius. Usually a
diffracted beam monochromator is present in front of the detector (5), selecting the Kα1,2 radiation
from the X-rays emitted by the tube so that only the corresponding lines are considered in the
emission profile. Alternatively, a primary beam monochromator may be used to select Kα1, radia-
tion, modifying the emission profile correspondingly. Position (1) would then correspond to the
focus of the primary beam monochromator.

the main components of the diffractometer together with the corresponding IRF
contributions.
Figure 4.2. illustrates the progressive improvement of a Rietveld fit upon adding
2θ 2θ 2θ
various contributions to the IRF, g2θ0 = g1 0
g2 0
g3 0 ... . The scale factor and the
background parameters were always adapted to give the correct peak height. The
TOPAS INP file is discussed below.
The data (xdd) are available in file xy_LaB6_rec-052_109-10_URD6L1_220316.xy in
xy-format (two columns: 2θ and number of counts). The keyword x_calculation_step 0.02
ensures that the calculated data match the experimental step size. The keyword chi2_-
convergence_criteria sets a stricter refinement convergence limit than the TOPAS default.
Six background parameters are set to refine. A Lorentz-polarization factor for a graphite
monochromator is defined:

xdd xy_LaB6_rec-052_109-10_URD6L1_220316.xy
r_wp 7.70633377
chi2_convergence_criteria 0.000001
x_calculation_step 0.02
bkg @ 180.020452 -14.8395972 33.2174362 -17.2567608 4.27653909 -0.474351033
LP_Factor( 26.6) ‘ Graphite diffracted beam monochromator
 4.2 Determination of the instrumental resolution function (IRF ) 101

40000 40.000 40.000

Only wavelength profile

110

Wavelength profile

Wavelength profile
+ slit + divergence
30000 30.000 30.000

20000 20.000 20.000

+ slit
10000 10.000 10.000

0 0 0
5.000 30 315.000 30 31
5.000 30 31

4000
321 4.000 4.000
Intensity/counts

3000 3.000 3.000

2000 2.000 2.000

1000 1.000 1.000

0 0 0
87 88 89 87 88 89
87 88 89
5000 431/5.000 5.000
510
4000 4.000 4.000

3000 3.000 3.000

2000 2.000 2.000

1000 1.000 1.000

0 0 0
140 141 142 143 141 142 143140 141 142 143
2θ/°

Figure 4.2: (110) (top), (321) (middle row) and (510)/(431) (bottom) reflections of the LaB6 standard
measurement considered in Section 4.2.1. Data points are shown as circles and fitted pattern as a
continuous line. Left: Rietveld fit with only structural model and wavelength distribution for the
Cu-Kα1,2 doublet as described in the text. Middle column: additional consideration of a convolution
due to the detector slit. Right: additional consideration of the contribution due to the divergence slit
and the axial divergence. Note that the scale factors were adapted for the left and middle column to
yield a calculated line that approximately fits the peak maxima. Lattice parameter and background
parameters were taken from an optimum fit considering all contributions.

The next lines provide the first contribution to the IRF: the emission profile
from the Cu anode in the X-ray tube as modified by the diffracted beam mono-
chromator. It describes the Cu-Kα1,2 doublet using the contents of the
102 4 Peak shapes: Instrument
microstructure

Cu-Ka4_Holzer.lam2 file provided in the lam subdirectory of the main TOPAS

directory (Hölzer et al., 1997; see Section 2.5.9). Appropriate description of the
emission profile is crucial, because a poor description of its shape cannot be
compensated by other line broadening contributions:

Lam ‘ Equivalent command: CuKa4(0.0001)

ymin_on_ymax 0.0001
la 0.579 lo 1.5405909 lh 0.4374157
la 0.080 lo 1.5410639 lh 0.643214
la 0.236 lo 1.5443990 lh 0.5128764
la 0.105 lo 1.5446855 lh 0.6872322

One might wonder why two major and two minor profile functions are used to
describe the effect of the Kα1 + Kα2 doublet. The two major contributions describe
the Kα1 and Kα2 lines at their literature positions. The two minor contributions
are slightly shifted to describe the experimentally observed asymmetry of both
lines.
The next lines define the primary and secondary radius (typically equal
for ordinary Bragg–Brentano diffractometers) in mm. These default to 173 mm
if not defined. The macro Specimen_Displacement corrects 2θ shifts caused by
the specimen being displaced from its ideal position (see Section 2.1.1). The
value is in mm and depends on primary radius Rp. Peak shape effects due to
defocusing caused by the specimen displacement aren’t considered by this
macro:

Rp 250 ‘ Primary radius (mm)

Rs 250 ‘ Secondary radius (mm)
Specimen_Displacement(@,0) ‘ Displacement (mm)

The peaks of the LaB6 phase can be generated in alternative ways. Using the keyword
hkl_Is either a Le Bail (Le Bail et al., 1988) or a Pawley (1981) (lebail 0 or omission of
the keyword) refinement can be performed:

hkl_Is
lebail 1 ‘ Omit this line for a Pawley refinement
phase_name "LaB6 line profile standard"
space_group Pm-3m
Cubic(@ 4.156)

2 An increased number of lines allows a more detailed description of the wave-length distribution.
Depending on the masking by convolution with other IRF contributions, a two-line classical descrip-
tion (e.g., CuKa2.lam) may suffice.
 4.2 Determination of the instrumental resolution function (IRF ) 103

In this way each peak is fitted with an optimum intensity (see Chapter 3). After
successful refinement, a list of hkl peaks is added automatically inside {} parentheses
below the generated command load hkl_m_d_th2 I. This list is reused in subsequent
refinements if it is left in the INP file. If the evaluated diffraction range or the lattice
parameters are changed significantly, the list of reflections and the keyword load
hkl_m_d_th2 I should be deleted to force the generation of a new list. Note that the
reflection intensities are treated as truly refined parameters for a Pawley refinement.
Alternatively, using the keyword str, one can perform a Rietveld refinement
(Rietveld, 1969):

str
phase_name "LaB6 line profile standard"
space_group Pm-3m
scale @ 0.00220081863
Cubic(@ 4.156579)
site La num_posns 1 x 0 y 0 z 0 occ La 1 beq bb 0.3
site B num_posns 6 x @ 0.2 y =1/2; z =1/2; occ B 1 beq = bb;

Note that the atomic displacement parameters of the La and B atoms have been
coupled together, and are thus forced to assume equal values.
In this example fits of almost identical quality can be achieved with the Le Bail,
Pawley or Rietveld methods. Generally, it is expected that the Le Bail or Pawley
approach will give better fits than the Rietveld approach, because they can compen-
sate for inappropriately modeled atomic structure or texture. Problems only occur in
cases of severe peak overlap, where improper partitioning between background and
peak intensity can be expected, and can lead to inappropriate peak profile para-
meters. This is unlikely to be an issue for the simple materials typically used for
measurements for the IRF.
Figure 4.2 (left) shows the Rietveld fit for three peaks using just the wavelength
distribution. We see that the width of the wavelength distribution increases on the 2θ
scale with increasing diffraction angle. In fact, the width increases with tanθ0, with
θ0 as the (half) diffraction angle of the peak.
The next contribution we consider is the receiving slit size of 0.52 mm. This can
be included with the macro:

Slit_Width( 0.52)

prior to the hkl_Is or str line. This macro convolutes each peak (at this stage only the
wavelength distribution) with a box (hat in TOPAS) function of 2θ0 -independent
width of 180°/π × (slit width)/(secondary diffractometer radius). Figure 4.2 (middle)
shows the effect of this additional convolution. Due to the 2θ0 independence of this
contribution and the increase of the width of the wave length contribution with
104 4 Peak shapes: Instrument
microstructure

2θ0 (Figure 4.2, left), the relative importance of the slit width decreases with
increasing 2θ0 .
The effects due to equatorial and axial divergence can be considered by:

Divergence( 0.46)
axial_conv
filament_length 10
sample_length 15
receiving_slit_length 12
primary_soller_angle @ 2.251032597 ‘ Only parameter refined
axial_n_beta 30

The macro Divergence applies an appropriate convolution for equatorial diver-

gence of 0.46° as defined by the primary divergence slit. The lines following
axial_conv indicate the filament length, sample length (actually the width of the
sample perpendicular to the beam direction), the length of the receiving slit (all
in mm) and the acceptance angle of the Soller slit (in °) in the primary beam. The
number after keyword axial_n_beta controls the numerical accuracy with which
the convolutions belonging to the axial_conv keyword are performed. Figure 4.2
(right) shows the improvement of the profile description achieved with these
commands, especially with respect to description of the asymmetry of reflections
at low 2θ0.
Note that in many cases the divergence-related asymmetry effects can instead be
described in an empirical way by refining the single argument of the macro
Simple_Axial_Model:

Simple_Axial_Model(@, 14.81471)

Two further improvements are possible, the effect of these is demonstrated in Figure
4.3. First, the effects of so-called tube tails are refined using the numbers in the macro
Tube_Tails (Bergmann et al., 2000; see Section 2.5.9). This considers diffraction by X-
rays not coming from the actual focus of the tube:

Tube_Tails(@, 0.04493,@, -0.98542,@, 1.24953,@, 0.00152)

Moreover, a tiny Cu-Kβ contribution to the diffraction line can be discerned. Its
intensity depends on the characteristics of the graphite monochromator and can be
refined together with its width and exact wavelength:

la @ 0.00198654126 lo @ 1.392030917 lh @ 0.4525269504

 4.2 Determination of the instrumental resolution function (IRF ) 105

200 200 200

As from Figure 110 Consideration 110 Consideration 110
4.2 of tube tails of tube tails
and Kβ

150 150 150

Sqrt (intensity/counts)

100 100 100

50 50 50

0 0 0

26 28 30 32 26 28 30 32 26 28 30 32
2θ/°

Figure 4.3: 110 peak of the LaB6 standard measurement considered in Section 4.2.1. Note the square
root (Sqrt) intensity scale in contrast to Figure 4.2. Left: As evaluated in Figure 4.2. Middle: Additional
consideration of tube tails. Right: Additional consideration of a Cu-Kβ contribution to the profile. Note
that the final fit still misses some features in the tails of the peak.

40000
110
Intensity/counts

30000

20000

10000 431/
321 510

20 30 40 50 60 70 80 90 100 110 120 130 140 150

2θ/°

Figure 4.4: Final fit to the LaB6 standard measurement considered in Section 4.2.1. The reflections
shown in Figures 4.2 and 4.3 are labeled.

This line is added as an additional line along with four other la lines of the original
wavelength distribution. The final fit to the whole diffraction pattern is shown in
Figure 4.4.
106 4 Peak shapes: Instrument
microstructure

Note that in this FP approach, only the acceptance angle of the primary beam
Soller slit, the parameters pertaining to the tiny Cu-Kβ line and parameters of the
tube tails have been refined. All other parameters were based on the actual geo-
metry of the instrument. This approach works well for a well-adjusted Bragg–
Brentano diffractometer. Unnecessary refinement of well-established instrument
parameters can lead to severe correlations during least squares refinement and to
unphysical refined values. This should be avoided as unphysical values can some-
times cause:
– improper extrapolation of the IRF outside the 2θ0 range available from the LaB6
measurements. Such extrapolation is sometimes unavoidable in case of low 2θ0
values, that is, at lower angles than the LaB6-100-reflection.
– long computation times and other artefacts.

Once one has arrived at a proper description of the IRF, it can be used for Rietveld (or
Le Bail or Pawley) fitting of diffraction data from substances that show line broad-
ening with respect to the IRF. This requires that their diffraction data have been
measured under conditions leading to the same IRF. If this is the case, the keywords
and macros determining the IRF should be copied with fixed parameters into the INP
file to be used for evaluating the broadened data. This can conveniently be done by
replacing the description of the LaB6 phase by descriptions of one or several new
phases in the INP file.

4.2.2 Bragg–Brentano geometry (Cu-Kα1,2 doublet): extended TCHZ approach

As a simple alternative to the treatment in Section 4.2.1, one can describe the IRF on
the basis of a pseudo-Voigt function (Thompson et al., 1987) along with a simplified
description of the beam-divergence-induced peak asymmetry:

‘ xdd level
Simple_Axial_Model(@, 12.02815)
‘ str or hkl_Is level
TCHZ_Peak_Type(!pku,0,!pkv, 0,pkw, 0.00624,!pkz, 0,!pkx, 0.02280,pky, 0.01927)

The TCHZ macro has to be entered at the str or hkl_Is level, where the LaB6
phase is described. The macro TCHZ_Peak_Type has 12 entries separated by
commas storing the six numerical values of parameters named pku, pkv, pkw,
pkz, pkx and pky (traditionally named U, V, W, Z, X and Y, see Section 2.5.5). The
first four values (pku, pkv, pkw, pkz) describe the 2θ0 dependence of the
Gaussian contribution. The effect of pkz can be described in terms of a combina-
tion of pku and pkw. As a result, pkz should not be refined at the same time as
pku and pkw without additional constraints (for details see TOPAS technical
 4.2 Determination of the instrumental resolution function (IRF ) 107

reference manual). pkx and pky describe the 2θ0 dependence of the Lorentzian
contribution. The 2θ0 -dependent shape of the pseudo Voigt function is then
determined as described by eq. (2.66).
The wavelength definition used in the FP approach (Section 4.2.1) should not be
used in combination with the TCHZ function (or only if the wavelength distribution
dominates the IRF). Instead a simplified emission spectrum should be used:

lam ‘ Equivalent to the line CuKa2(0.0001) plus a Kbeta component

ymin_on_ymax 0.0001
la 2 lo 1.5405909 lh 1
la 1 lo 1.544399 lh 1
la @ 0.00602 lo 1.392030917 lh 1

This approach is better as TOPAS interprets the lam commands differently when FP
is not used. In this case the U–Y (pku–pky) parameters are used to describe the
width of the strongest profile contribution and the lh values are factors describing
the relative width of each contribution. This works as non-equal widths of Kα1/2 and
Kβ components are usually masked by other broadening effects. Use of an emission
profile as in the FP approach (Section 4.2.1) can give unrealistically different widths
for the Kα1 and Kα2 components leading to a poor profile description. This becomes
irrelevant if only a single wavelength component is present (e.g., when using a
Johansson monochromator in the laboratory or monochromatic synchrotron
radiation).

4.2.3 Bragg–Brentano geometry with Cu-Kα1 radiation

In our third example of the determination of an IRF, an extended TCHZ

approach has been employed on LaB6 diffraction data recorded on a Bruker
diffractometer equipped with a Johansson monochromator in the primary beam
and a 3.5° position sensitive silicon strip detector. The complete TOPAS INP file
reads:

xdd LaB6-Standard-020712-2.raw
r_wp 3.38593058
chi2_convergence_criteria 0.000001
bkg @ 1714.9 -182.8 67.0 -151.7 76.9 -121.5 57.7 -80.4 51.7 -46. 9 32.4 -16.7
start_X 18
finish_X 118
LP_Factor( 27.3)
Specimen_Displacement(@, -0.04561)
Simple_Axial_Model(@, 7.73601)
lam
108 4 Peak shapes: Instrument
microstructure

ymin_on_ymax 1e-003
la 1 lo 1.540596 lh 1
hkl_Is
phase_name "LaB6"
TCHZ_Peak_Type(@, 0.00233,@, -0.00403,@, 0.00354,, 0,@, 0.05742,, 0)
r_bragg 0.180027153
space_group "Pm-3m"

In the lam section, the la value (set to 1) acts as an overall scale factor for
intensities. As a TCHZ_Peak_Type is used together with a single wavelength, the
value of the line width lh and/or lg has no effect on the peak shape and width.
The macro Simple_Axial_Model describes some minor divergence-related peak
asymmetry. The final Pawley fit is shown in Figure 4.5. Note that a Rietveld fit
to the same data leads to a somewhat improper fit of the angle-dependent
integrated intensities, likely caused by nonnegligible effects of absorption (see
Section 2.3.3).

140000 110
110
120000
Intensity/counts

100000
80000
60000
40000
20000
0

10 20 30 40 50 60 70 80 90 100 110 30 31
2θ/°

Figure 4.5: Final fit to the LaB6 standard measurement considered in Section 4.2.3 (Cu-Kα1 radiation).
Right: zoomed 110 reflection.

Use of the FP approach on data from a diffractometer with primary beam

monochromator and/or position-sensitive detectors may require more individual
treatment. The wavelength distribution of the Kα1 component may be narrower
and less Lorentzian shaped than is observed without a monochromator (Cline
et al., 2015). This can be considered by an additional Gaussian component of the
wave-length line with a width given by the parameter lg (not to be used with the
TCHZ approach):

lam
ymin_on_ymax 1e-003
la 1 lo 1.540596 lh 0.3 lg 0.2
 4.3 Treatment of sample-dependent line broadening 109

To a first approximation, the effect of the equatorial channel width of a position-

sensitive detector may be considered as a convolution with a box (hat) function with
2θ-independent width3:

Rp 217.5
Rs 217.5
prm c 0.1 min 0.000001 max 1
hat = Rad c / Rs;

where c is the width of a channel in mm. This effect corresponds to that of a receiving
slit in classical Bragg–Brentano geometry (see Section 4.2.1, Cline et al., 2015).
Note that misalignment of diffractometer components, especially of primary
beam monochromators, can lead to line-profile effects that are difficult to model.
This again illustrates the importance of a proper assessment of the instrumental
profile.

4.3 Treatment of sample-dependent line broadening

Like in Section 4.1 we first focus on the effects on the x = 2θ scale, whereas the x = TOF
scale will be considered within Section 4.4. Line broadening originating from the
sample can arise from multiple origins of differing complexity (see Figure 4.6). As
compared to the IRF (which can be regarded as a minimum observable line broad-
ening in view of eq. (4.1)), the effects can be small or large; that is, the measured
reflections can be slightly or considerably broader than a peak described by the IRF.
Parameters describing sample line broadening are often strongly correlated, and it is,
therefore, important to use a simple and appropriate model that is matched to the
data quality available.
We will restrict ourselves to size and microstrain as causes for structural line
broadening. These effects lead to modifications in the diffraction patterns that can be
described by broadening of the individual peaks, so that one can rewrite MS in eq.
(4.1) as:

MS = ðSize
StrainÞ, (4:5)

where the contributions from size and microstrain are assumed to be independent.
The situation is somewhat more complex in the case of irregular stacking of layers,
which is discussed separately in Chapter 10. Moreover, although microstrain broad-
ening can be significantly asymmetric (see Figure 4.6), only symmetric microstrain
broadening will be considered here. Both contributions can be either isotropic or

3 The macro Rad is a predefined constant equal to 57.2957795130823, that is, 180/Pi, where the
constant Pi is also predefined.
110 4 Peak shapes: Instrument
microstructure

anisotropic. Isotropy of a line broadening contribution means that shape and width
of the contribution varies solely as a function of the diffraction angle 2θ0 of the hkl
peak. Anisotropy means an additional dependence of width and/or shape on hkl.
Isotropic and anisotropic line widths are illustrated as a function of 2θ0 in Figure 4.6.
Figure 4.7 anticipates details from the next sections by highlighting the different 2θ0
dependences of size and microstrain broadening due to the dependence of their
widths (in angle-dispersive diffraction) on 1=cos θ0 and tan θ0 respectively. Roughly

Asymmetrically
Strong anisotropic broadened
line broadening peak
Width parameter

Strong isotropic
line broadening Symmetrically
broadened
peak

Weak isotropic
line broadening Symmetric peak
due to IRF
IRF

2θ0 2θ0 2θ

Figure 4.6: Left: Typical evolution of reflection widths with increasing diffraction angle for the
instrumental resolution (IRF), and for materials showing weakly or strongly isotropically broadened
peaks, as well as anisotropic line broadening, where the total widths cannot be described as a
continuous function in 2θ. Right: symmetric and asymmetric line broadening relative to the IRF.

3
Function value

Size: 1/cosθ0
1
Microstrain: tanθ0

0 20 40 60 80 100 120 140

2θ0/°

Figure 4.7: 2θ dependence of the functions 1/cos θ0 and tanθ0 illustrating the relative importance of
size and microstrain broadening at low and high angles.
 4.3 Treatment of sample-dependent line broadening 111

speaking, the size broadening is important at all angles, whereas microstrain broad-
ening becomes much more important at high diffraction angles.
The commands/macros described in the next sections are applied at the str or
hkl_Is phase level, whereas for multiphase mixtures the IRF is phase independent
and described at the global level. Nevertheless, specific macros like the
TCHZ_Peak_Type for the description of the IRF need to be defined (identically) for
each phase or via a construct such as for strs.
It should also be noted that from Version 5 on the so-called Whole Powder
Pattern Modeling (WPPM; Scardi & Leoni, 2002) approach can be applied within
TOPAS. This in particular opens the possibility to describe different MS contributions
on the x scale (2θ or also s) in terms of the Fourier transforms. This is useful because
there are many situations (e.g., explicit consideration of size distributions) where
analytical expressions for the line broadening contribution only exist on the Fourier
scale. Macros for the WPPM approach are available on http://topas.dur.ac.uk/topas
wiki/doku.php?id=wppm_macros&s[]=wppm. The WPPM approach will not be dealt
with in the present chapter.

4.3.1 Size broadening

As discussed in Section 1.2, size broadening is caused by a finite size of the crystallites
(or more precisely, of the coherently diffracting domains) contributing to the diffrac-
tion pattern. Typically these crystallites are not all identical but vary in size and
shape. This is frequently considered in terms of a crystallite size distribution, which
can be characterized by different parameters, the most important of these are average
size values.
On the 1/d scale the fwhm of the size broadening for a reflection hkl is propor-
tional to the reciprocal of the crystallite sizes. For convenience, we introduce Dhkl as
some average of the crystallite-size distribution perpendicular to the hkl lattice
planes:

Kfwhm
fwhmhkl
1=d = , (4:6)
Dhkl
where Kfwhm is a shape constant of the order of 1 (and set to 1 from hereon).
Transforming this expression to the 2θ scale via eq. (4.4), gives the famous
Scherrer equation derived in Section 1.2:

λ
fwhmhkl
2θ = , (4:7)
Dhkl cosθ0
where fwhmhkl
2θ is the fwhm (in radians) of the size broadening contribution to the
overall peak shape of the reflection hkl at the position 2θ0. Although use of the
Scherrer equation is frequently criticized, it is basically correct, and the criticism
112 4 Peak shapes: Instrument
microstructure

should more concentrate on the way in which the analysis is done.4 Practical
application of eq. (4.7) requires some description of the hkl-dependence of Dhkl and
of the hkl-dependence of the shape of the peaks. The peak shape depends on the
shape of the crystals and on the crystallite size distribution and is taken as hkl-
independent in most Rietveld applications. For approaches to consider more complex
shapes due to realistic size distributions, which also take into account specific
crystallite shapes, see, for example, David et al. (2010).
If the size is isotropic, the same size parameter value D0 applies for each reflec-
tion hkl:

Dhkl = D0. (4:8)

Adopting a Gaussian peak shape, size broadening can be realized by the following
commands (see Section 2.5):

prm D0 50
gauss_fwhm = 0.1 Rad Lam / (Cos(Th) D0);

with a starting value of D0 = 50 (in nm). This is equivalent to the built-in macro:

CS_G(@, 50)

For a Lorentzian shape, the equivalent commands would read:

prm D0 50
lor_fwhm = 0.1 Rad Lam / (Cos(Th) D0);

or:

CS_L(@, 50)

In the scripts above, the command gauss_fwhm (lor_fwhm) convolutes each peak of
the corresponding phase with a Gaussian (Lorentzian) with a fwhm as given in the
argument. The reserved parameter name Lam is the wavelength (see Section 4.2.1)
and the factor 0.1 converts from Å into the usual unit for D0 of nm.
In reality there is no crystallite shape and size distribution, which can lead to an
exact Gaussian size broadening. Typically, size broadening can be reasonably well

4 The most frequent flaws are not considering the IRF upon determination of fwhmhkl
2θ , not realizing
that the fwhm in eq. (4.7) is given in radians, the use of a single reflection and an inadequate
interpretation of the determined value Dhkl.
 4.3 Treatment of sample-dependent line broadening 113

approximated by an intermediate Gaussian–Lorentzian function, although broad

size distributions can also lead to super-Lorentzian profiles (which we won’t consider
further). This description for size broadening can be introduced by simply refining
the two parameters D0G and D0L simultaneously:

prm D0G 100

prm D0L 100
gauss_fwhm = 0.1 Rad Lam / (Cos(Th) D0L);
lor_fwhm = 0.1 Rad Lam / (Cos(Th) D0G);

The same is achieved by a combination of the predefined macros CS_G and CS_L:

CS_G(@, 100)
CS_L(@, 100)

or CS_G(D0G, 100) and CS_L(D0L, 100) if the refined parameters need to be acces-
sible for further calculations.
Getting a good fit after refining different values for D0G and D0L raises the
question of how to interpret the two values. It is clear that there is not a separate
Gaussian or Lorentzian distribution of domain sizes. Instead the ensemble exhibits
overall size broadening given by the shape and the total width of the function
resulting from the convolution of the Gaussian with the Lorentzian (i.e., a Voigt
function, which is, usually approximated in TOPAS by a pseudo-Voigt).5 A well-
defined average measure for the domain size, and thus a physically meaningful
parameter, is the volume average column height (length), called Dvol here. That
value corresponds to the integral breadth (see eq. (13.97) and following) of the overall
size broadening on the 1/d scale, which is reflection-order independent:
1
β1=d = . (4:9)
Dvol
Taking into account eq. (4.4) and the relations between the width parameters integral
breadth and fwhm, the value of Dvol can be calculated and reported by6:

CS_G(D0G, 50)
CS_L(D0L, 50)
prm Dvol = 1/Voigt_Integral_Breadth_GL (1/D0G, 1/D0L);:22.44

5 If a more accurate approximation to the Voigt function than the pseudo-Voigt is required, it can be
achieved with the TOPAS keyword more_accurate_Voigt.
6 Omitting “:22.44” would mean that the value is calculated by TOPAS but not reported. Report is
achieved by adding the “:” and a dummy number, which is replaced by the final value after
refinement (and the standard deviation, if do_errors is active).
114 4 Peak shapes: Instrument
microstructure

hkl-dependent size broadening due to anisotropically sized domains can be mod-

eled by having a hkl-dependent size value Dhkl instead of D0. Dhkl depends only
on the direction, but not on the reflection order n. Hence Dhkl = Dhʹkʹlʹ with hʹ = nh,
kʹ = nk, lʹ = nl. One way of ensuring such properties for Dhkl is to multiply an
average size parameter D0 with a spherical harmonics function using the spher-
ical_harmonics_hkl keyword (see Section 13.19)7:

prm D0 50
spherical_harmonics_hkl Ahkl
sh_order 4
prm Dhkl = Max(D0 Ahkl, 1);

TOPAS extends the sh_order 4 keyword by a symmetry adapted series of parameters

that are the coefficients of a spherical harmonics expansion describing the hkl-
dependence of the factor Ahkl. In the case of m
3m Laue symmetry this reads:

prm D0 50
spherical_harmonics_hkl Ahkl
sh_order 4 load sh_Cij_prm {
k00 !Ahkl_c00 1.00000
k41 Ahkl_c41 0.00000
}
prm Dhkl = Max(D0 Ahkl, 1);

Since Ahkl can be negative in some directions in space (i.e., for certain hkl), a positive
lower limit is imposed through the expression Max(D0 Ahkl, 1). The values of Dhkl
then can be used for size broadening with:

prm !zeta 1 min 0 max 1

prm DhklG = Dhkl/(1-zeta);
prm DhklL = Dhkl/zeta;
CS_G(DhklG)
CS_L(DhklL)

By using (1-zeta) and zeta as weighting factors it is possible to associate an hkl-

independent shape with Dhkl, which is Gaussian for zeta = 0 and Lorentzian for
zeta = 1. Note that by dividing the Dhkl by zeta or (1-zeta) the fwhm values of the
Gaussian and Lorentzian broadening executed by the CS_G and CS_L macros
become proportional to the weighing factors divided by Dhkl (see above and
eq. (4.7)). The parameter zeta resembles the mixing parameter η (or lor as

7 Laue symmetry is automatically considered in the spherical harmonics expansions although, in

general, the crystallite size does not need to be symmetry invariant.
 4.3 Treatment of sample-dependent line broadening 115

frequently used in TOPAS) of a pseudo-Voigt (see Section 13.17.4) but is not

identical to it, except for the fact that the extreme values of zeta = 0 and 1 lead
to a Gaussian and Lorentzian respectively.8 A similar approach will be used in
Section 4.3.2 for microstrain broadening.
hkl-dependent values for the volume average column height (length) can be
calculated as follows:

prm Dvolhkl = Dhkl/Voigt_Integral_Breadth_GL ((1-zeta), zeta);

Since Dvolhkl is different in different directions the easiest way to visualize its value is
to plot it as a surface, where the direction-dependent distance from the center of the
plot indicates the direction-dependent value of Dvolhkl:

normals_plot = Dvolhkl;
normals_plot_min_d 0.3

To generate such a plot (see, e.g., Figure 4.8), TOPAS performs a direction-
dependent interpolation between Dvolhkl values calculated for all reflections
hkl considered in the refinement. The command normal_plot_min_d 0.3 restricts
the number of reflections by considering only hkl with a minimum d spacing
of (in this case) 0.3 Å. Although the normals_plot keyword provides easy
access to plots, one should be aware that the Dvolhkl values cannot be read
directly from the plot. It is, however, possible to extract values with a con-
struct such as:

Figure 4.8: Example of a surface depicting the direction

b dependence of the coherently diffracting domain size
(e.g., of the volume average column height) drawn
a
using the normals_plot keyword.

8 zeta corresponds to ζ in Section 13.17.5 since, due to the 1/size dependence of the fwhm, it acts as
weighting parameter for the Gaussian and Lorentzian convolutions conducted by the macros CS_G
and CS_L.
116 4 Peak shapes: Instrument
microstructure

phase_out Dvolhkl_values.out load out_record out_fmt out_eqn {

"%3.0f" = H; " %3.0f" = K; " %3.0f" = L; " %11.4f\n" = Dvolhkl;}
}

This will write the Dvolhkl values for different hkl into a file called Dvolhkl_values.out.
Generally, direction-dependent domain size values should be interpreted with
care and shapes like in Figure 4.8 should not be equated with true crystallite shapes.
Use of microscopic methods is generally recommended to supplement line-broad-
ening analysis data. For interpretation of such data in terms of actual crystallite
shapes, see more advanced literature (e.g., Scardi & Leoni, 2001).

4.3.2 Microstrain broadening

Microstrain broadening is caused by local variations of d-spacings in the crystallites

of a diffracting specimen. A specific deviation from some reference d-spacing dhkl 0
(e.g., the average) is typically referred to as strain εhkl (see Figure 4.9):
dhkl − dhkl
εhkl = 0
. (4:10)
dhkl
0

With respect to the reflection position 2θhkl hkl

0 expected for the d spacing d0 , a peak due
to a spacing dhkl associated with a strain ε will shift by (with angles in radians!):
hkl

2θhkl − 2θhkl hkl

0 = − 2ε tanθ0
hkl
(4:11)

Here, eq. (4.3) has been used to describe the effect of a change in d spacing on the 2θ
scale.
The term microstrain means a coherent or incoherent superposition of the
diffraction from the regions in the specimen with different strain, thus, it is micro-
strain broadening that appears as Strain in eq. (4.5). The strain underlying micro-
strain broadening can be of elastic nature, that is, due to microstress/stress of second
kind as well as from the strain fields of dislocations or of nonelastic nature due to
local composition or concentration variations (see Leineweber, 2011).
In any real sample the strain will exhibit some kind of probability density
function on the εhkl scale. This can be mapped onto the 2θ scale (in radians) via eq.
(4.4) (Leineweber & Mittemeijer, 2010):

fwhmhkl hkl
2θ = 2 fwhmε tanθ0 (4:12)

The quantity fwhmhklε is an unusual measure for microstrain (in contrast to the square
 1=2
root of the variance of the microstrain σε = var εhkl , or ~ehkl corresponding to half of
the integral breadth of the microstrain distribution [Delhez et al., 1988]), but fwhmhkl
ε
can easily be related with the fwhm of the associated line broadening without taking
the actual shape of the peak into account.
 4.3 Treatment of sample-dependent line broadening 117

Crystallite Crystallite

d hkl d 0hkl

Strain
εhkl 0 εhkl

d hkl d0
hkl d spacing

2θ hkl 2θ0hkl 2θ

strain distribution
Crystallite

Microstrain or
(+ strain)
Crystallite
Crystallite

⟨2θhkl⟩ 0 2θ

Figure 4.9: Illustration of the effects of strain and microstrain on diffraction pattern for a given
reflection hkl. Top: Effect of strain, that is, a constant change of the d-spacing throughout (all
crystallites of) the sample in the sense of eq. (4.10), shown on the strain (ε), d spacing (d) and
diffraction angle (2θ) scales with the corresponding reflection indicated as a vertical line. On the 2θ
scale a shift with respect to the reference value 2θhkl0 occurs in accordance with eq. (4.11). Bottom:
Microstrain corresponds to inhomogeneous strain (e.g., different in different crystallites or varying
within the crystallites) leading to microstrain broadening. As an example three sets of lattice
planes with slightly different strain contributing to the overall profile (three vertical lines) are
shown, whereby here an average strain is superposed by microstrain. In the current chapter the
average d-spacing is taken as identical with the reference d-spacing. Moreover, only symmetric
microstrain broadening is considered.

Microstrain broadening can be modeled in TOPAS in various ways. Similar to

isotropic crystallite size broadening, isotropic microstrain broadening with a
Gaussian line shape can be introduced by:

prm E0 0.001
gauss_fwhm = 360 / Pi Tan(Th) E0;

which is alternatively achieved by the macro Strain_G:

prm E0 0.00100
prm arg_E0 = 360/Pi E0;:0.11459
Strain_G( arg_E0 )

The equivalent for a Lorentzian line shape is:

prm E0 0.001
lor_fwhm = 360 / Pi Tan(Th) E0;
118 4 Peak shapes: Instrument
microstructure

or:

prm E0 0.00100
prm arg_E0 = 360/Pi E0;:0.11459
Strain_L( arg_E0 )

For all cases the parameter E0 corresponds to fwhmhkl ε . Note that arguments of the
macros Strain_G/L have been defined in TOPAS such that they lead to unrealistic
values of strain measures (in contrast to CS_G/L). That has been compensated here by
the factors 360/Pi.
Microstrain broadening of intermediate Gaussian–Lorentzian shape can be mod-
eled with:

prm E0G 0.00100

prm E0L 0.00100
prm arg_E0G = 360/Pi E0G;:0.11459
prm arg_E0L = 360/Pi E0L;:0.11459
Strain_G( arg_E0G )
Strain_L( arg_E0L )

From the values E0G and E0L one can calculate the microstrain fwhmhkl
ε , which is the
fwhm of a (approximate) Voigt function describing the distribution of ε through:

prm E0 = Voigt_FWHM_GL(E0G, E0L);:0.00164

Anisotropic microstrain broadening can be modeled in different, more or

less equivalent fashions. The most prominent approach has been introduced by
Rodriguez-Carvajal et al. (1991) and is frequently used in the forms described by
Popa (1998) or Stephens (1999). The approaches have a sound statistical basis
(at least concerning the direction dependence of the variance of the micro-
strain), that is, they are related with the statistics of the microstrain distribu-
tion. Invariance of the line width with respect to the crystal’s Laue class is
typically adopted. Different macros are therefore available in TOPAS.INC for the
different Laue classes: Stephens_triclinic (
1), Stephens_monoclinic (2/m),
Stephens_orthorhombic (mmm), Stephens_tetragonal_low (4/m), Stephens_
tetragonal_high (4/mmm), Stephens_trigonal_low (
3), Stephens_trigonal_high
(
3m1), Stephens_trigonal_high_2 (
31m), Stephens_hexagonal (6/m and 6/mmm),
Stephens_cubic (m
3 and m
3m). Note that for all trigonal Laue classes the macros
assume the hexagonal setting of the unit cell. There are no ready-made macros
for rhombohedral unit cells, but symmetry restrictions have been reported by
Leineweber (2006). The user should note that these different macros employ
 4.3 Treatment of sample-dependent line broadening 119

different numbers of arguments due to the different number of independent

parameters to be refined.
In the following this approach is illustrated using X-ray data recorded from
tetragonal Pb3O4. This polymorph typically exhibits pronounced microstrain broad-
ening of the h00 reflections, whereas hh0 reflection are relatively narrow (for some
background, see Leineweber & Dinnebier [2010]). The data were recorded with the
same instrumental settings as the LaB6 data used for evaluation of the IRF in
Section 4.2.3 (including Cu-Kα1 radiation). After replacement of the diffraction
data (xdd keyword), fixing the parameters of the TCHZ_Peak_Type and the
Simple_Axial_Model macros, introducing an appropriate model for the crystal struc-
ture and introducing the macro Stephens_tetragonal_high, the INP file looks like (the
impurity lines are not shown in the listing):

xdd Pb3O4.raw
r_wp 8.96620702
chi2_convergence_criteria 0.000001
bkg @ 655.26 -349.0 313.1 -109.4 109.0 -27. 8 35.8 20.7 -24.6 24.8
start_X 18
finish_X 118
LP_Factor( 27.3)
Specimen_Displacement(@, 0.00334)
Rp 217.5
Rs 217.5
Simple_Axial_Model( 7.73601)
lam
ymin_on_ymax 1e-003
la 1 lo 1.540596 lh 1
str
phase_name "Pb3O4"
TCHZ_Peak_Type(, 0.00233,, -0.00403,, 0.00354,, 0,, 0.05742,, 0)
Stephens_tetragonal_high(zeta, 1.0, s400, 336.5,s004, 10.4,s220, -670.0,s202, 25.5)
r_bragg 4.546
space_group P42/mbc
scale @ 0.000285947836
Tetragonal(@ 8.813227,@ 6.564940)
site Pb num_posns 4 x 0 y 0.5 z 0.25 occ Pb 1 beq @ 0.9
site Pb1 num_posns 8 x 0.14 y @ 0.16381 z 0 occ Pb 1 beq @ 1.2
site O1 num_posns 8 x o1xx 0.68135 y =o1xx-0.5; z 0.25 occ O 1 beq bbo 3.0
site O2 num_posns 8 x @ 0.08847 y @ 0.62008 z 0 occ O 1 beq bbo 3.0

For simplicity a common isotropic displacement parameter for the two O sites (bbo) is
used. The symmetry restriction for the fractional coordinates is ensured by equating
the values for the fractional coordinates x and y. After refinement of all parameters
one gets the fit shown in Figure 4.10. The only line broadening with respect to the IRF
is achieved by the macro Stephens_tetragonal_high (with the usual “@” substituted
by user-set parameter names).
120 4 Peak shapes: Instrument
microstructure

24.71056
37.83708
29.09139
120000 30.32287
100000
Intensity/counts

80000

60000

40000

20000

20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
2θ/°

Figure 4.10: Final fit to powder diffraction data of Pb3O4 recorded with Cu-Kα1 using the macro
Stephens_tetragonal_high to describe anisotropic microstrain broadening. Details of the effects of
that macro are illustrated in Figure 4.11. Vertical lines indicate reflections of minor unknown
impurities treated as a peak phase.

In Figure 4.11 we can see an improving fit on introducing increasingly sophisti-

cated line broadening models: Using only the IRF gives too narrow fitted reflections.
Isotropic microstrain (e.g., use of the Strain_L macro) broadening gives a poor fit
since the line broadening is pronouncedly anisotropic. Only use of the macro
Stephens_tetragonal_high to describe the anisotropy of the microstrain broadening
leads to reasonable agreement.
The macro Stephens_tetragonal_high applies the following lines:

prm zeta 1.00000 min 0 max 1

prm s400 336.53970 min 0
prm s004 10.39557 min 0
prm s220 -670.02572
prm s202 25.49056
prm mhkl = s400 (H^4 + K^4) + s004 L^4 + s220 H^2 K^2 + s202 (H^2 L^2 + K^2 L^2);
prm pp = D_spacing^2 * Sqrt(Max(mhkl,0)) Tan(Th) 0.0018/Pi;
gauss_fwhm = pp (1-zeta);
lor_fwhm = pp zeta;

We can see that the macro convolutes each Bragg reflection with a Voigt-like
function. The width parameter E of each hkl reflection (on the 2θ scale in °) in
the sense of eq. (13.101) is given by the parameter pp, while the shape is defined by
the parameter zeta (variable ζ in eq. (13.101)). The shape of the microstrain broad-
ening is independent of hkl, but the width pp is direction (hkl) dependent.
 4.3 Treatment of sample-dependent line broadening 121

29.09378 30.3296
160000
Only IRF
211
120000
112
310
80000
220 202
40000 002
0
160000 26 27 28 29 29.09721
30 30.32212
31 32 33 34
Intensity/counts

IRF + isotropic
120000
microstrain broadening
80000
40000
0
160000 26 27 28 2929.09136
30 30.32279
31 32 33 34
IRF + anisotropic
120000 microstrain broadening

80000
40000
0
26 27 28 29 30 31 32 33 34
2θ/°

Figure 4.11: Part of the Pb3O4 data from Figure 4.10 with hkl indices for the different reflections,
showing the improvement of fit with increasing complexity of the line broadening model: top: only IRF
(scale factor adapted). Middle: isotropic microstrain broadening. Bottom: anisotropic microstrain
broadening as in Figure 4.10. Vertical lines indicate reflections of minor unknown impurities treated
as a peak phase.

The shape corresponds, in the present case, to a Lorentzian limit (zeta = 1).
The direction-dependence of the microstrain broadening is quantified by the
values of s400, s004, s220 and s202 (called sHKL parameters) as coefficients of
a symmetry invariant (here with respect to 4/mmm symmetry) fourth-order poly-
nomial in the Laue indices hkl. Following the original ideas of the theory (e.g.,
Stephens, 1999), this fourth-order polynomial (parameter mhkl) should be ≥ 0 for
each H, K and L used to calculate the value of pp. Except for high symmetries, it is
cumbersome to develop general limits for the sHKL parameters to ensure this. We
can avoid negative values pragmatically by calculating Sqrt(Max(mhkl,0)) instead
of Sqrt(mhkl). In fact, the present experimental data imply (nearly) vanishing line
broadening for hh0 reflections. This can be seen as s220 ≈ –2 × s400. More negative
values of s220 would result in mhkl < 0 for H = K. Note that underlying physical
models of microstrain distributions/lattice parameter distributions require mhkl ≥
0 (Leineweber, 2011).
122 4 Peak shapes: Instrument
microstructure

The direction dependence of the microstrain broadening is best analyzed by the

direction dependence of some suitable microstrain measure like fwhmhkl
ε . That value
can be calculated as (compare eq. (4.12)):
 2 pffiffiffiffiffiffiffiffiffiffi
dhkl mhkl
fwhmhkl
ε = × corr (4:13)
2 × 10000
where corr is the ratio fwhm=E in accordance with eq. (13.101), which is 1 for the present
case with ζ = zeta = 1. The factor 10000 in the denominator of eq. (4.13), takes into
account the factor 0.0018 instead of 180 (degrees) used in the calculation of pp. This
factor is used to obtain easy-to-handle values of sHKL, which could otherwise be very
small and incorrectly reported in TOPAS OUT files.9 Figure 4.12 shows the direction
dependence of fwhmhkl resulting from the refined values of sHKL, which is plotted by:

prm !corr = ((1-zeta)^5 + 2.69269 (1-zeta)^4 zeta +

2.42843 (1-zeta)^3 zeta^2 +4.47163 (1-zeta)^2 zeta^3 +
0.07842 (1-zeta)zeta^4 + zeta^5)^.2;:1.00000
normals_plot = D_spacing^2 mhkl^.5/20000 corr;
normals_plot_min_d 0.3

These lines include the factor corr = fwhm/E according to eq. (13.101). Note that in the
present case the parameter zeta refines to its upper limit of zeta = 1. Hence the values
of corr = 1.

Figure 4.12: Surface representing the direction-dependence of the extent of microstrain of Pb3O4 as
considered in Section 4.3.2. The maximum extent of fwhmε occurs along the h100i directions and
amounts fwhmε = 0.007.

9 TOPAS handles small floating point numbers correctly internally but converts them to fixed-point
number with a limited number of digits in the OUT file.
 4.3 Treatment of sample-dependent line broadening 123

As also discussed for an anisotropic size-broadening model in Section 4.3.1,

this easily-implemented sHKL model has some practical shortcomings. For exam-
ple, isotropy of the microstrain broadening is not easily recognizable from the
sHKL parameters (Leineweber, 2006; Leineweber, 2011). This has led to introduc-
tion of an alternative Cartesian parametrization of the microstrain broadening.
In this model one refines the coefficients of a fourth-order polynomial in the
components x1, x2 and x3 of a unit vector parallel to the diffraction vector
described using a standard Cartesian coordinate system. Note that the definition
of the unit-vector components x1, x2 and x3 is more complex for lower symme-
tries. The basis vectors of the Cartesian of a tetragonal crystal are usually chosen
along the [100], [010] and [001] directions. A core piece of the treatment is the
polynomial:
E
 4 4 E  2 2 2 2
Eε = Z1111 x1 + x2 + Z3333 x34 + 6Z1122
E
x12 x22 + 6Z1133
E
x1 x3 + x2 x3 (4:14)

defining the width parameter E on the strain ε scale. The coefficients Zijpq
E
are the
refinable parameters (the number of parameters, again, depends on the Laue class,
Leineweber, 2006). The polynomial is used in the following way to describe the
microstrain broadening of Pb3O4:

prm zeta 1.00000 min 0 max 1

prm z_e1111_scsc 50.49880 min 0
prm z_e1122_scsc -16.75325
prm z_e3333_scsc 0.48388 min 0
prm z_e1133_scsc 0.65699
prm !sc = 10^-3;
prm x1 = D_spacing H/Lpa;
prm x2 = D_spacing K/Lpb;
prm x3 = D_spacing L/Lpc;
prm ee_e = sc^2 (z_e1111_scsc (x1^4 + x2^4) + z_e3333_scsc x3^4
+ 6 z_e1122_scsc x1^2 x2^2 + 6 z_e1133_scsc (x1^2 x3^2 + x2^2 x3^2);
prm ee_2Th = (180/Pi)^2 (-2 Tan(Th))^2 ee_e;

E
In this series of commands, the Zijpq parameters (z_e1111_scsc and others) are
2
scaled by the parameter sc to keep their numerical values to a reasonable
magnitude. Parameters, which retain their “true” values lack the “_scsc” label,
for example, the (H, K and L dependent) squared width parameter E on the 2θ
scale (in °), which is used for convolution of each peak.
Using the additional lines:

prm !corr = ((1-zeta)^5 + 2.69269 (1-zeta)^4 zeta +

2.42843 (1-zeta)^3 zeta^2 +4.47163 (1-zeta)^2 zeta^3 +
0.07842 (1-zeta)zeta^4 + zeta^5)^.2;:1.00000
prm !z_fwhm1111_scsc = z_e1111_scsc corr^2;:50.49880
124 4 Peak shapes: Instrument
microstructure

prm !z_fwhm3333_scsc = z_e3333_scsc corr^2;:0.48388

prm !z_fwhm1122_scsc = z_e1122_scsc corr^2;:-16.75325
prm !z_fwhm1133_scsc = z_e1133_scsc corr^2;:0.65699

fwhm
one can calculate new (scaled) parameters Zijpq pertaining to:
  fwhm  2 2 
ðfwhmε Þ2 = Z1111
fwhm 4 fwhm 4
x1 + x24 + Z3333 fwhm 2 2
x3 + 6Z1122 x1 x2 + 6Z1133 x1 x3 + x22 x32 (4:15)
E
Using the Zijpq parameters, isotropy of the microstrain broadening is ensured by
E
 E
 E E E
 E
 E fwhm
Z1111 = Z2222 = Z3333 = 3Z1122 = 3Z1133 = 3Z2233 . The Zijpq parameters (or Zijpq or para-
meters defining the square root of the variance of the microstrain, Leineweber, 2011)
are dimensionless and are ideally suited to quantitatively relate observed microstrain
broadening with anisotropic tensor properties. It is possible, to construct one-to-one
relations between the sHKL and zE_ijpq_scsc parameters, meaning that the treat-
ments are equivalent and should lead to exactly the same quality of fit.
The spherical_harmonics_hkl keyword can also be used to describe anisotropic
microstrain broadening. The most convenient implementation is to fix the spherical
harmonic to an average value of 1 and scale it by an overall average microstrain.10
Note that from here on the fitting parameter zeta was fixed to 1:

prm !zeta 1 min 0 max 1

prm average 13.26144
spherical_harmonics_hkl ahkl
sh_order 4
prm !sc = 10^-3;
prm ee_e_scsc = average ahkl;
prm ee_2Th = sc^2 (180/Pi)^2 (-2 Tan(Th))^2 ee_e_scsc;
gauss_fwhm = (1-zeta) Max(ee_2Th,0)^.5;
lor_fwhm = zeta Max(ee_2Th,0)^.5;

After the first cycle, the third line automatically expands to the correct format for the
Laue symmetry. Refinement of the spherical harmonic coefficients and the parameter
average leads to:

prm !zeta 1 min 0 max 1

prm average 13.26144
spherical_harmonics_hkl ahkl
sh_order 4 load sh_Cij_prm {
y00 !ahkl_c00 1.00000
y20 ahkl_c20 -1.36314

10 Alternatively one can leave out the parameter average and remove the (set by default) “!” in front
of the parameter name ahkl_c20-ahkl_c00, which then takes the role of average; the other parameters
become average × ahkl_c20-ahkl_c44p.
 4.4 Neutron diffraction data with focus on the TOF method 125

y40 ahkl_c40 0.39057

y44p ahkl_c44p 1.82325
}

Note that the products such as average × ahkl_c20-ahkl_c44p can be directly related
to the parameters of the two other models described above (Leineweber, 2011), since
the number of parameters for a given symmetry is the same in each of them. This 1:1
relation wouldn’t exist, if, for example, the Eε (the square root of ee_e) had been
expanded in a similar way.

4.4 Neutron diffraction data with focus on the TOF method

When evaluating neutron diffraction data use of the neutron_data keyword is essen-
tial. This activates use of neutron scattering lengths. It is also important to ensure
that the correct isotope (e.g., 11B) is specified in the site list. For constant wavelength
neutron diffraction data, determination of the IRF can proceed using the TCHZ
approach of Section 4.2.3 using a suitable standard material.11 Constant wavelength
neutron IRFs are usually relatively simple. The situation is more complex for TOF
neutron diffraction data and the following should be kept in mind:
– Data are usually collected in several detector banks, each containing reflections
over a limited d-spacing range and each having a different IRF. TOPAS can
analyze these patterns simultaneously so that common (e.g., structural) para-
meters can be evaluated simultaneously from all banks.
– The profile shape is very dependent on the specific TOF instrument. It is therefore
advisable to consult the instrument scientist for guidance.
– Data are traditionally provided on the TOF scale. Appropriate calibration con-
stants are needed to convert TOF to d-spacing.
– The complex instrument detector geometry and neutron wavelength distribution
means it is essential to retain experimental standard uncertainties through the
calculations. The use of three column .xye files is recommended.

In the following INP file the IRF of a detector bank of HRPD, ISIS, UK is evaluated
using data from a CeO2 standard:

xdd hrp51690_1.xye
neutron_data
start_X 25000
finish_X 125000
x_calculation_step = Yobs_dx_at(Xo);
TOF_LAM(0.001)

11 Not La10B6, which has an enormous neutron absorption cross section!

126 4 Peak shapes: Instrument
microstructure

TOF_x_axis_calibration(t0,-11.31260,!t1, 48281.14,t2,-7.19610)
scale_pks = D_spacing^4;
bkg @ 9.37575017 7.38370967 0.924665853 0.114324576 0.285302449 0.0835731213
TOF_Exponential(a1, 193.96941,a2, 38.72409, 4, t1, +)
hkl_Is
phase_name CeO2
TOF_PV(b1, 56.95525, lor_b1, 0.269485101, t1)
Cubic(aa 5.413263)
space_group "Fm-3m"

The x_calculation_step command used ensures that the non-constant stepwidth in x

(TOF) is properly treated during analysis.12 The TOF_LAM(0.001) line determines the
intensity range for which the peak profile will be calculated (like ymin_on_ymax). The
macro TOF_x_axis_calibration defines the coefficients t0, t1 and t2 in eq. (2.3) relating
the d spacing with the x (TOF) coordinate in the xye file, which is repeated here for
convenience:

TOF = t0 + t1 d + t2 d2 (4:16)

where the numerical values of t0–t2 must correspond to d in Å. t1 relates to the

instrument flight path, t0 to a (usually electronic) zero-time offset and t2 is an
empirical quadratic correction factor. The variable names zero, difc and difa are
often used for t0, t1 and t2, respectively. In the INP file above the data were used to
fine-tune t0 and t2 (which should be done with care). The scale_peaks line gives the
Lorentz factor for TOF neutron diffraction.
The profile function is described using the macro TOF_Exponential, which
applies to all phases, and the macro TOF_PV, which applies to each phase since it
describes the sample contribution. TOF_PV defines a pseudo-Voigt of d-spacing
independent shape lor_b1 = η and fwhm which is proportional to the refined para-
meter b1 and the d-spacing. t1 is used as a scaling parameter for the refined parameter
b1. TOF_Exponential convolutes each peak with a (single) right-hand (“+”) exponen-
tial of d-spacing dependent width, which is described by the two fitting parameters a1
and a2. The functional form of the d-spacing dependence is determined by the
exponent “4.” Again, the constant t1 is used by the macro to scale the fitting para-
meters. More advanced macros exist to describe the details of the peak shapes, some
of them dedicated for specific instruments and based on a FP approach. See, for
example, the TOPAS wiki (http://topas.dur.ac.uk/topaswiki/doku.php?id=time_of_
flight_tof_isis_instrument_standard_files).
Once the standard material has been used to determine the IRF, its parameters
are held fixed when evaluating data measured on a real sample under the same
conditions. To impose size or microstrain broadening, one has to convolute the line

12 Typically such data are collected or provided in steps with constant Δd/d or ΔTOF/TOF.
 4.4 Neutron diffraction data with focus on the TOF method 127

broadening on the TOF scale instead of on the 2θ scale as considered in Section 4.3.
For size broadening, eqs. (4.7) and (4.4) lead to:
 hkl 2 1
fwhmhkl
TOF = t1 d . (4:17)
Dhkl
This commonly-accepted form neglects the (usually small) t2 term in eq. (4.16).
Analogously, for microstrain one can arrive at:

fwhmhkl hkl hkl

TOF = t1 d fwhmε . (4:18)

For isotropic size broadening analogues of the macros TOF_CS_L and TOF_CS_G can
be used. For example:

TOF_CS_L(D0,50)

applies the following:

prm D0 50
lor_fwhm = t1 .1 D_spacing^2 /D0;

The factor 0.1 again ensures D0 values in nm rather than in Å.

For intermediate Gaussian–Lorentzian shape, the same recipes as applied for
constant-wavelength X-ray diffraction data in Section 4.3.1 can be used. Similar
treatments for anisotropic size broadening are possible.
Currently, no ready-made macros are available in TOPAS.INC for microstrain broad-
ening. An analogue of what is achieved by the Strain_G macro (see Section 4.3.2) is
achieved by the following commands:

prm E0 0.01
gauss_fwhm = t1 D_spacing E0;

where E0 takes the role of the fwhm of the microstrain. Like in Section 4.3.2, the
treatment of microstrain can be extended to intermediate Gaussian–Lorentzian
shapes and to anisotropic microstrain broadening.
We’ll finish the chapter with an example of describing anisotropic strain broad-
ening in the orthorhombic cementite (Fe3C) making use of the CeO2 IRF determined
above (Leineweber, 2016). The measure of the microstrain broadening used was the
traditionally employed squareroot of the variance of the microstrain varðεhkl Þ. This is
possible due to the pure Gaussian shape of the microstrain broadening (having a
finite variance). The expression for the direction-dependent microstrain for the
orthorhombic symmetry of the cementite is:
128 4 Peak shapes: Instrument
microstructure

varðεhkl Þ = Z1111 x14 + Z2222 x42 + Z3333 x34 + 6Z1122 x12 x22 + 6Z1133 x21 x23 + 6Z2233 x22 x32 (4:19)

The INP file including the IRF and t0–t2 as determined from CeO2 (and therefore
fixed) is:

xdd hrp51694_b1_TOF.xye
neutron_data
x_calculation_step = Yobs_dx_at(Xo);
scale_pks = D_spacing^4;
bkg @ 2.75 0.71 -0.19 -0.05 -0.04 -0.11 -0.09 -0.01 -0.012 -0.04
start_X 25000
finish_X 125000
TOF_LAM(0.001)
TOF_x_axis_calibration(!t0,-11.31260,!t1, 48281.14,!t2,-7.19610)
TOF_Exponential(!a1, 193.96941,!a2, 38.72409, 4, t1, +)
hkl_Is
TOF_PV(!b1, 56.95525, !lor_b1, 0.269485101, t1)
phase_name Fe3C
a @ 5.091282
b @ 6.729827
c @ 4.525684
prm z1111_scsc 0.37466 min 0
prm z2222_scsc 0.52398 min 0
prm z3333_scsc 0.40968 min 0
prm z1122_scsc 0.04499
prm z1133_scsc 0.06015
prm z2233_scsc 0.71732

prm x1 = D_spacing H/Lpa;

prm x2 = D_spacing K/Lpb;
prm x3 = D_spacing L/Lpc;
prm !sc = 10^(-3);
prm vareps = sc^2 (z1111_scsc x1^4 +z2222_scsc x2^4 + z3333_scsc x3^4 + 6 z1122_scsc
x1^2 x2^2 + 6 z1133_scsc x1^2 x3^2 + 6 z2233_scsc x2^2 x3^2);
prm vartof = (t1)^2 D_spacing^2 vareps;
gauss_fwhm = 2(2 Ln(2) Max(vartof,0))^.5;
prm sze 548.03587 max 5000
lor_fwhm = D_spacing^2 t1 0.1/sze;
space_group "Pnma"

The factor 2(2 Ln(2))^.5 ensures interconversion between the fwhm and the square-
root of the variance. The final fits to the CeO2 and Fe3C data are shown in Figure 4.13.
 1=2
Using the normals_plot command, the direction dependence of σε = var εhkl
determined from the Zijpq parameters is shown in Figure 4.14. This reflects thermal
microstress leading to anisotropic thermal microstrain via the elastic anisotropy of
Fe3C (Leineweber, 2016).
 References 129

400 311
CeO2(→IRF )

222
200
Relative intensity

0
73000 74000 75000 76000 77000 78000 79000 80000 81000 82000 83000
35
Fe3C, 10 K
30 311
25 040/
20 212 230
15 132
222
10
132
5
0

73000 74000 75000 76000 77000 78000 79000 80000 81000 82000 83000
TOF/μs

Figure 4.13: Zoomed regions of the Pawley fits to TOF neutron powder-diffraction data from CeO2 (top:
used to determine the IRF) and Fe3C (bottom: exhibiting microstrain broadening).

Figure 4.14: Direction-dependence of the microstrain as

 1=2
quantified by var εhkl in Fe3C at 10 K due to thermal
microstrain. The maximum extent of the microstrain
b
amounts to 0.011.

References
Bergmann, J., Kleeberg, R., Haase, A., Breidenstein, B. (2000): Advanced fundamental parameters
model for improved profile analysis, in Proceedings of the 5th European Conference on Residual
Stresses, Delft-Noordwijkerhout, The Netherlands, September 29–30, 1999, A.J. Böttger, R.
Delhez, and E.J. Mittemeijer, Eds. Trans Tech Publications, 347–349, pp. 303–308.
Cheary, R.W., Coelho, A.A (1992): A fundamental parameters approach to X-ray line-profile fitting.
J. Appl.Cryst. 25, 109–121.
Cline, J.P., Mendenhall, M.H., Black, D., Windover, D., Henins, A. (2015): The optics and alignment of
the divergent beam laboratory X-ray powder diffractometer and its calibration using NIST
standard reference materials. J. Res. NIST 120, 173–222.
130 4 Peak shapes: Instrument
microstructure

David, W.I.F., Leoni, M., Scardi, P. (2010): Domain size analysis in the Rietveld method. Mater. Sci.
Forum 651, 187–200.
Delhez, R., de Keijser, T., Mittemeijer, E.J., Langford, J.I. (1988): Size and strain parameters from peak
profiles: sense and nonsense. Aust. J. Phys. 41, 213–227.
Hölzer, G., Fritsch, M., Deutsch, M., Härtwig, J., Förster, E. (1997): Kα1,2 and Kα1,2 X-ray emission lines
of the 3d transition metals. Phys. Rev. A 56, 4554–4568.
Klug, H.P, Alexander, L.E. (1974): X-ray diffraction procedures for polycrystalline and amorphous
materials, John Wiley and Sons, New York, USA, 992 pages.
Le Bail, A., Duroy, H., Fourquet, J.L. (1988): Ab-initio structure determination of LiSbWO6 by X-ray
powder diffraction,. Mat Res. Bull., 23, 447–452.
Leineweber, A. (2006): Anisotropic diffraction-line broadening due to microstrain distribution:
parametrization opportunities. J. Appl. Cryst. 39, 509–518.
Leineweber, A. (2011): Understanding anisotropic microstrain broadening in Rietveld refinement.
Z. Kristallogr. 226, 905–923.
Leineweber, A. (2016): Thermal expansion anisotropy as source for microstrain broadening of poly-
crystalline cementite, Fe3C. J. Appl. Cryst. 49, 1632–1644.
Leineweber, A., Dinnebier, R.E. (2010): Anisotropic microstrain broadening of minium, Pb3O4, in a
high-pressure cell: interpretation of line-width parameters in terms of stress variations. J. Appl.
Cryst. 43, 17–26.
Leineweber, A., Mittemeijer, E.J. (2010): Notes on the order-of-reflection dependence of microstrain
broadening. J. Appl. Cryst. 43, 981–989.
Pawley, G. S. (1981): Unit-cell refinement from powder diffraction scans. J. Appl. Cryst. 14, 357–361.
Popa, N.C. (1998): The (hkl) dependence of diffraction-line broadening caused by strain and size for
all laue groups in Rietveld refinement. J. Appl. Phys. 31, 176–180.
Rodriguez-Carvajal, J., Fernandez-Diaz, M.T., Martinez, J.L. (1991): Neutron diffraction study on
structural and magnetic properties of La2NiO4. J. Phys.: Condens. Matter 3 (1991) 3215–3234.
Rietveld, H.M. (1969): A profile refinement method for nuclear and magnetic structures. J. Appl. Cryst.
2, 65–71.
Scardi, P., Leoni, M. (2001): Diffraction line profiles from polydisperse crystalline systems. Acta Cryst.
A 57, 604–613.
Scardi, P., Leoni, M. (2002): Whole powder pattern modelling. Acta Cryst. A 58, 190–200.
Stephens, P.W. (1999): Phenomenological model of anisotropic peak broadening in powder
diffraction. J. Appl. Cryst. 32, 281–289.
Thompson, P., Cox, D.E., Hastings, J.B. (1987): Rietveld refinement of Debye-Scherrer synchrotron
X-ray data from Al2O3, J. Appl. Cryst. 20, 79–83.
Young, R.A. (1993): Introduction to the Rietveld method. In: The Rietveld method. In: edited by R.A.
Young, IUCr Book Series, Oxford University Press, UK, 1–39.
5 Quantitative phase analysis

5.1 Introduction
Quantitative phase analysis (QPA) is a method to determine the absolute weight
fractions of all crystalline phases present in a multiphase sample. Powder diffraction
as a phase-sensitive method is highly suited for this task and various procedures
using the intensities of single peaks, group of peaks or the entire powder pattern have
been developed over time. A significant breakthrough in QPA came when the suit-
ability of the Rietveld method as a precise phase-sensitive analytical technique for
QPA was realized. This is the focus of this chapter.
If all phases in the mixture are crystalline with known crystal structure, no
external or internal standard material is required, making Rietveld QPA an easy-
to-use standard-less method. Complications arise if an amorphous phase or a phase
with an unknown or only partially known crystal structure is present. In this case, an
external (or better) internal, well-defined standard can be used to renormalize the
weight fractions on an absolute basis. Once the amount of the amorphous or unknown
crystalline phase is quantified, this phase can be stored as a new standard (so-called
PONKCS) phase for further use. External and PONKCS standards depend on the
experimental conditions, which should be kept constant for a series of analyses. QPA
becomes more and more unreliable the more unknown phases are present. In all cases,
the intensity-affecting factors like absorption, overspill or preferred orientation should
be kept to a minimum.
In the following sections, the theory of QPA and its application either standard-
less or using internal, external or PONKCS standards is described. The integrated
intensity I of a reflection s = (hkl) for phase α in a multi-phase mixture measured on a
flat plate sample (Bragg–Brentano geometry) with “infinite thickness” is:
" #
I 0 λ 3 e4
Is, α =
32πR m2e c4
"     !#  
Ms 2 2
2 1 + cos 2θcos 2θm sin θ 2 Wα

j F s, α j exp − 2B α
(5:1)
2Vα2 sin2 θcosθ λ ρα μÛm

with instrument (blue), sample (green) and phase (red) dependent parameters:
incident beam intensity I0 , the distance R between specimen and detector, the X-ray
wavelength λ, the classical electron radius e2 =me c2 , the multiplicity Ms of reflection s
of phase α, the weight fraction Wα of phase α, the mass absorption coefficient μÛm of the
entire sample, the volume Vα of the unit cell of phase α in Å3, the structure factor Fs, α of
reflection s of phase α, the diffraction angle 2θ of reflection s of phase α, the diffraction
angle 2θm of the monochromator, the density ρα of phase α, and the overall atomic
displacement parameter Bα of phase α. Similar relations exist for other geometries.

https://doi.org/10.1515/9783110461381-005
132 5 Quantitative phase analysis

The relation can be greatly simplified by introducing two constants, one for the
experimental setup Cinstr , and a second one for the intensity of peak s for phase α Cs, α :
 
1 Wα
Is, α = Cinstr
Cs, α
2
. (5:2)
Vα ρα μÛm

In Rietveld analysis, all reflection intensities of a phase α are proportional to the

corresponding scale factor S:

Is, α ∝ Sα . (5:3)

This allows the introduction of a single scaling factor K that depends exclusively on
the instrumental condition and not on the sample/phases:
 
1 Wα
Sα = K
2
. (5:4)
Vα ρα μÛm

The phase density (g cm−3) can be calculated as:

Zα Mα
ρα = 1.6604
(5:5)
Vα
with the molar mass M (g mol−1) of one formula unit and Z the number of formula
units within a unit cell of phase α. The scale factor of phase α can be thus be
rewritten as:
 
1 Wα
Sα = K

(5:6)
Vα 1.6604
Zα Mα μÛm

5.2 External standard method

The external standard method was described by O’Connor and Raven (1988). From
eq. (5.6), the weight fraction of phase α can be calculated as:

Sα ðZMV Þα μ*m
Wα = (5:7)
K′
ðZMV Þα is an abbreviation of the product of Zα Mα Vα . K′ is an experimental constant
that is specific for the experimental setup used and depends only on instrumental
and data collection conditions. K′ can be determined from a single measurement of
an external standard, for which μÛm must be known:

Sα ðZMV Þα μ*m
K′ = (5:8)
Wα
With the knowledge of K′ and μÛm , the correct weight fractions for all phases of a
mixture for which the product ðZMV Þα and the Sα scale factor exists can be calculated.
 5.3 Rietveld method 133

The mass attenuation coefficient (MAC) µ of a sample (see Section 2.3.3) is

usually calculated from standard data in units of length−1. The tabulated values of the
MAC are usually those divided by the density of the material (Hubbell & Seltzer,
2004):
μ
μÛ = . (5:9)
ρ

It should be noted that the sample must fulfill the “infinite thickness” criterion
(Section 2.3.3) that is on the order of 0.1 mm for quartz and Cu-Kα radiation. The
(average) mass absorption coefficient μÛm of the entire sample (MAC) is defined as the
sum of linear attenuation coefficients of the elements multiplied by their weight
fractions. This sum runs over all phases i in the sample:
X
μÛm = wi μÛi . (5:10)
i

As a practical example, the MAC of a pure quartz (SiO2) sample for Cu-Kα radiation
is:

28.086 cm2
μÛQuartz = 60.6
28.086 + 2
15.999 g

2
15.999 cm2 cm2

+ 11.5 = 34.45 . (5:11)
28.086 + 2
15.999 g g
The same calculation for a pure alumina (Al2O3) sample is:

2
26.982 cm2
μÛAlumina = 48.6
2
26.982 + 3
15.999 g

3
15.999 cm2 cm2

+ 11.5 = 31.14 . (5:12)
2
26.982 + 3
15.999 g g
The MAC of a mixture of 50% alumina and 50% quartz would then be:

cm2 cm2 cm2

μÛm = 0.5
34.45 + 0.5
31.14 = 32.79 . (5:13)
g g g

5.3 Rietveld method

Under the assumption that all phases in a mixture are crystalline, the following
normalization relation can be used:
X
Wi = 1. (5:14)
i
134 5 Quantitative phase analysis

This relationship allows elimination of the instrument constant and the mass
absorption coefficient of the sample through:

Sα ðZMV Þα
Wα = P . (5:15)
i Si ðZMV Þi

Quantitative Rietveld refinement can therefore be performed without knowledge of K′

and μÛm (Madsen & Scarlett, 2008). The drawback of this standard-less method is that
the weight fractions of the phases are relative and not absolute numbers. They are
therefore insensitive to unknown or amorphous content. To overcome this problem,
an external or internal standard must be used.

5.4 Internal standard method

The amount of amorphous or unknown content can be determined by adding an
internal standard std:

Sα ðmeasÞ ðZMV ÞαðmeasÞ

Wα ðmeasÞ = Wstd ðmeasÞ . (5:16)
Sstd ðmeasÞ ðZMV Þstd ðmeasÞ

The absolute weight fractions of the known materials can then be can calculated by:

Wstd ðknownÞ
Wα ðabsÞ = Wα ðmeasÞ . (5:17)
Wstd ðmeasÞ

The weight fraction of the unknown or amorphous material follows directly from:

X
Wunknown ðabsÞ = 1.0 − Wk ðabsÞ . (5:18)
k

where k runs over all phases of the mixture except the unknown or amorphous
material.

5.5 PONKCS method

In cases where Bragg reflections from an unknown phase can clearly be identified, a
partial or no known crystal structure (PONKCS) phase can be created from a pure
sample or from a sample where the amount of the unknown phase is known (e.g., by
mixing a known amount of a standard with the sample). First, a set of intensities
 5.6 Some correction factors 135

(group of single peaks, Pawley or Le Bail) needs to be determined with an overall

fixed scale factor (=1.0). Second, the Lorenz-polarization correction must be
applied, which is done automatically in TOPAS using the appropriate keywords
and macros (see Chapter 2):

′ Imeas
Imeas = . (5:19)
LP

With the help of an internal standard std, an artificial value of ZM for the unknown
phase α can be calculated:

wα Sstd ðZMV Þstd

ðZMÞα = . (5:20)
wstd Sα Vα

The volume is either known in the case of Pawley or Le Bail fits, or set to Vα = 1 for
group of peaks. Note that the obtained ðZMÞα value has no physical meaning and is
only valid for the chosen experimental configuration. The “correct” ðZMÞα value
requires knowledge of the density of the unknown material:

ρα V α
ðZMÞαðtrueÞ = . (5:21)
1.6604

Peak intensities can then be scaled by:

ðZMÞαðtrueÞ
. (5:22)
ðZMÞα

5.6 Some correction factors

QPA is affected by many sources of error such as instrument configuration, particle

statistics, counting error, preferred orientation and microabsorption. The application
of correction factors like the Brindley-correction for spherical particles (Brindley,
1945) can be necessary for mixtures with strongly different mass absorption coeffi-
cients for the different phases. However, the assumption that all phases in a sample
consist of spherical particles of identical diameter is unrealistic, and should be used
with extreme caution. The Brindley model for correction of microabsorption effects
(Brindley, 1945) in TOPAS notation is:

prm !R 0.002 min 0 ‘ radius of the particle in [cm]

prm !PD 0.6 min 0 max 1 ‘ packing density.
Apply_Brindley_Spherical_R_PD( R, PD)
136 5 Quantitative phase analysis

The methods described above for quantitative phase analysis can also be applied
to powder diffraction data measured in Debye–Scherrer geometry, though there
are several important differences. On the positive side, preferred orientation
and graininess is typically less of a problem than in reflection geometry.
Measurement of an empty capillary will generally lead to a higher background
than a filled capillary due to the much lower absorption at low diffraction angles.
This leads to severe quantification errors in amorphous content. In order to
determine the background function reliably, it is therefore necessary to measure
a highly crystalline sample of similar absorption and packing density as the
sample under investigation or to use high energy radiation where absorption is
not significant.

5.7 Elemental composition

TOPAS has various built-in tools for working out the overall composition in a multi-
phase fit. For example, the following keywords exist:

Get_Element_Weight(Element) ‘ Returns the weight% a specific element in the sample

Get(sum_smvs_minus_this) ‘ Returns the sum of SMVs minus the phase where it is defined.
element_weight_percent $ELEMENT $NAME # ‘ An xdd dependent keyword that returns the
weight percent of an element within the corresponding str’s of the xdd

If the overall amount of an element is known from other sources (e.g., XRF, EDAX,
ICPMS analysis), it is possible to set up a constraint to force the refinement to reflect
this information using the keywords above (see the technical manual of TOPAS for
more information).
A more intuitive way to include a known weight% of an element is through a
restraint where the difference between known and refined weight% is minimized. In
the following example “zr” is the name given to the weight% of element Zr+4 in the
sample. A restraint is used to minimize the difference between the refined and the
known value (65 %). The refinement obeys the restraint according to the value set for
the keyword “penalties_weighting_K1”. A high value for penalties_weighting_K1
should be avoided as the restraints then behave more like constraints that might
contradict the information from the powder pattern. Restraints are covered in more
detail in Chapter 6.

penalties_weighting_K1 .1
xdd
element_weight_percent Zr+4 zr 0 ‘ After refinement 65.0275252
restraint = (zr - 65); : 0 ‘ After refinement 0.0275251
 5.8 Practical application 137

5.8 Practical application

In the remainder of the chapter we discuss several exercises on QPA in detail. The
mixtures considered consist of crystalline alumina, crystalline quartz and amor-
phous silica flour. The data are taken from the work of Madsen, Scarlett & Kern
(2011). XRPD data were measured using a Philips X’Pert diffractometer in Bragg–
Brentano geometry (173 mm radius, Cu-Kα radiation from secondary graphite mono-
chromator 2θm = 26.6°, fixed 1° divergence slit, 0.3 mm receiving slit, 1° antiscatter
slit, primary Soller slits with 2.5° opening). All data were recorded from 10°–140°
2θ in 0.02° 2θ steps for 3 seconds per step.

5.8.1 Determination of the IRF and background

As discussed in Chapter 4, it is a good practice to determine the instrumental

resolution function (IRF) from a line profile standard before a Rietveld refinement.
Here, Y2O3 was used as the line profile standard. The background was described by a
Chebyshev polynomial of eighth order and the line profile was fitted by a TCHZ-
Pseudo-Voigt function refining GW, LX and LY parameters only (Figure 5.1).

400
350
300
Sqrt (rel. intensity)

250
200
150
100
50
0
10 20 30 40 50 60 70 80 90 100 110 120 130
2θ/°

Figure 5.1: Rietveld plot of the line profile standard Y2O3 for determination of the IRF.

The background and the TCHZ pseudo-Voigt peak profile parameters need to be fixed
and copied to further refinements. The TOPAS script file for the Rietveld refinement of
the line profile standard Y2O3 is given below:

xdd Y1080223.xy
CuKa2(0.00001)
x_calculation_step 0.02
138 5 Quantitative phase analysis

bkg @ 428.40 -119.34 171.44 -86.61 75.03 -40.29 26.92 -8.74 1.36
LP_Factor( 26.6)
Specimen_Displacement(@, -0.00658)
Rp 173
Rs 173
axial_conv
filament_length 12
sample_length @ 18.247
receiving_slit_length 12
primary_soller_angle @ 2.629
axial_n_beta 30
str
phase_name "Y2O3 standard"
TCHZ_Peak_Type(, 0,, 0,@, 0.005555,, 0,@, 0.034624,@, 0.01553)
space_group Ia-3
scale @ 0.0001726791627
Cubic(@ 10.60344)
site Y1 num_posns 24 x @ -0.03233 y 0 z 0.25 occ Y+3 1 beq @ 0.338
site Y2 num_posns 8 x 0.25 y 0.25 z 0.25 occ Y+3 1 beq @ 0.382
site O1 num_posns 48 x @ 0.39102 y @ 0.15207 z @ 0.38113 occ O-2 1 beq @ 0.453

5.8.2 Standard-less Rietveld QPA

A diffraction pattern from a mixture of 50 weight% quartz and 50 weight%

alumina is subjected to unconstrained Rietveld refinement giving the fit shown
in Figure 5.2.

320
AI2O3 48.49%
300
280 Quartz 51.51%
260
240
Sqrt (rel. intensity)

220
200
180
160
140
120
100
80
60
40
20
0
–20
10 20 30 40 50 60 70 80 90 100 110 120 130 140
2θ/°

Figure 5.2: Rietveld plot of a 50/50 mixture of alumina and quartz.

 5.8 Practical application 139

The corresponding TOPAS INP file is shown below. In TOPAS, the weight fractions are
automatically reported in the MVW macro. The refined weight fractions of 48.5% and
51.5% are close to the expected values:

xdd NS_I1.xy
CuKa2(0.00001)
x_calculation_step 0.02
bkg @ 252.95 -159.07 155.55 -76.97 63.85 -46.05 25.43 -14.66 9.64
start_X 10
LP_Factor( 26.6)
Specimen_Displacement(@, -0.0430481624)
Rp 173
Rs 173
axial_conv
filament_length 12
sample_length 18.24674385
receiving_slit_length 12
primary_soller_angle 2.628582881
axial_n_beta 30
Absorption(@, 36.00689955)
str
e0_from_Strain( 8.402217427e-005,,,@, 0.03853369506)
TCHZ_Peak_Type(, 0,, 0,, 0.005555009,, 0,, 0.03462355,, 0.01552971)
phase_MAC 31.59020452
phase_name "Al2O3"
MVW( 611.767656, 254.7642651, 48.48921677)
space_group R-3c
scale @ 0.007759970864
Phase_LAC_1_on_cm( 125.9649831)
Phase_Density_g_on_cm3( 3.987469693)
Trigonal(@ 4.758757821,@ 12.9903446)
site Al1 num_posns 12 x 0 y 0 z 0.35214 occ Al+3 1 beq 0.34
site O1 num_posns 18 x 0.30656 y 0 z 0.25 occ O-2 1 beq 0.33
str
e0_from_Strain( 5.149340352e-005,,,@, 0.02361556489)
TCHZ_Peak_Type(, 0,, 0,, 0.005555009,, 0,, 0.03462355,, 0.01552971)
phase_MAC 35.81264014
phase_name "Quartz"
MVW( 180.2529, 112.973826, 51.51078323)
space_group P3221
scale @ 0.06309253838
Phase_LAC_1_on_cm( 94.88338204)
Phase_Density_g_on_cm3( 2.649438346)
Trigonal(@ 4.913002985,@ 5.404470483)
site Si1 num_posns 3 x 0.47074 y =0; z =2/3; occ Si+4 1 beq 0.8822
site O1 num_posns 6 x 0.41648 y 0.267569 z 0.791080 occ O-2 1 beq 1.5434

From the parameters in red in the INP file, the weight fractions can also be manually
calculated using eq. (5.16):
140 5 Quantitative phase analysis

WQuartz =
ð0.06309
180.253
112.974Þ
= 0.515 (5:23)
ð0.06309
180.253
112.974Þ + ð0.00776
611.768
254.764Þ

and

WAlumina =
0.00776
611.768
254.764
= 0.485. (5:24)
0.06309
180.253
112.974 + 0.00776
611.768
254.764

5.8.3 External standard method for determining unknown or amorphous content

To apply the external standard method, the diffractometer constant Kʹ must be

determined. A measurement of, for example, alumina (Figure 5.3) can be used.

220
AI2O3 100.00%
200
180
160
Sqrt (rel. intensity)

140
120
100
80
60
40
20
0
10 20 30 40 50 60 70 80 90 100 110 120 130 140
2θ/°

Figure 5.3: Rietveld plot of the diffraction data of alumina.

The TOPAS script file for the refinement is:

xdd Al2O31Cu-2.xy
CuKa2(0.00001)
x_calculation_step 0.02
bkg @ 252.95 -159.08 155.55 -76.97 63.85 -46.05 25.43 -14.66 9.64
start_X 10
LP_Factor( 26.6)
Specimen_Displacement(@, -0.06813183736)
Rp 173
Rs 173
axial_conv
filament_length 12
sample_length 18.24674385
 5.8 Practical application 141

receiving_slit_length 12
primary_soller_angle 2.628582881
axial_n_beta 30
Absorption(@, 28.13477304)
str
phase_name "Al2O3"
LVol_FWHM_CS_G_L( 1, 857.5025984, 0.89, 1190.952787,, 9981.763646,@, 1364.793828)
e0_from_Strain( 5.695107198e-005,,,@, 0.02611852478)
TCHZ_Peak_Type(, 0,, 0,, 0.005555009,, 0,, 0.03462355,, 0.01552971)
phase_MAC 31.59020452
MVW( 611.767656, 254.7217567, 100)
space_group R-3c
scale @ 0.01650053178
Phase_LAC_1_on_cm( 125.9860043)
Phase_Density_g_on_cm3( 3.988135128)
Trigonal(@ 4.758530635,@ 12.98941733)
site Al1 num_posns 12 x 0 y 0 z @ 0.35214 occ Al+3 1 beq @ 0.34
site O1 num_posns 18 x @ 0.30656 y 0 z 0.25 occ O-2 1 beq @ 0.33

From the parameters in red in the INP file, Kʹ can be calculated using eq. (5.8):
 3

0.0165
611.768
254.722 Å
31.59 cm2 =g
K′ = = 81224. (5:25)
1

Note that in TOPAS, a conversion factor of 1.66/100 is added leading to a Kʹ factor of

1348. The corresponding code in TOPAS to retrieve this value is:

K_Factor_WP(1348)

Alternatively, a Rietveld refinement of quartz (Figure 5.4) can be used to retrieve Kʹ.
The corresponding TOPAS script is:

xdd QuartzCu.xy
CuKa2(0.00001)
x_calculation_step 0.02
bkg @ 310.80 -150.21 157.92 -57.89 86.41 -28.46 32.75 0.084 17.95
start_X 10
LP_Factor( 26.6)
Specimen_Displacement(@, -0.008354646507)
Rp 173
Rs 173
axial_conv
filament_length 12
sample_length 18.24674385
receiving_slit_length 12
primary_soller_angle 2.628582881
axial_n_beta 30
Absorption(@, 51.37923198)
str
142 5 Quantitative phase analysis

e0_from_Strain( 3.350261639e-005,,,@, 0.01536474883)

TCHZ_Peak_Type(, 0,, 0,, 0.005555009,, 0,, 0.03462355,, 0.01552971)
phase_MAC 35.81264014
phase_name "Quartz"
MVW( 180.2529, 112.9561208, 100)
space_group P3221
scale @ 0.1168561698
Phase_LAC_1_on_cm( 94.89825444)
Phase_Density_g_on_cm3( 2.649853629)
Trigonal(@ 4.912656195,@ 5.404386419)
site Si1 num_posns 3 x @ 0.4707 y =0; : 0 z =2/3; occ Si+4 1 beq @ 0.8822
site O1 num_posns 6 x @ 0.4165 y @ 0.2676 z @ 0.7911 occ O-2 1 beq @ 1.5434

450 Quartz 100.00%

400
350
Sqrt (rel. intensity)

300
250
200
150
100
50
0
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120 125 130 135
2θ/°

Figure 5.4: Rietveld plot of the diffraction data of quartz.

From the parameters in red in the INP file, Kʹ can be calculated using eq. (5.8):
 3

0.11686
180.253
112.956 Å
35.81 cm2 =g
K′ = = 85204. (5:26)
1

Note that the values of Kʹ for alumina and quartz are not fully equal as they
theoretically should be.
We can use these Kʹ values to analyze materials with amorphous content. Here
we will analyze a mixture of alumina and 50 weight% of silica flour. Several options
exist for fitting the slowly oscillating background from the amorphous phase, a
constrained group of single peaks, a Pawley (Le Bail) phase, or a Rietveld phase
(see Chapters 2 and 3). The Rietveld-like approach uses the crystal structure of a
closely related compound (e.g., cristobalite) while applying a small crystallite size on
 5.8 Practical application 143

the order of 2–3 nm. This will only work if the positions and relative intensities of the
amorphous humps and the strongest Bragg reflections approximately match.
In the present case, a Pawley (or Le Bail) fit without known lattice parameters is
used (Figure 5.5). In order to distribute positions of peaks equally over the entire 2θ
range, an quasi one-dimensional orthorhombic unit cell in Pmmm with one large a
and two very small b and c (e.g., 0.1 Å) lattice parameters is used. Thereby, only
equidistant (h00) reflections are predicted in the powder pattern. An overall small
crystallite size (e.g., 1.5 nm) is set to ensure broad reflections. In order to have the first
peak position located at the beginning of the pattern, the corresponding d-value (or
multiples thereof) is taken as the lattice parameter a (here  4 × 8.8 Å). The scale
factor is fixed at unity. The TOPAS INP file is given below:

80
75
70
65
60
Sqrt (rel. intensity)

55
50
45
40
35
30
25
20
15
10
10 20 30 40 50 60 70 80 90 100 110 120 130
2θ/°

Figure 5.5: Fitting of a powder pattern of silica flour using a Pawley fit of a pseudo one-dimensional
orthorhombic phase. The background is denoted by a thin gray line.

xdd SilicaCu-2.xy
CuKa2(0.00001)
x_calculation_step 0.02
bkg @ 252.954 -159.07 155.55 -76.97 63.85 -46.05 25.43 -14.66 9.64
start_X 10
LP_Factor( 26.6)
Specimen_Displacement(, 0.08476106561)
Rp 173
Rs 173
axial_conv
filament_length 12
sample_length 18.24674385
receiving_slit_length 12
primary_soller_angle 2.628582881
axial_n_beta 30
144 5 Quantitative phase analysis

Absorption(, 37.51979846)
hkl_Is
phase_name "amorphous"
LVol_FWHM_CS_G_L( 1, 1.755642682, 0.89, 1.663253122,@, 1.868823733,,)
phase_MAC 31.59020452
scale 1
MVW( 0, 0.344, 0)
a 34.4 b 0.1 c 0.1
space_group Pmmm
hkl_m_d_th2 4 0 0 2 8.60000038 10.2776861 I @ 24.82109266
...
hkl_m_d_th2 42 0 0 2 0.81904763 140.264679 I @ 789.2849601

Alternatively, a constrained group of 15 single peaks with identical fwhm (defined by

an overall Lorentzian crystallite size parameter) and an overall scale factor fixed to
unity can be used (Figure 5.6).

80 10 22.63394 44.94668 96.39534 106.713 117.9061 137

29.82104 50.86944
75 36.33542 57.38428
70 63.33964
68.72082
65 75.98723
60 81.4587
Sqrt (rel. intensity)

55
50
45
40
35
30
25
20
15
10
10 20 30 40 50 60 70 80 90 100 110 120 130
2θ/°

Figure 5.6: Fitting a powder pattern of silica flour using a constrained group of single peaks.

The following TOPAS script shows only the peak phase that substitutes the Pawley
phase of the script above:

...
pk_grp phase_name "Peaks Phase:0"
scale 1.0

xo_Is
xo @ 10
peak_type fp
LVol_FWHM_CS_G_L( 1, 0.7044361144, 0.89, 0.9848078382,,,CS1, 1.106525661)
I @ 28.65612154
...
xo_Is
 5.8 Practical application 145

xo @ 137.5886737
peak_type fp
LVol_FWHM_CS_G_L( 1, 0.7044361144, 0.89, 0.9848078382,,,(CS1), 1.106525661)
I @ 1898.616265

If we return to the data from the powder containing 50% alumina and 50% silica
flour, a regular Rietveld refinement leads to 100 weight% of alumina (Figure 5.7). In
order to get the correct amount of alumina it is necessary to know the diffractometer
constant Kʹ and the mass absorption coefficient μÛm of the entire sample. In our case,
according to eq. (5.7) this amounts to:

0.007772
611.77
254.72
32.79

WAlumina = = 0.49 (5:27)
81224

140 amorphous 0.00%

130 AI2O3 100.00%
120
110
Sqrt (rel. intensity)

100
90
80
70
60
50
40
30
20
10
0
10 20 30 40 50 60 70 80 90 100 110 120 130
2θ/°

Figure 5.7: Rietveld plot of the 50/50 mixture of alumina and amorphous silica flour.

A variety of predefined keywords and macros to perform this task exist in TOPAS. The
amount of silica flour can then be easily deduced by subtracting the amount of
alumina from one as 51 weight%. The TOPAS script file for the Rietveld refinement
of the 50/50 mixture of alumina and silica flour is:

xdd NS_A1.xy
CuKa2(0.00001)
x_calculation_step 0.02
bkg 252.95 -159.07 155.55 -76.97 63.85 -46.05 25.43 -14.66 9.64
start_X 10
LP_Factor( 26.6)
Specimen_Displacement(@, -0.03934)
Rp 173
Rs 173
axial_conv
filament_length 12
146 5 Quantitative phase analysis

sample_length 18.247
receiving_slit_length 12
primary_soller_angle 2.6286
axial_n_beta 30
Absorption(@, 12.59461479)
str
phase_name "Al2O3"
e0_from_Strain( 5.483701953e-005,,,@, 0.025148992)
TCHZ_Peak_Type(, 0,, 0,, 0.005555009,, 0,, 0.03462355,, 0.01552971)
phase_MAC 31.59020452
MVW( 611.767656, 254.7418766, 100)
space_group R-3c
scale @ 0.007772
Phase_LAC_1_on_cm( 125.9760538)
Phase_Density_g_on_cm3( 3.987820139)
Trigonal(@ 4.758633079,@ 12.98988403)
site Al1 num_posns 12 x 0 y 0 z 0.35214 occ Al+3 1 beq 0.34
site O1 num_posns 18 x 0.30656 y 0 z 0.25 occ O-2 1 beq 0.33
hkl_Is
phase_name "amorphous"
LVol_FWHM_CS_G_L( 1, 1.646417734, 0.89, 1.559776067,, 1.752557378,,)
phase_MAC 31.59020452
scale @ 1.0
MVW( 0, 0.344, 0)
a 34.4 b 0.1 c 0.1
space_group Pmmm
hkl_m_d_th2 4 0 0 2 8.60000038 10.2776861 I 14.17381819
...
hkl_m_d_th2 42 0 0 2 0.81904763 140.264679 I 408.5852987

5.8.4 Internal standard for determining unknown or amorphous content

Nowadays amorphous content is most commonly determined using internal

standard methods. The unknown substance will be available in either pure
form (ideal case) or in a mixture (real life) and is spiked by an internal standard
of similar absorption. One might use alumina or quartz in our example, a less
absorbing material like LiF for an organic or TiO2 for a first-row transition metal
oxide. Here we will consider a mixture of crystalline and amorphous SiO2 spiked
by 50 weight% of alumina, which gave the Rietveld fit shown in Figure 5.8. The
refined Al2O3 weight% was 61.5% and SiO2 38.5%. We can calculate corrected
weight fractions as:

0.5
Walumina = 0.615 = 0.5 ðspiked phaseÞ
0.615
 5.8 Practical application 147

0.5
Wquartz = 0.385 = 0.31 (5:28)
0.615
and consequently the amorphous content is:

Wunknown = 1.0 − ð0.5 + 0.39Þ = 0.19. (5:29)

240 amorphous 0.00%

220 AI2O3 50.00%
200 Quartz 31.27%
180 Int Amor 18.73%
Sqrt (rel. intensity)

160
140
120
100
80
60
40
20
0
10 20 30 40 50 60 70 80 90 100 110 120 130
2θ/°

Figure 5.8: Rietveld plot of the mixture of 50 weight% alumina, 30 weight% quartz and 20 weight%
silica flour.

The TOPAS script file for the Rietveld refinement of the mixture is given below:

xdd NS_C1.xy
CuKa2(0.00001)
x_calculation_step 0.02
bkg 252.95 -159.07 155.55 -76.97 63.85 -46.05 25.43 -14.66 9.64
start_X 10
LP_Factor( 26.6 )
Specimen_Displacement(@, -0.04810104121)
Rp 173
Rs 173
axial_conv
filament_length 12
sample_length 18.24674385
receiving_slit_length 12
primary_soller_angle 2.628582881
axial_n_beta 30
Absorption(@, 19.58052177)
str
phase_name "Al2O3"
e0_from_Strain( 4.88542844e-005,,,@, 0.0224052295)
TCHZ_Peak_Type(, 0,, 0,, 0.005555009,, 0,, 0.03462355,, 0.01552971)
phase_MAC 31.59020452
MVW( 611.767656, 254.7217502, 61.52078011)
space_group R-3c
scale @ 0.007745045593
Phase_LAC_1_on_cm( 125.9860076)
148 5 Quantitative phase analysis

Phase_Density_g_on_cm3( 3.98813523)
Trigonal( 4.7585306, 12.98941719)
site Al1 num_posns 12 x 0 y 0 z 0.35214 occ Al+3 1 beq 0.34
site O1 num_posns 18 x 0.30656 y 0 z 0.25 occ O-2 1 beq 0.33
str
phase_name "Quartz"
e0_from_Strain( 9.401987686e-006,,,@, 0.004311877544)
TCHZ_Peak_Type(, 0,, 0,, 0.005555009,, 0,, 0.03462355,, 0.01552971)
phase_MAC 35.81264014
MVW( 180.2529, 112.9561208, 38.47921989)
space_group P3221
scale @ 0.03707566215
Phase_LAC_1_on_cm( 94.89825445)
Phase_Density_g_on_cm3( 2.649853629)
Trigonal( 4.912656195, 5.404386419)
site Si1 num_posns 3 x 0.470737 y =0; z =2/3; occ Si+4 1 beq 0.88
site O1 num_posns 6 x 0.41648 y 0.26757 z 0.79108 occ O-2 1 beq 1.54
hkl_Is
phase_name "amorphous"
LVol_FWHM_CS_G_L( 1, 1.755642682, 0.89, 1.663253122,, 1.868823733,,)
phase_MAC 31.59020452
scale 1
MVW( 0, 0.344, 0)
a 34.4 b 0.1 c 0.1
space_group Pmmm
hkl_m_d_th2 4 0 0 2 8.60000038 10.2776861 I @ 5.439444087
...
hkl_m_d_th2 42 0 0 2 0.81904763 140.264679 I @ 320.1672535

The correct weight% of the spiked phase can be added to the corresponding structure
phase in the INP file:

str
phase_name "Al2O3"
spiked_phase_measured_weight_percent 50

By adding the keyword corrected_weight_percent to all Peak, Rietveld, Pawley

and Le Bail phases and rerunning the refinement, the corrected weight% will be
automatically displayed:

str
phase_name "Quartz"
corrected_weight_percent 31.27375797

We can also add “weight_percent_amorphous 0” at the xdd level to directly report the
amorphous content.
 References 149

5.8.5 Application of the PONKCS method

For a successful application of the PONKCS method, the unknown phase must be present
in a pure form or in a significant amount in a mixture of known phases. In the following,
we reuse the amorphous SiO2 and 50 weight% alumina sample from the previous
section where the weight fraction of the amorphous phase was determined to be 19%.
In the PONKCS approach, the unknown amorphous phase is fitted with an
orthorhombic pseudo Pawley phase with the scale factor fixed to unity, all peak
intensities are fixed and the phase is stored as the so-called PONKCS phase. A pseudo
ZM value can then be calculated:

0.1873 0.007745 611.77
254.72 g

ðZMÞamorphous = = 1314.26 . (5:30)
0.5 1 0.344 mol
This value can now be reused for a different sample. Here we analyze a mixture where
5 weight% of the amorphous SiO2 was mixed with 45 weight% quartz and 50 weight%
of alumina (Figure 5.9). The previously-stored PONKCS phase is included in
the INP file with the scale factor as the only refined parameter and all intensities
fixed. Using the corrected_weight_percent keyword immediately reveals the correct
weight% of the PONKCS phase.

300 amorphous 5.58%

280 AI2O3 48.51%
260 Quartz 45.92%
240
220
Sqrt (rel. intensity)

200
180
160
140
120
100
80
60
40
20
0
10 20 30 40 50 60 70 80 90 100 110 120 130
2θ/°

Figure 5.9: Rietveld plot of the mixture of 50 weight% alumina, 45 weight% quartz and 5 weight%
silica flour.

References
Brindley, G.W. (1945): The effect of grain or particle size on X-ray reflections from mixed powders and
alloys, considered in relation to the quantitative determination of crystalline substances by X-ray
methods. Phil. Mag. 36, 347–369.
150 5 Quantitative phase analysis

Madsen, I.C., Scarlett, N.V.Y. (2008): Chapter 11 in Powder diffraction: theory and practice, Royal
society of chemistry, Dinnebier, R. E. and Billinge, S. (eds.)
Madsen, I.C., Scarlett, N.V.Y., Kern, A. (2011): Description and survey of methodologies for the
determination of amorphous content via X-ray powder diffraction. Z. Kristallogr. 226, 944–955.
Hubbell, J.H., Seltzer, S.M. (2004): Tables of X-ray mass attenuation coefficients and mass energy-
absorption coefficients (version 1.4). Available: http://physics.nist.gov/xaamdi, National
Institute of Standards and Technology, Gaithersburg, MD (Originally published 1995 as NISTIR
5632).
O’Connor, B.H., Raven, M.D. (1988): Application of the Rietveld refinement procedure in assaying
powdered mixtures. Powder Diffraction 3, 2–6.
6 Restraints, constraints and rigid bodies

Powder diffraction analysis is frequently limited by the low information content in

the experimental data. In particular, the ratio between the number of experimentally
observed intensities (or groups of intensities) and refined structural parameters is
usually much lower than achievable in single crystal diffraction. Introduction of non-
diffraction based information (e.g., from chemical or structural knowledge) can,
therefore, be extremely valuable. It can stabilize a refinement by either reducing
the number of parameters or by keeping parameters close to reasonable values.
Meaningless atomic positions due to flat minima of the minimization function can
be avoided, and the chance of refining to the correct global minimum increases. Ways
of including external information can include:
– Imposing linear or non-linear functional dependencies between parameters
(constraints).
– Grouping of atoms which move as a rigid or semi-rigid entity (rigid bodies – a
specific form of constraint).
– Guiding parameters to an expected value or expected range using a restraint or
penalty function that increases with increasing departure from the expected
value (restraints, sometimes called soft constraints).

Constraints and restraints are not limited to structural parameters but can be applied
to any of the parameters used during Rietveld refinement.
Constraints are incorporated into the least-squares model in a mathematically
precise way such that they must always be exactly obeyed. They can, therefore, be
thought of as “hard” information. Restraints, however, are treated in a similar way to
experimental observations and merely guide the refinement. They therefore act as
“softer” information and the degree to which they’re obeyed is balanced against the
model’s need to fit the experimental data. TOPAS does this by adding extra terms to
the objective function that is expressed as the sum of contributions from observations
(χ20 ), penalties (χ2P ) and restraints (χ2R ):

χ2 = χ20 + χ2P + χ2R (6:1)

with
N 
X  1
χ20 = K wi ðycalc, i − yobs, i Þ2 , K = PN ,
i=1 i=1 wi y2obs, i
(6:2)
X
NP X
NR
χ2P = KK1 KP Pp , χ2R = KK1 KR R2r ,
p=1 r=1

where ycalc, i and yobs, i are the calculated and observed data, respectively, at point i of
N data points; wi is the weighting given to data point i; Pp are penalty functions; NP is

https://doi.org/10.1515/9783110461381-006
152 6 Restraints, constraints and rigid bodies

the number of penalty functions; Rr are restraints; NR is the number of restraints; and
K1 , KP and KR are scaling terms applied to the penalty functions and restraints,
respectively, and discussed in Section 6.2. The distinction between penalties and
restraints is discussed below, though the terms are often used interchangeably. We’ll
use (soft-) restraints to describe both types.
The way in which penalties can help guide a refinement, restricting it to “sensible”
solutions, can be best understood through simple examples. One is the “anti-bump”
penalty which can be used to prevent atoms approaching more closely than is chemi-
cally plausible. This can be expressed by using a penalty that feeds into the χ2P term of
the objective function of the form:
(P 2
rij − r0 if rij < r0 and i ≠ j,
Pp = ABi = (6:3)
0 if rij ≥ r0

where r0 is the minimum approach distance, rij the distance between atoms i and j
including symmetry equivalent positions, and the summation is over all atoms of
type j. The penalty is zero when atoms are far apart but rises rapidly if atoms get
closer than r0.
A penalty function suitable for keeping a series of n atoms (e.g., an aromatic ring)
flat can be expressed as:

6 X n X n X n   
flat = bi × bj  bk  − tol 2
nðn − 1Þðn − 2Þ i = 1 j = i + 1 k = j + 1 (6:4)
 
if bi × bj  bk  > tol,

where tol is the allowed deviation from the least squares plane of the atoms before a
penalty is applied and b are Cartesian unit length vectors between the sites and the
geometric center of the n sites.
More sophisticated electrostatic potentials for ionic compounds (Figure 6.1), like
the Lennard-Jones or Born–Mayer potentials can be used as penalty functions by

Figure 6.1: Schematic electrostatic

potential as a combination of a repul-
sion term and an attractive Coulomb
term, which can be used as penalty
function.
 6.1 Constraints in TOPAS 153

applying what TOPAS calls the General Real Space (GRS) interaction (Coelho & Cheary,
1997). For a particular site i, the electrostatic potential Ui contains a Coulomb term Ci
and a repulsive term Ri and is written as:

Ui = Ci + Ri , (6:5)

where
e2 X n
Qi Qj
Ci = (6:6)
4πε0 j = 1 ≠ i rij

and
X n
Bij
Ri = (6:7)
rm
j = 1 ≠ i ij

for Lennard-Jones and

X
n
Ri = cij e − drij (6:8)
j=1≠i

for Born–Mayer, respectively. Here ε0 is the permittivity of free space, e is the charge
of an electron, Qi and Qj are the ionic charges of atoms i and j in units of e, rij is the
distance between atoms i and j including symmetry equivalent positions, and the-
summation is over all atoms to infinity. The repulsive constants Bij , m, cij and d are
characteristic of the atomic species and their surroundings. The penalty has the form
shown in Figure 6.1 and will guide atoms toward a sensible interatomic separation
during refinement.
We will see how these and other restraints are used in TOPAS in later sections of
this chapter.

6.1 Constraints in TOPAS

Parameters can be constrained to other parameters via a linear or nonlinear equation
of any complexity. Constraints are built into the refined model mathematically so that
they have to be obeyed exactly.
An easy example is the definition of an overall displacement parameter for a
group of atoms with an enhancement factor for hydrogen atoms:

prm B1 1.0 min 0 max 2

site C1 num_posns 8 x @ 0.0` y @ 0.0 z @ 0.0 occ C 1 beq =B1;
site C2 num_posns 8 x @ 0.0` y @ 0.0 z @ 0.0 occ C 1 beq =B1;
…
site H1 num_posns 8 x @ 0.1 y @ 0.1 z @ 0.1 occ H 1 beq =1.5 B1;
154 6 Restraints, constraints and rigid bodies

Constraints are also frequently used for defining occupancies of mixed sites. In the
following example the parameter zr is used in the equation “= 1–zr;” to define the Ti+4
site occupancy parameter1:

Site Zr x 0 y 0 z 0
occ Zr+4 zr 1 min=0; max = 1; beq 0.5
occ Ti+4 = 1-zr; beq 0.5

Many other constraints are built into the TOPAS macros. For example, the cell
parameter macro Cubic (@ 4.1) writes a series of constraint equations that force the
lattice parameters to obey a = b = c during refinement.
It is possible to assign minimum and maximum values to any parameter to
constrain the value to lie within a user-defined range.2 For example, one might
set a minimum and maximum value of ±0.1° on the zero point error of a well-
aligned diffractometer to prevent a refinement diverging to unreasonable values
in its early stages. Most of the TOPAS macros have this type of constraint built
in. One example is the Keep_Atom_Within_Box macro that sets min and max
values on fractional coordinates to force an atom to remain within a cubic
volume element centered on its starting value during refinement. In the follow-
ing example, the potassium site cannot move outside of a box with a length of
2 × size (here 0.1 Å) around its starting position:

…
prm !size 0.1
site K1 x kx 0.25 y ky 0.25 z kz 0.25 occ K+1 ok 0.5 min 0 max 1 beq B1
Keep_Atom_Within_Box (size)
…

6.2 Restraints & penalties in TOPAS

Soft-restraints allow more flexibility than constraints (Section 6.1) or rigid bodies
(Section 6.3). They are handled by the least squares in the same way as experimental
observations (they are “extra observations”). TOPAS has two ways of doing this –
using either penalty or restraint equations.

1 In TOPAS, equations always start with an equal sign and end in a semicolon.
2 Care must be taken if parameters hit their min/max limits at the end of a refinement as the model
may not be fully converged and other parameters may be affected. TOPAS flags this in the OUT file.
The origin of the problem should be investigated and, if necessary, parameters removed from the
model.
 6.2 Restraints & penalties in TOPAS 155

Imagine a simple case where one wants to use some form of soft-restraint to keep a
coordinate x1 close to a particular expected value, for example, close to 0.137. In
TOPAS one could use a penalty equation:

penalty = (x1 - 0.137)^2;

The parabolic form of the penalty means that it will increase the overall χ2 (eq. (6.1))
by an increasing amount as x1 deviates from 0.137 regardless of whether its value is
less than or greater than 0.137. The penalty will therefore guide the refinement
toward a model with x1 of 0.137 (low χ2P ) provided this doesn’t cause too great an
increase in χ20 .
Alternatively one could use a restraint equation:

restraint = (x1 - 0.137);

Internally TOPAS will square the restraint equation. It therefore ends up being
very similar to using the penalty equation. Don’t get too confused by the terms
“penalty” and “restraint.”3 Both methods try to steer the refinement toward a
solution that is consistent with a non-diffraction based observation. In general
penalty equations are more flexible as they’re not automatically squared and
can therefore take negative values (e.g., the potential of Figure 6.1). There can
be small differences in how penalties and restraints feed into the minimization
of χ2 , due to different weightings against the diffraction data. As a consequence
the minimization pathway and final minimum may differ. One difference is that
off-diagonal terms in the A matrix (see Section 2.7) are not calculated for
penalties unless:

approximate_A

is defined. This means that you might get quicker minimization using restraint
equations. If penalties have a significant contribution to χ2 than using
approximate_A may give faster convergence as the off diagonal terms are then
approximated by the BFGS method.
Penalties typically aren’t included in esd calculations (restraints are) though you
can override this with:

do_errors_include_penalties

3 For historical reasons even TOPAS syntax uses the terms inconsistently. For example only_penalties
will apply to both penalty and restraint equations; macros like Distance_Restrain typically use a
penalty equation.
156 6 Restraints, constraints and rigid bodies

TOPAS uses a separate A matrix for the data, penalties and restraints. It chooses how
to scale penalties or restraints against the data (i.e., how closely each is obeyed; the
terms K P and K R in eq. (6.2)) by considering the relative magnitudes of the inverse
error terms in each matrix. The overall relative importance of penalties and restraints
to data can be increased using the following keyword, which sets the value of K 1 in
eq 6.2:

penalties_weighting_K1 1

Normally the default value of 1 is appropriate. By using larger values of K1 penalties

and restraints will be more closely obeyed and they will start to mimic constraints.
If one has several penalty equations their individual relative weighting can be
changed by using an equation like:

penalty = w*(x1 - 0.137)^2;

The value w might be different for a soft-restraint on a bond angle compared to a bond
distance. Typically one would weight by 1/σ2 where σ is the standard uncertainty on
the quantity. The relative weighting for soft-restraints on distances (σ = 0.01) and
angles (σ = 1) might therefore be 10000 to 1.
To instruct the minimization procedure to minimize on penalty (and/or restraint)
functions only use:

only_penalties

This only_penalties switch is used either when there is no observed data or when one
wants to temporarily ignore the data. With sufficient restraints a structural model can
be refined without data, analogous to the DLS method of Baerlocher (Baerlocher, 1978).
Note that parameters that are not dependent on the penalties are not refined.
Penalties can also be a function of the iteration number and can be turned off or
decreased in importance once a certain iteration number is reached using the reserved
parameter Cycle_Iter that returns the current iteration within a cycle with counting
starting at zero. In the following example, the penalty is only applied for the first 10
iterations of the current refinement cycle:

penalty = If(Cycle_Iter < 9, (x1 - 0.137)^2, 0);

In general, the overall Rwp should not increase significantly when restraints are applied
to a model. If it does, it is a sign of deficiencies in the model and/or the measurement.
The possibilities for user defined penalties and constrains are essentially infinite. For
convenience there are many predefined restraint macros, some of which are described
in the following section.
 6.2 Restraints & penalties in TOPAS 157

6.2.1 Distance, angle and flatten restraints

Common macros in TOPAS that apply penalty equations include anti bumping, bond
lengths, bond angles and flatness restraints:

AI_Anti_Bump(…)
Distance_Restrain(…)
Angle_Restrain(…)
Flatten(…)

As an example, let’s consider how each of them might be used to control the geometry
of an aromatic cyclopentadienyl ring consisting of 5 carbon atoms C1…C5 (Figure 6.2). It
can be assumed that the bond lengths between two carbon atoms will be 1.4 Å and the
angle between three carbon atoms will be close to 108°. The molecule is assumed to be
flat with a maximum deviation from the plane of 0.01 Å. The positional parameters in
the TOPAS input file are:

Figure 6.2: Aromatic cyclopentadienyl ring consisting of 5 carbon atoms C1…C5. The insert shows a Z-
matrix description as discussed later.

site C1 x @ 0.31646` y @ 0.22174` z @ 0.36153` occ C 1 beq 2

site C2 x @ 0.31488` y @ 0.35432` z @ 0.32264` occ C 1 beq 2
site C3 x @ 0.18779` y @ 0.40617` z @ 0.34644` occ C 1 beq 2
site C4 x @ 0.10624` y @ 0.29993` z @ 0.39525` occ C 1 beq 2
site C5 x @ 0.18537` y @ 0.18763` z @ 0.40500` occ C 1 beq 2
158 6 Restraints, constraints and rigid bodies

When setting up restraints, care must be taken to restrain the correct atoms as the
list of atomic sites might contain atoms from different parts of the unit cell that are
in different molecules. An inadvertent restraint on an inter- rather than intra-
molecular C–C bond will cause catastrophic problems during refinement! To facil-
itate this, the command:

append_bond_lengths

writes bond lengths, angles and torsion angles to the OUT file after refinement
(which could be after 0 refinement cycles (iters 0) so that a model is unchanged). In
the listing each atom is uniquely identified by four numbers corresponding to the
symmetry operation used to generate it, and fractional offsets of the cell in which
the atom sits relative to the pivot atom. If all four numbers are zero they can be
omitted, otherwise they have to be specified for each atom in the restraint. Here the
five sites listed in the asymmetric unit are all part of the same molecule and the
output looks like:

{
C1:0 C2:0 0 0 0 1.38365
C5:0 0 0 0 1.42564 108.178
C3:0 0 0 0 2.24781 72.209 36.011
C4:0 0 0 0 2.26421 36.962 35.247 72.941
…
}

The most basic restraint would be to prevent atoms getting too close using the
AI_Anti_Bump macro between the five carbon sites. The relative weight given to the
penalty function is set to one and we penalizee approaches below 1.2 Å:

prm !bump_dist 1.2

prm !bump_weight 1
AI_Anti_Bump(C*, C*, bump_dist, bump_weight)

The closer the atoms are, the higher the penalty becomes. The AI_Anti_Bump
macro includes the penalty function given in eq (6.1), and its expression in
TOPAS syntax can be found in the TOPAS.INC file. Since anti-bumping restraints
force atoms to stay away from each other, they are mainly used in structure
determination using global optimization methods (see Chapter 7). Applying the
restraint only for the first few iterations of a refinement cycle can also be
beneficial.
A more powerful approach is to use the Distance_Restrain macro to restrain the
bond length between the carbon atoms C1 and C2 to 1.4 Å, with a tolerance of 0.02 Å
 6.2 Restraints & penalties in TOPAS 159

and a relative weight of 100.4 The tolerance means a penalty is only applied if the
refined distance lies outside the range 1.38–1.42 Å:

prm !bond_length 1.4

prm !bond_tol 0.02
prm !bond_weight 100
Distance_Restrain(C1 0 0 0 0 C2 0 0 0 0, bond_length, 1.384, bond_tol, bond_weight)

The actual value of the bond length (1.384 Å) is automatically added to the OUT file
after the first refinement run is completed. The penalty function used here is a
parabola [(r – 1.4)2 for r < 1.38 or r > 1.42], which means that the penalty increases
symmetrically for shorter and longer distances.
Similarly, the Angle_Restrain macro can be used to restrain the bond angle
between the carbon atoms C1, C2 and C5 to 108° with an angle tolerance of 2° and a
relative weight of 10 by a similar parabolic function:

prm !angle 108

prm !angle_tol 2
prm !angle_weight 10
Angle_Restrain(C5 0 0 0 0 C1 0 0 0 0 C2 0 0 0 0, angle, 108.18037`,angle_tol, angle_ weight)

Finally, restraining the five carbon atoms C1…C5 to a flat plane with a toler-
ance of 0.01 Å and a relative weight of 1000 can be performed by the Flatten
macro:

prm !flatten_tol 0.01

prm !flatten_weight 1000
Flatten(C1 C2 C3 C4 C5, 0.0107887918`, flatten_tol, flatten_weight)

which uses the penalty function defined in eq. (6.4). Note that more than three sites
must be used with this macro.
The relevant part of the corresponding TOPAS INP file for restraining the entire
five-ring using Distance_Restrain, Angle_Restrain and Flatten macros is given
below:

penalties_weighting_K1 1
prm !bond_tol 0.02
prm !bond_weight 10

4 The symmetry label and fractional offsets of all carbon sites in this example are zero. Although the
four zeros could be omitted, they are given here for completeness.
160 6 Restraints, constraints and rigid bodies

prm !angle_tol 2
prm !angle_weight 1
prm !flatten_tol 0.01
prm !flatten_weight 1000
Distance_Restrain(C1 0 0 0 0 C2 0 0 0 0, 1.4, 1.38365`, bond_tol, bond_weight)
Distance_Restrain(C2 0 0 0 0 C3 0 0 0 0, 1.4, 1.39125`, bond_tol, bond_weight)
Distance_Restrain(C3 0 0 0 0 C4 0 0 0 0, 1.4, 1.43036`, bond_tol, bond_weight)
Distance_Restrain(C4 0 0 0 0 C5 0 0 0 0, 1.4, 1.37359`, bond_tol, bond_weight)
Distance_Restrain(C5 0 0 0 0 C1 0 0 0 0, 1.4, 1.42564`, bond_tol, bond_weight)
Angle_Restrain(C1 0 0 0 0 C2 0 0 0 0 C3 0 0 0 0, 108, 108.19957`, angle_tol, angle_ weight)
Angle_Restrain(C2 0 0 0 0 C3 0 0 0 0 C4 0 0 0 0, 108, 107.87748`, angle_tol, angle_ weight)
Angle_Restrain(C3 0 0 0 0 C4 0 0 0 0 C5 0 0 0 0, 108, 107.68999`, angle_tol, angle_ weight)
Angle_Restrain(C4 0 0 0 0 C5 0 0 0 0 C1 0 0 0 0, 108, 107.95641`, angle_tol, angle_ weight)
Angle_Restrain(C5 0 0 0 0 C1 0 0 0 0 C2 0 0 0 0, 108, 108.18037`, angle_tol, angle_ weight)
Flatten(C1 C2 C3 C4 C5, 0.0107887918`, flatten_tol, flatten_weight)

6.2.2 Electrostatic potentials

Penalty equations can take any form – that is, they can be more complex than
the sum of differences squared that appear in restraint equations. For example
the equations in grs_interaction (which calculates a Lennard-Jones or Born–
Mayer potential according to eqs. (6.5)–(6.8) can’t be written as the sum of
differences between observed and calculated parameters. In TOPAS, these are
called using the predefined macros:

Grs_Interaction(…)
Grs_BornMayer(…)

The following example forces chemically sensible coordination environments

between an aluminum and several oxygen sites. Charges have been set to +3
and –2 for aluminum and oxygen, respectively. The expected bond length
is 2.6 Å between oxygen sites and 1.8 Å between aluminum and oxygen
sites. First, the Lennard-Jones potential (eq. (6.7)) for the repulsion term is
used:

prm !val_charge_Al 3
prm !val_charge_O -2
prm !dist_Al_O 1.8
prm !dist_O_O 2.6
prm !expo 5
Grs_Interaction(O*, O*, val_charge_O, val_charge_O, oo, dist_O_O, expo) penalty = oo;
Grs_Interaction(Al, O*, val_charge_Al, val_charge_O, alo, dist_Al_O, expo) penalty = alo;
 6.3 Rigid bodies 161

with the exponent of the repulsion part set to five.5 Alternatively, the Born–Mayer
equation (eq. (6.8)) for the repulsion term can be used:

prm !val_charge_Al 3
prm !val_charge_O -2
prm !dist_Al_O 1.8
prm !dist_O_O 2.6
prm !const 3
Grs_BornMayer(O*, O*, val_charge_O, val_charge_O, oo, dist_O_O, const) penalty = oo;
Grs_BornMayer(Al, O*, val_charge_Al, val_charge_O, alo, dist_Al_O, const) penalty = alo;

with the constant for the repulsion part set to three.

6.3 Rigid bodies

In many structures, groups of atoms (molecules or coordination polyhedra)
have strong local bonding and form a more or less rigid unit. Rather than
refining the constituent atoms independently it may be better to define them
as a single rigid body. Typical examples are the cyclopentadienyl anion or the
benzene ring. Rigid bodies have been a common tool in single crystal X-ray
diffraction analysis for more than 50 years (Scheringer, 1963), and are espe-
cially valuable when the quality of the data is low, the ratio of observations
to parameters is low and/or the structure is very complicated (e.g., proteins).
There are several general advantages for using rigid bodies, which are prob-
ably even more advantageous with powder data than single crystal
(Dinnebier, 1999):
– Since the group is forced to shift as a complete unit, meaningless changes in
internal geometry cannot occur.
– The number of refined parameters can be drastically reduced, allowing
them to be determined with much higher accuracy. This is in particular
useful in the case of powder data, where the ratio of the number of
independent observations (Bragg intensities) to refineable parameters is
typically low.
– The range of convergence to the correct structure is much larger than in normal
refinement.
– Hydrogen atoms can be included in the refinement process at an early stage. Only
their relative positions with respect to the other atoms are needed (a “riding
model”).

5 TOPAS allows the use of the wild card character “*” and the negation character “!” to simplify the
creation of lists of atom identifiers.
162 6 Restraints, constraints and rigid bodies

– Thermal parameters can be defined to describe the group as a whole. The use of
TLS matrices allows anisotropic refinement of the translational and librational
parts of the temperature factor with relatively few parameters.
– Rigid bodies can allow you to refine individual atomic positions even if disorder
is present. By using rigid bodies, it can be possible to model disorder even with
powder data (e.g., Behrens et al., 2008).

6.3.1 Definition of a rigid body

A rigid group of atoms can be set up using a variety of internal reference coordinate
systems. The most common are fractional, Cartesian, spherical or Z-Matrix formal-
isms. TOPAS supports each of these and even allows a mixture of them to be used
within a single INP file.
Whichever coordinate system is used, there must be a one-to-one match
between the sites defined in the rigid body (e.g., point_for_sites) and available
atomic sites in the str section of the INP file. The positions in the atomic site list
can be set to (0,0,0) as they are updated by TOPAS during refinement based on the
degrees of freedom of the rigid body. It is the responsibility of the user to check for
correct site symmetries and, if necessary, to restrict rigid body rotations and/or
translations and to set the fractional occupancy accordingly.
The graphical rigid body editor in TOPAS is of great help in setting up and
checking rigid bodies.

6.3.2 Cartesian coordinates

Atoms of a rigid body can be defined in the Cartesian coordinate system I = fi, j, kg
with unity axis length (usually one Angstrom) or in the equivalent fractional coordi-
nate system D = fa, b, cg. There is an infinite number of ways of defining the natural
basis of a crystal in terms of a Cartesian basis. In TOPAS, the conversion from
fractional to Cartesian coordinates in terms of the lattice vectors a, b and c is as
follows: (1) x-axis in the same direction as the a lattice parameter x k a; (2) z-axis
perpendicular to the a-b plane z k ðb × aÞ k c ; (3) y-axis in the direction defined by
the cross product of a and c y k a × ðb × aÞ (Fig. 6.3).
The conversion matrix M to convert the natural crystallographic coordinate
system D into the Cartesian coordinate system of the crystal I and vice versa:

I = MD and D = M − 1 I, s (6:9)
 6.3 Rigid bodies 163

b
y

γ
a
x
Figure 6.3: Orthonormalization of the crystallographic
z
coordinate system D into a Cartesian coordinate
c system I.

is thus given by:

0 1
1 0 0
B cosγ sinγ 0 C
B sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2 C
M¼B C. (6:10)
@ cosαcosβcosγ cosαcosβcosγ A
cosβ sin γ 1  cos2 β  sin γ

In TOPAS, a point in Cartesian space is denoted by the parameters ux, uy, uz. The origin
of the coordinate system is the basepoint of the rigid body and is normally the first
point_for_site defined. For practical reasons, in particular to ensure similar movement
of atoms at similar distances from the origin, the center of gravity of the rigid body is a
good origin choice. The origin doesn’t have to be at an atomic position, and in these
cases it can be convenient to place a dummy atom with zero occupancy at the origin.
There are an infinite number of ways of setting up a rigid body all of which have
pros and cons. It is usually advantageous to build a rigid body in a way that allows
direct refinement of the parameters (e.g., bond lengths and angles) of most interest.
As a first example of a rigid body in Cartesian coordinates, a regular octahedron
of ZrO6 with zirconium in the center can be defined in terms of a bond length r as
(Figure 6.4).

Figure 6.4: Rigid body of a regular ZrO6

octahedron.
164 6 Restraints, constraints and rigid bodies

prm r 2 min 1.8 max 2.2

rigid
point_for_site Zr ‘ The origin; ux, uy, uz not specified as default to 0
point_for_site O1 ux = r;
point_for_site O2 ux = -r;
point_for_site O3 uy = r;
point_for_site O4 uy = -r;
point_for_site O5 uz = r;
point_for_site O6 uz = -r;

The Cartesian coordinates, ux, uy and uz must be defined for each point_for_site unless
they are zero. The isotropic Zr–O bond length r is set as a refinable internal degree of
freedom in the limit between 2.0 ≤ r ≤ 2.2 Å.
If the octahedron is located on one or more symmetry elements, fewer atoms are
needed in the rigid body definition as the rest are created by space-group symmetry.
If the central Zr atom lies on a center of inversion, it is sufficient to define only the
three ligands O1, O3 and O5. Alternatively, the fractional occupancy of a particular
atomic site can be scaled appropriately; in this case sites O1 to O6 could be set to
occupancy 0.5.
More flexibility can be built into the ZrO6 group by making the three main axes
non-equidistant. This also distorts the 90° angle between three oxygen atoms in the
base plane, simulating compressive strain:

prm r1 2 min 1.8 max 2.4

prm r2 2 min 1.8 max 2.4
prm r3 2 min 1.8 max 2.4
rigid
point_for_site Zr
point_for_site O1 ux = r1;
point_for_site O2 ux = -r1;
point_for_site O3 uy = r2;
point_for_site O4 uy = -r2;
point_for_site O5 uz = r3;
point_for_site O6 uz = -r3;

To turn the octahedron into a tetragonal or even orthorhombic bispyramid, thus

keeping the angle between three oxygen atoms in the base plane rectangular, a
different definition of the rigid body is more useful:

prm angle 45 min 0 max 90

prm !s45 = Sin(Deg angle);
prm !c45 = Cos(Deg angle);

prm r1 1.8 min 1.8 max 2.4

prm r2 2.2 min 1.8 max 2.4
 6.3 Rigid bodies 165

prm r3 2.4 min 1.8 max 2.4

rigid
point_for_site Zr
point_for_site O1 ux = s45 r1; uy = c45 r1;
point_for_site O2 ux = -s45 r1; uy = c45 r1;
point_for_site O3 ux = s45 r2; uy = -c45 r2;
point_for_site O4 ux = -s45 r2; uy = -c45 r2;
point_for_site O5 uz = r3;
point_for_site O6 uz = -r3;

Using this approach, more and more degrees of freedom can be introduced. For
example, the polyhedron can be further distorted by changing the 45° angle.
The maximum number of degrees of freedom of the rigid body is always three
times the number of atoms, 3n, or 21 in our case. A perfect octahedron has six degrees
of freedom (three translational and three rotational) resulting in 15 fewer refinable
parameters. Refining one overall length leads to seven degrees of freedom. There are
9 degrees of freedom for an orthorhombic bipyramid and 10 degrees of freedom if one
angle in the equatorial plane is variable.
A general definition of a tetrahedron that can be distorted to a tetragonal
bisphenoid by changing the tetrahedral angle between the ligands (Figure 6.5) is
given below:

Figure 6.5: Rigid body description of a regular

SiBr4 tetrahedron.

prm r 2.17460 min 2.0 max 2.4

prm a 109.4712 ' = 2 Arccos(1/Sqrt(3));
prm s = Sin(Deg a/2);
prm c = Cos(Deg a/2);
166 6 Restraints, constraints and rigid bodies

rigid
point_for_site Si1
point_for_site Br1 ux = s r; uy = c r;
point_for_site Br1 ux = -s r; uy = c r;
point_for_site Br1 uy = -c r; uz = s r;
point_for_site Br1 uy = -c r; uz = -s r;

A more complicated example is para-hydroxybenzoate that consists of a central

benzene ring with hydroxy and carboxylate groups in the 1 and 4 positions. These
can both be twisted adding two additional internal degrees of freedom to the rigid body
(Figure 6.6). The most logical choice for the origin of the rigid body is in the center of
the benzene ring. Although not necessary, a dummy atom (e.g., called “X”) with zero
occupancy could be introduced at this position. Note that it needs to be defined both in
the point_for_site section of the rigid body and the site section of the str.

Figure 6.6: Rigid body of a para-hydroxybenzoate molecule.

prm !b_COH 1.377 min 1.33 max 1.40

prm !b_CC_aroma 1.392 min 1.37 max 1.42
prm !b_CC_single 1.540 min 1.50 max 1.60
 6.3 Rigid bodies 167

prm !b_CH 1.000 min 0.95 max 1.05

prm !b_CO_aroma 1.281 min 1.25 max 1.30

prm !s30 = 0.5;

prm !c30 = Sqrt(3) .5;

rigid
point_for_site X
point_for_site C1 ux = b_CC_aroma c30; uy = b_CC_aroma s30;
point_for_site C2 ux = b_CC_aroma c30; uy = -b_CC_aroma s30;
point_for_site C3 ux = -b_CC_aroma c30; uy = b_CC_aroma s30;
point_for_site C4 ux = -b_CC_aroma c30; uy = -b_CC_aroma s30;
point_for_site C5 uy = b_CC_aroma;
point_for_site C6 uy = -b_CC_aroma;

point_for_site H1 ux = (b_CC_aroma + b_CH) c30; uy = (b_CC_aroma + b_CH) s30;

point_for_site H2 ux = (b_CC_aroma + b_CH) c30; uy = -(b_CC_aroma + b_CH) s30;
point_for_site H3 ux = -(b_CC_aroma + b_CH) c30; uy = (b_CC_aroma + b_CH) s30;
point_for_site H4 ux = -(b_CC_aroma + b_CH) c30; uy = -(b_CC_aroma + b_CH) s30;

point_for_site O1 uy = -(b_CC_aroma + b_COH);

point_for_site H5 ux = b_CH c30; uy = -(b_CC_aroma + b_COH) - (b_CH s30);
point_for_site C7 uy = (b_CC_aroma + b_CC_single);
point_for_site O2 ux = b_CO_aroma c30; uy = (b_CC_aroma + b_CC_single) + (b_CO_aroma s30);
point_for_site O3 ux = -b_CO_aroma c30; uy = (b_CC_aroma + b_CC_single) + (b_CO_aroma s30);

The two internal torsions of the CO2– and the OH group can be defined by
rotations around a vector defined by two points. Take the carboxylate-group as
an example, the atoms O2 and O3 are rotated around a vector defined by the
atoms C5 and C7. Technically this is done by first translating the starting
position of the rotation vector to the origin together with the atoms to be
rotated. Afterwards, the rotation is performed around the rotation vector.
Finally, all rotated atoms are translated back to their original position. In
TOPAS notation:

Translate_point_amount(C5, -) operate_on_points "O2 O3"

rotate @ 0.0
Rotation_vector_from_points(C5, C7) operate_on_points "O2 O3"
Translate_point_amount(C5) operate_on_points "O2 O3"

or much more simply by the corresponding macro:

Rotate_about_points(@ 0.0, C5, C7, "O2 O3")

168 6 Restraints, constraints and rigid bodies

6.3.3 Fractional coordinates

Setting up a rigid body in fractional coordinates has the advantage that the coordi-
nates of an existing CIF file can be copied one to one to the rigid body. Any internal
degrees of freedom can then be added. A typical application would be a molecular
crystal structure for which a CIF file exists from low-temperature analysis that is to be
refined at higher temperature using powder data allowing determination of changes
of location, rotation and torsion angles of the molecules.
In TOPAS, a point in space with fractional coordinates is given by the parameters
ua, ub and uc. Starting location and orientation of the rigid body are predetermined
by the coordinates. Torsions can be defined and refined as in the example above.
In the following example, the fractional coordinates of a C–CF3 group are copied
into the rigid body:

rigid
point_for_site C1 ua 0.2151 ub 1.0328 uc 1.0323
point_for_site C2 ua 0.2302 ub 0.9579 uc 0.9878
point_for_site F1 ua 0.2917 ub 1.0360 uc 1.0743
point_for_site F2 ua 0.1324 ub 1.0038 uc 1.0592
point_for_site F3 ua 0.1841 ub 1.1079 uc 1.0137
…

In order to refine the rotation of the three fluorine atoms F1, F2 and F3 around the C1-
C2 axis, the following code must be added:

Rotate_about_points(@ 0, C2, C1, "F1 F2 F3")

6.3.4 Spherical coordinates

In some cases, such as molecules like C60, it can be beneficial to define a rigid
body in spherical coordinates. In order to introduce spherical coordinates in the
rigid body definition, they must be transformed to Cartesian coordinates in the
definition.
A simple rigid body, consisting of only one atom would look like6:

prm radius 1.0 min 0

prm theta 10 min 0 max 180
prm phi 10 min 0 max 360

6 In TOPAS, the following constants are predefined: Pi = π, Deg =2 π /360, Deg_on_2 = π/360, Rad =
360/2 π.
 6.3 Rigid bodies 169

rigid
point_for_site Ca ux = radius Sin(Deg theta) Cos(Deg phi);
uy = radius Sin(Deg theta) Sin(Deg phi);
uz = radius Cos(Deg theta);

or better defined as rotations:

…
rigid
point_for_site Ca
rotate theta qc 1 radius
rotate phi qa = Sin(Deg theta); qb = -Cos(Deg theta); radius

It is recommended to write a macro to facilitate the conversion.

6.3.5 Using internal rotations and translations to create a rigid body

Finally one can use tricks such as duplications, translations and rotations of sites
with built-in TOPAS macros to rapidly build very complex molecules without com-
plex trigonometry. As an example, let’s consider how to build a complex molecule
like the C10H8 molecule that consists of two hinged benzene rings as shown in
Figure 6.7.
First, a benzene ring is formed by duplicating a first point that is iteratively
rotated by 60° around the z-axis:

Figure 6.7: Rigid body of the final bent C10H8 molecule.

170 6 Restraints, constraints and rigid bodies

prm r1 1.3
prm r2 1.08
rigid
point_for_site C1 ux = r1;
Duplicate_rotate_z(C2, C1, 60)
Duplicate_rotate_z(C3, C2, 60)
Duplicate_rotate_z(C4, C3, 60)
Duplicate_rotate_z(C5, C4, 60)
Duplicate_rotate_z(C6, C5, 60)
…

This results in a regular benzene ring without hydrogen atoms (Figure 6.8).

Figure 6.8: Rigid body of a benzene ring without hydrogen atoms.

Four hydrogen atoms bonded to the carbon atoms C1…C4 are created by duplicating
one hydrogen atom at the position of C1 and translating it in the x-direction by the
C–H bond length. The following three hydrogen atoms are then formed by duplicat-
ing the first hydrogen atom that is iteratively rotated by 60° around the z-axis:

…
Duplicate_Point(H1, C1)
translate tx = r2; operate_on_points H1
Duplicate_rotate_z(H2, H1, 60)
Duplicate_rotate_z(H3, H2, 60)
Duplicate_rotate_z(H4, H3, 60)

This results in the first half of the molecule (Figure 6.9).

The other half of the molecule is created by duplicating the four C–H groups and
rotating the duplicated atoms by 140° around the vector formed by the C5 and C6
atoms. This produces the final molecule that was shown in Figure 6.7:
 6.3 Rigid bodies 171

Figure 6.9: Rigid body of a benzene ring with four hydrogen atoms.

…
Duplicate_Point(C21, C1)
Duplicate_Point(C22, C2)
Duplicate_Point(C23, C3)
Duplicate_Point(C24, C4)
Duplicate_Point(H21, H1)
Duplicate_Point(H22, H2)
Duplicate_Point(H23, H3)
Duplicate_Point(H24, H4)
Rotate_about_points(140, C5, C6, "C21 C22 C23 C24 H21 H22 H23 H24")

6.3.6 Z-matrix

The Z-matrix notation is commonly used by chemists to define a group of atoms

using an internal coordinate representation (Leach, 1996). Each atom in a molecule
is defined in terms of its label, bond length, bond angle and dihedral angle relative
to other atoms. The name arises because the Z-matrix always places the second
atom in a molecule directly along the Z axis from the first atom, which defines the
origin. The third atom is defined by a distance to the first or second atom and an
angle between the three atoms. The fourth and all consecutive atoms are defined by
a distance to one, an angle with two and a dihedral angle with three previously
defined atoms. Owing to this flexible definition there are various different ways of
setting up the Z-matrix. Particular care must be taken for cyclic compounds as the
172 6 Restraints, constraints and rigid bodies

closing of the ring is not defined (there is no requirement to specify a geometric

relationship between the last and the first atom), which sometimes leads to pro-
blems if individual bond lengths/angles are refined. A possible solution is the
introduction of dummy atoms (with zero occupancy) at the center of the ring that
can act as a kind of anchor.
Various tools are available that directly transform any molecule in CIF- or MOL-
format to Z-matrix format. Z-matrices can always be converted to Cartesian coordi-
nates and back, as the structural information contained is identical. In TOPAS, the
conversion of Z-matrix coordinates to Cartesian is as follows: (1) the first atom, if
defined using the z_matrix keyword, is placed at the origin; (2) the second atom is
placed on the positive z-axis; (3) the third atom is placed in the xz-plane.
We can illustrate the Z-matrix syntax in TOPAS using the regular ZrO6
octahedron discussed above in Z-matrix notation with the Zr atom located at
the origin. In a general line each site is specified by a distance relative to the
preceeding atom and two angles:

prm r 2 min 1.8 max 2.2

rigid
z_matrix Zr
z_matrix O1 Zr =r;
z_matrix O2 Zr =r; O1 90
z_matrix O3 Zr =r; O1 90 O2 90
z_matrix O4 Zr =r; O1 90 O2 180
z_matrix O5 Zr =r; O1 90 O2 270
z_matrix O6 Zr =r; O1 180 O2 0

A slightly more complicated example is a Fe2O7 double tetrahedron with the Fe–O–Fe
bond angle and the twist between the two tetrahedra as internal degrees of freedom
(Figure 6.10):

Figure 6.10: Rigid body description of a Fe2O7 double tetrahedron.

prm r 1.90
prm angle 120.00
prm twist 0.00
 6.3 Rigid bodies 173

rigid
z_matrix O1
z_matrix Fe1 O1 = r;
z_matrix O2 Fe1 = r; O1 109.5
z_matrix O3 Fe1 = r; O2 109.5 O1 120
z_matrix O4 Fe1 = r; O3 109.5 O2 240
z_matrix Fe2 O1 = r; Fe1 = angle;O2 180
z_matrix O21 Fe2 = r; O1 109.5 O2 = twist;
z_matrix O22 Fe2 = r; O1 109.5 O2 = twist + 120;
z_matrix O23 Fe2 = r; O1 109.5 O2 = twist + 240;

A third example is a rigid body for the cyclic pentamethyl-cyclopentadienyl C5(CH3)5–

anion, where three dummy atoms were defined to facilitate the definition of the inner
and outer carbon atom positions (Figure 6.11).

Figure 6.11: Rigid body of the pentamethyl-cyclopentadienyl C5(CH3)5– anion and three dummy atoms
X1, X2 and X3 in black.

rigid
z_matrix X2
z_matrix X1 X2 1
z_matrix C1 X2 1.42 X1 90
174 6 Restraints, constraints and rigid bodies

z_matrix C2 X2 1.42 X1 90 C1 72
z_matrix C3 X2 1.42 X1 90 C1 = 2 72;
z_matrix C4 X2 1.42 X1 90 C1 = 3 72;
z_matrix C5 X2 1.42 X1 90 C1 = 4 72;
z_matrix X3 C1 1 X2 90 X1 0.0
z_matrix C11 C1 1.50 X3 90 X2 180.0
z_matrix C21 C2 1.50 C1 126 X2 180.0
z_matrix C31 C3 1.50 C2 126 X2 180.0
z_matrix C41 C4 1.50 C3 126 X2 180.0
z_matrix C51 C5 1.50 C4 126 X2 180.0
z_matrix H11 C11 1.05 C1 109.5 X3 0.0
z_matrix H12 C11 1.05 C1 109.5 X3 120.0
z_matrix H13 C11 1.05 C1 109.5 X3 240.0
z_matrix H21 C21 1.05 C2 109.5 X3 0.0
z_matrix H22 C21 1.05 C2 109.5 X3 120.0
z_matrix H23 C21 1.05 C2 109.5 X3 240.0
z_matrix H31 C31 1.05 C3 109.5 X3 0.0
z_matrix H32 C31 1.05 C3 109.5 X3 120.0
z_matrix H33 C31 1.05 C3 109.5 X3 240.0
z_matrix H41 C41 1.05 C4 109.5 X3 0.0
z_matrix H42 C41 1.05 C4 109.5 X3 120.0
z_matrix H43 C41 1.05 C4 109.5 X3 240.0
z_matrix H51 C51 1.05 C5 109.5 X3 0.0
z_matrix H52 C51 1.05 C5 109.5 X3 120.0
z_matrix H53 C51 1.05 C5 109.5 X3 240.0

6.3.7 External degrees of freedom of a rigid body

A rigid group of atoms can be positioned uniquely in space by specifying six

external degrees of freedom: three translational parameters that define some
reference point of the group with respect to the origin of the crystallographic
coordinate system:
0 1
ta
t = @ tb A; (6:11)
tc

and three angles that define its orientation with respect to the three crystallographic
or Cartesian axes. The rotation angle is counter clockwise positive when looking
toward the origin of the coordinate system:
0 1
1 0 0
Ra ðωÞ = @ 0 cosω − sinω A
0 sinω cosω
 6.3 Rigid bodies 175

0 1
cosω 0 − sinω
B C
Rb ð ω Þ = @ 0 1 0 A (6:12)
sinω 0 cosω
0 1
cosω − sinω 0
Rc ðωÞ = @ sinω cosω 0 A. (6:12)
0 0 1

If the rigid body lies on a special position, some of these parameters will
have fixed values. In general, the number of independent positional para-
meters for a group of n atoms in crystal space is therefore reduced from 3n
to 6.
The conversion from a vector (e.g., atomic position) in Cartesian rigid body
coordinates s to fractional crystallographic coordinates u is given by:

u = M − 1 ðR  sÞ + t (6:13)

The rotation matrix R is the product of an arbitrary number of rotations around

the crystallographic or Cartesian axes. In TOPAS notation, the rotation around
the crystallographic a, b and c-axes (rx, ry and rz in degrees) of the rigid body is
defined by:

prm rx 0;
prm ry 0;
prm rz 0;
rotate rx qa 1
rotate ry qb 1
rotate rz qc 1

while rotation around the Cartesian i, j and k-axes of the rigid body is defined by:

prm rx 0;
prm ry 0;
prm rz 0;
rotate rx qx 1
rotate ry qy 1
rotate rz qz 1

For rotations around the crystallographic axes a predefined macro exists:

prm rx 0;
prm ry 0;
prm rz 0;
Rotate_about_axies( rx, ry, rz)

The translation along the unit cell axes (tx, ty and tz in fractional coordinates) is:
176 6 Restraints, constraints and rigid bodies

prm tx 0;
prm ty 0;
prm tz 0;
translate ta tx tb ty tc tz

or via the predefined macro:

prm tx 0;
prm ty 0;
prm tz 0;
Translate( tx, ty, tz)

In general, it is not recommended to limit the rotations from 0 to 360° or the

translations from 0 to 1. Doing so would require, for example, a parameter describing
a −4° rotation to refine from 0 to 356°. This is less likely to occur by least squares than
a 0 to –4° change. This is particularly important during structure determination as
discussed in Chapter 7.

6.3.8 Finding a starting orientation

One of the most difficult problems when setting up a rigid body is to define
its starting position and orientation. Sometimes the location of some or all of
the atoms of the rigid body is approximately known from structure determina-
tion or related compounds. Although it is mathematically possible, it is
usually impractical to hand-calculate starting values for translation, rotation
and torsions. An easier approach is to subject the degrees of freedom of the
rigid body to the global optimization method of simulated annealing to map
the body onto the known coordinates. This is done by minimizing the differ-
ence between the known approximate fractional coordinates and those of the
rigid body atoms. In the following example, the positions of the four atoms
C1, C2, C3 and N1 of the rigid body should match as closely as possible the
positions of the previously determined atoms CX1, CX2, CX3 and NX1. Since
we’re not refining against experimental data at this point, the only_penalties
keyword must be used (see Chapter 7 for more explanation of this method).

Auto_T(2)
…
Rotate_about_axies(@ 0, @ 0, @ 0)
Translate(x1 0.61961`_0.00021, @ -0.01596`_0.00084, =x1; : 0.61961`_0.00021 )
Rotate_about_points(@ 0, C1, C3, " C4")

Distance_Restrain(CX1 C1, 0.0, 0.25472, 0.0001, 10)

 6.3 Rigid bodies 177

Distance_Restrain(CX2 C2, 0.0, 0.22346, 0.0001, 10)

Distance_Restrain(CX3 C3, 0.0, 0.24508, 0.0001, 10)
Distance_Restrain(NX1 N1, 0.0, 0.19113, 0.0001, 10)

only_penalties

6.3.9 Rotation of rigid bodies around an arbitrary axis

Quite often molecules need to be rotated around arbitrary axes in space, for example C60
around <111> axes. The dependence of such a rotation on rotations around the Cartesian
axes x, y and z is usually non-linear. One option is to pre-rotate the entire rigid body in
order to align the rotation axes with one of the Cartesian/crystallographic axes, perform
the desired rotation around this axis and then back-rotate the rigid body into its original
orientation. For example, rotating axis r to the c-axis requires solving the equation
c r
= Ra ðωa ÞRb ðωb Þ (6:14)
c r
for ωa and ωb .
TOPAS offers an easier way of doing this. You simply define two atoms X1 and X2
on the rotation axis as dummy atoms (e.g., atom X1 at 0,0,0 and atom X2 at 1,1,1 for
the <111> axis; occ = 0) and list all atoms to be rotated in quotation marks using the
Rotate_about_points macro:

Rotate_about_points(@ 0, X1, X2, "…")

6.3.10 TLS matrices

One significant advantage of using rigid bodies is that thermal parameters can be
defined that refer to the group as a whole. In most Rietveld refinements it is sufficient
to refine a single overall isotropic temperature factor. For high-quality powder data
sets, it might be advantageous to try an anisotropic refinement of the thermal motion
by means of so-called TLS matrices. A comprehensive explanation and mathematical
treatment of TLS matrices is given by Willis and Pryor (1975) and by Downs (1992).
Following Downs, their meaning can be easily understood by separating the displa-
cement u of a rigid body into two parts, a translational component t and a librational
component λ × r:

u=t+λ×r (6:15)
178 6 Restraints, constraints and rigid bodies

While the translational component is the same for every part of the rigid body, the
librational component of motion represents that part of the rigid body that is located
at the end point of vector r. The vector λ is then defined as the direction of the
rotational axis with its magnitude representing the magnitude of its arc of rotation.
Both vectors originate from the same arbitrary origin. With respect to a Cartesian
basis, the equation can be written as:
0 1
tx
0 1 0 1B ty C
ux 1 0 0 0 rz ry B C 
@ uy A = @ 0 B tz C t
1 0 rz 0 rx AB C = ðI : AÞ (6:16)
B λx C λ
uz 0 0 1 ry rx 0 @ A
λy
λz

The atomic displacement parameters, U, for a given atom located at position r can be
obtained by taking the time average of the outer product # of the displacement
vector u for that atom:

U = hu#ui = ðt + λ × rÞ#ðt + λ × rÞ
= ht#ti + ðt#λ × rÞ + ðλ × r#tÞ + ðλ × rÞ#ðλ × rÞ
(6:17)
= ht#ti + ðt#ArÞ + ðAr#tÞ + ðAλÞ#ðAλÞ
≡ T + AS + SA + ALA

with the matrices T for the translational part, L for the librational part and S for the
screw motion (mixing part between translation and libration).
The mathematics of TLS matrices is not trivial and their use requires some care to
avoid divergence of the refinement. For practical purposes a few rules of thumb apply
when using a single rigid body:
– If the center of the rigid body is also the center of gravity, the components of the S
matrix (mixing term) can normally be set to zero.
– Refining only the diagonal components of the T matrix, constraining them to be
equal, and fixing all elements of L and S to zero is the same as refining an overall
isotropic temperature factor for the rigid body.
– Refining all independent elements of the T matrix, with L = S = 0, is the same as
refining an overall anisotropic temperature factor for the rigid body.
– For flat molecules like benzene rings, it is often sufficient to refine only the
components of the T matrix and the diagonal elements of the L matrix.
– If the rigid body is located on a symmetry element, some elements of the matrices
have to be set to zero or constrained accordingly.

TOPAS macros for using TLS matrices can, for example, be found in Halasz & Dinnebier
(2010). Figure 6.12 compares the adps from a single crystal refinement with those from
a TLS model using powder data. In this example the center of the rigid body was also
the center of mass (center of the shared C–C bond) and the S matrix was zero. Recently,
 6.3 Rigid bodies 179

Figure 6.12: Representations of atomic anisotropic displacement parameters of naphthalene as

calculated form molecular T and L matrices at 298 K. Overlap of ellipsoid from single crystal neutron
diffraction (pale gray) and X-ray powder diffraction (dark gray) (from Halasz & Dinnebier, 2010).

routines have become available (e.g., those by Jacco van der Streek via the TOPAS wiki)
to automatically generate TLS descriptions for TOPAS from .CIF files.

6.3.11 Example of a Rietveld refinement using rigid bodies

We now return to the Rietveld example of the double salt Mg(H2O)6RbBr3 that we
looked at in Section 2.10. We can use our knowledge of rigid bodies to greatly reduce
the number of refined parameters and still get a fit equivalent to that in Figure 2.52.
From the crystal structure of Mg(H2O)6RbBr3 (Figure 2.51) it is obvious that the
main building blocks of the structure are the two Mg(OH2)6 and RbBr6 octahedra
(Figure 6.13), which have different degrees of freedom.
To define the rigid body of a flexible octahedron, the Z-matrix notation is particu-
larly useful. Under the assumption that the center of symmetry and the angular frame
are preserved, the Mg(OH2)6 octahedron (Figure 6.13) might be distorted with up to
three different bond lengths turning it in an orthorhombic bisphenoid:

prm r1 2.10 min 2.0 max 2.2

prm r2 2.13 min 2.0 max 2.2
prm r3 2.08 min 2.0 max 2.2

rigid
z_matrix Mg1
z_matrix O1 Mg1 = r1;
180 6 Restraints, constraints and rigid bodies

z_matrix O2 Mg1 = r2; O1 = 90;

z_matrix O3 Mg1 = r3; O1 = 90; O2 90
z_matrix O4 Mg1 = r1; O1 180 O2 0
z_matrix O5 Mg1 = r2; O2 180 O3 0
z_matrix O6 Mg1 = r3; O3 180 O1 0

Figure 6.13: The two types of octahedra Mg(OH2)6 and RbBr6 present in the crystal structure of Mg
(H2O)6RbBr3. Symmetry equivalent atoms are shaded and marked by primes.

Due to the symmetry of the LT-phase, only part of the rigid body needs to be defined
as the remaining part is completed by symmetry (see Figure 6.13). Thus, the rigid
body for the Mg(OH2)6 octahedron can be built from one central magnesium atom
(Mg1) and three oxygen atoms (O1, O2 and O3). The central magnesium atom is
located on a fixed position with inversion symmetry. Only the three possible rotations
around the internal axes of the rigid body remain as external degrees of freedom. The
internal degrees of freedom are the length of the three principal axes. Thus only six
parameters are free to refine:

site Mg1 num_posns 4 x 0.50000 y 0.50000 z 0.50000 occ Mg+2 1 beq B1 1.8;
site O1 num_posns 8 x 0.58671 y 0.48270 z 0.63910 occ O-2 1 beq=B1;
site O2 num_posns 8 x 0.31996 y 0.39541 z 0.54806 occ O-2 1 beq=B1;
site O3 num_posns 8 x 0.40967 y 0.68389 z 0.53918 occ O-2 1 beq=B1;

prm r1 2.098 min 2.0 max 2.2

prm r2 2.131 min 2.0 max 2.2
prm r3 2.084 min 2.0 max 2.2
rigid
z_matrix Mg1
z_matrix O1 Mg1 =r1;
z_matrix O2 Mg1 =r2; O1 90
z_matrix O3 Mg1 =r3; O1 90 O2 90
 6.3 Rigid bodies 181

Translate_point_amount(Mg1, -) operate_on_points "O*"

rotate 15.877 qa 1 operate_on_points "O*"
rotate 18.146 qb 1 operate_on_points "O*"
rotate 30.729 qc 1 operate_on_points "O*"
Translate_point_amount(Mg1, +) operate_on_points "O*"

translate ta=1/2;
translate tb=1/2;
translate tc=1/2;

The rigid body for RbBr6 can be built from one central rubidium atom (Rb1)
and two bromine atoms (Br1 and Br2). Only one equatorial bromine atom is
necessary, as the other three are created by symmetry. For numerical reasons,
a very small deviation from the special position of the axial Br1 atom remains,
even if all symmetry restrictions for the rigid body rotation and translation are
obeyed. Therefore, the fractional site occupancy of Br1 must be set to ½. The y-
coordinate of the central rubidium atom and the rotation around the c-axis of
the rigid body remain as external degrees of freedom. The internal degrees of
freedom are the equatorial and the axial Rb–Br lengths. Thus a total of 4
parameters can be refined:

site Rb1 num_posns 4 x 0.50000 y -0.00225 z 0.75000 occ Rb+1 1 beq B2 1

site Br1 num_posns 8 x 0.50037 y -0.00186 z 1.00018 occ Br-1 0.5 beq=B2;
site Br2 num_posns 8 x 0.25361 y -0.25950 z 0.75012 occ Br-1 1 beq=B2;

prm r4 3.449 min 3.4 max 3.6

prm r5 3.476 min 3.4 max 3.6
rigid
z_matrix Rb1
z_matrix Br1 Rb1 =r4;
z_matrix Br2 Rb1 =r5; Br1 90

Translate_point_amount(Rb1, -) operate_on_points "Br*"

rotate 0 qa 1 operate_on_points "Br*"
rotate 0 qb 1 operate_on_points "Br*"
rotate @ 46.892 qc 1 operate_on_points "Br*"
Translate_point_amount(Rb1, +) operate_on_points "Br*"

translate ta=1/2;
translate tb @ -0.00225
translate tc=3/4;

Closer inspection of the crystal structure reveals that the parameter r4 (Rb1–Br1 bond
length) is directly correlated to the length of the c-axis, which can also be deduced
from the correlation matrix and the high standard deviation after preliminary refine-
ment. The number of degrees of freedom can therefore be reduced by equating r4 to
182 6 Restraints, constraints and rigid bodies

the length of the c-axis divided by four. In TOPAS notation this can be realized, for
example, by the equation prm r4=Get(c)/4;.
Note that when rigid bodies are used, the refinement of the coordinates is
exclusively controlled by the internal and external degrees of freedom of the rigid
body and not by refinement of the individual atoms in the atom list, where all
refinement flags on the coordinates must be turned off.

References
Baerlocher, C., Hepp, A., Meier, W.M. (1978): DLS-76: A program for the simulation of crystal
structures by geometric refinement. Institute of Crystallography and Petrography, ETH Zurich,
Switzerland.
Behrens, U., Dinnebier, R.E., Neander, S., Olbrich, F. (2008): Solid-state structures of base-free
rubidium and cesium pentamethylcyclopentadienides. Determination by high-resolution
powder diffraction. Organomet. 27, 5398–5400.
Coelho, A.A. Cheary, R. W. (1997): A fast and simple method for calculating electrostatic potentials.
Comp. Phys. Comm. 104, 15–22.
Dinnebier, R.E. (1999): Rigid bodies in powder diffraction. A practical guide. Powder Diffraction 14,
84–92.
Downs, R.T. (1992): Librational displacements of silicate tetrahedra in response to temperature and
pressure, Ph.D. thesis, Virginia Tech, Blacksburg VA 24061, USA (http://hdl.handle.net/10919/
39442).
Halasz, I., Dinnebier, R.E. (2010): Molecular motion by refinement of TLS matrices from high
resolution laboratory powder diffraction data: implementation in the program TOPAS and
application to crystalline naphthalene. MatSci. Forum 651, 65–69.
Leach, A.R. (1996): Molecular modelling: principles and applications, Addison Wesley Longman
Limited, Essex, England, 595 pages.
Scheringer, C. (1963): Least-squares refinement with the minimum number of parameters for
structures containing rigid-body groups of atoms. Acta Cryst. 16, 546–550.
Willis, B.T.M., Pryor, A.W. (1975): Thermal vibrations in crystallography, Cambridge University Press,
UK, 280 pages.
7 Solving crystal structures using the Rietveld
method

7.1 Introduction
Structure determination from powder diffraction data can in principle be divided into
algorithms working in reciprocal (diffraction) space (e.g., direct methods, Patterson
methods, charge flipping) and in direct (crystal) space (e.g., simulated annealing,
genetic algorithms). A comprehensive review on the different methods for structure
determination from powders is given in the textbook by David et al. (2006). Since that
text was published, the reciprocal space method of charge flipping (Oszlányi & Süto,
2004) has been successfully applied to powder diffraction data (Bärlocher et al.,
2007; Coelho, 2007), and is available in TOPAS.
The focus of this book is on Rietveld refinement where the minimization algo-
rithm is generally based on least squares with a limited radius of convergence.
Nevertheless, there is no reason, other than computational efficiency, why the mini-
mization algorithm used in Rietveld analysis can’t serve as a robust global optimizer
and this capability is now implemented in modern Rietveld codes. In this chapter, the
global optimization capabilities of TOPAS, with a focus on structure determination
from powder diffraction data, are described.

7.2 Local and global minima

Let us consider the monoclinic crystal structure (P21 ) of sodium-para-hydroxy-

benzoate (Dinnebier et al., 1999). We select the x- and the z-coordinate coordinate
of the single sodium atom as the only degrees of freedom. Figure 7.1 shows how the
RBragg (Section 2.8) changes as a function of the two coordinates (resolution ≈ 0.01
ÅÞ. The global minimum is located at x ≈ 0.947, and z ≈ 0.258 and is only slightly
lower than many of the local minima scattered over the complex hypersurface.
There is therefore a risk of a local minimization routine, such as that in a traditional
Rietveld least squares approach, getting trapped in one of the local minima. The
situation for nontrivial examples with many parameters is even more complicated
and gives a puckered hypersurface of the agreement factor (e.g., RBragg, Rwp, gof . . .)
with many local minima. In order to find the global minimum, the starting value
must be either in the well around the global minimum, or the hypersurface must be
probed in a more sophisticated way. Systematic grid search techniques are one
approach, but are limited to very simple problems as the number of calculations
necessary increases exponentially with the number of degrees of freedom.

https://doi.org/10.1515/9783110461381-007
184 7 Solving crystal structures using the Rietveld method

z 0.8
0.6
0.4
0.2

80

60

40
RBragg

20

0
0.2
0.4
0.6
x
0.8

Figure 7.1: RBragg value as a function of the x and z coordinate of the sodium atom in sodium-
para-hydroxybenzoate. All other parameters except for the linear scale factor were kept fixed. The
global minimum is denoted by a black sphere.

TOPAS uses the global optimization method of simulated annealing (Newsam

et al., 1992; Coelho, 2007). Just one macro is needed to switch between global
optimization mode and refinement mode making it a powerful extension of the
Rietveld method. In the following sections, the method of simulated annealing and
its implementation in TOPAS is explained in some detail.

7.3 The method of simulated annealing

The most common and most easily implemented global optimizer, though one of the
least efficient, is the Metropolis (Metropolis et al., 1953.) or simulated annealing (SA)
algorithm that tests multiple models against the diffraction data. The most usual
implementation is actually as a “regional” optimizer where the updates to para-
meters such as atomic position are constrained to be relatively close to the previous
values in such a way that the algorithm makes a random walk through the parameter
space. The algorithm can get out of a local minimum by “walking uphill” since
changes to the parameters that produce a worse agreement may be accepted accord-
ing to the Boltzmann criterion:
 7.3 The method of simulated annealing 185


Δχ2
−
kT
e (7:1)

with the Boltzmann constant k. The temperature T in this expression is a

fictitious temperature (i.e., it does not refer to any real temperature) and Δχ2 is
the change in the agreement produced by the trial update. The temperature
plays the role of tuning the probability of accepting a bad move. It is initially
chosen to have a high value, giving a high probability of escaping a local
minimum and allowing the algorithm to explore a large area of the parameter
space. Later during the run, the temperature is lowered (Figure 7.3), trapping
the solution into successively finer valleys in parameter space until it settles
into (hopefully) the global minimum (Figure 7.1). The calculation of the cost
function χ2 can be based on the entire profile, or on integrated intensities. For
the latter, the correlation between partially or fully overlapping reflections must
be taken into account (as outlined in Figure 7.2).
A possible cost function based on integrated intensities can be defined as a
double sum (David, 2004):
X Xh
 
i
χ2 = Is1 − cjFs1 j2 V − 1 s1, s2 Is2 − cjFs2 j2 (7:2)
s1 s2

over all reflections s1 and s2, with Is1 the observed intensity of reflection s1,
ðV − 1 Þs1, s2 the inverse of the correlation matrix reflecting the degree of overlap
between the reflections and cjFs1 j2 the calculated intensity of reflection s1, and c
is the scale factor.
Arbitrary parameters can be varied during SA. In the case of structure
determination these typically include internal and external degrees of freedom
(DOF) like translations (fractional coordinates or rigid body locations), rotations
(Cartesian angles, Eulerian angles or quaternions, describing the orientation of
molecular entities), torsion angles, fractional occupancies, temperature factors
and so on. In some cases it is useful to restrict certain parameters like torsion
angles to lie within reasonable values. Fractional or rigid body coordinates on
general positions should not be restricted to stay within one unit cell since this
could hinder free movement and, therefore, prevent the algorithm from finding
the correct solution.
At the beginning, the first χ2 is calculated from an initial, trial atomic1
structure, containing all atoms/molecules that are thought to be located inside
the unit cell, usually with the scale factor as the only variable. For structure
determination all unknown fractional or rigid body coordinates can initially be set

1 A trial structure can contain all atoms on arbitrary or partly known positions, where the latter can
be fixed or restrained.
186 7 Solving crystal structures using the Rietveld method

160 QuartzCu.xy_1 hkl_Phase 100.00 %

150
140
130
120
110
√(rel. intensity)

100
90
80
70
60
50
40
30
20
10
0
49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
2θ/°

Figure 7.2: Part of a Pawley-fit of a laboratory diffraction pattern (Cu-Kα radiation) of quartz showing
independent and partly overlapping reflections and the corresponding part of the correlation
matrix. The arrows point to groups of two and three overlapping reflections. Colours represent the
correlation between different variables on an artificial colour scale from black (0%) to white (100%
correlated).

to the origin (0, 0, 0). After variation of some or all of the DOF’s by the SA
algorithm, a new χ2new is calculated. According to the difference

Δχ2 = χ2new − χ2 (7:3)

two possibilities exist. If Δχ2 ≤ 0, the variation led to an improvement of the fit and is
therefore automatically accepted. If Δχ2 > 0, the fit is worse but the parameter shifts
are accepted with a probability according to eq. (7.1) that decreases with decreasing
temperature (or time through the SA run). In other words, the entire real space is
probed at the beginning of an SA where the “temperature” is high, whereas move-
ments are much more hindered once the temperature decreases until the entire
 7.4 The simulated annealing algorithm in TOPAS 187

system “freezes.” The actual temperature scheme and the type of movements might
be much more sophisticated in the actual algorithm.

1000
2

Average 𝜒2
T: best 𝜒 : average 𝜒

Best 𝜒
2

100
2

Temperature

10

104 105 106 107

Number of moves

Figure 7.3: Average and best χ2 (cost function) and “temperature” as a function of the number of
moves during a simulated annealing run according to the classical definition (from Dinnebier and
Müller, 2013; Copyright Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission).

Figure 7.3 shows a typical simulated annealing run for structure determination in
which the χ2 value falls dramatically in the first few thousand moves, indicating that
the scattering is dominated by the positioning of heavier atoms or large molecules.
Several million trial structures are usually generated before a minimum can be
reached. At the end of the simulated annealing run, Rietveld refinement can be used
to find the bottom of the global minimum valley more rapidly.

7.4 The simulated annealing algorithm in TOPAS

TOPAS typically uses a slightly different approach to “classical” SA in which Rietveld
refinement is performed at each step of the process. As such each individual step is
slower but fewer steps are required as each configuration will be least-squares
refined to the bottom of the local minimum in χ2. Simulated annealing is then
equivalent to a continuation of the refinement after convergence has been reached
using a specific “temperature” regime to scale parameter changes (Coelho, 2000). To
instruct TOPAS to continue refinement after convergence, the keyword:

continue_after_convergence

is used. In the simplest approach, several actions can be performed before continuing.
188 7 Solving crystal structures using the Rietveld method

(1) randomize_on_errors is a means of automatically randomizing parameters

based on the approximate errors in the parameters as given by:
sffiffiffiffiffiffiffiffiffiffiffiffi
0.02T 1
Δpi = Q SignðRandð − 1, 1ÞÞ , K= P
, (7:4)
KAii M
wm y2
m=1 obs, m

where T is the current temperature and Q is a scaling factor determined such that
convergence to a previous parameter configuration occurs 7.5% of the time on
average.
(2) rand_xyz adds a vector u = (Δx, Δy, Δz) to each site described by the fractional
coordinates (x, y, z), the direction of which is random and the magnitude in Å is: |u| =
T rand_xyz where T is the current “temperature” that can be set by:

Temperature !E

Only fractional coordinates xyz that are flagged as independent parameters are
randomized.
(3) val_on_continue2 supplies a means of changing the parameter value after the
refinement has converged in a user-controlled fashion. An example is given below:

site Pb1 x @ 0.1 val_on_continue=0;

y @ 0.2 val_on_continue=0;
z @ 0.3 val_on_continue=0;
occ Pb 1 beq 1

Here, the values of the fractional coordinates of a lead atom are reset to the origin
before any new refinement cycle during an SA run.
The simplest simulated annealing could thus look like:

continue_after_convergence
randomize_on_errors

This does not include a temperature regime. Setting up an optimum temperature

regime can be quite demanding. For most cases, the following macro can be used:

Auto_T(2)

2 When val_on_continue is defined then the corresponding parameter is not randomized according to
randomize_on_errors.
 7.4 The simulated annealing algorithm in TOPAS 189

which imposes a complex automated temperature program. It has shown to be

adequate for a wide range of simulated annealing examples.
There is a huge variety of keywords in the TOPAS scripting language to manip-
ulate and optimize the simulated annealing process. For example, some that speed
up SA include:

chi2_convergence_criteria 1E-5 ‘ Convergence is determined when the change in χ2 is

less than 1E-5 for three consecutive cycles
quick_refine 0.01 ‘ Removes parameters that influence χ2 in a smaller
manner than 0.01 during a refinement cycle

An in-depth discussion of all possibilities is beyond the scope of this book and the
prospective user is referred to the TOPAS technical reference.
For structure determination, one might consider a different weighting scheme
with stronger weighting of the more intense reflections, like:

weighting = 1 / Sqrt(Max(Yobs, 1));

For inorganic crystal structures, the identification of special positions is useful

during structure solution. This can be accomplished by a so-called “occupancy-
merge” procedure as proposed by Favre-Nicolin and Cerny (2004). For example, if
an atom refines to a position very close to a mirror plane it is likely that it actually lies
on the mirror. Its occupancy would then need to be halved as the atom and its mirror
image would be very close, effectively doubling the scattering from that site. In
TOPAS this is handled by considering sites as spheres with a radius r and updating
their occupancies if they approach more closely than 2r. The occupancy is automa-
tically adjusted to 1/(1 + intersection fractional volumes) so that any number of sites
can be merged. In the following example, special positions are identified when either
two oxygen or two lead atoms approach each other within a distance less than the
sum of their merging radii which are set to 0.7 Å:

occ_merge Pb* occ_merge_radius .7

occ_merge O* occ_merge_radius .7

An alternative to occ_merge is to refine the fractional occupancies during SA with a

minimum limit of zero and a maximum limit equivalent to the scattering power of the
strongest scatterer present. This increases the flexibility in structure determination
but requires later manual reassignment of atomic species.
SA runs usually continue until the maximum number of iterations defined by
iters is reached. Alternatively, the SA process can be stopped if e.g., the Rwp falls
under a user-defined limit like:
190 7 Solving crystal structures using the Rietveld method

iters = If(Get(r_wp) < 10, 0, 1000);

Usually no special algorithms are employed to prevent close contact of atoms or

molecules during the global optimization procedure. In general these are not neces-
sary, as the fit to the structure factors alone quickly moves the atoms or molecules to
regions of the unit cell where they do not grossly overlap with neighbouring molecules.
It is common to use a Pawley/Le Bail fit prior to SA to fix quantities like cell
parameters, peak shapes and background. This helps speed by reducing the number
of iterations for each convergence. In many Pawley/Le Bail fits the background at
high scattering angles, where lots of peak overlap occurs, correlates with the inten-
sity of the reflections and leads to a poor (low) background estimation. This may
prevent the SA from finding the correct crystal structure. It is therefore useful to
visually inspect and, if necessary, to correct the background before SA starts.

7.4.1 Example of a structure determination with simulated annealing

In the following example, the crystal structure of the mixed-valent oxide Pb3O4
(space group P42 =mbc, a = 8.81 Å, c = 6.57 Å) is determined by the global optimization
method of simulated annealing.
The powder pattern of Pb3O4 was measured with a Bruker D8-Advance powder
diffractometer (Cu-Kα1 radiation from a Ge(111)-Johannson primary beam monochro-
mator) in Bragg–Brentano geometry on a flat plate low background single crystal
sample holder at room temperature. The sample exhibits strong anisotropic line
broadening due to microstrain that can be empirically handled by a spherical
harmonics of fourth order applied to the Lorentzian fwhm (see Section 4.3.2). The
INP file for the Pawley fit (Figure 7.4) is given below:

360
340
320
300
280
Sqrt (rel. intensity)

260
240
220
200
180
160
140
120
100
80
60
40
20
0
–20
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120 125 130 135 140
2θ/°

Figure 7.4: Pawley fit of powder diffraction data from Pb3O4.

 7.4 The simulated annealing algorithm in TOPAS 191

iters 1000
xdd Pb3O4.raw
bkg @ 850.77 -581.44 611.07 -312.83 173.39 -81.07 1.37 3.36 38.58 54.48 50.61
start_X 10 finish_X 140
Specimen_Displacement(@, 0.00025`)
LP_Factor( 27.3)
Rs 217.5
Simple_Axial_Model(, 1)
Slit_Width( 0.1)
Divergence( 1)
lam
ymin_on_ymax 0.00001
la 1 lo 1.540596 lh 0.401844
hkl_Is
phase_name "Pb3O4 Pawley fit, Bruker D8 Advance, Bragg-Brentano"
LVol_FWHM_CS_G_L(1, 142.03880`, 0.89, 151.56239`,CSG, 203.37399`,CSL, 602.52323`)
prm p1 0.37733` min 0.0001
spherical_harmonics_hkl sh1
sh_order 4 load sh_Cij_prm {
y00 !sh1_c00 1.00000
y20 sh1_c20 -1.03954`
y40 sh1_c40 0.16276`
y44p sh1_c44p 1.48377`
}
lor_fwhm = Max(0.0001, sh1 p1 Tan(Th));
Tetragonal(@ 8.813233, @ 6.565034)
space_group "P42/mbc"
load hkl_m_d_th2 I
{
1 1 0 4 6.231897 14.20049 @ 23.87964
…
0 0 8 2 0.820629 139.65792 @ 45.94778
}

From the cell volume, the number of formula units per unit cell can be estimated to be
Z = 4, equivalent to 12 lead and 16 oxygen atoms. With respect to the possible site
symmetries, this could imply 1–4 oxygen and 2–3 unique lead positions. Here we
assume 2 oxygen and 2 lead positions, but it might become necessary to probe other
combinations.
To set up an SA run, the following changes to the INP file are performed:
– The macro Auto_T(2) is placed at the beginning of the INP file. All parameters
with a refinement flag set on are now subject to global optimization.
– All refinement flags are turned off.
– Switch from WPPF to Rietveld mode by replacing hkl_Is by str.
– Change the weighting scheme to weighting = 1/Sqrt(Max(Yobs, 1));
– Add a scale factor and turn its refinement flag on: scale @ 0.0001.
192 7 Solving crystal structures using the Rietveld method

– Add the atomic sites, placing them arbitrarily at the origin and turning the
refinement flags of the positional parameters on:
– site Pb1 x @ 0 y @ 0 z @ 0 occ Pb 1 beq 1
– site Pb2 x @ 0 y @ 0 z @ 0 occ Pb 1 beq 1
– site O1 x @ 0 y @ 0 z @ 0 occ O 1 beq 1.5
– site O2 x @ 0 y @ 0 z @ 0 occ O 1 beq 1.5
– Identify special positions by merging occupancies between lead atoms and
between oxygen atoms that come closer than 0.7 Å:
– occ_merge Pb* occ_merge_radius .7
– occ_merge O* occ_merge_radius .7
The updated input file after a successful SA run looks like:

iters 1000000
Auto_T(0.1)
xdd Pb3O4.raw
bkg 850.77 -581.44 611.07 -312.83 173.39 -81.07 1.37 3.36 38.58 54.48 50.61
start_X 10 finish_X 140
Specimen_Displacement(, 0.00025)
LP_Factor( 27.3)
Rs 217.5
Simple_Axial_Model(, 1)
Slit_Width( 0.1)
Divergence( 1)
lam
ymin_on_ymax 0.00001
la 1 lo 1.540596 lh 0.401844

weighting = 1 / Sqrt(Max(Yobs, 1));

str
phase_name "Pb3O4 SA, Bruker D8 Advance, Bragg-Brentano"
LVol_FWHM_CS_G_L( 1, 142.03880, 0.89, 151.56239,!CSG, 203.37399,!CSL, 602.52323)
‘ Spherical harmonics of 4th order for anisotropic Lorentzian microstrain broadening
prm !p1 0.37733 min 0.0001
spherical_harmonics_hkl !sh1
sh_order 4 load sh_Cij_prm {
y00 !sh1_c00 1.00000
y20 !sh1_c20 -1.04925`
y40 !sh1_c40 0.00982`
y44p !sh1_c44p 1.46390`
}
lor_fwhm = Max(0.0001, sh1 p1 Tan(Th));

space_group "P42/mbc"
Tetragonal( 8.813233, 6.565034)
r_bragg 4.44686519
scale @ 0.000222785226
site Pb1 x @ -0.16280 y @ 0.14335 z @ -0.51714 occ Pb 0.56788 beq 1
site Pb2 x @ 0.00200 y @ 0.49685 z @ 0.26607 occ Pb 0.28815 beq 1
site O1 x @ 0.07062 y @ 0.63909 z @ 0.00085 occ O 0.50301 beq 1.5
 7.4 The simulated annealing algorithm in TOPAS 193

site O2 x @ -0.19949 y @ 0.68456 z @ 2.24855 occ O 0.55529 beq 1.5

occ_merge Pb* occ_merge_radius .7
occ_merge O* occ_merge_radius .7
view_structure ‘ Launches the structure viewer during SA

During an SA run, the Rwp is used as the indicator for the progress of the global
optimization as shown in Figure 7.5. We see from this plot that a low Rwp solution was
found around eight times during the SA run. TOPAS automatically returns to the
lowest Rwp solution at the end of the run (note that the final point on Figure 7.5 is the
lowest Rwp) and this is written to the output file.

Figure 7.5: Rwp during the SA run on Pb3O4.

The best fit at the end of this process is shown in Figure 7.6, and the corresponding
crystal structure of Pb3O4 is shown in Figure 7.7.

360
340
320
300
280
Sqrt (rel. intensity)

260
240
220
200
180
160
140
120
100
80
60
40
20
0
–20
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120 125 130 135 140
2θ/°

Figure 7.6: Best Rietveld fit during SA on Pb3O4.

By inspecting the structure visually and looking at the fractional coordinates and
sites occupancies it’s clear that atoms have refined close to special positions. These
can be identified using, for example, the International Tables for Crystallography
Vol. A (Aroyo, 2016):
194 7 Solving crystal structures using the Rietveld method

Figure 7.7: Crystal structure of Pb3O4 after an SA run. Although atoms are formally located on general
positions, special positions can be easily seen by eye.

prm o2xx -0.19949

site Pb1 num_posns 8 occ Pb 1 x @ 0.16280 y @ 0.14335 z=1/2; beq 1 ‘ 8 h m..
site Pb2 num_posns 4 occ Pb 1 x=0 y=1/2; z=1/4; beq 1 ‘ 4 d 2.22
site O1 num_posns 8 occ O 1 x @ 0.07062 y @ 0.63909 z=0; beq 1.5‘ 8 h m..
site O2 num_posns 8 occ O 1 x =o2xx; y =-o1xx+1/2; z=1/4; beq 1.5‘ 8 g ..2

Although the SA algorithm in TOPAS is extremely powerful, there is no guarantee that

SA will find the global minimum. As a rule of thumb, crystal structures up to a
complexity of 20 structural degrees of freedom can frequently be solved if good quality
powder diffraction data with reflections up to at least d ≈ 1.75 Å (better 1.2 Å) are
available. The number of structural degrees of freedom can be greatly reduced by using
rigid bodies and/or fixed atomic coordinates of partly known crystal structures (e.g.,
from charge flipping).

References
Aroyo, I.M. (ed.) (2016): International tables for crystallography. Volume A: space-group symmetry.
IUCr Series, Kluwer Academic Publishers, Dordrecht, The Netherlands.
Baerlocher, C. McCusker, L.B., Palatinus, (2007): Charge flipping combined with histogram matching
to solve complex crystal structures from powder diffraction data. Z. Kristallogr. l. 222, 47–53.
Coelho, A.A., (2000): Structure solution by simulated annealing. J. Appl. Cryst. 33, 899–908.
Coelho, A.A. (2007): A charge-flipping algorithm incorporating the tangent formula for solving
difficult structures. Acta Cryst. A36, 400–406.
 References 195

David, W.I.F (2004): On the equivalence of the Rietveld method and the correlated integrated
intensities method in powder diffraction. J. Appl. Cryst. 37, 621–628.
David, W.I.F., Shankland, K., McCusker L.B., Baerlocher, Ch. (2006): Structure determination from
powder diffraction data. IUCr Monographs on Crystallography, Oxford University Press, UK.
Dinnebier, R.E., Von Dreele, R., Stephens, P.W., Jelonek, S., Sieler, J. (1999). Structure of sodium
para-hydroxybenzoate, NaO2C-C6H4OH by powder diffraction: application of a phenomenologi-
cal model of anisotropic peak width. J. Appl. Cryst. 32, 761–769.
Dinnebier, R.E., Müller, M. (2013): Modern Rietveld refinement, a practical guide. Chapter 2 in Modern
Diffraction Methods, Mittemeijer, E. and Welzel, U. (eds.) Wiley-VCH Verlag GmbH & Co. KG
Weinheim, Germany.
Favre-Nicolin, V. & Cerny, R. (2004): Fox: modular approach to crystal structure determination from
powder diffraction. Mater. Sci. Forum 443–444, 35–38.
Metropolis, N., Rosenbluth, A.W., Rosenbluth, M.N., Teller, A.H. (1953): Equation of state calculations
by fast computing machines. J. Chem. Phys. 21, 1087–1092.
Newsam, J.M., Deem, M.W., Freeman, C.M. (1992): Direct space methods of structure solution from
powder diffraction data. NIST Special Publication 846 Accuracy in Powder Diffraction II, Prince,
E. and Stalick,]. K. (eds.), 80–91.
Oszlányi, G., Süto A. (2004): Ab initio structure solution by charge flipping. Acta Cryst. A60, 134–141.
8 Symmetry mode refinements

8.1 Introduction
Many materials undergo phase transitions as a function of external variables such as
temperature, pressure or changes in their chemical environment. Powder diffraction
is a particularly powerful tool for studying these transitions as it is relatively easy to
design cells for in situ or operando studies, and one doesn’t have to worry about
issues such as crystals shattering, which is a major difficulty in similar single crystal
experiments.
We can broadly divide phase transitions into two classes: reconstructive and
non-reconstructive. During a reconstructive transition there are large changes in the
bonding and sufficient atomic rearrangement such that there is no simple relation-
ship between the structures of the two phases. An example might be a CsCl structure
transforming to rock salt under pressure. If one is analyzing powder diffraction data
before and after the phase transition, the data need to be treated as two separate
problems.
At a non-reconstructive phase transition the changes in structure are more
subtle. They could involve small movements in atomic positions away from high-
symmetry sites, such as the cooperative tilting of octahedra illustrated in Figure 8.1;
they could involve metal atoms moving away from the center of an otherwise
undistorted coordination polyhedron; they could involve atomic sites that are a
50:50 random mixture of two elements at high temperature ordering on cooling; or
they could involve the ordering of magnetic moments as discussed in Chapter 9.
Understanding this type of phase transition is an extremely important area of
condensed matter science as they are often associated with changes in physical
properties such as changes from insulator to metal to superconductor, paraelectric
to ferroelectric, second harmonic generation inactive to active or paramagnet to anti/
ferro/ferri-magnet.
Analyzing the powder diffraction pattern of a sample before and after a non-
reconstructive phase transition can again be tackled as two separate problems.
However, the close structural relationship between the two phases means that there
are often more efficient ways of determining and understanding the structures of low-
symmetry phases (often called child phases; more formally hettotypes) relative to their
high-symmetry parent (aristotype). We discuss some of these methods in this chapter.

8.2 Symmetry lowering phase transitions

Non-reconstructive phase transitions often involve a material changing from a high-

symmetry parent structure to a lower-symmetry child structure. The structures will

https://doi.org/10.1515/9783110461381-008
 8.2 Symmetry lowering phase transitions 197

Cooling Cooling
Pt

Fe
Disordered, fcc Ordered, L10

Figure 8.1: Left: a phase transition involving octahedral tilting in WO3. Right: a phase transition
involving site ordering in FePt (A1 (fcc) → L10 or CuAu type).

therefore have a group–subgroup relationship. High-symmetry forms are usually

associated with high temperature (or low pressure) and low symmetry forms with
low temperature (or high pressure), though there is no strict thermodynamic
requirement for this. Materials therefore typically lose symmetry on cooling. We
will use WO3 as an example throughout this chapter. Even though it has never been
observed experimentally, one can imagine a high-symmetry Pm
3m form of WO3 with
perfectly regular WO6 octahedra sharing corners with 180° W–O–W bond angles
between them (Figure 8.1 left or Figure 8.2). The structure has W on the 1a Wyckoff
site at (0, 0, 0) and O on 3d at (½, 0, 0) and therefore no refineable coordinates. On
cooling, WO3 undergoes a series of phase transitions that can be described as W
moving away from the center of the WO6 octahedra (a second order Jahn–Teller
distortion) and tilts of the WO6 octahedra around different crystallographic direc-
tions leading to non-180° W–O–W bond angles. The structures encountered on
cooling are depicted in Figure 8.2. Even a material as compositionally simple as
WO3 can have a remarkable structural complexity, and seven different phases have
been structurally characterized.

??K 1173K 1073K

Pm3m P4/nmm P4/ncc P21/c

993K 623K ~230K ~190K

Pbcn P21/n P1 Pc

Figure 8.2: The phase transitions of WO3. Blue WO6 octahedra undergo different internal distortions
and coupled rotations.
198 8 Symmetry mode refinements

If we want to describe the structure of each of the phases of WO3 using a

conventional crystallographic approach, we need to determine: (1) the unit cell or
basis of each phase; (2) the space-group symmetry; (3) the appropriate origin choice;
(4) the fractional coordinates of each atom; and (5) decide which coordinates are free
to change by symmetry. For simple cases we can use International Tables to help us.
Under the entry for each space group there is a list given of “maximal isomorphic
subgroups of lowest index” and “maximal non-isomorphic subgroups.” Isomorphic
subgroups are those with the same space group as the parent but a larger unit cell (so
there are fewer symmetry elements per unit volume). The term index gives the ratio
between the number of symmetry operations in the group and subgroup. The term
maximal means that there are no other groups between the parent and child, and this
is guaranteed by the term “lowest index.” Non-isomorphic subgroups have a differ-
ent space group. Under heading I, International Tables list translationengleiche
subgroups (gleiche = German for same) that involve losses of rotational symmetry
and under II klassengleiche subgroups, where translational symmetry elements are
lost. Category IIa lists subgroups with the same (conventional) unit cell or basis,
which arise by loss of centring, and IIb those with a larger unit cell.
For example, the highest symmetry characterized structure of WO3 has
space-group type P4/nmm with a unit cell of ~√2a × ~√2a × ~a relative to the
hypothetical cubic Pm
3m parent (we will omit the “approximately equal to”
signs and a from hereon). To describe the structure we need to specify a W at
(¼,¼, z) [z ≈ 0] and an O at (¼,¼, z) [z ≈ ½]. There are therefore two free unit
cell parameters and two free fractional coordinates. The distortion relative to the
cubic phase is due to W moving away from the center of the WO6 octahedron
parallel to the c-axis. The next change that occurs on cooling is a tilt of
octahedra around the c-axis that lowers the symmetry to P4/ncc with a unit
cell that is twice as large along c (i.e., √2 × √2 × 2). This space group is listed in
International Tables as a maximal non-isomorphic subgroup of P4/ncc of type
IIb. We need to specify W at (¼,¼, z) [z ≈ ¼], O at (x, –x, ¼) [x ≈ 0] and O at
(¼,¼, z) [z ≈ 0] to describe the structure. The change in the W z-coordinate from
~¼ to ~0 between the phases is due to an origin shift between the two space
groups. Even for this relatively straightforward example there is significant work
in relating the two structures and determining an appropriate structural
description.
If we look at the (hypothetical) highest temperature transition from Pm
3m to
the observed P4/nmm structure we hit a further complication: P4/nmm is not
listed as a maximal subgroup of Pm
3m and we would have to describe the
relationship between the two using intermediate groups. Even for a “simple”
compound like WO3 the possible transformations between different subgroups of
the parent can become bewilderingly complex. In fact, if you work out all the
possible subgroups between a Pm
3m parent and a 2 × 2 × 2 P1 child structure,
there are 1427 possibilities!
 8.3 Symmetry adapted distortion modes 199

8.3 Symmetry adapted distortion modes

8.3.1 Overview

Fortunately there are powerful group theory tools to help understand this type of
structural distortion. These are sufficiently well-developed that they can be applied
to structural problems without the user having a deep theoretical understanding of the
methods. This is particularly true if one of the web-based tools such as the ISODISTORT
software suite (http://stokes.byu.edu/iso/isodistort.html) or the Bilbao crystallo-
graphic server (http://www.cryst.ehu.es/) is used. In the following section, we will
discuss the ISODISTORT symmetry adapted distortion mode approach implemented in
TOPAS, and cover enough of its language to help understand its practical application.
The basic idea of the symmetry mode approach is that we describe a distorted
child structure in terms of the parent structure plus the amplitudes of a family of
symmetry-motivated distortions (Campbell, 2006 & 2007; Kerman, 2012). These
distortions define the pattern and magnitude of atomic displacements and hence
the symmetry of the material. We will find that this approach has a number of
advantages:
1. We can often describe child structures with far fewer parameters than using
traditional xyz coordinates. This can help us to extract detailed structural infor-
mation from powder data.
2. The magnitude of the parameters we use will naturally be defined in the range
0 to ~4, where a large magnitude indicates a large structural change. This can help
us focus on parameters that are most important in describing the phase transition.
3. Our new parameter set will allow us to rapidly explore different possible symme-
tries for child structures in a systematic way.
4. We can use well-developed, freely available, user-friendly computational infra-
structure to take most of the hard work out of the process and help eliminate
otherwise difficult-to-avoid errors.

The first of these advantages is probably the most powerful and most useful. It arises
from the fact that the free energy change associated with any phase transition can be
expressed in terms of the order parameters of the irreducible representations of the
parent space-group symmetries (we’ll explain the meaning of this sentence later;
Landau and Lifshitz, 1970). Often, only a small number of these order parameters
(frequently only one) dominate the free energy. These are the parameters we will use
in the symmetry mode description, and we will therefore be refining parameters of
direct energetic significance.
Before we delve into the details it might be useful, particularly for those from a
chemistry background, to consider a simple molecular analogy to the process. H2O is
a bent molecule with C2v symmetry and a 104.5° bond angle. If we wanted to describe
a distorted H2O molecule (e.g., one where the O–H bonds were of different lengths
200 8 Symmetry mode refinements

lowering the symmetry, or where the bond angle was different) we could specify xyz
coordinates for each of the atoms. Alternatively, we could remember the vibrational
properties of a H2O molecule that are usually described using normal modes. In this
approach we find that H2O has a symmetric bond stretch that has symmetry label
(irrep) a1, an asymmetric bond stretch (b2) and a bending mode (a1). If we wanted to
describe a statically distorted molecule we could do so in terms of the undistorted
(parent) coordinates plus the amplitude of one or more frozen vibrational modes.
Note that the a1 symmetric stretch and a1 bend would not lower the symmetry from
C2v, but the b2 stretch would. a1 is called the totally symmetric irreducible
representation.

8.3.2 Definitions and terminology

The symmetry operations of a space group are conventionally described by a set

of matrix operations called a representation. When these are expressed in their
mathematically simplest block diagonal form this is called an irreducible repre-
sentation or irrep. A representation has an order defined as the number of
matrices and a dimension defined by the dimension of the matrices. A phase
transition can always be associated with a k point or wave vector in reciprocal
space. This k point describes where superlattice reflections will be seen follow-
ing the phase transition. For example, a phase transition associated with k point
(½, 0, 0) will give a child with extra (h/2, k, l) superlattice reflections or a
doubled a-axis. For a given parent space group, specifying a k point defines a
set of irreps. Each of these maps the symmetry elements onto a set of irreducible
matrices. When a material undergoes a phase transition some of its symmetry
elements are lost and some remain. Those that remain define what is called the
isotropy subgroup or distortion symmetry. Since each irrep describes a set of
parent symmetries that can be broken, they give us a “recipe” for lowering
symmetry. The language used is that we superpose one (or more) irreps on
the parent space group and this takes us to a specific isotropy subgroup. Within
a given isotropy subgroup, many different distortions are possible due to the
different values that the individual degrees of freedom can adopt.
One convenient way to think about any distortion is in terms of a distortion
vector. For the simple case of a single atom moving away from a high-symmetry
site in 1, 2 or 3 dimensions this is very intuitive. If we’re thinking of a more complex
case involving multiple atoms then we can use the same idea, but the distortion
vector is harder to visualize as it lies in a higher dimensional “distortion space” (more
formally the carrier space in which the matrices of the space group representation
operate). The distortion vector is invariant under the symmetry operations of the
isotropy subgroup. As with any vector, if we can define a basis for it, we can describe
it in terms of vector components along each of the basis vector directions. In a
 8.3 Symmetry adapted distortion modes 201

traditional crystallographic approach we use parameters such as the unit cell, xyz
fractional coordinates or site occupancies as the basis. In the symmetry mode
approach we take advantage of the fact that we can use irreps, which are simply
linear combinations of traditional crystallographic parameters, as a complete and
unique basis for the distortion vector.
There’s one final aspect of the distortion vector that it is useful to understand
called the order parameter direction or OPD. An OPD is a specific direction (or sub-
space) of the generalized distortion space within which each distortion vector repre-
sents structures with the same symmetry. Remember that each different vector within
this subspace represents a different specific distortion. A three dimensional irrep has
its most general OPD expressed as (a; b; c). This will give rise to a distortion symmetry
known as the kernel. The kernel is the lowest distortion symmetry associated with an
irrep. The same three-dimensional irrep may have a one-dimensional OPD where, for
example, b and c happen to be zero or b = –a expressed as (a; 0; 0) or (a; –a; 0),
respectively. These will lead to an intermediate distortion symmetry that is a super-
group of the kernel and a subgroup of the parent. A simple analogy to this is to think
of moving a single atom in a cubic structure away from (0, 0, 0). If we move it in a
general direction to (x, y, z), we will destroy certain symmetry elements; if we move it
to (x, x, x), we would retain more symmetry. The variable parameters of an OPD are
called branches.
In the kernel we can define a distortion mode as a vector component along one of
the irrep basis vectors; in one of the higher symmetries it may be a linear combination
of different irrep basis vectors. The ISODISTORT definition of an order parameter is a
distortion vector along a specific OPD of a specific irrep at a specific k-point of the
parent symmetry. The parameters we use to specify it are the individual mode
amplitudes (one for each branch of the OPD) and these are therefore order parameter
components.
We are now at the point where we’ve defined the various terms involved in a
symmetry mode description and have found that we have a way of defining structural
parameters (mode amplitudes) that relate directly to the order parameters used to
express the free energy change at a phase transition. We will find in the examples
below that these parameters often describe meaningful structural changes in a
material such as rotations of polyhedral groups, recognizable distortions of polyhe-
dra or Jahn–Teller distortions moving metals away from high-symmetry coordination
environments. As such they can be extremely useful in describing structural changes.

8.3.3 Symmetry mode labels and TOPAS implementation

To recap, we’ve discussed that the recipe to define an isotropy subgroup or

distortion symmetry is to choose a k point that defines a set of irreps. Selecting
an irrep will produce a series of possible order parameter directions ranging
202 8 Symmetry mode refinements

from high to low (the kernel) distortion symmetries. Each of these produces a set
of symmetry modes whose amplitudes define a specific structural distortion. The
irrep chosen defines the distortion symmetry (it is called a primary irrep),
though it will often give rise to additional irreps (called secondary irreps).
These secondary irreps alone may not destroy sufficient symmetry to lead to
the isotropy subgroup in question. When describing modes it is useful to give
each a unique label. A small section of TOPAS INP code exemplifying these
labels and showing how mode amplitudes relate to xyz fractional coordinates is
given below for the room temperature WO3 structure that we will discuss later.
The code was automatically generated using the procedure in Section 8.4.1 on
the ISODISTORT webserver:

'{{{mode definitions
'etc
prm a7 1.12853 min -4.00 max 4.00 'Pm-3m[1/2,1/2,1/2]R4+(a,0,b)[O:d]Eu(a)
prm a8 1.11797 min -4.00 max 4.00 'Pm-3m[1/2,1/2,1/2]R4+(a,0,b)[O:d]Eu(b)
'etc
‘}}}
'{{{mode-amplitude to delta transformation
'etc
prm O_1_dx = +0.03125*a8 - 0.03125*a10 + 0.03125*a11 - 0.03125*a12;: 0.03494
prm O_1_dy = -0.04419*a13 + 0.04419*a14;: 0.00000
prm O_1_dz = +0.03125*a7 + 0.03125*a9 + 0.03125*a11 + 0.03125*a12;: 0.03527
'etc
'}}}
'{{{distorted parameters
'etc
prm O_1_x = 1/4 + O_1_dx;: 0.28494
prm O_1_y = 3/4 + O_1_dy;: 0.75000
prm O_1_z = 3/4 + O_1_dz;: 0.78527
'etc
'}}}
'{{{mode-dependent sites
'etc
site O_1 x = O_1_x;:0.28494 y = O_1_y;:0.75000 z = O_1_z;:0.78527 occ O 1 beq 0
'etc
'}}}

At the top of the script two displacive symmetry modes are defined with TOPAS
parameter names a7 and a8 and amplitudes 1.13 and 1.12 (the 22 other modes have
been omitted for brevity). In the second section of the file, we see that a8 feeds into
parameter O1_dx that defines a displacement of this site in fractional coordinates.
a8 also feeds into O2_dx, O3_dy and O4_dy (omitted here). The next section has
equations defining the final coordinates of each site in terms of the position derived
from the undistorted parent plus the mode-dependent shifts. Finally, these values are
used to specify the atomic sites. Most of the time these equations don’t need to be
 8.3 Symmetry adapted distortion modes 203

read by the user, so they are normally hidden away in jEdit (https://community.dur.
ac.uk/john.evans/topas_academic/jedit_main.htm) '{{{. . . '}}} folds by default.
After each mode amplitude we see the mode label in ISODISTORT language.
This lists the parent symmetry (Pm
3m) the k-point (here (½, ½, ½) or the R-point of
Pm
3m), the three-dimensional R4+ irrep using Miller and Love (1967) notation and
the two-dimensional OPD [(a, 0, b)]. This is followed by the parent Wyckoff site of
the atom in question and the spectroscopic label for the Wyckoff site point group
irrep that induces the distortion. In some cases an additional order parameter
number is appended (_1, _2) to distinguish different modes with the same local
point group irrep. The final labels “a” and “b” distinguish the two branches of the
two-dimensional order parameter.

8.3.4 Symmetry mode description of WO3

We can see how these ideas work in practice using our WO3 example. The first structural
distortion on cooling WO3 is an off-centring of the W inside the WO6 octahedra. This
gives rise to superstructure peaks in the powder pattern at (h/2, k/2, l) positions,
corresponding to the M point of reciprocal space. The subsequent distortions involve
tilting of octahedra and give rise to (h/2, k/2, l/2) type superstructure reflections
corresponding to the R point.
If we take the parent structure and superpose an M point irrep then there
are seven choices: M1+ , M2+ , M3+ , M4+ , M5+ , M3− and M5− ; each would lower
symmetry in a different way. M3− is the appropriate choice (we discuss later
how you would determine this experimentally) and there are then six choices for
the OPD as in the table below. Each results in different degrees of freedom in
the child resulting in a different space group, basis and origin combination. The
first part of each line (e.g., P1, P2) is an OPD label and the other parts are as
described above. The most general OPD is S1(a; b; c) that has a doubled cell and
space group I222 (the kernel). This child has three different M3− modes and we
can explore what each one does using the graphical tools in ISODISTORT.1 We
find that they displace W by different amounts along the a-, b- or c-axis of the
parent cubic cell. If we choose identical amplitudes for each of the modes then
W atoms move parallel to the three-fold axis of the parent cell and the symmetry

1 Assuming you have some familiarity with ISODISTORT, you can explore these distortions using the
default SrTiO3 example: On the ISODISTORT home page click “Get started quickly with a cubic
perovskite parent”. On the “ISODISTORT: search” page select “M” as the k point in method 2 and
click OK. On the pull down menu of the “ISODISTORT: irreducible representation” page select “M3−”
and click OK. On the “ISODISTORT: order parameter direction” page select S1 and click OK. On the
“ISODISTORT: distortion” page select “view distortion”. Move the M3− slider bars for Ti sites. This
mimics WO3, which we can think of as a perovskite without an A site cation.
204 8 Symmetry mode refinements

is higher. This mimics the third OPD choice P3(a; a; a) and corresponds to


space-group symmetry I 43m. This is an example of how the symmetry mode
basis lets you easily explore different symmetries and how they are related.
Similarly, two modes with equal amplitude move W along parent <110> direc-
tions corresponding to the second choice (P2(a; 0; a); space-group symmetry I4/
mcm) and a single mode moves W along a single axis [P1(a;0;0), space group
P4/nmm with basis ({(1,1,0),(–1,1,0),(0,0,1)})]. This is the actual distortion
observed above 1173 K with the √2 × √2 × 1 cell discussed above.

P1 (a;0;0) 129 P4/nmm, basis={(1,1,0),(-1,1,0),(0,0,1)}, origin=(0,1/2,0), s=2, i=6, k-active=

(1/2,1/2,0)
P2 (a;0;a) 140 I4/mcm, basis={(0,0,2),(2,0,0),(0,2,0)}, origin=(3/2,0,1/2), s=4, i=12, k-active=
(1/2,1/2,0),(0,1/2,1/2)
P3 (a;a;a) 217 I-43m, basis={(2,0,0),(0,2,0),(0,0,2)}, origin=(1/2,1/2,1/2), s=4, i=8, k-active=
(1/2,1/2,0),(1/2,0,1/2),(0,1/2,1/2)
C1 (a;0;b) 72 Ibam, basis={(0,0,2),(2,0,0),(0,2,0)}, origin=(1/2,0,1/2), s=4, i=24, k-active=
(1/2,1/2,0),(0,1/2,1/2)
C2 (a;b;a) 121 I-42m, basis={(0,0,2),(2,0,0),(0,2,0)}, origin=(1/2,1/2,1/2), s=4, i=24, k-active=
(1/2,1/2,0),(1/2,0,1/2),(0,1/2,1/2)
S1 (a;b;c) 23 I222, basis={(2,0,0),(0,2,0),(0,0,2)}, origin=(1/2,1/2,1/2), s=4, i=48, k-active=
(1/2,1/2,0),(1/2,0,1/2),(0,1/2,1/2)

The next two distortions that occur on cooling involve rotations of the WO6
octahedra. If we superpose R4+ distortions on the parent we get the following
OPD choices:

P1 (a;0;0) 140 I4/mcm, basis={(1,1,0),(-1,1,0),(0,0,2)}, origin=(0,0,0), s=2, i=6, k-active=

(1/2,1/2,1/2)
P2 (a;a;0) 74 Imma, basis={(1,0,1),(0,2,0),(-1,0,1)}, origin=(0,0,0), s=2, i=12, k-active=
(1/2,1/2,1/2)
P3 (a;a;a) 167 R-3c, basis={(-1,1,0),(0,-1,1),(2,2,2)}, origin=(0,0,0), s=2, i=8, k-active=
(1/2,1/2,1/2)
C1 (a;b;0) 12 C2/m, basis={(0,0,-2),(0,2,0),(1,0,1)}, origin=(0,1/2,1/2), s=2, i=24, k-active=
(1/2,1/2,1/2)
C2 (a;a;b) 15 C2/c, basis={(-1,2,-1),(-1,0,1),(1,0,1)}, origin=(0,1/2,1/2), s=2, i=24, k-active=
(1/2,1/2,1/2)
S1 (a;b;c) 2 P-1, basis={(0,1,1),(1,0,1),(1,1,0)}, origin=(0,0,0), s=2, i=48, k-active=
(1/2,1/2,1/2)

In a similar way to W displacements, an (a; b; c) OPD has three branches each of

which rotates octahedra around a single <100> axis of the parent cell. An (a; a; a) OPD
corresponds to rotations around the three-fold axis and (a; 0; a) to rotations around
<110>.
We can then understand the P4/ncc high temperature phase of WO3 (Figure
8.2) in terms of superposing M3− (a; 0; 0)⊕R4+ (b; 0; 0) on the parent and the
subsequent P21/c phase in terms of M3− (a; 0; 0)⊕R4+ (c; b; b). We can think of
these irreps as being primary and sufficient to define the symmetry of the child
 8.4 Examples of symmetry mode refinements 205

phase. When they are active other secondary irreps are also activated (X4− , X5− ,
M5− , R3+ , R5+ ). These may or may not have significant amplitude in the distorted
structure.

8.4 Examples of symmetry mode refinements

8.4.1 WO3 850 °C P4/ncc and 25 °C P21/n Rietveld refinements

The script below is an INP file for a symmetry mode refinement of laboratory powder
diffraction data collected on P4/ncc WO3 at 850 °C:

xdd d8_03901_850c_03.xy
finish_X 90
x_calculation_step = Yobs_dx_at(Xo); convolution_step 4
bkg @ 209.298187 -54.7924369 13.8094009 -1.59425645 8.68555976 1.26939144
LP_Factor(!th2_monochromator, 0)
CuKa2(0.0001)
Specimen_Displacement(height,-0.06930)
Simple_Axial_Model(axial, 8.50263)

str
space_group P4/ncc:2 'transformPp a,b,c;0,0,0
a lpa 5.280882
b lpa 5.280882
c lpc 7.847197
al 90.00000
be 90.00000
ga 90.00000
scale @ 0.000242874444
r_bragg 11.0005319
TCHZ_Peak_Type(pku,-0.007,pkv,0.009,pkw,-0.00437,!pkz,0.0,pky,0.141,!pkx,0.0)
'{{{mode definitions
prm a1 0.52070 min -2.00 max 2.00 'Pm-3m[1/2,1/2,0]M3-(a;0;0)[W:a:dsp] T1u(a)
prm a2 0.01301 min -2.00 max 2.00 'Pm-3m[1/2,1/2,0]M3-(a;0;0)[O:d:dsp] A2u(a)
prm a3 0.75857 min -2.83 max 2.83 'Pm-3m[1/2,1/2,1/2]R4+(a,0,0)[O:d:dsp] Eu(a)
'}}}
'{{{mode-amplitude to delta transformation
prm W_1_dz = +0.06649*a1;: 0.03462
prm O_1_dx = +0.04702*a3;: 0.03567
prm O_2_dz = +0.06649*a2;: 0.00087
'}}}
'{{{distorted parameters
prm !W_1_x = 1/4;: 0.25000
prm !W_1_y = 1/4;: 0.25000
prm W_1_z = 3/4 + W_1_dz;: 0.78462
prm O_1_x = 1/2 + O_1_dx;: 0.53567
prm O_1_y = 1/2 - O_1_dx;: 0.46433
206 8 Symmetry mode refinements

prm !O_1_z = 1/4;: 0.25000

prm !O_2_x = 1/4;: 0.25000
prm !O_2_y = 1/4;: 0.25000
prm O_2_z = 0 + O_2_dz;: 0.00087
prm !W_1_occ = 1;: 1.00000
prm !O_1_occ = 1;: 1.00000
prm !O_2_occ = 1;: 1.00000
'}}}
'{{{mode-dependent sites
site W_1 x = W_1_x; y = W_1_y; z = W_1_z; occ W = W_1_occ; beq bval -0.99268
site O_1 x = O_1_x; y = O_1_y; z = O_1_z; occ O = O_1_occ; beq bval -0.99268
site O_2 x = O_2_x; y = O_2_y; z = O_2_z; occ O = O_2_occ; beq bval -0.99268
'}}}

This is a relatively straightforward case in which there are three symmetry mode
amplitudes (a1–a3) that can be refined. As discussed above, a1 and a2 describe
distortions of the WO6 octahedra and a3 describes a tilt of the octahedra around
the parent c-axis. The mode amplitudes are converted to fractional atomic coordi-
nates by the equations given. The plots in Figure 8.3 show Rietveld fits using three
different models. The lowest plot shows the fit if no mode amplitudes are refined and
cell parameters are exactly related to their cubic values. We see that this model fails
to produce the observed peak splittings and position shifts due to the distortion to
tetragonal symmetry. Interestingly, the only peak in the correct position is the (202)
at ~41° 2θ – you might like to think why this is! This fit is 100% equivalent to fitting
the parent Pm
3m model to the data, and is another example of how a symmetry mode
refinement lets us easily simulate fitting a higher-symmetry model.

60000
50000
Rel. intensity

40000

30000

20000
10000

20 25 30 35 40 45 50 55 60 65
2θ/°

Figure 8.3: Rietveld fits to data collected on P4/ncc WO3 at 850 °C using the three models discussed
in the text.

The middle plot shows the fit when the cell parameters are allowed to distort but
mode amplitudes are kept at zero. We see that most of the strong peaks are now
correctly fitted, but several superstructure reflections have zero calculated intensity.
 8.4 Examples of symmetry mode refinements 207

This model is equivalent to fitting a P4/mmm model with a 1 × 1 × 1 unit cell (one of the
maximal non-isomorphic subgroups of Pm
3m). The top fit shows the effect of refining
the three mode amplitudes. We now get an excellent fit to the diffraction data.
You might notice from the TOPAS script that the temperature factors have refined
to negative values, which makes no physical sense. This is because the data were
collected in Bragg–Brentano geometry on a small sprinkled sample that was not
larger than the illuminated area at all angles, and not densely packed. Hence, the
criteria for validity of the usual 2θ-dependent intensity scaling (see Section 2.2) were
not fulfilled. As a result, there are small systematic errors in the peak intensities as a
function of 2θ that are “mopped up” by the negative temperature factor.
This example demonstrates advantages 2–4 of the symmetry modes that we
introduced in Section 8.3.1. It doesn’t, however, give any reduction in the number
of parameters used (advantage 1): in the symmetry mode description we have three
refineable mode amplitudes and in a conventional description we would have three
xyz parameters; all of them are necessary to describe the structure.
We can, however, demonstrate parameter reduction using the more complex
room temperature P21/n structure. The lower plot in Figure 8.4 shows a Rietveld fit
to a 30 °C data set using a conventional crystallographic description with 24 xyz
parameters (Rwp = 9.0%) and the upper plot shows a fit using just five symmetry
mode amplitudes (Rwp = 9.2%). We can see that the five-parameter fit is essentially as
good as a 24-parameter fit. The five modes are sufficient to describe the key distor-
tions of the material and are: tungsten X5− and M3− (W off-center displacements in
different directions), 2 × oxygen R4+ and a 1 × oxygen M3+ (octahedral rotations). Even
if we try and do simultaneous fitting of X-ray and neutron data (which is more
sensitive to O displacements) we get a very good fit with five parameters and an
excellent fit with seven. We’ll discuss these combined refinements in more detail in
the following sections.

18000
16000
14000
Rel. intensity

12000
10000
8000
6000
4000
2000
0
20 25 30 35 40 45 50 55 60 65
2θ/°

Figure 8.4: Rietveld fits to P21/n WO3 data collected at 30 °C using (lower plot) a 24-parameter
conventional xyz refinement and (upper plot) a five-parameter symmetry mode refinement.
208 8 Symmetry mode refinements

8.4.2 Using a genetic algorithm (GA) to identify symmetry

In the previous section, we discussed using symmetry modes in cases where

we’ve already known the symmetry of the material and which distortions are the
most important. How would we determine this in unknown cases?
One way of demonstrating what’s possible is to forget for a moment that we
know that the true symmetry of room temperature WO3 is P21/n, and think
whether it’s possible to solve the structure with no symmetry assumptions. If
WO3 had a 2 × 2 × 2 unit cell and P1 symmetry, there would be 96 possible mode
amplitudes. If we could detect which of these are actually necessary to describe
the diffraction data, we would essentially simultaneously determine the space
group of the material (via which branches of the OPD are needed) and the
specific distortions of the structure (from the amplitudes of the modes). It turns
out that there are several ways to do this. One is to use a Genetic Algorithm or
GA. In this approach we can flag whether modes in a structural model are active
(“turned on” or 1) or inactive (“turned off” or 0) with a string of 96 1s and 0s
that we can think of like a piece of genetic code (001011000 . . .). We then test
the model against the diffraction data by performing a refinement in which the
selected mode amplitudes (prms a1 to a96 in TOPAS syntax) are allowed to refine
and recording the Rwp of each model. We can do this for a “population” of
several different models and rank the “fitness” of each according to the Rwp. We
can then apply some of the rules of Darwinian selection to our population of
structures to produce a new offspring population. The rules might include
“survival of the fittest” (deleting some of the highest-Rwp structures from the
population); “mating” some of the better models (e.g., taking half of the string of
0s and 1s from one model and half from another to make a new structure) or
“mutating” (randomly changing some of the 1s to 0 and vice-versa). We can then
refine each member of this offspring population to obtain its Rwp, and keep
repeating the process until we get a good fit to the diffraction data. If we add
a small penalty to the Rwp fitness criterion that scales with the number of modes
turned on, we can rapidly find structures that are simple in terms of their
number of degrees of freedom, but still provide a good fit to the diffraction data.
There are simple PYTHON scripts published for performing this procedure,
and within a few minutes on a normal desktop computer we find that for WO3 a
seven-mode model will give an excellent simultaneous fit to both X-ray and
neutron data. From the resulting model we can use an algorithm such as
FINDSYMM (http://stokes.byu.edu/iso/findsym.php) to identify that the true sym-
metry of this P1 symmetry mode description is P21/n. The method can be explored
via an online tutorial at: http://community.dur.ac.uk/john.evans/topas_work
shop/tutorial_GA_wo3.htm.
 8.4 Examples of symmetry mode refinements 209

8.4.3 Using inclusion runs to identify symmetry

We can also tackle the problem of identifying the important modes in WO3 in a way
that is conceptually simpler (Lewis, 2018). Figure 8.5 shows the Rietveld Rwp from a
simple “experiment” in which each of the 96 possible modes amplitudes was refined
individually and the one that gave the lowest Rwp fit to X-ray and neutron data
selected (here it was an oxygen R4+ mode describing an octahedral tilt that reduced
Rwp from 30.1 with no mode amplitudes refined to 27.0%). We then simultaneously
refined this mode amplitude along with each of the other 95 in turn (i.e., a series of
two-parameter fits). This process was repeated until all the 96 mode amplitudes were
included in the fit. As such, the plot in Figure 8.5 shows the lowest Rwp achievable
using this recipe as a function of the number of modes in the model. We see that Rwp
reaches low values when just seven modes are used and that improvements beyond
this are much less marked. Reassuringly, these modes turn out to be the same ones
identified by the GA and uniquely define the structure as P21/n. We also find that this
plot is fully “reversible” and we see an identical behavior of Rwp versus the number of
modes if we start with 96 modes on and sequentially eliminate the ones with the
lowest impact on Rwp.

35
Symmetry mode inclusion
30 xyz inclusion
xyz exclusion
25
Rwp/%

20

15

10

5
0 10 20 30 40 50 60 70 80 90
Number of mode amplitudes or xyz coordinates in model

Figure 8.5: Rwp values obtained as a function of the number of structural degrees of freedom using
either symmetry modes or xyz parameters. Values from simultaneous fit to X-ray and neutron data.

Figure 8.5 also includes data from a similar “experiment” using a traditional xyz
parameter set instead of mode amplitudes (with equivalent constraints to limit
coordinate shifts to those in the mode analysis). We again see that Rwp reduces as
more parameters are included in the model, but that we need far more parameters to
achieve an equivalent fit to the data. If we repeat this experiment or perform the
reverse mode exclusion procedure, we always obtain similar plots to those in
Figure 8.5, though the precise shape of the plot can change as the differentiation
210 8 Symmetry mode refinements

between the contributions of each parameter to Rwp is less marked. It is therefore

much easier to make a small “mistake” in the sequence of parameter choice and
follow a different pathway across the R-factor hypersurface.
Figure 8.5 provides a quantitative feel for how much more efficient symmetry
mode refinements can be in terms of the number of parameters employed.

8.4.4 Exhaustive searching using a subgroup tree

A final way to tackle the WO3 problem is to use a feature of ISODISTORT that
produces a subgroup tree containing an exhaustive list of all the possible child
structures. If one considers a parent structure (here Pm
3m WO3 in a 1 × 1 × 1 cell) and
a low-symmetry base structure with sufficient degrees of freedom to fully explain the
diffraction data (for example a 2 × 2 × 2 description in P1), it is possible to derive all
the possible subgroups between the parent and base child. In this case there are
1427 of them. Each of these candidate structures can be tested against the diffraction
data to determine its Rwp, giving a comprehensive “map” of the Rwp surface for all
possible models. It is then relatively simple to make informed decisions about the
“best” structural model, where best typically means the simplest description that
describes all the important features of the diffraction data. This approach has been
described in the literature (Lewis et al, 2016) and there are online tutorials describing
how to do it in practice (http://community.dur.ac.uk/john.evans/topas_workshop/
tutorial_exhaustive_symmetry.htm).

8.4.5 Other order parameters – occupancies, rotational modes and magnetism

In this chapter we’ve focussed on examples where the order parameters are symmetry
modes that describe atomic displacements and have the tensor properties of micro-
scopic (because they act on atoms) polar vectors. It’s possible to use a similar
language to describe site occupancies (microscopic scalar parameters), magnetic
ordering (microscopic axial vectors or pseudo-vectors) or the rotational properties
of rigid molecular groups (microscopic axial vectors). Some of these ideas are
explored in Chapter 9 on magnetic Rietveld refinement.

8.5 Example of a Rietveld refinement using symmetry modes

We’ll finish with the Rietveld example of the double salt Mg(H2O)6RbBr3 that we
looked at previously in Sections 2.10 and 6.3.11. In our earlier analyses on this
material we didn’t discuss the fact that it undergoes a phase transition at high
temperature. With our knowledge of symmetry modes we can develop a simple
 8.5 Example of a Rietveld refinement using symmetry modes 211

description of the low-temperature (LT) structure relative to the parent high tempera-
ture (HT) structure using only a small number of parameters.
If the Mg(OH2)6 unit of the room temperature phase of Mg(H2O)6RbBr3 was
replaced by an atom A, the resulting ARbBr3 structure would be analogous to a
pffiffiffi pffiffiffi
2 × 2 × 2 supercell of a distorted cubic perovskite with a = 6.94189 Å, and Rb on
the B site. The phase transition takes Mg(H2O)6RbBr3 to a cubic (Pm
3m) HT structure
analogous to an ideal perovskite. The HT structure exhibits four-fold orientational
disorder of the Mg(OH2)6 octahedral tilts as illustrated in the right-hand panel of
Figure 8.6. These become orientationally ordered in the LT structure, so that the tilts
alternate along the long axis of the supercell.

LT HT
a a
c c

Mg+2
Br–1
Rb+1
O–2

b b
a a

Figure 8.6: Comparison of the LT (left) and HT (right) crystal structures of Mg(H2O)6RbBr3 as viewed
along the b-axis (top) and the c-axis (bottom) showing green RbBr6 and brown Mg(OH2)6 octahedra
(from Dinnebier et al., 2008).

In the following paragraphs we will go through a cookbook recipe for a symmetry

mode refinement of LT-Mg(H2O)6RbBr3. We will again use the ISODISTORT
(Campbell, 2006) package to calculate the symmetry modes for us, and to automati-
cally produce a TOPAS STR file. We will reference the LT structure to an idealized
212 8 Symmetry mode refinements

hypothetical cubic parent structure in which the Mg(OH2)6 octahedra are ordered and
aligned parallel to the cell axes, though the actual HT structure is orientationally
disordered (Figure 8.6).
Firstly, the ordered high-symmetry parent structure should be uploaded to the
ISODISTORT server from a CIF-file. ISODISTORT includes a viewer (Figure 8.7) to
check the structure has read in correctly.

parent structure of HT-Mg

Figure 8.7: Screen shot of ISODISTORT showing ordered cubic (Pm3m)
(H2O)6RbBr3.

We can use ISODISTORT method 3 “Search over arbitrary k points for specified space
group and lattice” to create a description of the LT C2=c structure. In the “Select space
group symmetry box” we specifiy “15 C2-c C2h-6.” We then choose the basis for the
child structure as:
0 1 0 10 1
a 1 1 0 a
@bA ¼ @ 1 1 0 A@ b A (8:1)
c child 0 0 2 c parent

and click on OK. On the next screen we select the second choice, “15 C2/c, basis =
{(1,1,0),(–1,1,0),(0,0,2)}, origin=(0,0,1/2), s = 2, I = 24,” and click OK. On the follow-
ing page we can either select “View distortion” to visualize the structure in the child
setting (Figure 8.8), “TOPAS.STR” to write out an STR file in TOPAS format or “View
diffraction” to visualize the calculated powder pattern of the material. In the “View
distortion” option you can adjust the values of the horizontal slider bars to inves-
tigate the effect of each symmetry mode on the structure. For example, you should
see that changing the amplitudes of the O Γ4+ or X1− modes gives rise to rotations of
the Mg(OH2)6 octahedra. In the “View diffraction” window you can see how these
modes influence the powder diffraction data.
 8.5 Example of a Rietveld refinement using symmetry modes 213

Figure 8.8: Screenshot of ISODISTORT showing the parent and child structures of Mg(H2O)6RbBr3 and
all displacive and strain modes allowed by symmetry. The parent cell is in pink and the child cell in
purple.

For our purposes we can write out a TOPAS.STR file to allow us to refine the LT
structure relative to the HT parent. The contents of TOPAS.STR replace the normal
structural coordinates in the INP file that was given in Section 2.10. We find that there
are 13 possible displacive mode amplitudes which correspond to the 13 free fractional
coordinates in a conventional description. After refinement the relevant part of the
INP file reads:

str
phase_name double_salt
scale @ 8.50799022e-005
a @ 9.641327
b @ 9.865327
c @ 13.786095
be @ 90.08790
CS_L(@, 802.58724)
CS_G(@, 819.13894)
Strain_G(@, 0.12267)
space_group C12/c1

'{{{mode definitions
prm a1 -0.03397 min -2.00 max 2.00 'Pm-3m[0,1/2,0]X4-(0;0;a)[Br:d:dsp] Eu(a)
prm a2 0.07955 min -2.00 max 2.00 'Pm-3m[0,1/2,0]X5-(0,0;0,0;a,-a)[Br:d:dsp] A2u(a)
214 8 Symmetry mode refinements

prm a3 -0.18402 min -2.00 max 2.00 'Pm-3m[0,1/2,0]X5-(0,0;0,0;a,-a)[Br:d:dsp] Eu(a)

prm a4 0.03595 min -1.41 max 1.41 'Pm-3m[0,1/2,0]X5-(0,0;0,0;a,-a)[Rb:a:dsp] T1u(a)
prm a5 0.34429 min -3.46 max 3.46 'Pm-3m[0,0,0]GM1+(a)[O:f:dsp] A1(a)
prm a6 0.01731 min -2.45 max 2.45 'Pm-3m[0,0,0]GM3+(a,0)[O:f:dsp] A1(a)
prm a7 2.36162 min -2.83 max 2.83 'Pm-3m[0,0,0]GM4+(a,-a,0)[O:f:dsp] E(a)
prm a8 0.15038 min -2.83 max 2.83 'Pm-3m[0,0,0]GM5+(a,b,b)[O:f:dsp] E(a)
prm a9 0.02173 min -2.83 max 2.83 'Pm-3m[0,0,0]GM5+(a,b,b)[O:f:dsp] E(b)
prm a10 -1.61761 min -2.83 max 2.83 'Pm-3m[0,1/2,0]X1-(0;0;a)[O:f:dsp] E(a)
prm a11 0.07180 min -2.83 max 2.83 'Pm-3m[0,1/2,0]X4-(0;0;a)[O:f:dsp] A1(a)
prm a12 0.17459 min -2.83 max 2.83 'Pm-3m[0,1/2,0]X5-(0,0;0,0;a,-a)[O:f:dsp] E_1(a)
prm a13 0.33838 min -2.00 max 2.00 'Pm-3m[0,1/2,0]X5-(0,0;0,0;a,-a)[O:f:dsp] E_2(a)
'}}}
'{{{mode-amplitude to delta transformation
prm Br_1_dx = +0.03601*a2 + 0.03601*a3;: -0.00376
prm Br_1_dy = -0.03601*a2 + 0.03601*a3;: -0.00949
prm Br_1_dz = -0.03601*a1;: 0.00122
prm Rb_1_dy = +0.07203*a4;: 0.00259
prm O_1_dx = +0.02079*a5 + 0.01470*a6 + 0.02547*a8 + 0.02547*a10 - 0.02547*a11;: -0.03179
prm O_1_dy = -0.02079*a5 - 0.01470*a6 + 0.02547*a8 + 0.02547*a10 + 0.02547*a11;: -0.04295
prm O_1_dz = +0.01801*a7 + 0.01801*a9 + 0.02547*a12;: 0.04737
prm O_2_dx = +0.02079*a5 + 0.01470*a6 + 0.02547*a8 - 0.02547*a10 + 0.02547*a11;: 0.05427
prm O_2_dy = -0.02079*a5 - 0.01470*a6 + 0.02547*a8 - 0.02547*a10 - 0.02547*a11;: 0.03579
prm O_2_dz = +0.01801*a7 + 0.01801*a9 - 0.02547*a12;: 0.03848
prm O_3_dx = -0.03601*a7 + 0.03601*a9;: -0.08426
prm O_3_dy = -0.05093*a13;: -0.01723
prm O_3_dz = +0.02079*a5 - 0.02940*a6;: 0.00665
'}}}
'{{{distorted parameters
prm !Mg_1_x = 0;: 0.00000
prm !Mg_1_y = 1/2;: 0.50000
prm !Mg_1_z = 0;: 0.00000
prm Br_1_x = 3/4 + Br_1_dx;: 0.74624
prm Br_1_y = 1/4 + Br_1_dy;: 0.24051
prm Br_1_z = 3/4 + Br_1_dz;: 0.75122
prm !Br_2_x = 0;: 0.00000
prm !Br_2_y = 0;: 0.00000
prm !Br_2_z = 0;: 0.00000
prm !Rb_1_x = 0;: 0.00000
prm Rb_1_y = 0 + Rb_1_dy;: 0.00259
prm !Rb_1_z = 1/4;: 0.25000
prm O_1_x = 0.35500 + O_1_dx;: 0.32321
prm O_1_y = 0.14500 + O_1_dy;: 0.10205
prm O_1_z = 0 + O_1_dz;: 0.04737
prm O_2_x = 0.35500 + O_2_dx;: 0.40927
prm O_2_y = 0.14500 + O_2_dy;: 0.18079
prm O_2_z = 1/2 + O_2_dz;: 0.53848
prm O_3_x = 1/2 + O_3_dx;: 0.41574
prm O_3_y = 0 + O_3_dy;: -0.01723
prm O_3_z = 0.85500 + O_3_dz;: 0.86165
'}}}
 8.5 Example of a Rietveld refinement using symmetry modes 215

'{{{mode-dependent sites
site Mg_1 x = Mg_1_x; y = Mg_1_y; z = Mg_1_z; occ Mg 1 beq bm 2.19023
site Rb_1 x = Rb_1_x; y = Rb_1_y; z = Rb_1_z; occ Rb 1 beq =bm;
site Br_1 x = Br_1_x; y = Br_1_y; z = Br_1_z; occ Br 1 beq bbr 1.67475
site Br_2 x = Br_2_x; y = Br_2_y; z = Br_2_z; occ Br 1 beq =bbr;
site O_1 x = O_1_x; y = O_1_y; z = O_1_z; occ O 1 beq bo 0.07540
site O_2 x = O_2_x; y = O_2_y; z = O_2_z; occ O 1 beq =bo;
site O_3 x = O_3_x; y = O_3_y; z = O_3_z; occ O 1 beq =bo;
'}}}

From the TOPAS script above we can see that two modes have particularly large
amplitude: a7 (O Γ4+ ) and a10 (O X1− ). These are therefore the most important in
describing the differences between the parent and child structures. We can visualize
the effect of all the refined mode amplitudes in ISODISTORT using the so-called
method 4 “Mode decomposition of a distorted structure by loading a distorted structure
(child structure) from CIF file.” This gives an easy way of comparing the LT and HT
structures. In the window shown in Figure 8.9 we select the child basis from a pull
down list of options then click OK.

Figure 8.9: Screenshot of ISODISTORT showing the Method 4 window for comparing the HT and LT
structures.
216 8 Symmetry mode refinements

The child structure can then be visualized as shown in Figure 8.10. By clicking
“animate” in the ISODISTORT viewer we can play a simple movie between the parent
and child structures. By pressing “z” on the keyboard we can zero all mode ampli-
tudes; by pressing “r” we can return them all to the values refined for the LT structure.

Figure 8.10: Screenshot from ISODISTORT showing the child structure of Mg(H2O)6RbBr3 and the
magnitudes of all displacive and strain modes allowed by symmetry.

It is worth noting that the distortions we called rotations above will in fact distort
polyhedra to some extent. For example, the Mg(OH2)6 X1− rotations move oxygen
atoms perpendicular to the Mg─O bonds so actually cause a slight expansion of the
octahedra. To properly describe rotations of fully rigid bodies, a more sophisticated
approach is needed using so-called rotational rigid body modes (Müller et al., 2014).
Here a rotational axis of the rigid body is defined by a vector in space with the length
of the vector describing the magnitude of rotation. This procedure is more complex
and beyond the scope of this book. The interested reader is referred to the paper by
Müller et al. (2014), which contains a description of the application of rigid body
modes in TOPAS using our Mg(H2O)6RbBr3 example, or the work of Liu et al. (2018).

References
Campbell, B.J., Stokes, H.T., Tanner D.E., Hatch, D.M. (2006): ISODISPLACE: An internet tool for
exploring structural distortions. J. Appl. Cryst., 39, 607–614.
 References 217

Campbell, B.J., Evans, J.S.O., Perselli, F., Stokes, H.T. (2007): Rietveld refinement of structural
distortion-mode amplitudes. IUCr Comput. Commun. Newsletter, 8, 81–95.
Dinnebier, R.E., Liebold-Ribeiro, Y., Jansen, M. (2008): The low and high temperature crystal
structures of [Mg(H2O6)]XBr3 double salts (X = Rb, Cs). Z. Anorg. Allg. Chem. 634, 1857–1862.
Kerman, S., Campbell, B.J., Satyavarapu, K.K., Stokes, H.T., Perselli, F., Evans, J.S.O (2012): The
superstructure determination of displacive distortions via symmetry-mode analysis. Acta
Cryst. A 68, 222–234.
Landau, L.D., Lifschitz, E.M. (1970): Lehrbuch der theoretischen Physik V. Statistische Physik.
Akademie Verlag, Berlin, D.
Lewis, J.W., Payne, J.L., Evans, I.R., Stokes, H.T., Campbell, B.J., Evans, J.S.O. (2016): An exhaustive
symmetry approach to structure determination: phase transitions in Bi2Sn2O7. J. Am. Chem. Soc.
138, 8031–8042.
Lewis, J.W. (2018): Symmetry methods for understanding structures of inorganic functional materials.
PhD thesis, Durham University, UK.
Liu, H., Zhang, W., Halasyamani, P.H., Stokes, H.T., Campbell, B.J., Evans, J.S.O., Evans, I.R. (2018):
Understanding the behavior of the above-room-temperature molecular ferroelectric 5-6-
dichloro-2-methylbenzimidazole using symmetry adapted distortion mode analysis. J. Am.
Chem. Soc. 140, 13441–13448.
Miller, S.C., Love, W.F. (1967): Tables of irreducible representations of space groups and
co-representations of magnetic space groups, Pruett Press, Boulder (USA).
Müller, M., Dinnebier, R.E., Dippel, A., Stokes, H.T., Campbell, B.J. (2014): A symmetry-mode
description of rigid-body-rotations in crystalline solids: a case study of Mg[H2O]6RbBr3. J. Appl.
Cryst. 47, 532–538.
9 Magnetic refinements

9.1 Introduction
One significant advantage of neutron over X-ray (powder) diffraction is its sensitivity
to the magnetic structures of materials. For magnetically ordered materials neutrons
can therefore determine both the nuclear structure and the configuration of magnetic
moments. The most general and powerful approach for describing magnetic struc-
tures uses the crystallographic cell to describe the nuclear structure and a Fourier
series to describe magnetic moments, which often have a larger periodicity. This
allows an elegant and unified description of both commensurate and incommensu-
rate magnetic structures, which is described in detail elsewhere (Von Dreele &
Rodriguez-Carvajal, 2008). The current TOPAS implementation of magnetic diffrac-
tion relies, however, on the use of a magnetic unit cell and is restricted to commen-
surate structures (though incommensurate structures can be approximated using
large supercells). We will, therefore, restrict our discussion to this method. We give a
brief introduction to magnetic scattering, introduce some of the key ideas of magnetic
symmetry and Shubnikov groups, highlight some potential pitfalls in magnetic
structure analysis and then discuss different ways to perform magnetic refinements
in TOPAS.

9.2 Magnetic diffraction

Thermal neutrons produced by either reactor or spallation sources have wavelengths
in the 1 to 4 Å range and are ideally suited for diffraction experiments. Even longer or
shorter wavelengths (cold and hot neutrons) can be produced using moderators of
appropriate temperature. Neutrons have a spin of ½ and an intrinsic magnetic
moment of 1.9 nuclear magnetons, meaning that they will interact with the magnetic
moments produced by unpaired electrons in materials. When these moments1 order
below a critical temperature2 to give ferro-, ferri- or antiferromagnetic long range
order we therefore see magnetic contributions to diffraction patterns. The magnitude
of the effect is significant, and for many transition metal or rare earth containing
samples magnetic and nuclear intensities can be comparable.
We will only consider the case of unpolarized neutrons where the beam contains
all neutron spin orientations. In this situation the structure factor (ignoring Debye–
Waller terms discussed in Section 2.2.3) can be expressed (Bacon, 1975) as the sum of
two terms – one describing the nuclear scattering (see eq. (2.21)) and one the

1 We will often use the term “spin” for brevity.

2 The Curie temperature, TC, for ferromagnets and Néel temperature, TN for antiferromagnets.

https://doi.org/10.1515/9783110461381-009
 9.2 Magnetic diffraction 219

magnetic. In cases where the nuclear and magnetic cells are different (see below), it’s
important that the two terms are appropriately scaled:


2
jF ðsÞj2 = jFnuc ðsÞj2 +
Fmag ðsÞ







X  
2
X  
2





=
bj e2πisxj
+
qj pj ðsÞe2πisxj
. (9:1)

j

j



2
The second term,
Fmag ðsÞ
, contains two quantities we haven’t met before: qj and
pj . q is the magnetic interaction vector that is defined as:

q = εðε  KÞ − K. (9:2)

The various vectors in eq. (9.2) are illustrated in Figure 9.1. ε is a unit vector in the
direction of the scattering vector s discussed in Section 1.1. It is perpendicular to the
scattering plane and in the plane of the incident and diffracted beams. K is a unit
vector in the direction of the spin.3 From this definition we can deduce that q is in the
plane of ε and K and perpendicular to ε. Its magnitude is given by sin α where α is the
angle between the scattering vector and the spin:

jqj = sin α. (9:3)

θ
Incident beam
θ
ε α Dif
Scattering K fra
cte
vector direction Spin db
ea
m

Figure 9.1: The relationship between the vectors given in eq. (9.2). After Bacon (1975).

We therefore see that magnetic intensities are sensitive to spin direction: when the
spin is perpendicular to ε, jqj = 1, but when parallel jqj = 0 meaning that no magnetic
scattering is observed.
The quantity p is given by:

e2 γ
p= 2Sf ðsÞ, (9:4)
2  4πε0 me c2
where γ is the magnetic moment of the neutron in nuclear magnetons, e2 =me c2 the
classical radius of the electron (e, charge on electron; me , mass of electron; c, speed of
light) and S the spin quantum number. For a spin only ion in which the orbital moment

3 Note that the spin direction is opposite to the magnetic moment.

220 9 Magnetic refinements

is quenched, 2S is equal to the magnetic moment in Bohr Magnetons (BM). In cases

where spin and orbital contributions need to be considered, 2S is replaced by gJ where g
is the Landé factor. The final term in eq. (9.4), f ðsÞ, is the magnetic form factor. This is
similar to the form factor familiar from X-ray diffraction (Section 2.2.1) and arises as the
electrons that give rise to the magnetic moment are distributed over a volume of space
comparable to the neutron wavelength. This is in marked contrast to nuclear scattering
where the nuclei act as essentially point scatterers. Since the electrons responsible for
magnetic moments are in high principal quantum number orbitals – the outer shells –
magnetic neutron scattering factors fall off more rapidly with ~s = s=2 = sin θ=λ
than X-ray scattering factors. Taking Mn2+ as an example, at sin θ=λ = 0.32 (60° 2θ for
λ = 1.54 Å) the X-ray and magnetic form factors have fallen to about 60% and 20% of
their value at θ = 0, respectively. The effect on magnetic intensities (which depend on f 2)
is even more dramatic: in our Mn2+ example intensities will typically fall to 50% by
sin θ=λ = 0.15 (~27° 2θ for λ = 1.54 Å).
Equation (9.4) also gives us a feel for the relative intensities of nuclear and
magnetic scattering. Substituting values4 we find p = 0.54Sf × 10 − 12 cm. For low
sin θ=λ and typical S values (up to 5/2 for transition metals), p will be the same order
of magnitude as bn . For example, for Fe3+ bn = 0.96 × 10− 12 cm and p = 1.35 × 10− 12 cm at
θ = 0.
The form of eq. (9.1) also lets us understand where we will see magnetic peaks in
the diffraction patterns of simple materials. For a simple ferromagnetic material like
that in Figure 9.2a, qj will be the same for all atoms and magnetic intensity will
simply add to the nuclear peaks. For a simple antiferromagnetic material such as that
shown in Figure 9.2b, qj will alternate from + to – along the a-axis. The magnetic
contributions at 2θ values corresponding to nuclear reflections will therefore be
precisely zero. However, the larger magnetic unit cell means that there will be
magnetic peaks at 2θ values corresponding to a doubled a-axis – (½, 0, 0), (½, 1, 0),
(½, 1, 1) and so on using the nuclear cell. We therefore see nuclear-only and magnetic-
only peaks in the neutron diffraction pattern. This type of magnetic ordering is there-
fore very easy to spot in a neutron diffraction experiment.

9.3 Magnetic symmetry

One significant difference between normal and magnetic crystallography is the need
for an extended language to describe symmetry. In conventional crystallography one
of the 230 space groups types is used to generate each atom in the unit cell from the
asymmetric unit. In magnetic crystallography we need to do the same with the

4 e = 1.602 × 10−19 C, γ= 1.9132 nuclear magnetons (1 nuclear magneton is −9.662×10−27 JT−1),

m= 9.1094 × 10−31 kg, c = 2.998 × 108 ms−1; ε0 = 8.855 × 10−31 J−1C2m−1.
 9.3 Magnetic symmetry 221

Ferromagnet
45
40

Rel. intensity
35 Magnetic only
30 Total
(a) 25
20
15
10
5
0
c 5 10 15 20 25 30 35 40 45 50 55
b 2θ/°
a
45
40
Rel. intensity 35
30
(b) 25
20
15
10
5
0
Antiferromagnet 5 10 15 20 25 30 35 40 45 50 55
2θ/°

Figure 9.2: Neutron diffraction patterns from (a) a ferromagnetically ordered material: the red line
shows the magnetic contribution to the diffraction pattern and the blue line the total diffraction
pattern; and (b) an antiferromagnetically ordered material: each peak has either a magnetic or
nuclear contribution. Note that the red magnetic intensities fall off rapidly with 2θ due to the
magnetic form factor f .

magnetic moments or spin directions and must consider new symmetry operations that
can, for example, change spins from “up” to “down.” Much of the language needed to
do this was developed for so-called two-color or black and white three dimensional
space groups by Belov and others (Belov, 1955; Belov, 1957) and by Zamorzaev
(Zamorzaev, 1953, 1957a, 1957b) building on ideas introduced by Heesch and
Shubnikov (Heesch, 1930; Shubnikov, 1951). Some of the historical developments are
covered by Wills (2017). These groups contain not only operations that rotate or
translate atoms in space, but operations that can change their color from black to
white. There are 1651 two-color space groups and they are often called Shubnikov
groups. As described below, when we use them for magnetism we relate the change
in color of a site (black to white) to the flip of a spin at that site (e.g., up to down).
Before we go into more detail, it’s instructive to consider the effect of some simple
symmetry elements on magnetic moments or spins. Figure 9.3 contrasts the effect of a
mirror plane on two types of vector that we encounter in structural science. On the left
we represent a polar vector, such as an electric dipole in a molecule, by an arrow with
charges at either end. Here we see the “common sense” behavior that an arrow
perpendicular to a mirror plane is flipped, whereas an arrow parallel to it retains
the same orientation.
Magnetic dipoles show the opposite and initially counter-intuitive behavior. In
the center of Figure 9.3 we see that an arrow perpendicular to the mirror plane
remains pointing in the same direction, whereas one parallel to the plane is flipped.
222 9 Magnetic refinements

+
−

m m mʹ
−
−

+
+

Electric dipole Magnetic dipole

polar vector axial vector

Figure 9.3: The action of a mirror plane on (left) a polar vector and (right) an axial vector such as a
magnetic moment or spin. We can imagine the moment as being generated from a current loop.

We can understand this by considering the moment in a classical sense as being

generated by a circulating current loop.5 We therefore have to consider the effect of
the symmetry operation on the direction of this current. In the ↑ case the mirror
operation retains the current loop direction, whereas for → the current is reversed
leading to the spin being flipped.6 This type of vector is known as an axial vector. It’s
important to remember that the arrow we conventionally draw to depict a moment is
a somewhat misleading depiction of its symmetry properties.
There will be cases where we need a mirror plane that also flips the direction of a
↑ spin. To do this, we have to introduce a new symmetry operator that is usually
called time reversal or spin reversal and is labeled 1ʹ. Physicists tend to favor the term
time reversal, which we can interpret from Figure 9.3 in terms of the reverse in
direction of the current loop giving rise to the spin – the electrons move backwards
in time when their direction is changed. A symmetry element such as a mirror that
combines reflection and spin reversal is called a primed element (mʹ); a normal
mirror is said to be unprimed (m). Symmetry operations are often called “black” if
primed and “white” if unprimed. Slightly confusingly, magnetic space group dia-
grams usually color symmetry elements red for primed and black for unprimed.
There are some concepts that arise from the introduction of time or spin reversal
that we’ll need to understand. First, we consider what happens to the 32 crystal-
lographic point groups (which describe site symmetries in extended structures) when
they’re combined with time reversal. This process creates magnetic point groups. There
are three ways in which this combination is done. In the first case, the 1ʹ operator is not
present giving 32 magnetic point groups that are the same as the crystallographic point
groups. They are called trivial magnetic groups or monochrome/colorless groups as all

5 Current is conventionally the flow of positive charge. The right-hand grip rule reminds us that the
moment is in the direction of our thumb if our fingers are curled in the direction of current flow. Spin
is in the opposite direction.
6 Magnetic moments or spins have the symmetry properties of a rotating cylinder: a center of
inversion and a mirror plane perpendicular to their axis.
 9.3 Magnetic symmetry 223

the operators are white. In the second case, 1ʹ can be present as a separate element
meaning every operator in the point group is both black and white. These 32 groups are
therefore called gray groups. Finally, we can combine 1ʹ with individual operations
which leads to 58 nontrivial or black-white point groups. For example, ignoring the gray
groups, point group 4/m will give rise to four magnetic point groups 4/m, 4/mʹ, 4ʹ/mʹ
and 4ʹ/m – one trivial and three nontrivial.
Many magnetic point groups aren’t possible for a magnetically ordered site in a
crystal. First, the gray groups all contain 1ʹ as an operator. This means that every site
must contain both a moment and a reversed moment (↑ and ↓ giving ↕). This is only
possible if the time-averaged moment on the site is zero, and gray groups will describe a
paramagnetic site. Some trivial and black-white point groups also require zero moment.
For example, in 4ʹ/m a spin can’t lie along the 4ʹ axis as the operation would reverse its
direction; equally if it were perpendicular to the 4ʹ (in the mirror plane) its direction
would be reversed by the mirror. 4ʹ/m is therefore said to be a nonadmissible magnetic
point group. In total there are 31 admissible magnetic point groups. These are listed
along with the restrictions they impose on spin direction in Table 9.1.

Table 9.1: The 31 admissible magnetic point groups and the restrictions they impose on spin
direction. After Cracknell, 1975.

Type  Type  Direction

1 1 Any
m′ Any in plane
2′ 2′=m′ ? 2′
m mm′m′ m′m2′ ?m
2 2=m 2′2′2 m′m′2 k 2
4 
4 4=m 42′2′ 4m′m′ 
42′m′ 4=mm′m′ 
k 4 or 4
3 
3 32′ 3m′ 
3m′ 
k 3 or 3
6 
6 6=m 62′2′ 6m′m′ 
6m′2′ 6=mm′m′ 
k 6 or 6

We can apply similar considerations to develop the magnetic space groups or

Shubnikov groups. In doing so, it is useful to introduce a small amount of the formalism
needed to understand the nomenclature used in the literature and in TOPAS input files.
The 1651 magnetic space groups (we’ll give them symbol G) are classified into four
different types depending on how they are derived from one of the 230 crystallographic
space groups (symbol F) as summarized in Table 9.2. For type 1 magnetic space groups
none of the operators contain time reversal, and they are equivalent to the correspond-
ing crystallographic space group. We can express this as G = F. A type 2 magnetic space
group contains 1ʹ as an operator and will have both an unprimed and a primed copy of
each operator: G = F + F1ʹ. As for magnetic point groups, these are called gray groups
and describe paramagnetic materials in zero applied field.
224 9 Magnetic refinements

Table 9.2: A summary of different magnetic space group types, and how they might be remembered
by those of a similar age to the authors.

Type G= Description Number Comment

 F colorless or  All operators unprimed

monochrome

 F+Fʹ gray or  Each operator primed and

paramagnetic unprimed

 D+(F–D)ʹ black-white  D equi-translation:

first kind translations not associated
BW with spin reversal. Magnetic
cell equals crystal cell.

 D+(F–D)ʹ black-white  D equi-class: translations

second kind associated with spin
BW reversal. Magnetic
primitive cell larger than
crystal primitive cell.

Types 3 and 4 magnetic space groups are slightly more complex. They can both
be expressed as G = D + (F – D)1ʹ where D is a subgroup of index 2 of G; that is, D is a
space group that contains half the symmetry operations of G. In words, this equation
says that the magnetic space group G has half its symmetry operations unprimed
(those in subgroup D) and the remaining half primed (those not in D). The distinction
between type 3 and type 4 is in the type of subgroup chosen for D. In type 3, D is an
equi-translation subgroup, meaning that the lattice translations don’t involve spin
reversal. Overall, exactly half the point operators are white and half are black so type
3 space groups will have a colored point group. In type 4, D is an equi-class subgroup,
meaning that translational symmetry operations are either white or black and exactly
half the lattice translations involve spin reversal. Type 4 magnetic space groups are
therefore said to have a colored lattice, which means that the primitive magnetic cell
will be larger than the primitive crystal cell.
The 1651 magnetic space groups are given labels similar to those used for crystal-
lographic space groups. For type 1 the magnetic space group label is the same as the
conventional label. For type 2 the symbol 1ʹ is appended to the label. For type 3 the
generators of the magnetic point group that include time reversal are primed. For
type 4 there are two common conventions: the BNS (Belov, Neronov, Smirnova) and
OG (Opechowski and Guccione) schemes. In the BNS convention a subscript is
appended to the first letter of the crystallographic space group symbol of D to indicate
the type of colored lattice. For example, a primitive monoclinic lattice can give rise to
Pa, Pb and Ca (see Figure 9.4). This is then followed by the point group generator
 9.3 Magnetic symmetry 225

symbols. No primes are needed as the combination of translation and point symme-
tries results in each point operator being present with and without time reversal (the
magnetic point group is gray). In the OG convention the crystallographic space group
of F is used for naming, and the equivalent labels are P2a, P2b and PC. In this
convention some of the point group generators need to be primed. As an example,
the BNS symbol Camma (67.509; TOPAS symbol C_amma) and OG symbol Pcmmmʹ
(47.11.357) are the same magnetic space group. The BNS symbol is based on D (Cmma)
and OG symbol on F (Pmmm). Each magnetic space group is given a serial number of
the form n.m (BNS) or n.m.p (OG) and tables relating the different symbols are
available online (e.g., at http://stokes.byu.edu/iso/magneticspacegroups.php).
TOPAS uses the keyword mag_space_group along with the BNS number. These and
the corresponding BNS symbol can be found in the file shubnikovgroups.txt.

x
P Pa (P2a) Pb (P2b) Ca (PC)

Figure 9.4: Magnetic lattices arising from a primitive monoclinic lattice. Labeled with BNS (OG)
symbols. Lattice vectors coupled with time reversal are shown in red.

The crystallographic properties of magnetic space groups have been tabulated in an

electronic book “Magnetic Space Groups” that has a similar layout to the regular
International Tables. This is freely available online (Litvin, 2008). Part of a represen-
tative entry is shown in Figure 9.5 for magnetic space group P2bmʹmaʹ (OG symbol;
BNS symbol is Pamna serial 53.330). The figure on the top left shows the lattice
diagram that has the conventional cell of F. The lattice vectors coupled with spin
inversion (here b) are shown in red and those not coupled are shown in black. To the
right of this are listed: the short international (Hermann–Maugin) symbol of the
magnetic space group, the point group symbol and the crystal system, followed by
the serial index and the long international symbol. The figure in the bottom right
shows the symmetry elements with primed elements in red. The figure in the bottom
left is the general position diagram. Positions at +z are shown in red and –z in blue.
Magnetic moments are colored the same as the general positions they are associated
with. In-plane moment directions are shown by an arrow and the + or – symbol at the
end of the arrow indicates whether it is inclined above or below the z-plane. Note that
the lower diagrams don’t show the full unit cell for this type 4 magnetic space group.
The full cell is, however, easy to visualize as the general position diagram is periodic
in unprimed translations (here a and c) but has spins inverted in the primed transla-
tion (here b).
226 9 Magnetic refinements

P2b m'ma' mmm1' Orthorhombic

51.15.401 P2b 21'/m'2/m2'/a'

x
– –
,+ , +
, ,
– + + –
– + + – ,
,
, ,
+ – – +
– –
,+ , +
, ,
– + + –

Multiplicity,
Wyckoff letter,
site symmetry
(0,0,0) + (0,1,0)' +

16 l 1 (1) x,y,z [u,v,w] (2) x+1/2,y,z [u,v,w] (3) x,y,z [u,v,w] (4) x+1/2,y, z [u,v,w]

(5) x,y,z [u,v,w] (6) x+1/2,y,z [u,v,w] (7) x,y,z [u,v,w] (8) x+1/2,y, z [u,v,w]

8 k m'.. 1/4,y,z [0,v,w] 1/4,y,z [0,v,w] 3/4,y,z [0,v,w] 3/4,y,z [0,v,w]

8 j .m'. x,1/2,z [u,0,w] x+1/2,1/2,z [u,0,w] x,1/2, z [u,0,w] x+1/2,1/2, z [u,0,w]

8 i .m. x,0, z [0,v,0] x+1/2,0,z [0,v,0] x,0, z [0,v,0] x+1/2,0, z [0,v,0]

8 h .2. 0,y,1/2 [0,v,0] 1/2,y,1/2 [0,v,0] 0,y,1/2 [0,v,0] 1/2,y,1/2 [0,v,0]

8 g .2. 0,y,0 [0,v,0] 1/2,y,0 [0,v,0] 0,y,0 [0,v,0] 1/2,y,0 [0,v,0]

4 f m'm'2 1/4, 1/2,z [0,0,w] 3/4,1/2, z [0,0,w]

4 e m'm2' 1/4,0, z [0,v,0] 3/4,0,z [0,v,0]

4 d .2/m'. 0,1/2,1/2 [0,0,0] 1/2,1/2,1/2 [0,0,0]

4 c .2/m. 0,0,1/2 [0,v,0] 1/2,0,1/2 [0,v,0]

4 b .2/m'. 0,1/2,0 [0,0,0] 1/2,1/2,0 [0,0,0]

4 a .2/m. 0,0,0 [0,v,0] 1/2,0,0 [0,v,0]

Figure 9.5: Extracts of an entry from “Magnetic Space Groups” by Litvin (2008). Reproduced with
permission.

Other aspects of the tables are similar to conventional crystallographic tables.

For each Wyckoff position the site symmetry is given for the first position quoted in a
block of equivalent positions. For each position the symmetry-allowed magnetic
 9.4 Ambiguities in magnetic structures 227

moments are also given. For example, in P2bmʹmaʹ Wyckoff site 4c has site symmetry
.2/m. giving the two positions 0,0,½ [0,v,0] and ½,0,½ [0,v,0] meaning that the
moments are required to have equal magnitude (v) and lie along the b-axis for both
sites. Equivalent information is included in electronic files for both OG and BNS
conventions from the ISO-MAG resource (http://stokes.byu.edu/iso/magneticspa
cegroups.php). The Bilbao crystallographic server (http://www.cryst.ehu.es/) also
contains an excellent set of online interactive tools for exploring magnetic symmetry.

9.4 Ambiguities in magnetic structures

When performing any Rietveld analysis it is important to remember that the informa-
tion content in a powder diffraction pattern can be low, and that results should be
interpreted with caution. This is particularly true with magnetic data, where one
usually relies on a relatively low number of reflections at low sin θ=λ, which are often
collected at relatively low resolution, so prone to peak overlap.
There are also a number of cases where neutron diffraction simply can’t distin-
guish between alternate magnetic models. For example, Figure 9.6 shows two very
different magnetic structures. In each of them the magnetic cell is quadrupled in
one direction relative to the nuclear cell. Figure 9.6a is a structure in which all sites
have identical magnetic moments arranged in blocks of two-up then two-down
(↑↑↓↓ …). Figure 9.6b has zero moment on every second site and up/down
ordering between magnetic sites (↑0↓0 …). The magnetic (powder) diffraction
pattern from these two arrangements differs only by a scale factor. If the moment
pffiffiffi
in Figure 9.6b is 2 larger than in Figure 9.6a, the two configurations are
indistinguishable.

(a)
120
100
Rel. intensity

c 80
b 60
a
40
20
0
(b)
0 5 10 15 20 25 30 35 40 45 50 55
(c) 2θ/°

Figure 9.6: Two apparently different magnetic structures (a) and (b) give rise to identical nuclear and
magnetic diffraction patterns. The magnetic diffraction is highlighted in red.

Problems also arise in magnetic structures that need more than one propagation
vector to describe them (so-called multi-k structures). In this language the
228 9 Magnetic refinements

propagation vector k relates the reciprocal space cells of the nuclear and magnetic
structures: in Figure 9.6b the relationship would be k = ¼a*. In multi-k structures
the spin configuration can’t be unambiguously determined by diffraction data,
though introducing symmetry constraints or restricting solutions to those with
equal moments on each site can reduce the number of possible solutions.
The difficulties in determining magnetic structures are even greater with powder
diffraction data on high-symmetry samples. Since an individual experimental peak
may be the contribution of different symmetry-related reflections, information on the
direction of moments is lost. As described by Shirane (Shirane, 1959), for collinear
magnetic structures this means that the moment direction can’t be determined for
systems higher than orthorhombic. For cubic systems there is no information avail-
able on the moment direction. For tetragonal, rhombohedral and hexagonal systems
only the angle of moments relative to the unique c-axis can be determined, and no
information is available on their orientation in the ab-plane. Partial peak overlap
means that similar ambiguities will arise when analyzing low-resolution powder data
on orthorhombic, monoclinic and triclinic samples with only small metric distortions
from higher symmetry.

9.5 LaMnO3 TOPAS worked example

We will explore the various ways in which magnetic diffraction patterns can be
analyzed in TOPAS using a simulated set of powder neutron diffraction data for the
perovskite LaMnO3 provided by John Evans and Branton Campbell. LaMnO3 is of
interest as it is the parent phase of the so-called Giant Magneto Resistive (GMR)
perovskites that show a large change in their electrical conductivity under an
applied magnetic field – an effect that can be exploited in data storage devices.
The ideal perovskite structure of LaMnO3 can be described as an infinite network of
corner sharing MnO6/2 octahedra with the La in 12-coordinate cuboctahedral coor-
dination by O (Figure 9.7). The real structure undergoes two important distortions
relative to this (see Chapter 8 for more discussion on this concept): a Jahn–Teller
distortion driven by the d4 configuration of Mn(III), which leads to octahedra with
four shorter and two longer bonds, and cooperative tilts of the octahedra around the
axes of the cubic cell (Glazer tilt system a+b–b–) such that its space-group symmetry
is Pnma (the non-standard Pbnm setting is frequently used in the literature).
LaMnO3 also undergoes magnetic ordering below its Néel temperature, TN,
of ~139 K and adopts the so-called A-type structure in which each Mn is ferromag-
netically coupled to the four neighboring Mn sites in the ac-plane, but these planes
couple antiferromagnetically along the b-axis (Moussa, 1996). The resulting mag-
netic space group is Pnʹmaʹ, number 62.448 in the BNS convention. The magnetic
moment on each Mn is around 4 Bohr magnetons (BM).
 9.5 LaMnO3 TOPAS worked example 229

MnO6/2
octahedra
b La
O

a Mn
c

Figure 9.7: Magnetic structure of LaMnO3.

9.5.1 Single phase nuclear and magnetic Shubnikov approach

The simplest way to perform a magnetic Rietveld refinement is to use a single phase
for both nuclear and magnetic contributions and provide TOPAS with the magnetic
spacegroup using the key word mag_space_group. This is specified as a serial number
(here 62.448), and the corresponding symbol (Pnʹmaʹ) is stored in the file shubni-
kovgroups.txt. Atoms that contribute to the magnetic diffraction are flagged with the
keywords mlx, mly and mlz, which contain the components of the moment in Bohr
Magnetons per Angstrom along each crystallographic axis. It is usually convenient to
include the macro MM_CrystalAxis_Display that reports the overall moment in Bohr
Magnetons.
A simple script for the magnetic refinement is given below. The top panel of
Figure 9.8 shows the fit without any magnetic contribution (mlx = mly = mlz = 0) and
the lower panel with the magnetic contribution. We can see that the magnetic
contribution to the diffraction pattern is largest at low 2θ and decreases rapidly
with 2θ due to the form factor.

xdd maglamno3.xy
x_calculation_step = Yobs_dx_at(Xo); convolution_step 4
bkg @ 9.968 -0.002 0.018 -0.024 -0.004 -0.018
lam ymin_on_ymax 0.0001 la 1.0 lo 1.54 lh 0.5
neutron_data
LP_Factor( 90)
str
phase_name "LaMnO3"
mag_space_group 62.448
r_bragg 0.289311571
a lpa 5.747293
230 9 Magnetic refinements

b lpb 7.693035
c lpc 5.536693
site La x @ 0.04911 y 0.25 z @ 0.49217 occ La 1 beq bval 1.00845
site Mn x 0 y 0 z 0 occ Mn 1 beq bval 1.00845
mlx @ 0.67 mly @ 0.03 mlz @ 0.00 MM_CrystalAxis_Display( 3.87, 0.19, 0.00)
site O1 x @ 0.48792 y 0.25 z @ 0.57470 occ O 1 beq bval 1.00845
site O2 x @ 0.30668 y @ 0.03846 z @ 0.22619 occ O 1 beq bval 1.00845
scale @ 0.0100387974
CS_L(@, 49.82149)

110
100
90
Rel. intensity

80
70
60
50
40
30
20
10
10 20 30 40 50 60 70 80 90
2θ/°
110
100
90
Rel. intensity

80
70
60
50
40
30
20
10
10 20 30 40 50 60 70 80 90
2θ/°

Figure 9.8: Rietveld fits to simulated neutron diffraction data of LaMnO3 with (top) no magnetic
scattering contribution and (bottom) magnetic scattering included.

9.5.2 Separate nuclear and magnetic phases

In many cases, magnetic ordering leads to a unit cell that is larger than the nuclear
cell, though there are typically minimal changes in the nuclear structure on magnetic
ordering. If we adopted the approach in the previous section, then we would need to
describe coordinates of all the atoms in this larger cell during refinement (to reliably
refine atomic coordinates, it would also be necessary to set up appropriate equations
 9.5 LaMnO3 TOPAS worked example 231

to describe their relationship to the paramagnetic structure). It is often more con-

venient to describe the magnetic diffraction using a separate magnetic-only phase
that contains only the atoms that contribute to the magnetic diffraction. This can be
done by including a second phase in the TOPAS INP file flagged with the keyword
mag_only_for_mag_sites as shown below. In this case, since only the Mn atom carries
a magnetic moment, and it lies on the special position (0, 0, 0), we could refine
atomic coordinates on La and O sites in the nuclear-only structure alongside the Mn
moment. With this approach the scale factor of the magnetic-only phase should be
appropriately related to the nuclear phase. In this example, the nuclear and magnetic
phases have the same unit cell and scales should be equated. In a general case, the
scale of the magnetic-only phase should be (Vnuc/Vmag)2 times that of the nuclear
phase (eq. (5.2)).
This approach has a secondary benefit in that the magnetic scattering can be
displayed separately from the nuclear scattering as in Figure 9.9. This makes it easy
to identify the different contributions to the overall diffraction pattern.

110
100
90
Rel. intensity

80
70
60
50
40
30
20
10
0
10 20 30 40 50 60 70 80 90
2θ/°

Figure 9.9: Rietveld fit using a second phase to describe magnetic scattering with the keyword
mag_only_for_mag_sites. Magnetic peaks are highlighted in green.

str
phase_name "LaMnO3_magnetic"
mag_space_group 62.448
mag_only_for_mag_sites
a =lpa_nuc;:5.747294
b =lpb_nuc;:7.693034
c =lpc_nuc;:5.536694
site Mn x 0 y 0 z 0 occ Mn 1 beq bval1 1.00974
mlx @ 0.67 mly @ 0.02 mlz @ 0.00 MM_CrystalAxis_Display( 3.88, 0.18, 0.00)
scale =scale_nuclear;:0.0100330659
CS_L(@, 49.40491)
232 9 Magnetic refinements

9.5.3 Symmetry modes approach

When the magnetic structure has a larger unit cell or lower symmetry than the
nuclear structure, one has to derive a description of the nuclear structure in the
magnetic cell and space group. It is also necessary to decide which magnetic moment
components mlx/mly/mlz are free to refine by symmetry. Both these processes can be
time-consuming and it is very easy to make mistakes. Fortunately, we can use ideas
very similar to those discussed in Chapter 8 to help us with this, and describe the
magnetic structure using symmetry modes.
With our LaMnO3 example we can use the Pnma nuclear structure as the parent
structure of the magnetic phase transition. We can upload this to the ISODISTORT
suite of software and specify that we want to consider magnetic distortions for the Mn
site and atomic distortions for all sites. As the magnetic cell of LaMnO3 is identical to
the nuclear cell we can choose the gamma k-point, and select the mΓ4+ irrep. This
gives rise to a single choice for the order parameter direction, which uniquely
specifies the magnetic space group as 62.448, Pnʹmaʹ (we discuss in Section 9.5.4
how to make this choice if you didn’t know it). We can then write out a description of
the magnetic structure in TOPAS STR format that, after refinement, has the form
shown in the table below.

str
phase_name "LaMnO3_magnetic"
mag_space_group 62.448 'Pn'ma'
a 5.74730
b 7.69290
c 5.53670
al 90.00000
be 90.00000
ga 90.00000
scale @ 0.0100372179
'{{{mode definitions
prm !a1 0.00000 min -2.00 max 2.00 'Pnma[0,0,0]GM1+(a)[La:c:dsp] A'_1(a)
prm !a2 0.00000 min -2.00 max 2.00 'Pnma[0,0,0]GM1+(a)[La:c:dsp] A'_2(a)
prm !a3 0.00000 min -2.00 max 2.00 'Pnma[0,0,0]GM1+(a)[O1:c:dsp] A'_1(a)
prm !a4 0.00000 min -2.00 max 2.00 'Pnma[0,0,0]GM1+(a)[O1:c:dsp] A'_2(a)
prm !a5 0.00000 min -2.83 max 2.83 'Pnma[0,0,0]GM1+(a)[O2:d:dsp] A_1(a)
prm !a6 0.00000 min -2.83 max 2.83 'Pnma[0,0,0]GM1+(a)[O2:d:dsp] A_2(a)
prm !a7 0.00000 min -2.83 max 2.83 'Pnma[0,0,0]GM1+(a)[O2:d:dsp] A_3(a)

prm mm1 7.74131 min -8.00 max 8.00 'Pnma[0,0,0]mGM4+(a)[Mn:a:mag]Ag_1(a)

prm mm2 0.38675 min -8.00 max 8.00 'Pnma[0,0,0]mGM4+(a)[Mn:a:mag]Ag_2(a)
prm mm3 0.00285 min -8.00 max 8.00 'Pnma[0,0,0]mGM4+(a)[Mn:a:mag]Ag_3(a)
'}}}

'{{{mode-amplitude to delta transformation

prm La_1_dx = + 0.08700*a1;: 0.00000
prm La_1_dz = + 0.09031*a2;: 0.00000
 9.5 LaMnO3 TOPAS worked example 233

prm O1_1_dx = + 0.08700*a3;: 0.00000

prm O1_1_dz = + 0.09031*a4;: 0.00000
prm O2_1_dx = + 0.06152*a5;: 0.00000
prm O2_1_dy = + 0.04596*a6;: 0.00000
prm O2_1_dz = + 0.06386*a7;: 0.00000

prm Mn_1_dmlx = + 0.08700*mm1;: 0.67349

prm Mn_1_dmly = + 0.06499*mm2;: 0.02513
prm Mn_1_dmlz = + 0.09031*mm3;: 0.00026
'}}}

'{{{distorted parameters
prm La_1_x = 0.04907 + La_1_dx;: 0.04907
prm !La_1_y = 1/4;: 0.25000
prm La_1_z = 0.49220 + La_1_dz;: 0.49220
prm !Mn_1_x = 0;: 0.00000
prm !Mn_1_y = 0;: 0.00000
prm !Mn_1_z = 0;: 0.00000
prm O1_1_x = 0.48787 + O1_1_dx;: 0.48787
prm !O1_1_y = 1/4;: 0.25000
prm O1_1_z = 0.57474 + O1_1_dz;: 0.57474
prm O2_1_x = 0.30670 + O2_1_dx;: 0.30670
prm O2_1_y = 0.03846 + O2_1_dy;: 0.03846
prm O2_1_z = 0.22625 + O2_1_dz;: 0.22625

prm Mn_1_mlx = 0 + Mn_1_dmlx;: 0.67349

prm Mn_1_mly = 0 + Mn_1_dmly;: 0.02513
prm Mn_1_mlz = 0 + Mn_1_dmlz;: 0.00026

prm !La_1_occ = 1;: 1.00000

prm !Mn_1_occ = 1;: 1.00000
prm !O1_1_occ = 1;: 1.00000
prm !O2_1_occ = 1;: 1.00000
'}}}

'mode-dependent atoms
site La_1 x = La_1_x; y = La_1_y; z = La_1_z; occ La = La_1_occ; beq bval1 1.0
site Mn_1 x = Mn_1_x; y = Mn_1_y; z = Mn_1_z; occ Mn = Mn_1_occ; beq bval1 1.0
mlx = Mn_1_mlx; mly = Mn_1_mly; mlz = Mn_1_mlz; MM_CrystalAxis_Display(3.87, 0.19, 0.00)
site O1_1 x = O1_1_x; y = O1_1_y; z = O1_1_z; occ O = O1_1_occ; beq bval1 1.0
site O2_1 x = O2_1_x; y = O2_1_y; z = O2_1_z; occ O = O2_1_occ; beq bval1 1.0
CS_L(@, 49.82248)

The format of this structural description should be familiar from Chapter 8.

The parameters a1–a7 describe displacive mode amplitudes. By refining them, the
atomic coordinates will shift from the parent positions as allowed by the space group
symmetry. Refining a1–a7 is equivalent to refining 7 independent xyz parameters in a
conventional crystallographic description (the seven structural @ symbols in Section
9.5.1). The parameters mm1–mm3 describe magnetic mode amplitudes. The equations
in the “mode-amplitude to delta transformation” section show how the changes in
234 9 Magnetic refinements

a1–a7 and mm1–mm3 determine the shifts of atomic coordinates and the change in
magnetic moments from zero along the different crystallographic axes. The shifts
in both coordinates and magnetic moments are then transformed to actual values in
the “distorted parameters” section (actual = parent + shift) and the resulting para-
meters are used to describe the structure in the “mode-dependent sites” section.
During most refinements many of these equations can be ignored and they are
therefore hidden away in jEdit '{{{…'}}} folds by default.
In the LaMnO3 example, we find that parameter mm1 refines to a large value
(7.74) and that it describes a magnetic moment of 3.87 BM along the a-axis. The fit is
100% equivalent to those shown in Figures 9.8 and 9.9. The values of mm2 and mm3
are close to zero and fixing them to zero has no impact on the quality of the fit. This
tells us that although moments are allowed to have components along b and c by
symmetry, the (simulated) data shows that these components are negligible.

9.5.4 Symmetry modes – determining the Shubnikov group

If you’re working on a material like a perovskite, you may have a good idea of
possible magnetic structures and their symmetry from literature analogues. It is
then relatively straightforward to test these different models against your data. In
other cases, however, you may have no idea of the possible magnetic structures and
their symmetries. In these cases symmetry modes offer one possible way of solving
the magnetic structure.
Let’s assume that we have successfully identified a unit cell that will describe all
the magnetic diffraction peaks and that it is a simple multiple (e.g., a 2a × 1b × 1c) of
the nuclear cell. We can then use ISODISTORT to produce a description of the
magnetic structure in space group P1 (BNS number 1.1). We can then refine all of
the allowed magnetic modes in TOPAS to see if it is possible to get a good fit to the
magnetic diffraction. If we can, this tells us that this P1 model has sufficient degrees
of freedom to describe the magnetic structure. It will, however, almost certainly have
too many degrees of freedom and in fact be a subgroup description of the true
structure. If we can choose which of the magnetic modes are actually required to fit
the data, and which can be fixed at 0.0 without degrading the fit, then this will tell us
about the true magnetic symmetry.
If we take the case of LaMnO3, we can use a method analogous to that in the
previous section to produce an STR description of the magnetic structure in magnetic
space group 1.1. We find there are four unique Mn atoms in the unit cell, which means
there are 12 possible magnetic modes (3 degrees of freedom on each Mn, described by
modes m1 to m12) that belong to four different irreps (mΓ1+ , mΓ2+ , mΓ3+ and mΓ4+ ). If
we refine all of these mode amplitudes, we find that we get an excellent fit to the
diffraction data with Rwp = 3.927%. However, on inspecting the output file we find
 References 235

that one of the amplitudes – m10, Pnma[0,0,0]mGM4+(a)[Mn:a]Ag_1(a) – is signifi-

cantly larger than the others as in the table below.

prm m1 -0.12234 'Pnma[0,0,0]mGM1+(a)[Mn:a]Ag_1(a)

prm m2 0.02348 'Pnma[0,0,0]mGM1+(a)[Mn:a]Ag_2(a)
prm m3 1.93525 'Pnma[0,0,0]mGM1+(a)[Mn:a]Ag_3(a)
prm m4 0.08904 'Pnma[0,0,0]mGM2+(a)[Mn:a]Ag_1(a)
prm m5 -0.06189 'Pnma[0,0,0]mGM2+(a)[Mn:a]Ag_2(a)
prm m6 -0.64994 'Pnma[0,0,0]mGM2+(a)[Mn:a]Ag_3(a)
prm m7 -0.27857 'Pnma[0,0,0]mGM3+(a)[Mn:a]Ag_1(a)
prm m8 0.02419 'Pnma[0,0,0]mGM3+(a)[Mn:a]Ag_2(a)
prm m9 0.05751 'Pnma[0,0,0]mGM3+(a)[Mn:a]Ag_3(a)
prm m10 -7.50541 'Pnma[0,0,0]mGM4+(a)[Mn:a]Ag_1(a)
prm m11 -0.01403 'Pnma[0,0,0]mGM4+(a)[Mn:a]Ag_2(a)
prm m12 -0.31751 'Pnma[0,0,0]mGM4+(a)[Mn:a]Ag_3(a)

If we now reset the amplitude of all the magnetic modes to zero and fix them, then
refine only the amplitude of m10 we get a fit with Rwp = 3.92%, which is essentially
identical to that with all the modes turned on. This tells us that the magnetic ordering
can be described using just the mΓ4+ irrep. It is usual for magnetic ordering transitions
to involve a single irrep. If we return to ISODISTORT, we can upload the parent
structure, superpose the mΓ4+ irrep and find that there is a single choice of order
parameter direction that leads to space group 62.448 Pnʹmaʹ, and gives us the correct
high symmetry description of the structure that we used in earlier sections. We have
therefore managed to solve the structure and determine the magnetic space group.
For more complex examples, there are ways of automatically testing which modes
are required. Some of these are explored in the online tutorial at http://community.dur.
ac.uk/john.evans/topas_workshop/tutorial_GA_magnetic.htm.

References
Bacon, G.E. (1975): Neutron diffraction, Clarendon Press, Oxford, UK, 636 pages.
Belov, N.V., Neronova, N.N. & Smirnova, T.S. (1955): The 1651 Shubnikov groups. Trudy Inst.
Kristallogr. Acad. SSSR 11, 33–67.
Belov, N.V., Neronova, N.N., Smirnova,T.S. (1957): Shubnikov groups. Sov. Phys. Crystallogr. 2,
311–312.
Cracknell, A.P. (1975): Magnetism in crystalline materials, Pergamon Press.
Heesch, H. (1930): Über die vierdimensionalen Gruppen des dreidimensionalen Raumes. Z.
Kristallogr. 73, 325–345.
Litvin, D.B. (2008): Tables of crystallographic properties of magnetic space groups. Acta. Cryst. A64.
419–424; supplementary information PZ5052. http://www.bk.psu.edu/faculty/Litvin/download.
html.
Moussa, F., Hennion, M., Rodriguez-Carvajal, Moudden, J.H., Pinsard, L., Revcolevschi, A. (1996):
Spin waves in the antiferromagnet perovskite LaMnO3: a neutron-scattering study. Phys. Rev. B
54, 15149–15155.
236 9 Magnetic refinements

Opechowski, W., Guccione, R. (1965): Magnetism, edited by G.T. Rado and H. Suhl, Vol. 2A, ch.3, New
York, Academic Press.
Shirane, G. (1959): A note on the magnetic intensities of powder neutron diffraction. Acta Cryst. 12,
282–285.
Shubnikov, A.V. (1951): Symmetry and antisymmetry of finite figures. In Russian, USSR Academy of
Sciences, Moscow.
Von Dreele, R.G., Rodriguez-Carvajal, J. in Dinnebier, R., Billinge, S.J.L. (eds.) (2008): Powder
diffraction: theory and practice, RSC publication, Cambridge (UK), 574 pages.
Wills, A.S. (2017): A historical introduction to the symmetries of magnetic structures. Part 1. Early
quantum theory, neutron powder diffraction and the coloured space groups. Powder Diffraction,
32, 148–155.
Zamorzaev, A.M. (1953): Dissertation, Leningrad State University, Russia .
Zamorzaev, A.M. (1957a,b): Generalisation of Federov groups. Kristallografiya 2, 15–20 or Sov. Phys.
Crystallogr. 2, 10–15.
10 Stacking disorder

10.1 Introduction
Many crystal structures contain two-dimensional layered structural motifs (atoms, ions
or molecules) periodically repeated in the third dimension, giving the crystal a distinct
stacking sequence. In many real crystals (including technologically important materials
like battery electrodes and alloys), the stacking order can deviate from ideal periodicity.
The stacking sequence becomes nonuniform, and a microstructure is created. Irregular
stacking of layers is typically connected with the occurrence of what are called stacking
faults, and can have a huge impact on the diffraction pattern. In this chapter we will
discuss how this type of structural defect can be modeled in TOPAS.

10.2 Description of stacking orders

Two basic types of packings are possible for close-packed spheres: cubic close
packing (ccp) and hexagonal close packing (hcp). Both packing types can be distin-
guished by their stacking sequence. In a ccp-structure that contains only one atom
type (e.g., copper metal), copper metal, the layers are stacked perpendicular to the
[111] direction in an αβγ-stacking sequence, with metal position indicated by small
Greek letters (Figure 10.1, left). A hcp-structure, (e.g., magnesium metal), exhibits
layers stacked perpendicular to the [001] directions with an αβαβ-stacking sequence
(Figure 10.1, right).
A useful description of a stacking sequence can be achieved by using stacking
vectors, which describe the transition from a given layer to the subsequent one (Figure
10.1 magenta and blue arrows). For convenience the unit cells (Figure 10.1 black lines),
with lattice parameters a, b, c, α, β, γ are transformed into pseudo-orthorhombic or
pseudo-trigonal unit cells, with lattice parameters a', b', c', α' = β' = 90°, γ' = 90° or
120°, in which the layers are perpendicular to the c'-axis (Figure 10.1 dashed gray
lines). The stacking vectors, Si, are given in fractional coordinates of the transformed
unit cell. Each stacking vector consists of three components (eq. (10.1)): Sxi , Syi and
Szi , with Sxi and Syi describing the shift within the a'b '-plane and Szi describing the
shift parallel to c':
0 1
Sxi
Si ¼@ Syi A: (10:1)
Szi

A ccp-structure can be described by using only one stacking vector, S1, whereas a
hcp-structure can be constructed by using two stacking vectors, S1 and S2, that are
repeated in an alternating fashion (Figure 10.1, magenta, blue):

https://doi.org/10.1515/9783110461381-010
238 10 Stacking disorder

γ β
S1 S1

β α
S1 S2
α β
S1 S1

γ α
S1
S2
β β
S1 S1

α α

Figure 10.1: Stacking sequence in ccp (left) and hcp-packing (right). Unit cell edges are displayed in
black and transformed unit cells in gray dashed lines. The stacking order is indicated by stacking
vectors S1 and S2 describing layer to layer transitions. Positions of metal atoms are indicated by
lowercase Greek letters.

0 1 0 1 0 1
1=3 1=3 1=3
ccp: S1 ¼@1=3 A hcp: S1 ¼@1=3 A; S2 ¼@ 1=3 A: (10:2)
1=3 1=2 1=2

Many inorganic materials, like NaCl or MgO, also exhibit close packed struc-
tures. In these structures layers of metal cations, denoted by small Greek letters
and layers of anions, denoted by capital Roman letters, are arranged in an
alternating fashion (Figure 10.2, left). Layered structures like Mg(OH)2 (brucite)
or CdCl2 can be derived directly from close packings by removing half of the
cation layers (Figure 10.2, right). As a consequence the stacking sequence trans-
forms from CαBγAβCαBγAβ … (NaCl-type) to CαB□AβC□BγA□ … (CdCl2-type)
with “□” indicating a cation vacancy. Usually the vacancies are not explicitly
shown and layers are indicated by round brackets. Therefore the stacking order
in a CdCl2-type lattice can be expressed as: (CαB)(AβC)(BγA). Like close packed
structures, layered structures exhibit several types of packings. In Figure 10.3
three examples are shown: C19-type (CdCl2-type), C6-type (CdI2/brucite-type) and
CrOOH-type (sometimes denoted a 3R-type). There are many more basic stacking
types in other layered compounds. For a good overview see, for example,
Hulliger (Hulliger, 1976).
 10.3 Types of stacking faults 239

α α
S1
C C
S1
β
S1
A A
γ γ
S1
B B
S1
α
S1
C C

β β
S1
A A
γ S1
S1
B B
α α

Cation Anion

Figure 10.2: Stacking sequence in a binary MX ccp-structure (left) and in a related MX2 structure. Unit
cell edges are displayed as black solid lines, edges of transformed unit cells in gray dashed lines. The
stacking order is indicated by stacking vectors S1 describing layer to layer transitions. Positions of
metal cations are indicated by lowercase Greek letters, anion positions are indicated by capital
Roman letters.

C A A
β γ γ
A B B
S1 S2 S3
B A B
α γ α
C B C
S1 S2 S3
A A C
γ γ β
B B A
C19-type C6-type 3R-type

Figure 10.3: Examples of stacking types in layered structures. MX6/3-octahedra with M = metal cation
and X = anion are indicated by black solid lines.

10.3 Types of stacking faults

Several types of stacking faults can occur in a structure based on close packed
spheres. In a ccp-structure the stacking order can have two different orientations:
αβγ (stacking vector S1, Figure 10.4, top left) and γβα (stacking vector S2). When two
differently oriented crystals are intergrown, the stacking vector switches from S1 to
240 10 Stacking disorder

γ
S2
α γ
Twin b S1 hcp-
S2 stacked
β β section
S2 S2
γ γ
S1 S1 ccp-
β stacked
β
S1 Twin a S1 section
α α
S1 S2
γ β
S1
hcp-
α
γ stacked
S2
S1 section
ccp- β
S1
β
S1
stacking
α
α S2
S2 hcp-fault β
β S1 ccp-
S1 stacked
α
α ccp- S1 section
S1 stacking γ
γ

Figure 10.4: Types of stacking faults. Transitions between different stacking vectors leading to:
twinning (top left), a single hcp-fault in a ccp-stacking (bottom left) and crystallographic intergrowth
of ccp- and hcp-stacked sections (right).

S2 at the interface. This interface can be considered as a stacking fault. The switch of
the stacking vectors creates a local hcp-like stacking, for example, βγβ (Figure 10.4,
top left) in the stacking order. This type of intergrowth can be also described as
twinning.
If the stacking vector switches from S1 to S2 and afterwards back to S1, then ccp-
like packing is interrupted by a hcp-like transition (Figure 10. 4, bottom left). This can
be considered as a local stacking fault.
Sometimes local faults exhibit a certain extension, for example, after several ccp-
like layer-to-layer transitions more than one hcp-like transition occurs before the
stacking order switches back to ccp-stacking (Figure 10.4, right). Hence distinct
structural motifs are crystallographically intergrown and the interfaces between the
intergrown sections can be considered as stacking faults.
In each of these cases the stacking vector can be described by a finite array of
distinct vectors. For close packed spheres, this array contains only two vectors1:

1 Sz = 1/3 assumes c' is three times the layer spacing.

 10.3 Types of stacking faults 241

0 1 8 0 1 0 19
Sx < 1=3 1=3 =
S ¼@ Sy A2 S1 ¼@ 1=3 A; S2 ¼@ 1=3 A : (10:3)
: ;
Sz 1=3 1=3

These types of faulting can be found, for example, in highly alloyed TRIP steels
(Martin et al., 2011).
The same considerations can be applied to layered structures. The main differ-
ence is that far more than two basic packing types exist. If, for example, the packing
types presented in Figure 10.3 form a microstructure, the array that describes the
stacking vector consists of three vectors:
0 1 8 0 1 0 1 0 19
Sx < 1=3 0 1=3 =
S ¼@ Sy A 2 S1 ¼@ 1=3 A; S2 ¼@ 0 A; S3 ¼@ 1=3 A : (10:4)
: ;
Sz 1=3 1=3 1=3

If more packing types contribute to the microstructure of a layered compound, the

array of possible stacking vectors will be larger. If the structure remains closely
related to a close packing, then this array will always be finite.
In layered structures with a large interlayer spacing, weak interlayer interactions
or with a high rcation: ranion ratio, the stacking order of the layers often deviates from
close packing. Hence layers can be randomly displaced from their ideal position by
shifts within the layer plane or by rotation within the layer plane, leading to turbostratic
like disorder (Figure 10.5). Accordingly a random component, Δ with Δx, Δy and Δz, can
be added to the stacking vector. As turbostratic disorder doesn’t affect the interlayer
spacing, Δz = 0:
01 0 1
Sx Δx
S ¼ Si þ Δ ¼@ Sy Aþ@ Δy Awith 0 < Δx ; Δy < 1 and Δx ; Δy 2 fR g: (10:5)
Sz 0

+Δy
A A' A'
+Δx

A A A

Figure 10.5: Ideal stacked layers (left); top layer randomly dislocated by a shift in the ab-plane
(middle); and top layer randomly dislocated by a rotation in the ab-plane.

Since the x and y components can assume all real values between 0 and 1, the
stacking is described by an infinite array of distinct vectors.
Some layered compounds can take up atoms, ions or molecules in-between the
sheets in a process called intercalation. Intercalation can be inhomogeneous, that is,
242 10 Stacking disorder

d(001)

Ni O O/Cl/Br/I N

Figure 10.6: Disorder in α/β-Ni(OH)2 by intercalation of ions and molecules ≡ interstratification.

some layers are intercalated and others are not, or homogenous. Intercalation is
usually accompanied by an increase of the interlayer distance (Figure 10.6). When
the distribution of intercalated particles is random or inhomogeneous then intercala-
tion appears as a form of stacking faulting that is called interstratification.
Interstratification affects the z-component of the stacking vector and often goes
along with turbostratic-like disorder in materials such as brucite-type magnesium
and nickel hydroxides (Radha et al., 2003; Ramesh et al., 2003; Ramesh et al., 2008).

10.4 Modeling the influence of stacking faults on diffraction

One of the most fundamental and significant theoretical works on the effects of
stacking disorder on XRPD patterns was carried out by B.E. Warren (Warren, 1941).
He considered complete random stacking (turbostratic disorder) in a layered struc-
ture with Δz = 0. As rotations of layers (Figure 10.5, right) don’t cause any additional
reflections in an XRPD pattern, considering only random translations (Figure 10.5,
right) is sufficient. In his work, Warren included only sharp 00l and diffuse hk0
reflections. The 2θ-dependent reflection intensities (i.e., the broadened reflections)
were derived from the Debye formula. For this purpose, three vectors are introduced:
vectors a1 and a2, describing the scheme of repetition in each layer and vector a3,
which is perpendicular to a1 and a2 and therefore describes separation between the
layers (Figure 10.7, red arrows). Lattice points describing the transformed unit cell
with lattice parameters a', b', c', α', β' and γ' (see above) are indicated by m1 and m2
and an individual layer is denoted by the layer number m3 (Figure 10.7, green
numbers).
The position of an atom, n, in layer m3 can be described by the vector Rn(m3) that
includes the basis vector, rn, that contains the fractional coordinates x', y', z', for the
atoms n (Figure 10.7, blue balls and arrows):

Rn ðm3 Þ = m1 a1 + m2 a2 + m3 a3 + Δxðm3 Þ a1 + Δyðm3 Þ a2 + rn . (10:6)

 10.4 Modeling the influence of stacking faults on diffraction 243

rn
m1 Layer no. m3:
a' a1
α m3 = 4
b' a2

m2 γ m3 = 3

β m3 = 2
c' a3
α m3 = 1
a1
a'
m1
rn

Figure 10.7: Description of the crystal structure of ccp-Cu in the notation used by Warren
(Warren, 1941), m1 and m2 are lattice points, m3 denotes the number of an individual layer, a1 and a2
are vectors describing the scheme of repetition in each layer, a3 is a vector describing the separation
between the layers, rn is the basis vector for atoms of type n.

The intensity for a particular scattering vector s = (hkl) is then given by:
 
X h i 2
 2πiðhx0 n þky0 n þlz0 n Þ 
I s ¼ f n ðsÞe 
 n 
sin2 ðπN 1 a1 hÞ sin2 ðπN 2 a2 kÞ
  (10:7)
sin2 ðπa1 hÞ sin2 ðπa2 kÞ
 
X h i2
 2πiðhΔxðm3 Þa1 þkΔyðm3 Þa2 þlm3 a3 Þ 
 f m3 ðsÞe 
m 
3

where N1a1 and N2a2 are the dimensions of the layers, which is assumed to be a
parallelogram.
For hk0 reflections, the randomness does not play any role and the intensity
is that of a crystalline reflection. The phase factors of these reflections are,
however, completely random in the summation over m3 resulting in incoherent
scattering from the individual layers (Figure 10.8, left). For a two-dimensional
layer the reciprocal lattice becomes a series of parallel lines perpendicular to the
layer, called truncation rods (Warren, 1941) (Figure 10.8, middle). For stacking
along c*, these rods intersect the plane at the hk0 points. This leads to aniso-
tropic broadening of the hk0 reflections. Warren demonstrated that these reflec-
tions are always broadened toward higher diffraction angles leading to
characteristic Warren-type triangular or saw-tooth peak shapes (Figure 10.8,
right). The unaffected 00l reflections remain sharp.
One powerful tool to simulate and visualize the effects of stacking faults on
diffraction patterns is the DIFFaX software package (Treacy et al., 1991). Simulations
can be carried out in DIFFaX by using transformed unit cells of the type described
above with layers stacked perpendicular to the c-axis. Faulted structures are generated
244 10 Stacking disorder

00I Reflection C*
hk0 Plane
(10)

(003) b* (004)
(002) h=0
(001) h=1

h=2

40 60
k=2
a 2:1 Layer hk0 Reflection a* k = 0 k=1 2θ/°

Figure 10.8: Left/middle: real and reciprocal space of a turbostratically disordered layer structure
(Moore et al., 1997; Ufer et al., 2012). Right shows a small region of the diffraction pattern of carbon
black (adapted from Briscoe and Warren, 1942).

by defining individual layers that can be chemically identical or different. Transitions

from one layer to the next are described by using stacking vectors (e.g., eq. (10.1)) with
associated probabilities. Patterns can be simulated using two fundamental program
modes: explicit – an XRPD pattern from one pre-defined or random sequence of layers
and stacking vectors is simulated; or recursive – an averaged diffraction pattern is
calculated from all possible stacking sequences consistent with a transition probability
matrix.
Transition probabilities used to describe the degree of faulting are expressed
through a square probability matrix. In this matrix the elements Pij describe the
probability of layer i being followed by layer j. The description is often set up so the
diagonal terms Pii correspond to the probability of a faultless stacking and off-
diagonal terms describe a stacking fault, though this doesn’t have to be the case.
An example of a transition probability matrix including three stacking vectors is
given in Table 10.1:

Table 10.1: Example of a 3 × 3 transition probability matrix for a 3-layer system (Pn1 + Pn2 + Pn3 = 1).

transition from↓/to→ layer  layer  layer 

layer  P P P

layer  P P P
layer  P P P

When generating stacking sequences the position/type of the preceding layer is

always taken into account. Hence, the transition probability matrix can be non-
symmetric. The whole approach can be considered as recursive. As a statistical
constraint, all transition probabilities within each row of the matrix must sum up to
unity:
 10.4 Modeling the influence of stacking faults on diffraction 245

X
n
Pij = 1. (10:8)
j=1

A demonstration of the two DIFFaX program modes is given in Figure 10.9.

Diffraction profiles of polycrystalline copper with faultless ccp-stacking (black
line) and a hexagonal intergrowth (colored lines) were simulated by applying a
2 × 2 transition probability matrix with P12 = P22 = 0.02 (implying P11 = P21 = 0.98)
and stacking vectors as given in eq. (10.3). For the simulations, the ccp unit cell of
copper was transformed to a hexagonal one with c' along the cubic [111] direction.
Simulations in recursive mode using 500 layers show that the 003 reflection (111 in
the original cubic unit cell) is slightly broadened and is less intense (Figure 10.9,
green line) than a unfaulted sample. In addition, a weak feature appears at 40.6° 2θ
that can be indexed as the 010 reflection of hcp-stacked copper, suggesting small
sections of hexagonal intergrowths. When an equivalent simulation is carried out
in the explicit mode with a single stack of 5000 layers (blue line) the pattern is
essentially identical. This shows that the statistics of a 5000 layer stack is equiva-
lent to the array of weighted stacking sequences that is modeled in the recursive
approach. If, however, the number of layers is reduced from 5000 to 500, signifi-
cant “ripples” appear in the pattern (red line). Each time a 500 layer stack is
calculated, slightly different ripples appear due to the slightly different supercell

DIFFaX TOPAS

(003)
(Rel. intensity)0.5

(Rel. intensity)0.5

(010)

Ideal Ideal
Recursive Nv = 500 Nv = 500 Nstr = 100

Explicit Nv = 5000 Nv = 5000

Explicit Nv = 500 Nv = 500

40 41 42 43 44 45 46 40 41 42 43 44 45 46
2θ/° 2θ/°

Figure 10.9: DIFFaX and TOPAS simulations of diffraction profiles for ccp-Cu. Top to bottom: unfaulted
ccp-packing (black), 2% hexagonal intergrowth 500 layer recursive calculation (green), 5000 layers
explicit (blue) and 500 layers explicit (red). Corresponding TOPAS simulations on the right.
Hexagonal intergrowths cause the appearance of a peak close to the ccp-forbidden (010) reflection.
All indices are given in the notation of the transformed hexagonal unit cell.
246 10 Stacking disorder

generated. If a series of these patterns is averaged the resulting pattern closely

approximates the recursive or 5000 stack simulations.

10.5 TOPAS implementation

While DIFFaX provides a straightforward simulation tool for single-phase sys-
tems, it doesn’t offer the power of a full Rietveld approach. There have been a
number of Rietveld-compatible stacking-fault methods published, each with their
own advantages and disadvantages (e.g., Leoni et al., 2004; Casas-Cabanas et al.,
2016; Reynolds, 1985; Ufer et al., 2004; Wang et al., 2012). Other approaches to
derive the layer constitution from vastly stacking faulted samples exist, for exam-
ple, by using two-dimensional projections of the crystal structure and geometric
considerations (Leineweber et al., 2012).
TOPAS v6 and onwards (Coelho et al., 2016) contains powerful tools for
DIFFaX-like modeling of stacking-faulted patterns using a super cell approach.
This allows stacking faults to be analyzed alongside all the other features of
TOPAS. An extended unit cell is defined that contains a large (several 100 to
several 1000) number, Nv, of layers that are stacked in the c'-direction and
individual faults can be built into the supercell. With a sufficiently large c'-axis,
the large number of hkl reflections of the supercell sum to give an excellent
approximation of the broadened smooth peak shapes observed experimentally.
As an example, diffraction profiles of polycrystalline copper with faultless ccp-
stacking and hexagonal intergrowths have been simulated using TOPAS and are
shown alongside the DIFFaX simulations in Figure 10.9. To achieve this the
space group was reduced to P1 and the unit cell of copper was transformed as
described in Table 10.22: As copper is situated at the origin in both the original
and the transformed unit cell, the fractional coordinates don’t need to be
transformed.
The simplest TOPAS process for simulating stacking-faulted XRPD patterns
is comparable to the explicit mode of DIFFaX. First of all the number of
generated layers in a stacked sequence, and therefore the extension of the
unit cell, has to be defined:

number_of_stacks_per_sequence Nv 500 ' Generation of a supercell including 500 layers

c = Get(generated_c);:0 ' Works out c-axis from number of layers and heights

The keyword layer identifies a site as belonging to a layer called $layer_name, which
is A in our case. Here, this first layer consists of only one copper atom located at the

2 Note that the same transformation must be carried out for DIFFaX simulations.
 10.5 TOPAS implementation 247

Table 10.2: Transformation of the ccp-structure of copper to a pseudo-hexagonal supercell.

Copper Original unit cell Transformed unit cell

Space group 
Fm3m P
a′ = b′ = p1ffiffi2 a= 2.56 Å;
a = b = c = . Å pffiffi
Unit cell c′ = 33 a= 2.09 Nv Å
α = β = γ =  °
α = β = 90; γ = 120
Volume . Å ⅓·Nv·. Å
Stacking
[] []
direction
Number of
 Nv
layers

origin. In general, the z-coordinate of the atomic sites belonging to a layer must be
divided by the total number of layers, Nv:

site Cu1 x = 0; y = 0; z = 0.0/Nv; occ Cu 1 beq 0.2 layer A

TOPAS has certain rules that govern the behavior of sites marked with the layer
keyword: a site marked with layer cannot take part in restraints and it is not displayed
in the view_structure window. These can be overcome using the keyword
generate_these.
Next the stacking vector (eq. (10. 1)) is defined as a transition from one layer to the
subsequent one. For creating a pure ccp-type stacking only one layer type and one
stacking vector (Figure 10.1, left) is necessary:

prm !h 2.0925 'this is the interlayer spacing, see Tab. 10.2

Transition(A, h)
to A = 1; ' Translate from layer type A to layer type A with a probability of 1
a_add = 1/3; b_add = -1/3;' For A → A stacking vector: sx =⅓, sy = -⅓,sz = h/(h·Nv)

In order to introduce a hexagonal intergrowth fault, we define a second type of

layer, layer B, which is chemically identical to layer A. A second stacking vector
and associated transition probabilities must be defined:

prm !PAA 0.98 ' PAB needs not to be defined as PAB = 1 - PAA (eq. 10. 8)
prm !PBA 0.98 ' PBB needs not to be defined as PBB = 1 - PBA (eq. 10. 8)
Transition(A, h)
to A = PAA; ' Translate from layer type A to layer A with a probability of 0.98
a_add = 1/3;
b_add = -1/3; ' For A → A apply a stacking vector with sx = ⅓, sy = -⅓, sz = h/(h·Nv)
248 10 Stacking disorder

to B = 1-PAA; ' Translate from layer type A to layer type B with a probability of 0.02
a_add = -1/3;
b_add = 1/3; ' For A → B apply a stacking vector with sx = -⅓, sy = ⅓, sz = h/(h·Nv)
Transition(B, h)
to A = PBA; ' Translate from layer type B to layer type A with a prob. of 0.98
a_add = 1/3;
b_add = -1/3; ' For B → A apply a stacking vector with sx = ⅓, sy = -⅓, sz = h/(h·Nv)
to B = 1-PBA; ' Translate from layer type B to layer type B with a prob. of 0.02
a_add = -1/3;
b_add = 1/3; ' For B → B apply a stacking vector with sx = -⅓, sy = ⅓, sz = h/(h·Nv)

The pattern simulated with this approach is shown in red on the right hand
side of Figure 10.9. The use of only 500 layers causes ripples just as observed
in the DIFFaX simulations. These can be reduced by using 5000 layers (blue
line) to improve the statistics. While the use of 5000 layers improves the
simulation, the significantly larger c axis leads to many more hkl reflections
in the simulation. In the current case the number of reflections over a 5–150°
range increases from 24946 to 249431 and the simulation time (on a standard
PC) increases significantly from 0.29 to 2.59 s.
It is, however, possible to mimic the DIFFaX recursive mode using the
TOPAS number_of_sequences Nstr keyword. By setting Nstr to 100, TOPAS will
average the patterns of 100 individual sequences. Internally TOPAS uses the
same set of hkl reflections for each of the sequences and the speed penalty for
calculating the pattern of 100 sequences compared to 1 is minimal (0.29 to 0.63
s in this example). This gives the green pattern on the right of Figure 10.9 that
closely matches the 5000 layer simulations or the DIFFaX recursive calcula-
tion. The specific language needed is included in the INP file below, which
actually performs a Rietveld fit to the data simulated in DIFFaX in recursive
mode (the green pattern in Figure 10.9, left).

seed ' Reinitiates random number generator each time

xdd "Cu-rec500.xy" ' Simulated in DIFFaX, green in Fig. 10.9 left
start_X 20
LP_Factor(0)
bkg @ 0.0333011786` 0.0289287388` 0.0405216319`
Zero_Error(@,-0.00878`)
lam ymin_on_ymax 0.00001 la 1 lo 1.5418 lh 1e-5
str
prm s 3484.09010` min 1e-15
scale = s 1e-6/ (Nv Nstr); ' Scale factor proportional to layers used
phase_name "Cu-2%-faults"
space_group P1
a !lpa 2.563066
b !lpa 2.563066
 10.5 TOPAS implementation 249

c = Get(generated_c);:1046.230000 ' c-axis derived from total number of layers

ga 120

prm !PAA 0.980

prm !PBA 0.980
prm !h 2.09246
generate_stack_sequences {
number_of_sequences Nstr 100 ' 100 stacking sequences
number_of_stacks_per_sequence Nv 500 ' Each sequence has 500 layers
Transition(A, h)
to A = PAA; a_add = 1/3; b_add = -1/3;
to B = 1-PAA; a_add = -1/3; b_add = 1/3;
Transition(B, h)
to A = PBA; a_add = 1/3; b_add = -1/3;
to B = 1-PBA; a_add = -1/3; b_add = 1/3;
}
site Cu1 x = 0; y = 0; z = 0/Nv; occ Cu 1 beq bval 0.0 layer A
site Cu1 x = 0; y = 0; z = 0/Nv; occ Cu 1 beq bval 0.0 layer B
peak_buffer_based_on = Xo;
peak_buffer_based_on_tol 0.1 ' Limit number of hkl's used
SF_smooth(@, 5.28134`, 1) ' Smooth peaks if Nv small, section 10.6
TCHZ_Peak_Type(, -0.07754,, 0.13069,, -0.03604,, 0,, 0.15771,, 0.00010) ' IRF

In this file we also make use of the peak_buffer_based_on_tol command to speed up

calculations. This reduces the number of individual hkl peaks and different peak
shapes that need to be summed to calculate the powder pattern. In this example the
use of the peak buffer reduces the number of hkl reflections used from 24946 to 2430
and 249431 to 2872 for Nv = 500 and Nv = 5000, respectively, with negligible impact on
the calculated pattern. More details of how this is done are given by Ainsworth et al.
(2016) and Coelho et al. (2016). The final Rietveld fit to the DIFFaX-simulated data is
shown in Figure 10.10.

50

40
Sqrt (rel. intensity)

30

20

10

20 40 60 80 100 120 140

2θ/°

Figure 10.10: Full Rietveld fit of DIFFaX-simulated data for 2% faulted Cu. Note the effectively
continuous blue bar of hkl tick marks.
250 10 Stacking disorder

10.6 Determining fault probabilities by Rietveld refinement

The speed of stacking fault calculations in TOPAS means that it is possible to

determine stacking fault probabilities from experimental data, despite the stacking
fault probabilities not being directly refinable parameters. One way of doing this is to
make use of the #list command (discussed in detail in Chapter 12) to test a series of
different fault probabilities (Ainsworth et al., 2016). Examples for dealing with more
sophisticated faulting models have been given by, for example, Bette et al., (2015)
and Kudielka et al., (2017).
For example, the INP file below performs a Rietveld refinement on a simulated
data set of an intergrowth of 70% cubic diamond and 30% hexagonal lonsdaelite
layers (discussed in the DIFFaX manual or Ainsworth, 2016). The script scrolls
through 27 different values of the stacking probability pa and does a Rietveld fit for
each. The lowest Rwp is found for pa = 0.7, as shown in Figure 10.11.

35
(a) Rwp (c)
30
pa = 0.70
25 Smooth
Rwp/%

20 Nv = 100
15 (d)
10 pa = 0.70
5 No smooth
Nv = 100
0
(rel. intesity)0.5

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Fault probability pa
(e) pa = 0.70 Nstr = Nv
= 200
(b) Rietveld
(Rel. intensity)0.5

(f) pa = 0.05 Nstr = Nv

= 200

(g) pa = 0.95 Nstr = Nv

= 200

20 40 60 80 100 120 140 38 40 42 44 46 48 50

2θ/° 2θ/°

Figure 10.11: (a) Rwp for a series of Rietveld fits to a simulated data set of an intergrowth of
diamond and lonsdaelite. The minimum Rwp is achieved with pa = 0.7 leading to the Rietveld fit in
panel (b). The right hand plots show simulated patterns for different stacking fault probabilities.
Plots (e) to (g) show simulations with Nstr = 200 Nv = 200 and smoothing applied. Panel (d) shows
the “ringing” observed in the calculated pattern when a small number of layers (Nv = 100) is
simulated; panel (c) shows how smoothing allows a small value of Nv to simulate a much larger
supercell. In each panel red lines are TOPAS calculated patterns and (b, c, e) blue DIFFaX-
simulated “observed” data.
 10.6 Determining fault probabilities by Rietveld refinement 251

num_runs 27
#list paval {
0.9999 0.9500 0.9000 0.8500 0.8000 0.7500 0.7300 0.7200 0.7100
0.7000 0.6900 0.6800 0.6700 0.6500 0.6000 0.5500 0.5000 0.4500
0.4000 0.3500 0.3000 0.2500 0.2000 0.1500 0.1000 0.0500 0.0010
}
seed
xdd diffax_dia.xye
weighting 1
LP_Factor(0) bkg 1
start_X 10 finish_X 149.6
rebin_with_dx_of 0.02
lam ymin_on_ymax 0.0001 la 1 lo 1.5405754 lh 1e-5
Zero_Error( , 0.00238)
str
space_group P1
a 2.518156 b 2.518156 c = Get(generated_c); ga 120
prm !pa =paval(Run_Number);:0.70000 prm !h 2.05870
prm s 4983.64843 min 1e-15 scale = s 1e-6 / (Nv Nstr);
generate_stack_sequences {
number_of_sequences Nstr 200
number_of_stacks_per_sequence Nv 200
Transition(A, h)
to A = pa; a_add = 2/3; b_add = 1/3;
to B = 1-pa; a_add = 0; b_add = 0;
Transition(B, h)
to A = 1-pa; a_add = 0; b_add = 0;
to B = pa; a_add =-2/3; b_add =-1/3;
site C1 x =-1/3; y =-1/6; z = -0.125/Nv; occ C 1 beq 1 layer A
site C2 x = 1/3; y = 1/6; z = 0.125/Nv; occ C 1 beq 1 layer A
site C3 x = 1/3; y = 1/6; z = -0.125/Nv; occ C 1 beq 1 layer B
site C4 x =-1/3; y =-1/6; z = 0.125/Nv; occ C 1 beq 1 layer B
peak_buffer_based_on = Xo; peak_buffer_based_on_tol 0.1
TCHZ_Peak_Type( , 0.09408, , -0.07795, , 0.02058, ,0, , 0.11131, , 0.02457) 'IRF
SF_smooth( , 5.7, 1)
}

This INP file contains one more “trick” to help speed up refinements by allowing
the use of a small number of layers Nv through the SF_smooth macro. As shown
in panel (d) of Figure 10.11, if one uses Nv = 100 the finite number of hkl
reflections generated leads to a “ringing” in the calculated pattern caused by
the peak shapes of the individual reflections. The SF_smooth macro applies a
convolution based on the calculated intensity of adjacent hkl reflections to
smooth out the profile of some reflections, while leaving others sharp. In this
way a small number of layers can effectively mimic a much larger number,
which speeds refinements up enormously.
252 10 Stacking disorder

References
Ainsworth, C.M., Lewis, J.W., Wang, C.H., Coelho, A.A., Johnston, H.E., Brand, H.E.A., Evans, J.S.O.
(2016): 3D transition metal ordering and Rietveld stacking fault quantification in the new
oxychalcogenides La2O2Cu2–4xCd2xSe2. Chem. Mat. 28, 3184–3195.
Bette, S., Dinnebier, R.E., Freyer, D. (2015): Structure solution and refinement of stacking-faulted NiCl
(OH). J. Appl. Cryst. 48, 1706–1718.
Biscoe, J., Warren, B.E. (1942): An X‐ray study of carbon black. J. Appl. Phys. 13, 364–371.
Casas-Cabanas, M., Reynaud, M., Rikarte, J., Horbach, P., Rodríguez-Carvajal, J. (2016): FAULTS: a
program for refinement of structures with extended defects. J. Appl. Cryst. 49, 2259–2269.
Coelho, A.A., Evans, J.S.O., Lewis, J.W. (2016): Averaging the intensity of many-layered structures for
accurate stacking-fault analysis using Rietveld refinement. J. Appl. Cryst. 49, 1740–1749.
Hulliger, F. (1976): Structural chemistry of layer-type phases, Springer Netherlands, Dordrecht.
Kudielka, A., Bette, S., Dinnebier, R.E., Abeykoon, M., Pietzonka, C., Harbrecht, B. (2017): Variability
of composition and structural disorder of nanocrystalline CoOOH materials. J. Mater. Chem. C 5,
2899–2909.
Leineweber, A., Kreiner, G., Grüner, D., Dinnebier, R., Stein, F. (2012): Crystal structure, layer defects,
and the origin of plastic deformability of Nb2Co7. Intermetallics 25, 34–41.
Leoni, M., Gualtieri, A.F., Roveri, N. (2004): Simultaneous refinement of structure and microstructure
of layered materials. J. Appl. Cryst. 37, 166–173.
Martin, S., Ullrich, C., Šimek, D., Martin, U., Rafaja, D. (2011): Stacking fault model of ε-martensite
and its DIFFaX implementation. J. Appl. Cryst. 44, 779–787.
Moore, D.M., Reynolds, R.C. (1997); X-ray diffraction and the identification and analysis of clay
minerals, 2nd edition, Oxford University Press, Oxford; New York, 378 pages.
Radha, A.V., Vishnu Kamath, P., Subbanna, G.N. (2003): Disorder in layered hydroxides: synthesis
and DIFFaX simulation studies of Mg(OH)2. Mater. Res. Bull. 38, 731–740.
Ramesh, T.N., Jayashree, R.S., Kamath, P.V. (2003): Disorder in layered hydroxides: diffax simulation of
the X-ray powder diffraction patterns of nickel hydroxide. Clay Clay Miner. 51, 570–576.
Ramesh, T.N., Kamath, P.V. (2008): Planar defects in layered hydroxides: simulation and structure
refinement of β-nickel hydroxide. Mater. Res. Bull. 43, 3227–3233.
Reynolds, R.C., Jr. (1985): NEWMOD: A computer program for the calculation of one-dimensional
patterns of mixed-layered clays, Hannover, NH (US).
Treacy, M.M.J., Newsam, J.M., Deem, M.W. (1991): A general recursion method for calculating
diffracted intensities from crystals containing planar faults. Proc. R. Soc. London, Ser. A 433,
499–520.
Ufer, K., Kleeberg, R., Bergmann, J., Dohrmann, R. (2012): Rietveld refinement of disordered Illite-
Smectite mixed-layer structures by a recursive algorithm. II: Powder-pattern refinement and
quantitative phase analysis. Clay Clay Miner. 60, 535–552.
Ufer, K., Roth, G., Kleeberg, R., Stanjek, H., Dohrmann, R., Bergmann, J. (2004): Description of X-ray
powder pattern of turbostratically disordered layer structures with a Rietveld compatible
approach. Z. Kristallog. – Crystalline Materials 219, 519–527.
Wang, X., Hart, R.D., Li, J., McDonald, R.G., van Riessen, A. (2012): Quantitative analysis of turbos-
tratically disordered nontronite with a supercell model calibrated by the PONKCS method.
J. Appl. Cryst. 45, 1295–1302.
Warren, B.E. (1941): X-Ray diffraction in random layer lattices. Phys. Rev. 59, 693–698.
11 Total scattering methods

11.1 Introduction
A powder pattern can be regarded as the projection of three-dimensional reciprocal
space onto a one-dimensional axis (1/d, 2θ, sinθ=λ or Q = 4π sinθ=λ), as discussed in
Chapter 1. Geometrically this means that all intensity in a spherical shell with inner
radius 1/(d+Δd) and outer radius 1/d will be summed up to the corresponding step-
scan intensity of width 1/(Δd) in the powder pattern. Everything that is present in the
incident and diffracted beam paths will potentially contribute to the total intensity
measured. The powder pattern therefore contains contributions both from the envir-
onment (instrument) and from the sample: scattering from the sample holder, from
air, from inelastic Compton scattering and fluorescence all combine to give what is
commonly called “background.”
The coherent elastic scattering from the sample, which is what we are usually
interested in, consists of two parts: Bragg scattering at the reciprocal lattice points
and diffuse scattering due to deviations from the average periodic structure (local
structure) in between the reciprocal lattice points. Both types of elastic scattering
contain a wealth of structural information.
The Rietveld method retrieves information solely about the time- and space-
averaged periodic crystal structure by fitting the Bragg intensities. A powder pattern
for Rietveld analysis should, therefore, be measured with a focus on high resolution
in reciprocal space and only up to the angle beyond which no Bragg scattering is
visible above the background.
The total scattering method is an alternative approach for analyzing the elastic
scattering by Fourier transforming the entire powder-diffraction data (Bragg peaks
and diffuse features between them) to the so-called atomic pair distribution function
(PDF). The real-space PDF can give information about local structure, and will
include contributions from amorphous components, disorder and from lattice
dynamics. The nature of Fourier transforms means that data is needed over a wide
range of reciprocal space to give rise to high resolution in real-space. Powder-
diffraction patterns therefore need to be measured so that the coherent scattering
(Bragg and diffuse) is collected over as much of reciprocal space as possible.
A combination of Rietveld refinement and PDF analysis will reveal the maximum
amount of structural information from a powder pattern, but one must be aware
that the measurement strategies for the two techniques are somewhat different, often
requiring that two powder patterns are measured with different setups or instruments.
In the following sections we will discuss the origin and form of PDF functions and
how they are analyzed in TOPAS. At the time of writing, PDF analysis and its TOPAS
implementation are developing rapidly. Methods and macros are being updated
regularly. We give an introduction to what’s possible, and the reader should check,

https://doi.org/10.1515/9783110461381-011
254 11 Total scattering methods

for example, the TOPAS wiki for links to the latest features (http://topas.dur.ac.uk/
topaswiki/doku.php?id=pdf_fitting). There is also a gitHub site containing a
TOPAS pdf.inc file, which contains various macros that are useful in PDF fitting
(https://github.com/pachater/topas).

11.2 Total scattering and atomic pair distribution function (PDF)

analysis

Powder-diffraction data for total scattering studies can be measured in much the same
way as for regular studies, but it’s usual to collect data to a Q value1 of 30–50 Å−1 and
with small statistical uncertainties. This makes synchrotron or neutron radiation from
spallation sources the methods of choice. A typical data set recorded using a two-
dimensional area detector at the NSLS synchrotron is shown in Figure 11.1. In the
laboratory, Mo-Kα or Ag-Kα radiations are the best choice. Ag-Kα radiation offers the
highest Qmax (approximately 22 Å–1 at 160° 2θ), while Mo-Kα radiation provides a
compromise between Qmax and flux.
Corrections must be made for contributions to the scattered intensity from effects
like Compton scattering, fluorescence, scattering from the sample holder and so on
(Egami & Billinge, 2012). Instrument effects are usually removed by measuring an
empty sample holder – typically a glass, amorphous silica or Kapton capillary
(Figure 11.1). This background signal is then subtracted from the sample signal.
Some scaling may be required if absorption by the sample significantly influences
the instrument contribution that reaches the detector.
The resulting coherent scattering function I(Q) is a continuous function of Q
(Figure 11.2) with sharp features where there are Bragg peaks, and broad features in
between. I(Q) is normalized by dividing by the total scattering cross-section of the
sample to give S(Q), the total-scattering structure function. In the case of X-ray
scattering, the sample scattering cross-section is the square of the average atomic
form-factor, hf ðQÞi2 , which becomes very small at high-Q. Thus, during the normal-
ization process the experimental data at high-Q are amplified. This has the effect that
even weak signals at high-Q, which are usually insignificant in a conventional
Rietveld analysis, can become important in a total-scattering experiment. Because
the signal at high-Q is weak, it is important to collect the data in that region with
good statistics. One way of doing this with one-dimensional detectors is to increase
the counting time with increasing Q, ideally in a way inversely proportional to the
global decrease in intensity.
The final dimensionless total-scattering structure function SðQÞ has an average
value of unity (Egami & Billinge, 2012) and is given by:

1 Q = 2π=d = 2πs = 4π sinθ=λ.

 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
2000 2000 2000 2000

1800 1800 1800 1800

1600 1600 1600 1600

1400 1400 1400 1400

1200 1200 1200 1200

1000 1000 1000 1000

Rows

Rows
800 800 800 800

600 600 600 600

400 400 400 400

200 200 200 200

0 0 0 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Columns Columns

–0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Intensity *10E5 Intensity *10E4
11.2 Total scattering and atomic pair distribution function (PDF) analysis

Figure 11.1: Screenshots of the Fit2D program (Hammersley et al., 1996) showing the two-dimensional powder pattern of nickel (left) and the background
pattern of an empty capillary (right) measured at a wavelength of λ = 0.1839 Å at beamline X17A (NSLS) using a PerkinElmer amorphous silicon detector.
The red areas denote masked regions omitted from the integration (due to primary beam stop, dead pixels, etc).
255
 256

I(Q)

10 20 Q/Å-1
S(Q)

5 10 15 20 Q/Å-1
F(Q)
11 Total scattering methods

5 10 15 20 Q/Å-1
G(r)

5 10 15 20 25 r/Å

Figure 11.2: Screenshot of the PdfGetX3 program that is part of the xPDFsuite (Yang et al., 2014). From top to bottom: (1) raw integrated powder pattern I(Q)
from Figure 11.1; (2) total-scattering structure function S(Q); (3) normalized reduced total scattering structure function F(Q) and (4) atomic pair distribution
function (PDF) G(r). F(Q) = Q[S(Q)− 1] is called the reduced structure function and tends to zero at high Q.
 11.2 Total scattering and atomic pair distribution function (PDF) analysis 257

I ð QÞ h f 2 ðQÞi
SðQÞ = − + 1, hSðQÞi = 1, (11:1)
h f ðQÞi2 h f ðQÞi2

with hf 2 ðQÞi the mean of the squared form factors and hf ðQÞi2 the square of the
average form factor. For monoatomic materials this simplifies to:

I ðQÞ
SðQÞ = . (11:2)
h f ðQÞi2

In addition to the form factor fall off, the coherent intensities (the features) in I(Q) and
hence S(Q) decrease with increasing Q due to the Debye-Waller factor, which comes
from thermal and quantum zero-point motion of the atoms, and any static disorder in
the material. By a Q value of 30–50 Å−1 (depending on temperature and the stiffness
of the sample) there are usually no significant features in I(Q), and there is no need to
measure data to higher-Q.
From the total scattering structure function S(Q) one can calculate the reduced
pair distribution function, G(r).2 G(r) is related to S(Q) through a sine Fourier transform
according to:

ð
Qmax
2
GðrÞ = Q½SðQÞ − 1 sinðQrÞdQ. (11:3)
π
Qmin

It has slope − 4πρ0 r at low-r where ρ0 is the atomic number density in atoms/Å3, and
oscillates around zero at high-r (Figure 11.2, panel 4). The values for the integration
limits Qmin and Qmax are limited by the experimental setup. Qmax is usually lowered
below the experimentally measured maximum to prevent noise in high-Q data from
adversely affecting the Fourier transform. G(r) is one of several functions called the
PDF (see footnote 3).
Formally, G(r) is the autocorrelation (Section 13.7) of the atomic density. This is
obtained by taking the atomic density (the atoms at their respective positions) of the
molecule, cluster or crystal and convoluting it with a replica of itself. This object is
then averaged over all orientations to obtain the PDF. The PDF therefore contains
peaks at positions, r, that separate pairs of atoms in the solid with high probability
(Figure 11.3). For a completely random distribution of atoms, the PDF would be flat
(if atoms are allowed to overlap). For nonrandom distributions the PDF contains
valuable information on both short-range nearest-neighbor interactions and on pair-
correlations extending to much higher values of r. In fact, with high Q-space resolu-
tion data, PDFs can be measured out to hundreds of nanometers (thousands of

2 There are a number of different naming conventions in use for different “flavors” of radial
distribution functions. Unfortunately different communities (and often continents!) use the same
X(r) label for different functions. Keen (2001) has provided a useful collation of the different
formalisms.
258 11 Total scattering methods

4.26 Å

2.84 Å

2.46
35
30
1.42 Å 25

1.42

4.26
20

G(r) (Å–1)

4.92
15

3.76
2.84
10
2.46 Å

5.11
5
0
3.76 Å –5

4.92 Å 0 1 2 3 4 5
5.11 Å
r(Å)

Figure 11.3: Two-dimensional graphene layer and the corresponding PDF, with peaks corresponding
to pairs of atoms separated by distance r. The circles highlight radial distances from the central atom
where other atoms are found.

Angstroms) with the structural information remaining quantitatively reliable

(Levashov, Billinge & Thorpe, 2005).
There is a large number of other correlation functions related to G(r) that are
commonly used.3 G(r) has a number of advantages over other functions. First, it
arises directly from the experimental S(Q) with no assumption made about the atomic
number density ρ0 that appears in other expressions. Indeed, ρ0 is contained experi-
mentally in the low-r slope of G(r). The uncertainties in the data are also constant as a
function of r meaning that difference plots between observed and calculated G(r)
functions should show constant scatter with r. Finally, oscillations in G(r) continue
with equal amplitude up to r = ∞ for crystalline materials, meaning that departures
from this behavior (Section 11.3.2) give direct visual information on various sample or
experimental factors.
One useful and intuitive correlation function is the radial distribution function
(RDF), R(r), which is related to G(r) by:

RðrÞ
GðrÞ = − 4πρ0 r. (11:4)
r

3 For example, gðrÞ is the atomic pair distribution function and gives the probability of finding 2 atoms
at distance r. It is defined via GðrÞ = 4πρ0 rðgðrÞ − 1Þ such that gðrÞ ! 0 as r ! 0 and gðrÞ ! 1 as
r ! ∞. ρðrÞ = ρ0 gðrÞ is the atomic pair density function. ρðrÞ ! 0 as r ! 0 and ρðrÞ ! ρ0 as r ! ∞.
GðrÞ, gðrÞ, and ρðrÞ are all commonly abbreviated as PDF.
 11.2 Total scattering and atomic pair distribution function (PDF) analysis 259

The RDF is important because it is most directly related to the physical struc-
ture: R(r)dr gives the number of atoms in an annulus of thickness dr at distance
r from another atom. This means that the coordination number or the number of
neighbors, NC, around a chosen atom is given by:
rð2

NC = RðrÞdr, (11:5)
r1

where r1 and r2 define the beginning and ending positions of the RDF peak corre-
sponding to the coordination shell in question. One disadvantage of the RDF is that it
depends on r2 at high r making it inconvenient for plotting and visualization.
The RDF suggests a simple way that PDFs can be calculated from atomic models.
We can set up a model consisting of a large number of atoms situated in an arbitrary
box at positions r with respect to the origin. The relative atomic positions will give
rise to a set of atomic distances rjk, where rjk = |rj − rk|, for each pair of atoms in the
structure. We can therefore express the RDF as a sum of delta functions:

1X n X n  
RðrÞ = δ r − rjk , (11:6)
n j=1 k=1

where the double sum runs twice over all atoms in the sample (Figure 11.3). If several
types of atoms are present, the expression for R(r) is:

1X n X n
fj fk  
RðrÞ = δ r − rjk , (11:7)
n j = 1 k = 1 h f i2

where the fs are the form factors, evaluated at Q = 0, for the jth and kth atoms and h f i
is the sample average form factor.4
When eqs. (11.4) and (11.6) are used to calculate the PDF from a structural model,
the δ-functions are expressed as Gaussians with a width dependent on various
sample-related factors. The equation is also modified to account for various experi-
mental effects (Olds et al., 2018) as outlined in the following sections.
Since we can calculate R(r) and therefore G(r) from any atomic configuration, an
experimental G(r) can easily be fitted using a Rietveld-like approach, or analyzed
using methods such as simulated annealing. This is often called real space Rietveld
refinement. The size of box used for modeling depends on the type of system being
studied. If one is looking for small local deviations from an ordered crystal structure,
it’s common to use a box that corresponds to a single unit cell, often reduced to lower
symmetry. To model longer-range effects (or amorphous samples) a large-box
approach might be used.

4 Or scattering lengths for neutrons.

260 11 Total scattering methods

11.3 Parameters influencing the PDF

The appearance of an experimental PDF will depend strongly on a number of parameters

related to the structure and microstructure of the sample, and also on the experimental
conditions used. Each PDF peak is characterized by position, width and area. All of these
can be affected by the instrument resolution function and by PDF processing.
TOPAS calculates PDFs as a summation of Gaussian functions (Coelho, 2015). The
minimum/maximum extent of the Gaussians along the X-axis (r) defaults to the value
where y falls to 10−3 of its maximum height, but can be modified by changing the
expression:

pdf_ymin_on_ymax 0.001

11.3.1 Peak position in a PDF

The peak positions in a calculated PDF are determined by the bond lengths and angles
in the structural model. If fractional coordinates are used, these will depend on the size
of the simulation box containing the atoms (often a single unit cell in small-box
modeling of crystalline materials). It should be noted that the range of convergence
for coordinates is much lower for a PDF than for regular Rietveld refinement, since
coordinate shifts change peak positions in the PDF rather than intensities.
Any zero point can normally be satisfactorily described by a simple linear func-
tion. Fixed offsets are sometimes needed to correct for different conventions used by
different packages to produce the PDF. Note that the offset is applied in r-space so
that the correction needs to be at the TOPAS str level:

str
prm z0 0.00216 min -.1 max .1 val_on_continue = Rand(-0.05, 0.05);
prm z1 -0.03542 min -1 max 1 val_on_continue = Rand(-0.05, 0.05);
pdf_zero = z0 + z1 0.01 X;

11.3.2 Peak area of a PDF

The areas of PDF peaks are determined by coordination number, occupancies and the
scattering power of the atom pair. For materials in which the coherent domains have
a finite size, peak areas will decrease as a function of r due to a decreasing number of
pair–pair distances. For example, for nanoparticles there can clearly be no peaks at r
values greater than the particle size. Functions to describe this so-called peak
 11.3 Parameters influencing the PDF 261

damping have been calculated for various sample shapes (e.g., Kodama et al., 2006;
Korsunskiy et al., 2007). An example showing how to describe the damping function
of 5 nm spherical particles in TOPAS is given below:

prm !r 5 `radius of spherical nanoparticle in nm

scale_phase_X = IF X > 2 r
THEN 0
ELSE (Pi X^2 ((0.25 (X/r)^3)-(3 X/r)+4)) / (4 Pi X^2 1)
ENDIF ;

A phenomenological peak damping can be achieved using a unit area Gaussian

centered at r = 0 with a fwhm that can be refined to give some kind of size measure:

prm damp_fwhm 51.08423 min 1e-6 max = 10 X2;

prm damp = Gauss(0, damp_fwhm);
scale_phase_X = damp;

The resolution dQ in Q-space in which the data are collected also influences the
intensity of peaks in the PDF and is another source of peak damping. This can be
modeled by a normalized Gaussian function (Toby & Egami, 1992) with dQ as an
adjustable or fixed parameter (Figure 11.4). In TOPAS this is realized by:

prm !dQ 0.01 min 0.001 max 0.5 ' Instrumental resolution of S(Q) data
scale_phase_X = Exp(-0.5 X^2 (dQ/2.35482)^2);

dQ = 0.5
0.8 dQ = 0.1
Damping factor

0.6

0.4

0.2

0
0 20 40
r/Å

Figure 11.4: Gaussian damping factor with different resolutions dQ.

262 11 Total scattering methods

Since the low-r region is relatively unaffected by the Q-space resolution, PDF collec-
tions are often optimized for Qmax at the expense of dQ. The resulting damping causes
peak decay that is similar to that in a nanoparticle. Qualitative estimates of particle
size from visual inspection of PDFs should therefore be treated with caution.
For multiphase samples one will see peaks in a PDF from each component. It is
therefore possible to perform quantitative phase analysis in a similar manner to that
described in Chapter 5. The scale factor of each phase is proportional to its mole fraction.5

11.3.3 r-space resolution of a PDF

Since the density of interatomic distances in a PDF increases rapidly with r (Figure 11.3),
it becomes difficult to associate a PDF peak with a specific atom–atom pair past the first
few Å, even in simple structures. The ability to distinguish different distances is related
to the PDF resolution, Δr, which is dominated by the Qmax used in the Fourier transform.
According to Farrow et al. (2011) this can be approximated by:
π π
Δr  unless Qmax > pffiffiffiffiffiffiffiffiffi (11:8)
Qmax hu2 i
pffiffiffiffiffiffiffi
where hu2 i is the root-mean-squared atomic displacement. The Qmax limitation
relates to the fact that local thermal vibrations will cause an inherent smearing out
of interatomic distances and hence a broadening of peaks. High Q values may there-
pffiffiffiffiffiffiffi
fore not help resolution for materials with large values of hu2 i. Qmax values that can
be reached using laboratory instruments are 8 Å−1 for Cu-Kα1, 17 Å−1 for Mo-Kα1 and 22
Å−1 for Ag-Kα1, while at a high-energy synchrotron or neutron TOF beamline Qmax
values > 50 Å−1 might be measurable.
As we’ve just discussed, the sample-related peak width of a PDF is determined by
vibrations (phonons), and disorder. In TOPAS this can be modeled in several ways.
The simplest is to use an isotropic temperature factor, beq. This approach neglects the
fact that atomic motions of directly–bonded atoms are likely to be highly correlated,
which will lead to an overestimation of peak widths at low-r compared to high-r.
A better approximation might be to allow beq to vary as a function of r (X in the
TOPAS equation) using the error function (see Chapter 13):

prm erf_a 0.3 min 0

prm b1 1.0 min 0
…
site Ni1 x 0 y 0 z 0 occ Ni 1 beq = b1 Erf_Approx( erf_a X);

5 In TOPAS v6, the weight fractions reported for PDF fits in the gui and default MVW macros that are
defined for Rietveld refinement should therefore not be used.
 11.3 Parameters influencing the PDF 263

In many samples, we might expect different broadenings for different types of inter-
actions. For example, in relatively rigid molecules long range intramolecular distances
might be significantly stiffer than much shorter intermolecular (VDW controlled)
distances. To account for this, the keyword pdf_gauss_fwhm can be used to write
specific width equation for pairs specified by pdf_for_pairs. If all of the pairs possible
are described by pdf_for_pairs (e.g., pdf_for_pairs * *) then the associated beqs are not
used and they become redundant. An example is given below:

pdf_for_pairs C1 "C2 C3 C4" ' C2 to C4 in a different molecule to C1

pdf_gauss_fwhm = 0.1;

Note that the keyword pdf_for_pairs simply overwrites fwhms of the PDF peaks from
beq values, and values set by pdf_gauss_fwhm are used instead.

11.3.4 Processing artefacts

The fact that S(Q) can only be Fourier transformed over a finite Q range gives rise to
termination ripples in the PDF (Chung and Thorpe, 1997). As outlined in Chapter 13, the
Fourier transform of a box function is a sinc function with termination ripples. It is
important that these termination ripples aren’t misinterpreted as pair–pair distances.
Termination ripples can be reduced using, for example, a Lorch (or super-Lorch)
function (Soper & Barney, 2012). The function is applied at the Fourier-transform step
and effectively smears the density in G(r) space over a finite volume removing
ripples. There is, however, a price to pay in that the real-space resolution is degraded
making it harder to distinguish closely separated pair distances. If this type of function
is used when producing the experimental PDF, an appropriate correction (e.g.,
convolute_SoperLorch(!d_zero, 0.08) in TOPAS) must be applied to the calculated PDF.
Alternatively, it is possible to convolute additional functions like the sinc func-
tion directly into the PDF peak profile to describe the ripples. This can be done using
the keyword pdf_convolute. The following example (courtesy of Phil Chater) will
convolute a sinc function into the PDF profile:

prm !Step_Size = 0.02;

prm !q2 0.25 min= Step_Size / 10; max = Step_Size / 2;
prm !q3 0 min= -Pi; max = Pi;
pdf_for_pairs * *
pdf_gauss_fwhm @ 0.1
local conv_max = (5 Qmax - Mod(5 Qmax,1))/5 2 Pi / Qmax;
pdf_convolute = If(Abs(X) > Yobs_dx_at(Xo),(Sin(Qmax) X)/(X)),Qmax);
min_X = Min(-Xo,-conv_max) ;
max_X = Max( Xo, conv_max) ;
264 11 Total scattering methods

11.4 Example: Ni standard

A Ni standard sample was measured at ambient conditions at beamline 28-ID-2 of the

National Synchrotron Light Source II using the rapid acquisition PDF mode (Chupas
et al., 2003), with an X-ray wavelength of 0.1827 Å using a PerkinElmer two-dimensional
detector (2048 × 2048 pixels and 200 × 200 μm pixel size) mounted orthogonal to the
incident beam path with a sample-to-detector distance of 219.431 mm. Calibration and
integration of the two-dimensional images were performed using the program Fit2D
(Hammersley et al., 1996), and Fourier transformation to the PDF was performed using
PDFgetX3 in xPDFsuite (Juhas et al., 2013; Yang et al., 2014).
The INP file below shows the TOPAS commands needed to perform a PDF fit to
the experimental G(r). The only significant difference to a standard Rietveld INP file
is the inclusion of the line pdf_data. The data are assumed to be in a two (.xy) or
three (.xye) column format file in PDFgetX3.gr format. If other formats are used, an
appropriate scaling should be applied.6 The INP file uses the r-dependent displace-
ment parameter discussed in Section 11.3.3 and a linear correction for peak posi-
tions. In general, the absolute scaling of the G(r) data coming out of PDFgetX3 is
arbitrary and will later be accounted for by the scale factor of the model. The fit is
shown in Figure 11.5.

15
10
5
Rel. intensity

0
–5
–10
–15
–20
5 10 15 20 25 30 35 40 45 50 55 60 65 70
r/Å

Figure 11.5: Fit of the PDF of a nickel reference material. Note the termination ripples apparent at low r.

6 Different software packages produce PDFs with different normalizations. TOPAS expects the PDF to
vary from − 4πρ0 r at low-r to 0 at high-r. This is the .gr format of PDFgetX2 and PDFgetX3 (with a scale
factor applied), the .mdor01 format produced by GudrunN when the sample is set to normalize to <b>2,
the .dofr format produced by GudrunX when “Divide by <F>2” is selected in the normalization or the
dofr.xy format of DAWN. The equivalent function in the large-box RMCPROFILE package is Dnorm(r).
Alternative X(r) formats can be fitted using TOPAS scale_phase_X and fit_obj commands. For example,
if the data are in the Gʹ(r) format of Keen (2001), the following commands can be used: prm number_-
density = num_atoms / Get(cell_volume); scale_phase_X = If(X>0.01,1/(4 Pi number_ density X),0);
fit_obj = If(X>0.01,1,0);
 11.5 Calculation of a PDF 265

#include "pdf.inc" ' Read pdf.inc, by Phil Chater if not already in your local.inc

xdd "Ni_standard_real-space.xy"
start_X 1.4
finish_X 70.0
pdf_data

' Refine Qresolution damping and set the window for the Fourier truncation of the data
dQ_damping(dQ, 0.09817)
convolute_Qmax_Sinc(!Qmax, 22)
str
phase_name "Ni"
space_group Fm-3m
Cubic( @ 3.523588 )
site Ni1 x 0 y 0 z 0 occ Ni 1
scale @ 0.99676674 ' 1.0 for a fully normalised G(r)
' Define beq by PDFfit2 method, using Uiso, delta1 for correlated motion of nearby atoms,
' and qbroad for Q-dependent broadening
beq_PDFfit2(uiso, 0.00632, !rcut, 0.0, !sratio, 1.0, delta1, 0.82068, !delta2, 0.0,
qbroad, 0.01422)

11.5 Calculation of a PDF

A PDF can be simulated in a similar way to a regular powder diffraction pattern (see
Chapter 12). The following TOPAS script calculates a PDF for nickel from r = 0 to 100 Å
with a step width of dr = 0.01 Å without any damping (Figure 11.6):

12
10
8
6
Rel. intensity

4
2
0
–2
–4
–6
–8
–10
–12
–14
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
r/Å

Figure 11.6: Simulated PDF of nickel without damping.

iters 0
yobs_eqn !Ni_PDF.xy = 1; min 0 max 100 del 0.01
pdf_data
Out_X_Ycalc(Ni_PDF.xy)
266 11 Total scattering methods

str
phase_name "Ni"
space_group Fm-3m
Cubic(3.524)
scale 1
prm z0 0.05000 prm z1 0.50000 pdf_zero = z0 + z1 0.01 X;
site Ni1 x 0 y 0 z 0 occ Ni 1 beq = 0.75 Erf_Approx( 0.3 X);

11.6 Joint refinements | Bragg + PDF

One useful feature of TOPAS is the ability to combine PDF fitting with Rietveld
refinement and/or the global optimization method of simulated annealing. With
disordered materials this approach can, for example, help constrain bond lengths
in the Rietveld fit to their better-defined PDF values. The PDF dataset can either be
derived from the powder pattern used for Rietveld refinement, or from one measured
with PDF-optimized parameters (good statistics at high Qmax). Both patterns are then
evaluated by refining common parameters derived from both. Data sets are weighted
according to the summed intensities, but this can be modified by the user.
A simple constrained PDF+Rietveld refinement of the Ni standard is shown
in Figure 11.7, and the corresponding TOPAS script is given below. The cubic
lattice parameter and the isotropic displacement parameter are jointly refined
parameters:

32000
30000
28000
26000
24000
22000
20000
Rel. intensity

18000
16000
14000
12000
10000
8000
6000
4000
2000
0
–2000
0 10 20 30 40 50 60 70
2θ/° or r/Å

Figure 11.7: Rietveld (lower) and PDF plots (upper) from a combined Rietveld and PDF refinement of Ni.
 11.6 Joint refinements | Bragg + PDF 267

#include "pdf.inc" ' Read pdf.inc, by Phil Chater if not already in your local.inc

'------[ corefined parameters ]------

prm aNi 3.52427 ' lattice parameter of Ni
prm bNi 0.42202 min 0.0 ' displacement parameter of Ni

'------------[ PDF data ]------------

xdd "Ni_standard_real-space.xy"
xdd_sum !sum1 = Abs(Yobs);
weighting = 1/sum1;
start_X 1.4
finish_X 70.0
dQ_damping(dQ, 0.09279)
convolute_Qmax_Sinc(!Qmax, 22)
pdf_data

str
phase_name "Ni PDF"
space_group Fm-3m
Cubic(=aNi;)
scale @ 0.997
site Ni num_posns 4 x 0 y 0 z 0 occ Ni 1
beq_PDFfit2(uiso, =bNi/8/Pi^2;, !rcut, 0.0, !sratio, 1.0, delta1, 0.63, !delta2, 0.0,
qbroad, 0.02)

'------------[ XRD data ]------------

xdd "Ni_standard_2th.xy"
xdd_sum !sum2 = Abs(Yobs);
weighting = 1/sum2;
x_calculation_step 0.02
convolution_step 5
bkg @ 210.26 -64.69 -22.25 3.36 -1.68 -0.35 0.62 -0.15 -2.88 2.51 -3.18
start_X 3
LP_Factor( 90)
lam
ymin_on_ymax 0.0001
la 1 lo 0.1827 lh 1e-005
str
phase_name "Ni XRD"
LVol_FWHM_CS_G_L( 1, 6.16204, 0.89, 5.97526,@, 6.89316,@, 138.30578)
e0_from_Strain( 0.00110,@, 0.46394,@, 0.07777)
scale @ 0.000216628033
space_group Fm-3m
Cubic(=aNi;)
site Ni num_posns 4 x 0 y 0 z 0 occ Ni 1 beq=bNi;
268 11 Total scattering methods

References
Coelho, A.A., Chater, P.A., Kern, A. (2015): Fast synthesis and refinement of the atomic pair distri-
bution function. J. Appl. Cryst. 48,869–875.
Chung, J.S., Thorpe, M.F. (1997): Local atomic structure of semiconductor alloys using pair distribu-
tion functions. Phys. Rev. B 55,1545–1153.
Chupas, P.J., Qiu, X., Hanson, J.C., Lee, P.L., Grey, C.P., Billinge, S.J.L. (2003): Rapid-acquisition pair
distribution function (RA-PDF) analysis. J. Appl. Cryst.36, 1342–1347.
Egami, T. Billinge, S.J.L. (2012): Underneath the Bragg peaks, structural analysis of complex materi-
als, Volume 16, 2nd edition, Pergamon, Oxford (UK), 422 pages.
Farrow, C.L., Shaw, M., Kim, H., Juhás,P., Billinge, S.J.L. (2011): Nyquist-Shannon sampling theorem
applied to refinements of the atomic pair distribution function. Phys. Rev. B 84, 134105.
Hammersley, A.P., Svensson, S.O., Hanfland, M., Fitch, A.N., Hausermann, D. (1996): Two-dimensional
detector software: from real detector to idealised image or two-theta scan. High Pressure Research
14,235–248.
Juhas, P., Davis, T., Farrow, C.L., Billinge, S.J.L. (2013): PDFgetX3: a rapid and highly automatable
program for processing powder diffraction data into total scattering pair distribution functions.
J. Appl. Cryst. 46, 560–566.
Keen, D.A. (2001): A comparison of various commonly used correlation functions for describing total
scattering. J. Appl. Cryst. 34, 172–177.
Kodama, K., Tikubo, S., Taguchi, T., Shamoto, S. (2006): Finite size effects of nanoparticles on the
atomic pair distribution functions. Acta. Cryst. A 62, 444–453.
Korsunskiy, V.I., Neder, R.B., Hofmann, A., Dembski, S., Graf, C., Rühl, E. (2007): Aspects of the
modelling of the radial distribution function for small nanoparticles. J. Appl. Cryst. 40, 975–985.
Levashov, V.A., Billinge, S.J.L., Thorpe, M.F. (2005): Density fluctuations and the pair distribution
function. Phys. Rev. B 72, 024111.
Olds, D., Saunders, C.N., Peters, M., Proffen, T., Neuefeind, J., Page, K. (2018): Precise implications
for real-space pair distribution function modeling of effects intrinsic to modern time-of-flight
neutron diffractometers. Acta Cryst. A 74, 293–307.
Soper, A.K., Barney, E.R. (2012): On the use of modification functions when Fourier transforming total
scattering data. J. Appl. Cryst. 45, 1314–1317.
Toby, B.H., Egami, T. (1992): Accuracy of pair distribution function analysis applied to crystalline and
non-crystalline materials. Acta Cryst. A 48,336–346.
Yang, X., Juhas, P., Farrow, C.L., Billinge, S.J.L. (2014): xPDFsuite: an end-to-end software solution for
high throughput pair distribution function transformation, visualization and analysis, arXiv
1402.3163v3.
12 Multiple data sets

A major strength of powder diffraction lies in its ability to measure diffraction patterns
or PDFs under the influence of external variables such as temperature, pressure, time,
changing chemical environment, exposure to light and so on. Recent progress in
detectors in the laboratory (silicon strip position sensitive detectors) and at the syn-
chrotron (large two-dimensional area detectors) allows the collection of huge numbers
of powder patterns during in situ or operando powder diffraction experiments.
There are, in principle, two ways to analyze such a series of powder patterns:
sequential and parametric. In sequential analysis, Rietveld analysis of each powder
pattern is done separately based on the results of the previous refinement, with the
entire set of all relevant parameters refined separately for each pattern. Further
analysis of the values of these parameters, for example, fitting to empirical or physi-
cally based functions, is then performed after the Rietveld refinements.
In contrast, in parametric refinement (sometimes called surface refinement), all
powder patterns of a series are analyzed simultaneously with selected parameters
refined across all the powder patterns using an appropriate functional form. The
parametric approach can greatly reduce the number of parameters and can allow the
refinement of nonstructural parameters like transition temperatures or critical expo-
nents. It also allows simple physical models to be imposed on the refinement, which
helps constrain it to “sensible” solutions when there is more than one equivalent or
similar model. The caveat is, of course, that the validity of the applied functions must
be carefully checked.
In addition, if structural data are available, powder patterns or PDFs can easily be
simulated as a function of internal variables like a particular torsion angle, crystallite
size and so on. This can be a powerful method of assessing where key information is
contained in the data.
In this chapter, we will show how powder pattern simulation, sequential and
parametric refinement can be realized using the scripting language of TOPAS.

12.1 Simulation of angle dispersive powder patterns

Simulating powder patterns is a powerful way to visualize how changes in the
instrumental configuration, the application of specific correction functions or small
changes of the crystal structure will change a data set. Simulation can be very useful
in designing experiments and data analysis strategies.

12.1.1 Simulation of a single powder pattern

To simulate a pattern in TOPAS, the keyword yobs_eqn is used in place of the xdd
keyword. It replaces the observed data by an equation. The name given to the equation

https://doi.org/10.1515/9783110461381-012
270 12 Multiple data sets

!name is used for identifying the plot of the equation in the GUI. The simulated powder
pattern(s) can be saved, for example, as an .xy file using the Out_X_Ycalc macro. In the
following example, an angle dispersive, equal step powder pattern of the LaB6 line
profile standard (NIST SRM 660a LaB6) is created from 12° to 120° 2θ with a step of 0.01°
2θ and saved to the file “D8_Mo_LaB6_capillary-simulated.xy” (Figure 12.1).

110
100
90
Sqrt (rel. intensity)

80
70
60
50
40
30
20
10
15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115
2θ/°

Figure 12.1: Simulated powder pattern of the LaB6 line profile standard in Debye–Scherrer geometry
using Mo-Kα1 radiation.

The instrumental parameters describe a Bruker D8 advance powder diffractometer in

Debye−Scherrer geometry with Mo-Kα1 radiation (λ = 0.7093 Å) from a Ge(220)
primary beam monochromator (angle for Lorentz polarization factor of 20.5°),
and a silicon strip position sensitive detector. Parameters for the peak shape
function (TCHZ_Peak_Type) and for the asymmetry due to axial divergence
(Simple_Axial_Model) are taken from a previously determined instrumental resolu-
tion function of this diffractometer. The background is simulated by a Chebyshev
polynomial of 10th order. The number of iterations (iters) is set to zero as no
parameter is refined. The entire INP file for simulation of the powder pattern is
given here:

' Create a simulated angle dispersive powder pattern of LaB6

iters 0
yobs_eqn !calc.xy = X; min 12 max 120 del .01
Out_X_Ycalc( D8_Mo_LaB6_capillary-simulated.xy )

bkg 67.72 -85.24 64.89 -44.30 31.82 -21.40 12.48 -6.41 2.38 -0.58 -0.41
LP_Factor( 20.5)
Zero_Error(, 0.0)
Rs 217.5
Simple_Axial_Model( 9.5)
lam
 12.1 Simulation of angle dispersive powder patterns 271

ymin_on_ymax 1e-005
la 1 lo 0.7093 lh 0.2695
str
TCHZ_Peak_Type(, 0.0106885,, -0.0011298,, 0.000521,, 0,, 0.041852,, 0)
phase_name "LaB6 simulated"
space_group Pm-3m
scale 0.0003
cubic( 4.154782417)
site La num_posns 1 x =0;: 0.0 y =0; : 0.0 z =0; : 0.0 occ La+3 1 beq 0.39
site B num_posns 6 x 0.19986 y =1/2; : 0.5 z =1/2; : 0.5 occ B 1 beq 0.24

For a more realistic pattern, random noise can be added. In the following
script, Poisson statistics are assumed (typical for single counter detectors), and
noise proportional to the square root of the intensity is added to the calculated
profile:

xdd_out sim_noise.xy load out_record out_fmt out_eqn

{
" %11.6f " = X;
" %11.6f\n" = Rand_Normal(Ycalc, Sqrt(Ycalc));
}

12.1.2 Simulation of a series of powder patterns

It is often useful to simulate a series of powder patterns as a function of one or more

variables. Instead of manually changing and rerunning the INP file, TOPAS batch
commands can perform all the simulations in a single step. TOPAS calls each
simulation a “run”.
In the following example, an increasing Lorentzian strain broadening is
added to the previous LaB6 simulation, creating a series of 10 simulated
powder patterns (Figure 12.2). The number of simulations (runs) is defined
by the variable num_runs and the current simulation number is
defined by Run_Number. These variables control the way TOPAS executes
the INP file:

' Create a series of 10 simulated patterns with increasing Lorentzian strain broadening
num_runs 10
iters 0
' Save INP file as an OUT
out_file = Concat(String(INP_File), ".out");
system_after_save_OUT { copy INP_File##.out INP_File##_##Run_Number##.out }

yobs_eqn !calc##Run_Number##.xy = X; min 12 max 120 del .01

' Save each simulated powder pattern in the XY file format
Out_X_Ycalc( D8_Mo_LaB6_capillary-simulated_##Run_Number##.xy )
272 12 Multiple data sets

bkg 67.72 -85.24 64.89 -44.30 31.82 -21.40 12.48 -6.41 2.38 -0.58 -0.41
LP_Factor( 20.5)
Zero_Error(, 0.0)
Rs 217.5
Simple_Axial_Model( 9.5)
lam
ymin_on_ymax 1e-005
la 1 lo 0.7093 lh 0.2695
str
phase_name "LaB6 series simulated"
TCHZ_Peak_Type(, 0.0106885,, -0.0011298,, 0.0005207,, 0,, 0.041852,, 0)
lor_fwhm = 0.1 Run_Number Tan(Th);
space_group Pm-3m
scale 0.0002823909984
Cubic( 4.154782417)
site La num_posns 1 x =0;: 0.0 y =0; : 0.0 z =0; : 0.0 occ La+3 1 beq 0.39
site B num_posns 6 x 0.19986 y =1/2; : 0.5 z =1/2;: 0.5 occ B 1 beq 0.24

The line out_file creates an output (OUT) file for each simulation in case the simulated
data are needed again. The OUT file comprises the INP file but with parameter values
updated. In this example, the OUT file is created using the name of the INP file plus
the extension “.out” then copied to a unique name. More complex examples can be
performed using the #list command discussed in Section 12.2.

110
100
90
Sqrt (rel. intensity)

80
70
60
50
40
30
20
10
0
13 14 15
2θ/°

Figure 12.2: Blow up of a series of simulated powder patterns of LaB6 in Debye–Scherrer geometry
using Mo-Kα1 radiation with increasing Lorentzian strain broadening.

12.1.3 Scripting and the command line mode

For more complicated examples needing loops or parameter input from other
sources or software, more elaborate scripting languages like Perl or Python might
be used. In this situation, TOPAS can be called stand-alone from the windows
 12.1 Simulation of angle dispersive powder patterns 273

command line or from a powershell. Specific parameters and instructions can then
be automatically passed to TOPAS using macros and #define commands.
An example of the command line format is as follows:

C:\topas_v5> tc C:\directory\file.inp "macro filename {C:\directory\d9_02592} macro rangeuse {22}

macro information {prm !te 300 } #define CMD_LINE"

The corresponding INP file, file.inp, can contain any of the usual TOPAS scripting
language, but would need the additional sections:

#ifdef !CMD_LINE 'use these lines when in gui mode

macro filename {d8_022592}
macro rangeuse { 12}
macro information {prm !te 300 }
#endif

information

' Instead of giving the raw filename and range number in the INP file use the words
' "filename" and "rangeuse" which get replaced by information in the macros above or
' from command line.

RAW(filename)
range rangeuse

To execute this command line instruction, one needs to be in the main TOPAS
directory. tc calls the program tc.exe, which is the command line version of
TOPAS. It reads instructions from the INP file file.inp, which is best stored in a
different directory. Any text placed in the quotation marks on the command
line is passed to TOPAS. Here the first two macros define the name of the
Bruker RAW data file to be analyzed and the data set range number within that
file. The next macro passes information to the refinement as parameters – here
the temperature at which the data were recorded. The final section “#define
CMD_LINE” is a convenient trick that allows the same INP file to be run by
TOPAS in either command line mode (tc.exe) or gui mode (topas.exe or ta.exe).
This turns out to be very useful when testing INP files. The INP file contains a
small section “#ifdef !CMD_LINE . . . #endif”. This syntax means that TOPAS
will read this part of the INP file if CMD_LINE has not (!) been defined and will
ignore it otherwise. When ta.exe or topas.exe is run macros filename, rangeuse
and information are therefore defined via the INP file. When tc.exe is run, the
command line flag #define CMD_LINE means that they are instead defined
from the command prompt instruction.
274 12 Multiple data sets

12.2 Sequential Rietveld refinement

12.2.1 Introduction

Evaluation of a large numbers of powder patterns measured as a function of an

external variable is potentially very time consuming. Fortunately, TOPAS can auto-
mate the data handling in most cases. For instance, if the step change of the external
variable (e.g., temperature) between data sets is small, the refined parameters of one
diffraction pattern can be used as starting values for the following diffraction pattern
in an automated manner, provided that the important parameters don’t change
significantly. This can happen if, for example, a phase transition occurs. This
approach is called sequential Rietveld refinement. If there are major step changes,
TOPAS offers some flexibility through conditional statements to change the type and
number of strs used as a function of the external variable(s).

12.2.2 Sequential refinement

From TOPAS v6 onward sequential refinements can be realized within a single INP
file using a few keywords and macros.1 In order to tell the program which data sets
(powder patterns) should be analyzed, a one-dimensional array (list) of strings (#list
list_name) is used. Each item in the list denotes a filename of a powder pattern, and
the position in the list is given an index (starting at zero) which allows access to the
corresponding powder pattern (e.g., list_name(2))2:

#list File_Name
{
file1.xy
file2.xy
file3.xy
}

The number of runs that should be performed (usually equivalent to the number of
data files) is given by the constant num_runs. Once started, the running variable
Run_Number contains the running index from 0 until num_runs–1. This variable
allows access to the powder pattern to be analyzed from the #list by list_name
(Run_Number) (e.g., xdd File_Name(Run_Number)):

1 The same process can be readily achieved using scripting and the command line mode in v5 and
earlier.
2 Hint: The contents of a directory can be conveniently copied into a text file by typing e.g. dir *.raw >
filelist.txt at the command prompt of the windows command interpreter cmd.exe.
 12.2 Sequential Rietveld refinement 275

num_runs 3
…
xdd File_Name(Run_Number)

Parameters like temperature or time, which might be needed for identification

or calculation purpose, may also be defined as a function of Run_Number. In
the following example, the starting temperature is 25 °C which is increased by
2.6667 °C for each 10 s run until it reaches 450 °C after which it stays
constant:

prm !Nr= Run_Number;

prm !Seconds= 10 Nr;
prm !DegCelsius = If( (Nr * 2.6667 + 25) > 450, 450, Nr * 2.6667 + 25);

If temperature values do not have a simple functional dependency on Run_number,

and cannot be calculated, they can be included in the #list as a second column of real
numbers. Any additional metadata and/or parameters can be included in the same
way:

#list File_Name Temperature

{
file1.xy 25.0
file2.xy 27.8
file3.xy 39.5
}

Most frequently, the output of the current run should serve as input for the
following run. Therefore, an OUT file must be created after each convergence
and then copied to the INP filename. It sometimes helps to keep a copy of each
OUT file for quality control, which can be achieved, for example, by converting
the Run_number to a string to be included in the filename. In TOPAS syntax this
can all be achieved with the following:

out_file = Concat(String(INP_File), ".out"); ‘ create an output file

system_after_save_OUT
{
copy INP_File##.out INP_File##Run_Number##.out ‘save an individ. copy of the output file
copy INP_File##.out INP_File##.inp ‘ create the next input file
}
276 12 Multiple data sets

Two different keywords can be used, depending on whether the system commands
should be executed before (system_before_save_OUT) or after (system_after_save_OUT)
the OUT file is created.
There are various ways to report refined values of parameters of interest for
each Run_Number to an external file. The least controlled is the command
out_prm_vals_ on_convergence, which writes each parameter in the model to a
file. If only a small number of parameters are of interest, then the out command
can be used either directly or via macros. For example, the macro Append_0(file)
creates a new file at Run_Number = 0 then appends the data of the following runs.
The following macro can be called from different phases with a different filename
for each phase:3

macro Parameters(file, time, temperature)

{
Append_0(file)
Out_String("\n") ' puts new line in file
Out(time, " %V")
Out(temperature, " %V")
. . .
Out(Get(cell_volume), " %V")
Out(Get(r_bragg), " %V")
}

Additional parameters of interest can always be added to a previously used results

file. For example, the following script first calculates the integral breath and then
writes it to the results file. If a parameter doesn’t exist, TOPAS will stop writing the
results file. This can be circumvented with the command Prm_There(), which tests if
the parameter exists, before it is used for calculation or report:

LVol_FWHM_CS_G_L( 1, 14.22392`, 0.89, 15.58663`,prm_CL, 22.07950`,prm_CG, 49.73158`)

prm prm_IB = 1 / IB_from_CS(CV(, prm_CL), CV(, prm_CG));
if Prm_There(prm_IB) { Out( prm_IB, " %V") }

Once all refinements are done, the results files can be read by a scientific plotting
program like ™Origin, ™Grapher, ™Excel or gnuplot to analyze the data posteriori.
Section 12.3.3 provides some simple gnuplot tricks for automating this.
For multiphase samples, phases that disappear during an experiment often
refine to give unrealistically broad Bragg reflections, which can correlate with the
background leading to incorrect phase fractions. This makes the definition of a lower
size limit useful:

3 The %V in the output command is a c++ like format descriptor. The various options are described in
the Technical Reference.
 12.2 Sequential Rietveld refinement 277

macro lower_size_limit { 11 }
…
LVol_FWHM_CS_G_L( 1, 7.202120694, 0.89, 10.06647809,,,@, 11.30909565 min lower_size_limit)

Alternatively, all phases with fractions lower than, for example, 0.7 weight% can be
automatically removed from the refinement using the following:

for strs { Remove_Phase 0.7 }

A more elegant way is to define the occurrence of phases using precompiler condi-
tional statements. In the following example, phase 1 exists from 158 K < t < 198 K and
phase 2 from 198 K < t < 226 K. The phases are then automatically turned on and off
during the refinement:

' Phase boundaries

#prm t = 100 + (Run_Number 4); ‘Temperature = 100 + Run_Number * 4
#if Or(t < 158, t > 226);
#define Phase1
#elseif And(t > 158, t < 198);
#define Phase1
#define Phase2
#elseif And(t > 198, t < 226);
#define Phase2
#endif
…
#ifdef Phase1
str
…
#endif

We’ll see in Section 12.3 that parametric refinement can remove the need for such
tricks, allowing a consistent model for all data sets.

12.2.3 Example of a sequential Rietveld refinement

The phase composition and particle size of carbonyl iron powder (CIP) was investigated
on heating in a 2 bar stream of nitrogen gas (König, 2017). CIP contains 0.7–1 weight%
carbon, and, depending on process conditions, nitrogen and oxygen below 1 weight%.
A series of synchrotron X-ray powder diffraction patterns was collected as the CIP
sample was heated from room temperature (RT) to 450 °C at a heating rate of 10 °C/min.
The measurement time was fixed to 10 s/scan. Two hundred and fifty-five patterns were
recorded as the sample was heated, then another 93 powder patterns while the final
278 12 Multiple data sets

temperature was held constant. α-Fe, Fe3C, Fe4N and Fe3O4 were found in varying
amounts as crystalline phases. Full quantitative sequential Rietveld refinements as a
function of time and temperature were performed. A typical Rietveld plot is shown in
Figure 12.3 and selected refined quantities in Figure 12.4.

360 Fe 77.51 %
340
320 Fe4N 11.76 %
300 Fe3O4 1.21 %
280
Sqrt (rel. intensity)

260 Fe3C 9.53 %

240
220
200
180
160
140
120
100
80
60
40
20
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
2θ/°

Figure 12.3: Typical Rietveld plot showing a full quantitative analysis of a single powder pattern of CIP
at high temperature in a hot nitrogen gas stream. The square root of the intensity is displayed for
better visualization of the minor phases.

100 140
95
120
90
Crystallite size from IB [nm]

85 100
Fe
80
Weight%

80
75
20 60
15
40
Fe4N
10
20
5 Fe3C
Fe3O4
0 0
0 500 1000 1500 2000 2500 3000 3500 0 500 1000 1500 2000 2500 3000 3500
Time/s Scan number

0 150 300 450 450 0 150 300 450 450

Temperature/°C Temperature/°C

Figure 12.4: Results of the sequential Rietveld refinements of CIP in a hot nitrogen gas stream as a
function of time and temperature. Left: weight fractions of all phases. Right: crystallite sizes of iron
particles from integral breath data.

The TOPAS script for the sequential quantitative Rietveld refinements is given below.
Only two phases (α-Fe and Fe4N) are shown. In addition to the parameters listed in the
macro Parameters(), the integral breath and two anisotropic microstrain parameters
 12.2 Sequential Rietveld refinement 279

700
650
600
Sqrt (rel. intensity)

550
500
450
400
350
300
250
200
150
100
50
4 5 6 7 8 9 10 11 12 13 14 15
2θ/°

Figure 12.5: A 3D Rietveld plot for a sequential Rietveld refinement series of CIP in a hot nitrogen gas
stream depending on scattering angle and time/temperature.

for α-Fe are reported for each run. The 3D Rietveld plot for the sequential Rietveld
refinement series is shown in Figure 12.5.

#list File_Name
{
CIP_3_SDD986_10s-00266.chi.xy
CIP_3_SDD986_10s-00267.chi.xy
…
CIP_3_SDD986_10s-00613.chi.xy
}
num_runs 348
out_file = Concat(String(INP_File), ".out");
system_after_save_OUT
{
copy INP_File##.out INP_File##Run_Number##.out
copy INP_File##.out INP_File##.inp
}
prm !Nr= Run_Number;
prm !Seconds= 10 Nr;
prm !DegCelsius = If( (Nr * 2.6667 + 25) > 450, 450, Nr * 2.6667 + 25);

' lower crystallite domain size limit of the crystalline (!) phases in nm
macro lower_size_limit{ 11 }

macro Parameters(file, time, temperature)

{
Append_0(file)
Out_String("\n")
Out(time, " %V")
Out(temperature, " %V")
Out(Get(weight_percent), " %V")
Out(Get(a), " %V")
280 12 Multiple data sets

Out(Get(b), " %V")

Out(Get(c), " %V")
Out(Get(al), " %V")
Out(Get(be), " %V")
Out(Get(ga), " %V")
Out(Get(cell_volume), " %V")
Out(Get(r_bragg), " %V")
}
xdd File_Name(Run_Number)
x_calculation_step 0.02
bkg @ 338.64 -66.14 32.72 -53.26 17.48 -14.59 -5.33 -2.01 6.04 7.99 -3.68
start_X 3
finish_X 16.3
LP_Factor( 90)
Zero_Error(@, -0.01674` min -0.02 max 0.02)
exclude 4 4.1
lam
ymin_on_ymax 0.001
la 1 lo 0.20717 lh 1e-005
str ‘α-Fe
Stephens_cubic(, 1 min =0;,prm_S400, 67953.87 min =0;,prm_S220, 0.0 min =0;)
TCHZ_Peak_Type(, 0,, -0.00124,, 0.00123,, 0,, 0.0106,, 0.00426)
LVol_FWHM_CS_G_L( 1, 14.22392`, 0.89,15.58663`,prm_CL, 22.07950`,prm_CG, 49.73158`)
phase_name "Fe"
space_group Im-3m
scale @ 0.000441464724`
Cubic(@ 2.859845`)
site Fe1 num_posns 2 x 0 y 0 z 0 occ Fe 1 beq B1 0.46191` min =0; max =1;
prm prm_IB = 1 / IB_from_CS(CV(, prm_CL), CV(, prm_CG));

Parameters("Fe_CIP3.dat", Seconds, DegCelsius)

if Prm_There(prm_S400) { Out( prm_S400, " %V") }
if Prm_There(prm_S220) { Out( prm_S220, " %V") }
if Prm_There(prm_IB) { Out( prm_IB, " %V") }
str ‘ Fe4N
LVol_FWHM_CS_G_L( 1, 7.00282`, 0.89, 9.79000`,,,@, 11.00 min lower_size_limit)
TCHZ_Peak_Type(, 0,, -0.00124,, 0.00123,, 0,, 0.0106,, 0.00426)
phase_name "Fe4N"
space_group Pm-3m
scale @ 4.39949789e-006`
Cubic(@ 3.774593`)
site Fe1 num_posns 3 x 0.5 y 0 z 0 occ Fe 1 beq B2 1.0
site Fe2 num_posns 1 x 0.5 y 0.5 z 0.5 occ Fe 1 beq =B2;
site N1 num_posns 1 x 0 y 0 z 0 occ N 1 beq =B2;

Parameters("Fe4N_CIP3.dat", Seconds, DegCelsius)

…
 12.3 Parametric Rietveld refinement 281

12.3 Parametric Rietveld refinement

In the previous sections, we’ve discussed sequential refinements where individual data
sets are analyzed independently. There are, of course, many situations where it’s
necessary to analyze an ensemble of data collected under evolving conditions (a
“surface” of diffraction data) in a more sophisticated way. For example, you might
want to derive some parameters from all the data sets rather than from a single (noisy)
pattern. A trivial example might be the diffractometer zero point during a series of time-
resolved experiments. The instrument misalignment leading to the zero point is
unlikely to change during the course of the experiment, and it is therefore best
determined from all data sets simultaneously. Similarly, when analyzing a variable
temperature experiment in Bragg–Brentano geometry, one might need to refine a
height correction to allow for thermal expansion of the sample mount. The correction
is likely to change smoothly with temperature, and is probably best described using a
simple empirical function derived from all the data. The fitting of a smooth function to
describe parameters from a surface of diffraction data is called parametric Rietveld
refinement. Since zero point and sample height both correlate with cell parameters,
describing these parameters parametrically would lead to reduced uncertainties in cell
parameters, that is, we have a method to extract “good information” from “bad data.”
The flexibility of the TOPAS scripting language means that it’s possible to para-
metrize a wide range of different parameters. Some of these and typical functions
used to describe them are given in Table 12.1.
The parametric approach is most powerful in cases where there is a significant
correlation between different refined parameters, or multiple models that could fit an
individual data set equally well. In this situation, the parametric equations can guide
the refinement toward a unique or physically sensible minimum. The second major
strength comes from the ability to refine “non-crystallographic” parameters, which
wouldn’t be defined by a single data set. Examples include kinetic parameters, critical
exponents of phase transitions or temperature calibration curves for in situ experiments.
When looking for subtle changes in a parameter of interest (e.g., a specific site
occupancy or atomic coordinate), the best way to use the parametric approach is
probably to parameterize the variables you’re not interested in and to freely refine
the parameter of most interest.
The danger of the parametric approach is that a physical model is imposed on the
refinement as a mathematical constraint and it is important to ensure that it is valid.
One way to assess this is to compare the R-factor for each data set from a parametric
fit with that from sequential refinement. If the parametric fit is significantly worse,
then the model may not be describing the experiment well. That said, identifying an
inappropriate model is actually a major strength of the parametric approach: In
fitting an individual data set, refined model parameters can adopt physically non-
sensical values and incorrectly fit features of the data. This is prevented in the more
demanding parametric approach. Discrepancies in plots of Rwp over the refined
282 12 Multiple data sets

Table 12.1: Examples of functions used in parametric refinements.

Quantity Expression Comment

Cell parameters or adps c1 θ1 Simple Einstein-like model with

aðT Þ = a0 +
eðθ1 =T Þ − 1
a, c and θ as refinable
parameters. Ensures a physically
sensible zero gradient at T =  K.

Fractional coordinates x ðT Þ = x0 ð1 + c1 T + c2 T 2 + c3 T 3 Þ Simple polynomial form.

Critical behavior  β Site occupancy, magnetic moment

ν ðT Þ = c1 1 − T
Tc
or other quantity related to an
order parameter approaching a
phase transition.


Kinetic parameters fracðt Þ = c1 1 − e − kfrac t + c2 Simple rate expression. k(t − t) or

k(t – t)n to describe more
cellðt Þ = c1 1 − e − kcell t + c2
complex evolution.

Zero point zeroðt Þ = const Zero point correction unchanging

over data surface.

Sample height height ðT Þ = h0 ð1 + c1 T + c2 T 2 Þ Sample height as a smooth

function of temperature.

Offset between furnace set point

Temperature error ΔT = c0 1 + c1 Tset + c2 Tset
2 and sample temperature as a
smooth function.

surface can therefore reveal inadequacies in the model that wouldn’t otherwise be
identified (Magdysyuk et al., 2014).

12.3.1 Parametric quantitative refinement

As an example of parametric refinement we’ll consider analysis of a set of laboratory

powder data recorded on cooling a sample of WO3 from 300 to 90 K. Over this tempera-
ture range WO3 undergoes two transitions from a RT P21/n monoclinic structure to
triclinic P1 at ~230 K then to monoclinic Pc at ~190 K (Figure 12.6). These are displacive
transitions that give rise to relatively minor structural changes and therefore relatively
small changes in the powder diffraction pattern (Figure 12.6). In fact, if we consider a
small portion of the diffraction pattern around 24° 2θ, all three phases of WO3 have peaks
in this 2θ region, with similar intensities (Figure 12.7). This makes quantitative analysis
of a single data set difficult as it’s possible for the “incorrect” phase to distort to fit the
diffraction data. For example, even though the material is predominantly in the mono-
clinic P21/n form at RT, the P1 structural model, which has more degrees of freedom, can
distort to fit the experimental data as well or better than a P21/n model.
 12.3 Parametric Rietveld refinement 283

~230 K ~190 K

1/ 1

Figure 12.6: Illustrations of the crystal structures of the different polymorphs of WO3 at temperatures
below 300 K. Tungsten is located in the center of the blue octahedra.

Figure 12.8(a) and (c) show the phase fractions (as weight%) and cell volumes of
each phase when the data are analyzed sequentially. Fifty-six parameters are used at
each temperature meaning 5600 parameters for all data sets. Each refinement was
started from what we know (from the later parametric analysis) to be an ideal model
and simulated annealing was performed at each temperature to ensure the lowest
Rwp solution was found. As such this is the “best possible” sequential fitting. It is
clear that the results don’t make physical sense: abrupt changes and reversals in the
phase evolution and cell parameters are observed. We note that less careful (“nor-
mal”) sequential fitting models give more marked discrepancies.
Figure 12.8(b) and (d) show equivalent plots from a parametric approach in
which all patterns were evaluated simultaneously using three simple assumptions:
(1) The cell parameters of each individual phase show a smooth variation with
temperature with the functional form given in Tab. 12.1, with coefficients a0, c1 and
θ refined from the diffraction data; for non-90° angles a second-order polynomial was
used. (2) The peak shape description was set up such that each individual phase had
an identical peak shape at all temperatures. (3) Fractional atomic coordinates were
refined from all data sets simultaneously. In this way a single Rietveld refinement
was performed with 1167 parameters being refined simultaneously from all 100 data
sets. This led to the phase fractions and cell volumes shown in Figure 12.8(b) and (d).
The extracted phase factions make much more chemical sense and vary in a smooth
manner with temperature, even though they were not constrained in any way.
Treating quantities of minor interest (here the cell) parametrically has given better
information on the quantities of interest (phase fractions). The parametrically fitted
cell parameters also make physical sense in that they give rise to comparable volume
coefficients of expansion for each phase as expected.
The success of this approach has two basic origins. The most important is the fact
that the three-phase model, with each phase constrained to have cell parameters that
evolve with temperature in a physically sensible way, must simultaneously fit every
experimental data set. This prevents, for example, the high-temperature P21/n phase
distorting to fit minor discrepancies in the low-temperature data as happens in
the sequential analysis. If it did, the distortions would prevent accurate fitting of
 p21n_t0000 61.37 %
3000 pb1_t0000 32.74 %
284

pc_t0000 5.89 %
2500
2000
1500
1000
500

Rel. intensity
0
–500
25 30 35 40 45 50 55 60 65 70 75
2θ/°
12 Multiple data sets

3500

3000

2500 90 K
2000

1500

Rel. intensity
1000

500 300 K
0
22.5 23 23.5 24 24.5 25
2θ/° 2θ/°

1
1/
Figure 12.7: Powder diffraction data of WO3. Top: 300 K Rietveld fit. Bottom left: reflections contributing to peaks at ~24° 2θ. Bottom right: parametric Rietveld
fit on cooling 300–90 K showing phase evolution.
 100 424
1 423
80
422
60 421
420
40

Weight%
419
1/

Cell volume/Å3
418
20
417
0 416
100 150 200 250 300 100 150 200 250 300
(a) Temperature/K (c) Temperature/K

100 424
423
80
422
60 421
420
40

Weight%
419

Cell volume/Å3
418
20
417
0 416
100 150 200 250 300 100 150 200 250 300
(b) Temperature/K (d) Temperature/K

Figure 12.8: Phase fractions (a) and (b) and cell volumes (c) and (d) for sequential (top) and parametric (bottom) refinements of WO3.
12.3 Parametric Rietveld refinement
285
286 12 Multiple data sets

high-temperature data, where more of the P21/n phase is present. Secondly, the
parametric approach to modeling peak shapes means that peak shape parameters
for an individual phase are controlled by the temperature regions of the diffraction
surface where it is actually present. This prevents peak shapes for a phase becoming
excessively broad in regions of the data where the phase is not present, which would
allow it to “mop up” minor errors in fitting the experimental background or peak
tails. In essence the parameters describing an individual phase are determined by
regions of the data where the phase is present rather than regions where it is not.
The INP file needed to perform this refinement makes use of many of the
scripting ideas we’ve met before. In particular, it makes extensive use of #define
and #ifdef . . . #endif statements to control refinement; it uses the local keyword
instead of prm to allow reuse of parameter names in different sections of the
file; it uses “for xdds” and “for strs” sections to feed information to individual
data sets and it uses macros to insert blocks of text into multiple sections of the
file. We can break the file down into the six separate sections shown below,
where each section has been folded in jEdit so that the ~12000 line input file
looks remarkably simple:

'--------------------------------------------------------------
' Parametric fit of WO3 data collected on cooling 300 K to 90 K
' See http://community.dur.ac.uk/john.evans/topas_workshop/tutorial_surface_new.htm
'--------------------------------------------------------------

'{{{ 1. R-factors, refinement choices, iters, etc [16 lines]

'{{{ 2. Select data sets to analyse: define filenames, temperatures, metadata [119 lines]
'{{{ 3. Refinement instructions for each dataset [11701 lines]
'{{{ 4. Overall refined parameters for every data set [20 lines]
'{{{ 5. for xdds section with for strs inside it [75 lines]
'{{{ 6. Some overall simplifying macros [34 lines]

If we expand each of the six sections in turn what we see is as follows:

Section 1 contains overall instructions for the refinement and #define flags
telling TOPAS specific instructions. For example, #param_cell is a flag telling
TOPAS to read the section of the INP file describing a smooth cell parameter
across the data; if it’s not defined, then cell parameters are refined indepen-
dently. #define write_out flags that parameters should be written to an output
file on convergence. To make later plotting of results easier, there is also a
command that deletes any old parameter output file “results.txt”, then writes a
column header line to a new file:

'{{{ 1. R-factors, refinement choices, iters, etc

r_wp 12.7080712 r_exp 8.94672233 r_p 9.82045561 r_wp_dash 18.7386884 r_p_dash 16.1765072
r_exp_dash 13.1923908 weighted_Durbin_Watson 102.986766 gof 1.42041641
chi2_convergence_criteria 0.001
 12.3 Parametric Rietveld refinement 287

continue_after_convergence
conserve_memory
iters 10000
'#define riet_out
#define write_out
#define param_cell
Backup_INP ' Writes a .bck backup of the INP in case of divergence
system_before_save_OUT { del results.txt }
out "results.txt" append
Out_String(" Temp(K) Rwp(%) Perc_P21n Esd Scale_P21n Esd a_P21n Esd b_P21n Esd c_P21n Esd
alpha_P21n ESD beta_P21n Esd gamma_P21n Esd Vol_P21n Esd Perc_Pb1 Esd Scale_Pb1 Esd a_Pb1 Esd
b_Pb1 Esd c_Pb1 Esd alpha_Pb1 ESD beta_Pb1 Esd gamma_Pb1 Esd Vol_Pb1 Esd Perc_Pc Esd Scale_Pc Esd
a_Pc Esd b_Pc Esd c_Pc Esd alpha_Pc ESD beta_Pc Esd gamma_Pc Esd Vol_Pc Esd")
'}}}

Section 2 defines the filename, temperature, range number and any other
metadata for each data set using macros. A flag #define use_t0001 on each
line tells TOPAS to analyze this specific data set, where t0001 is a simple label.
Individual data sets can then be included/excluded rapidly by commenting
these lines in/out of the INP file. The philosophy is similar to the #list
command:

'{{{ 2. Select data sets to analyse: define filenames, temperatures, metadata

'#define use_some 'flag this to just fit 10 ranges
#define use_t0001 macro filename_t0001 {d9_02592} prm !t0001 300 macro r_t0001 { 1 }
#define use_t0002 macro filename_t0002 {d9_02592} prm !t0002 297 macro r_t0002 { 2 }
#define use_t0003 macro filename_t0003 {d9_02592} prm !t0003 295 macro r_t0003 { 3 }
.
.
#define use_t0100 macro filename_t0100 {d9_02592} prm !t0100 90 macro r_t0100 {100}
#endif

'}}}

Section 3 contains the “guts” of the INP file – 11701 lines in this example!4
Within an #ifdef use_t0001 . . . #endif section instructions are given for Rietveld
refinement in the usual TOPAS format. The data set is defined using the
macro“filename_t0001” and temperature is defined as a local parameter; both
use information specified in Section 2. These are followed by local parameters
defining an overall temperature factor and sample height. Finally, there are
lines that give the option of outputting the R-factor to a file and/or writing
ASCII files of observed calculated and difference patterns to separate direc-
tories for later plotting:

4 The whole INP file contains 51731 “words” – almost double Shakespeare’s longest play!
288 12 Multiple data sets

'{{{ 3. Refinement instructions for each dataset

/* {{{ (Foldable) Information for dataset number -> 0001 */
#ifdef use_t0001
RAW(filename_t0001)
local te = t0001;:300.00000 range r_t0001
bkg @ 62.1081942` -58.3869804` 33.8200838` -23.7210244` 14.0400682` -5.0985472`
local individual_b -1.01942` min -4 max 10 'single bvalue for all phases
local height -0.08002` min = -0.5; max = 0.5; del 0.001 th2_offset = -2
57.2957795130823 (height) Cos(Th) / Rs;

'{{{ write out xdd specific results here; phase-specific info done in the for_strs section
#ifdef write_out
out "results.txt" append
Out(te, "\n %11.5f")
Out(Get (r_wp), " %11.5f")
#endif
'write out rietveld fits here
#ifdef riet_out
Rietveld_Plot(riet_plots\rie_t0001.xyd)
xdd_out obs_plots\obs_t0001.xy load out_record out_fmt out_eqn { " %11.5f " = X; " %11.5f
\n" = Yobs; }
xdd_out calc_plots\cal_t0001.xy load out_record out_fmt out_eqn { " %11.5f " = X; " %11.5f
\n" = Ycalc; }
xdd_out diff_plots\dif_t0001.xy load out_record out_fmt out_eqn { " %11.5f " = X; " %11.5f
\n" = Yobs-Ycalc; }
#endif
'}}}

'{{{ P21/n structure information

'{{{ Pb1 structure information
'{{{ Pc structure information

#endif

Section 3 also contains information on each of the phases in a series of str

sections. These are folded away in the script above, but are shown in unfolded
form below. In this example there are two options set up for describing the unit
cell: either with a parametric approach (#ifdef param_cell) using the cell para-
meter expression given in Tab. 12.1 or (#else) free refinement at each tempera-
ture. Fractional atomic coordinates aren’t given within the str section, but are
read in from the later Section 5 using “for strs”:

'{{{ P21/n structure information

str
phase_name p21n
#ifdef param_cell
 12.3 Parametric Rietveld refinement 289

local mlpa = zero_ma + (grad_ma/(Exp(theta_ma/t0001)-1));:7.30115`

local mlpb = zero_mb + (grad_mb/(Exp(theta_mb/t0001)-1));:7.53897`
local mlpc = zero_mc + (grad_mc/(Exp(theta_mc/t0001)-1));:7.69259`
local mlpbeta = beta_m0 * (1+beta_m1*1e-6*t0001+beta_m2*1e-9*t0001^2);:90.84889`
a = mlpa;:7.301151`
b = mlpb;:7.538973`
c = mlpc;:7.692591`
al 90.
be = mlpbeta;:90.84889`
ga 90
#else
a @ 7.302503 mlpaminmax
b @ 7.537414 mlpbminmax
c @ 7.690997 mlpcminmax
al 90.
be @ 90.86670 mbetminmax
ga 90.
#endif
space_group "P121/n1"
scale @ 1.38975664e-05`
weight_percent 82.160`
Phase_Density_g_on_cm3( 7.27438`)
r_bragg 100
output_lines
'}}}}

Section 4 of the file is used to define overall parameters that are simulta-
neously refined from all data sets. In this case, the parameters are coefficients
of the functions used to describe smooth unit cell parameters via the equations
of Tab. 12.1:

'{{{ 4. Overall parameters for every data set

#ifdef param_cell
'overall variables controlling cell dimensions here for P21/n (m), P-1 (t) and Pc (p) phases
'cell edges
prm zero_ma 7.29410` prm grad_ma 0.00447` prm theta_ma 147.30008`
prm zero_mb 7.50500` prm grad_mb 0.02154` prm !theta_mb = theta_ma;:147.30008`
prm zero_mc 7.65447` prm grad_mc 0.02417` prm !theta_mc = theta_ma;:147.30008`
prm zero_ta 7.30852` prm grad_ta 0.00275` prm !theta_ta = theta_ma;:147.30008`
prm zero_tb 7.48849` prm grad_tb 0.02081` prm !theta_tb = theta_ma;:147.30008`
prm zero_tc 7.65227` prm grad_tc 0.02814` prm !theta_tc = theta_ma;:147.30008`
prm zero_pa 5.27198` prm grad_pa 0.00669` prm !theta_pa = theta_ma;:147.30008`
prm zero_pb 5.15781` prm grad_pb 0.00490` prm !theta_pb = theta_ma;:147.30008`
prm zero_pc 7.65858` prm grad_pc 0.01523` prm !theta_pc = theta_ma;:147.30008`
'angles
prm beta_m0 91.08706` prm beta_m1 -4.74099` prm beta_m2 -13.25062`
prm alph_t0 88.92748` prm alph_t1 -15.83299` prm alph_t2 63.55158`
prm beta_t0 89.71326` prm beta_t1 136.08812` prm beta_t2 -365.55447`
290 12 Multiple data sets

prm gamm_t0 89.96058` prm gamm_t1 64.99380` prm gamm_t2 -94.22736`

prm beta_p0 91.78854` prm beta_p1 -19.65621` prm beta_p2 66.08487`
#endif
'}}}

Section 5 contains information that applies to all the data sets within a “for xdds
{. . .}” block, information that applies to all structures “for strs {. . .}” and
information for each individual structure “for strs n to n {. . .}”. Using this format
means that a change on a single line in the INP file (e.g., finish_X 80) will be
applied to all the data sets. Defining atomic sites with parameter names also
means that a single set of coordinates is fitted across the entire data surface.
Alternatively coordinates could be refined using a smooth function with a
language analogous to that for the unit cells in Sections 3 and 4. Or, in the
other extreme, an @ flag could be used so they refine independently for each
data set. The coordinates in this example are contained within a macro that
prevents them moving too far from literature values and allows them to be reset
in a continue_after_convegence process:

'{{{ 5. for xdds section with for strs inside it

for xdds {
x_calculation_step = Yobs_dx_at(Xo); convolution_step 4
start_X 20
finish_X 80.0
LP_Factor(!th2_monochromator, 0)
CuKa2(0.0001)
Zero_Error(!zero, 0.0) 'refining height for each phase so not needed
Simple_Axial_Model(!axial, 5.74444)
scale_pks = (Exp(-2*individual_b/(2 D_spacing)^2)); 'bvalue for all phases in each
dataset

for strs {
TCHZ_Peak_Type(pkuc,-0.78214`,pkvc, 0.53092`,pkwc,-0.07771`,!pkxc,0.0001,pkyc,
0.16135`,!pkzc, 0.0001)
}
macro A1(param,val,val2) {x param val min = val2 - 0.05; max = val2 + 0.05;
val_on_continue = val2;}
macro A2(param,val,val2) {y param val min = val2 - 0.05; max = val2 + 0.05;
val_on_continue = val2;}
macro A3(param,val,val2) {z param val min = val2 - 0.05; max = val2 + 0.05;
val_on_continue = val2;}
for strs 1 to 1 {
'ideal val2 coordinates from Woodward P21/n
site W1 A1( xW1, 0.25258`, 0.25130) A2( yW1, 0.03122`, 0.02770) A3( zW1, 0.27932`,
0.28650) occ W 1.0 beq !bval 0
site W2 A1( xW2, 0.24584`, 0.24810) A2( yW2, 0.03034`, 0.03420) A3( zW2, 0.78571`,
0.78150) occ W 1.0 beq !bval 0
site O3 A1(!xO3, 0.00143, 0.00143) A2(!yO3, 0.03550, 0.03550) A3(!zO3, 0.21526,
 12.3 Parametric Rietveld refinement 291

0.21526) occ O 1.0 beq !bval 0

site O4 A1(!xO4, 0.99858, 0.99858) A2(!yO4, 0.46301, 0.46301) A3(!zO4, 0.21782,
0.21782) occ O 1.0 beq !bval 0
site O5 A1(!xO5, 0.28165, 0.28165) A2(!yO5, 0.26251, 0.26251) A3(!zO5, 0.28384,
0.28384) occ O 1.0 beq !bval 0
site O6 A1(!xO6, 0.20905, 0.20905) A2(!yO6, 0.25985, 0.25985) A3(!zO6, 0.73375,
0.73375) occ O 1.0 beq !bval 0
site O7 A1(!xO7, 0.27950, 0.27950) A2(!yO7, 0.04128, 0.04128) A3(!zO7, 0.00461,
0.00461) occ O 1.0 beq !bval 0
site O8 A1(!xO8, 0.28259, 0.28259) A2(!yO8, 0.48854, 0.48854) A3(!zO8, 0.99449,
0.99449) occ O 1.0 beq !bval 0
Strain_G(strainp21n, 0.12499`)
}
for strs 2 to 2 {
.
.
}
for strs 3 to 3 {
.
.
} 'end of for xdds
'}}}

Section 6 contains macros that simplify the input file. For example, the macro
output_line is included in each of the strs of Section 4. By changing the macro, one
can instantly change the information written to results.txt at the end of each refine-
ment. There are also macros to automatically control non-parameterized cell para-
meters in the case of divergence:

'{{{ 6. Some overall simplifying macros

' These set min and max values for cells if allowed to refine individually
' Useful for resetting things to "sensible" values if refinement diverges
' val_on_continue values set at typical values for each phase from parametric fit
prm !tol 1.1
macro mlpaminmax { min =7.301/tol; max = 7.301*tol; val_on_continue = 7.301;}
macro mlpbminmax { min =7.539/tol; max = 7.539*tol; val_on_continue = 7.539;}
macro mlpcminmax { min =7.692/tol; max = 7.692*tol; val_on_continue = 7.692;}
macro tlpaminmax { min =7.310/tol; max = 7.310*tol; val_on_continue = 7.310;}
macro tlpbminmax { min =7.508/tol; max = 7.508*tol; val_on_continue = 7.508;}
macro tlpcminmax { min =7.678/tol; max = 7.678*tol; val_on_continue = 7.678;}
macro plpaminmax { min =5.274/tol; max = 5.274*tol; val_on_continue = 5.274;}
macro plpbminmax { min =5.159/tol; max = 5.159*tol; val_on_continue = 5.159;}
macro plpcminmax { min =7.662/tol; max = 7.662*tol; val_on_continue = 7.662;}
macro pbetminmax { min 91.6 max 91.8 val_on_continue = 91.68;}
macro mbetminmax { min 90.0 max 92.0 val_on_continue = 90.86;}
macro tlpalminmax { min 87.0 max 90.0 val_on_continue = 88.87;}
macro tlpbeminmax { min 90.0 max 92.0 val_on_continue = 90.85;}
macro tlpgaminmax { min 90.0 max 92.0 val_on_continue = 90.79;}
292 12 Multiple data sets

' Output key information on each phase to a file through this macro
macro output_lines {
#ifdef write_out
Out(Get(weight_percent), " %11.6f", " %11.6f")
Out(Get(scale), " %11.9f", " %11.9f")
Out(Get(a), " %11.6f", " %11.6f")
Out(Get(b), " %11.6f", " %11.6f")
Out(Get(c), " %11.6f", " %11.6f")
Out(Get(al), " %11.6f", " %11.6f")
Out(Get(be), " %11.6f", " %11.6f")
Out(Get(ga), " %11.6f", " %11.6f")
Out(Get(cell_volume), " %11.6f", " %11.6f")
#endif
}
'}}}

One of the challenges of producing a file like this is, of course, writing the
11701 lines needed in Section 4 for the 100 data sets. However, this is much
less daunting than it might seem! In essence, the information within each of
the #ifdef use_t0001 . . . #endif folds is identical to that needed for any single
Rietveld refinement. The only difference is that it contains various local para-
meters and equations containing the temperature (or any other external vari-
able) as t0001. Once a single Rietveld has been set up one just needs to
reproduce this section of the file once for each data set with the t0001 string
swapped to t0002, t0003, . . . , t0100. This can be easily done with a simple
computer program or PYTHON script. It can also be done in seconds on a linux
system (e.g., the ubuntu shell that is included in Windows 10) using the built
in function sed, which performs simple text editing functions from a shell
prompt.
In this case, all that’s needed is a text file (single.riet) that contains the Section 3
information for a single refinement. Then create a file (e.g., expand.sh) containing
the following lines:

sed '1,$s/t0000/t0001/g' single.riet

sed '1,$s/t0000/t0002/g' single.riet
.
.
sed '1,$s/t0000/t0100/g' single.riet

These commands can be executed with “sh expand.sh >all_riet.txt.” This will create
a single text file all_riet.txt containing the 11701 lines of TOPAS script needed.
expand.sh and Section 2 information on filenames can be generated in seconds
using column editing in jEdit from any list of experimental details (e.g., dir *.xye >
files.list).
 12.3 Parametric Rietveld refinement 293

12.3.2 Non-crystallographic parameters

The parametric approach also allows the refinement of noncrystallographic para-

meters. An example is the extraction of smooth temperature calibration curves for
nonambient experiments as described by Stinton et al. (2007). In this approach internal
standards such as Si or Al2O3, which have known thermal expansion, are mixed with
the sample of interest. During parametric analysis of variable temperature data the cell
parameters of the standards aren’t refined freely, but are instead calculated using
expressions of the type given in Tab. 12.1 with coefficients known from the literature. If
there is any discrepancy between the furnace set temperature and the actual set
temperature (Tsample = Tset + ΔT) then calculated and observed peak positions of the
standards will not match. This can be corrected by refining ΔT either individually from
each data set or (better) using a function that varies smoothly with temperature. In a
published example of analyzing 871 data sets on ZrP2O7 to understand its unusual
thermal expansion, using this latter approach both the true phase transition tempera-
ture and cell parameters could be extracted more reliably with the parametric approach
than sequential refinements. Figure 12.9 (left) shows cell parameters extracted sequen-
tially as scattered open points and those from a parametric refinement as closed points.

8.30 0.90
Tc set Tc refined
8.30
0.85
8.29
Fractional occupancy

8.29 0.80
Unit cell/Å

8.28

8.28 0.75
frac (t) = c1 [1 – exp (–kfrac t)] + c2
8.27
0.70
8.27

8.26 0.65
500 550 600 650 0 20000 40000 60000 80000 100000
Refined temperature/K Time/s

Figure 12.9: Left: unit cell parameters of ZrP2O7 close to a displacive phase transition derived either
sequentially (open points) or parametrically (closed points) on warming (red) and cooling (blue).
A parametric temperature calibration gives significantly lower uncertainty in cell parameters. The
uncorrected phase transition temperature, Tc, was ~520 K, whereas the refined Tc of 567–571 K
agreed perfectly with DSC data. Right: a comparison between parametrically (red solid line) and
sequentially (blue points) refined site occupancies for an order–disorder phase transition in
ZrWMoO8. The parametric approach allows direct Rietveld refinement of the rate constant, kfrac.

A second published example followed the kinetics of an oxygen migration process in

ZrWMoO8 using time-resolved X-ray data. Here either site occupancies or accompanying
294 12 Multiple data sets

cell parameter changes could be expressed using simple rate laws, allowing rate con-
stants to be refined directly from a surface of data. In fact, in unpublished work, it was
possible to parametrically analyze >1000 data sets collected as a function of time and
temperature (4D data) using a simple Arrhenius expression to describe the temperature
dependence of rate constants. This allowed the activation energy of the process to be
determined directly by Rietveld refinement.

12.3.3 Plotting tricks

Thanks to the speed of TOPAS, one of the slowest parts of either sequential or
parametric work is often visualizing the refined parameters to see if they make
physical sense. This is particularly time consuming if you need to compare
multiple refinement models. The text files of refined parameters produced by
the methods in this chapter can, of course, be plotted in any standard program
(e.g., Origin, Excel, etc.). For speed we recommend the plotting program gnuplot
(http://www.gnuplot.info/), which is freely available for most operating systems.
One good feature is that it can be driven by a simple scripting language just like
TOPAS, and can produce multiple plots very quickly. For example, the simple
script below will automatically produce plots both on the screen and saved as
individual .gif files for the 28 columns of the file “results.txt” in the WO3
example. It can be adapted to autoplot any results file by editing just five
lines. The file can be run either by double clicking in windows or using “load
‘filename.gnu’” at the gnuplot command:

# Script to plot results file from multiple refinements

# Plot columns with no error first then columns with errors
# File has a dummy header line labelling the columns
# Temperature in col 1 Then data are in col 2 to 14; col 2 has no esd
# Change the column numbers to your example

reset

file = 'results.txt'
xcol = 1 # x in column 1 is temperature
startcol = 2 # first column to plot is column 2
numnoesd = 1 # 1 column without an esd (Rwp in column 2)
numesd = 27 # 27 columns with an esd

set style line 1 lt rgb "red" linewidth 2 pt 7

set key autotitle columnheader font "Arial,12" Left noenhanced

set term win font "Arial,24"
set ytics mirror
set xlabel 'Temperature/K'
 References 295

#plot each column to the screen then to a gif file for columns with no esd
do for [col=startcol:startcol+(numnoesd-1):1] {
plot file using xcol:col w linesp ls 1
set term gif size 640,480 font "arial,12"; set output 'pic_col_'.col.'.gif'; replot; set
term win
pause -1 "<cr> to see the next plot"
}

#plot each column to screen then to a gif file for columns with associated esd
do for [col=startcol+numnoesd:startcol+numnoesd+2*(numesd-1):2] {
plot file using xcol:col w linesp ls 1
replot file using xcol:col:col+1 w errorbars ls 1 notitle
set term gif ; set output 'pic_col_'.col.'.gif'; replot; set term win
pause -1 "<cr> to see the next plot"
}

References
König, R., Müller, S., Dinnebier, R.E., Hinrichsen, B., Müller, P., Ribbens, A., Hwang, J., Liebscher,
R., Etter, M., Pistidda, C. (2017): The crystal structures of carbonyl iron powder – revised using in
situ synchrotron XRPD. Z. Kristallogr. Cryst. Mater. 232, 835–842.
Magdysyuk,O.V., Müller, M., Dinnebier, R.E., Lipp, C., Schleid, T. (2014): Parameterization of the
coupling between strain and order parameter for LuF[SeO3]. J. Appl. Cryst. 47, 701–711.
Stinton, G.W., Evans, J.S.O. (2007): Parametric Rietveld refinement. J. Appl. Cryst. 40, 87–95.
13 Appendix: Mathematical basics

The aim of this chapter is to remind the reader of some basic mathematical tools
which are commonly used in crystallography and, in particular, in Rietveld analysis.
This not intended as a substitute for a math (e.g., Papula, 2014) or general crystal-
lography textbook (e.g., Giacovazzo, 2011; Müller, 2013). For brevity we give a
collection of formulas and mathematical concepts rather than provide full definitions
or proofs. We assume the reader is familiar with basic vector algebra, analysis and
analytical geometry.
Conventions: Tensors of rank one or greater are always in bold. Transposed
matrices are denoted by an additional bar. Lowercase letters are used for one-
dimensional tensors (vectors) and uppercase letters for multidimensional tensors
(matrices). Positive rotations are always counterclockwise when looking toward the
origin of the rotation axis/vector.

13.1 Complex numbers

A complex number z codes a length and an angle, like a vector in a two-dimensional

Cartesian coordinate system spanned by the two axes of the real and imaginary
numbers (Figure 13.1).

z = x + iy = jzjðcos ’ + i sin ’Þ (13:1)

pffiffiffiffiffiffiffi
with the complex number i = − 1. The complex axis has a phase shift of π=2

(or 90°) with respect to the real axis. The conjugate complex number z is
defined as:

z* = x − iy = j zjðcos ’ − i sin ’Þ. (13:2)

The length of the complex vector can be calculated as:

pffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffi
j zj = zz* = x2 + y2 . (13:3)

The angle is determined by:

y 
’ = arctan if x > 0,
x
y
’ = arctan + π if x < 0 and y ≥ 0,
x
y
’ = arctan − π if x < 0 and y < 0
x

https://doi.org/10.1515/9783110461381-013
 13.2 Power series 297

Imaginary axis
z
y

|z| i sin(φ)
|z|

φ
x Real axis
0 –φ
|z| i sin(–φ)

|z|
Figure 13.1: The vector (pointer) z and its conjugate complex z*
–y z* in the complex plane. Dashed lines represent the Cartesian
|z|cos(φ) coordinates.

π
’= if x = 0 and y > 0,
2
π
ϕ= − if x = 0 and y < 0 (13:4)
2

For x = y = 0 , ’ is undetermined.

13.2 Power series

Derivatives of nonlinear and transcendental functions are important in the Rietveld

method. Such functions can be approximated by a special power series (Taylor series)
around the current values of their parameters. Such a series is usually terminated
after the first or second term. When used in least squares this truncation means that
shifts applied to a model won’t lead to a fully minimized solution, but to a hopefully
better approximation.
A general representation of a power series with the center of the series x0 is
given by:

X
∞
P ð xÞ = an ðx − x0 Þn (13:5)
n=0

with real numbers a0 , a1 , . . . as coefficients. Under several boundary conditions, it is

in principle possible to approximate a function f ðxÞ by a power series. The most
common power series of this kind is the Taylor series:

X
∞ ðnÞ
f ðx0 Þ
f ð xÞ = ðx − x0 Þn , (13:6)
n=0
n!
298 13 Appendix: Mathematical basics

where the coefficient an is determined by the n-th derivative f ðnÞ ðx0 Þ of the approxi-
mated function around the point x0 with f ð0Þ ðx0 Þ = f ðx0 Þ (Figure 13.2). If the deriva-
tives of the original function cannot be determined analytically, numerical
approximations can be used. If the development point x0 is at the origin of the
function, the so-called MacLaurin series is obtained:

X
∞ ðnÞ
f ð0Þ
f ð xÞ = xn . (13:7)
n=0
n!

Figure 13.2: Taylor expansion of a sine-function around the origin as development point x0 using
derivatives of second, fourth, sixth and eighth order.

MacLaurin series for the exponential, sine and cosine functions are given by:

P
∞
xn
 n
ex = n! , e = lim 1 + n1 = 2.718281 . . .
n=0 n!∞
P∞ 2n + 1
sin x = ð − 1Þn ðx2n + 1Þ! ,
n=0
P
∞ 2n
cos x = ð − 1Þn ðx2nÞ! . (13:8)
n=0

A common estimate for the error between the original and approximation function
using a Taylor expansion up to the nth term is given by the remainder term after
Lagrange:

f ðn + 1Þ ðδxÞ n + 1
Rn ð xÞ = x with ð0 < δ < 1Þ. (13:9)
ðn + 1Þ!
 13.4 Summing pointers 299

13.3 The Euler formula

Using a MacLaurin series, the famous Euler formula (Figure 13.3) can be calculated
by:

X
∞
ði’Þn ði’Þ1 ði’Þ2 ði’Þ3 ði’Þ4 ði’Þ5 ði’Þ6 ði’Þ7
ei’ = =1+ + + + + + + +
n=0
n! 1! 2! 3! 4! 5! 6! 7!
’2 ’3 ’4 ’5 ’6 ’7
1 + i’ − −i + +i − −i + ...
2! 3! 4! 5! 6! 7!
   
’2 ’4 ’6 ’3 ’5 ’7
= 1− + − + ... +i ’− + − + ... ,
2! 4! 6! 3! 5! 7!
(13:10)

Imaginary axis
i
i sin(φ)

–1 φ 1 Real axis
0 cos(φ)

Figure 13.3: The unity vector (pointer) in the

–i complex number plane.

which is the same as:

ei’ = cos ’ + i sin ’ (13:11)

from which it follows:

1  i’  1  i’ 
cos ’ = e + e − i’ and i sin ’ = e − e − i’ . (13:12)
2 2

13.4 Summing pointers

A pointer (e.g., amplitude) consists of a magnitude (length) and a phase. Summing

pointers can be done by vector-like addition in two-dimensional space using the
complex number plane (Figure 13.4).
300 13 Appendix: Mathematical basics

Imaginary axis
φ3
f3

f2
F
φ2
f1
φ
φ1
Real axis
0 Figure 13.4: Summation of complex pointers f1,
f2 and f3 to yield complex F.

The resulting complex amplitude F:

F = jF j ei’ = jF jðcos ’ + i sin ’Þ (13:13)

can be calculated by complex summation over all complex pointers fj :

X
n X
n  
F= fj ei’j = fj cos ’j + i sin ’j . (13:14)
j=1 j=1

The sum above basically represents a Fourier series.

13.5 Fourier transformation

The Fourier integral transforms a function f ðxÞ of the variable x into an integral over
cosine und sine functions F ðsÞ of the variable s:

ð
∞ ð
∞

F ð sÞ = f ðxÞ e2πisx dx = f ðxÞ ðcosð2πsxÞ + i sinð2πsxÞÞdx (13:15)

−∞ −∞

while the back transformation is given by:

Ð
∞
f ð xÞ = FðsÞ e − 2πisx dx
−∞
Ð
∞
= F ðsÞðcosð2πsxÞ − i sinð2πsxÞÞds. (13:16)
−∞

Since the exponent of the exponential function is dimensionless, s is reciprocal to x.

If x is in position space, s is in “Fourier space” or “reciprocal space”, if x is time, s is a
frequency.
 13.6 Convolution 301

The function f ðxÞ must satisfy the conditions of Dirichlet:

– The interval of definition can be divided into a finite number of intervals in
which f ðxÞ is continuous and monotonous.
– The limits of any point of discontinuity are defined on both sides
lim f ðxν − εÞ = f ðxν − 0Þ and lim f ðxν + εÞ = f ðxν + 0Þ:
ε!∞ ε!∞
+Ð∞
– The integral jf ðxv Þjdx must converge.
−∞

The Fourier transformation (denoted by $Þ is a linear operation:

αf ðxÞ + βgðxÞ $ αF ðsÞ + βGðsÞ, (13:17)

with the complex numbers α, β and f ðxÞ $ FðsÞ and gðxÞ $ GðsÞ:
Shifting of a function f ðxÞ on the x-axis by a constant u to the right simply leads to
an additional phase factor before the Fourier transform. Using the substitution
w = x − u with the same derivative dx = dw leads to:

ð
∞ ð
∞

f ðx − uÞe2πisx dx = f ðwÞe2πisðw + uÞ dw = e2πisu F ðsÞ. (13:18)

−∞ −∞

The Fourier transform of a differentiated function leads to an additional pre-factor:

∂n
f ðxÞ $ ð − 2 π isÞn F ðsÞ. (13:19)
∂xn

13.6 Convolution
Convolution or folding is a basic concept in crystallography and of particular impor-
tant in powder diffraction analysis. The process of convolution is one in which the
product of two functions f ðxÞ and gðxÞ is integrated over all space:

ð
∞

hðx′Þ = f ðxÞ gðx′ − xÞdx = f  g, (13:20)

−∞

where hðx′Þ is the convolution product, x′ is the variable of integration in the same
domain as x and  denotes the convolution process. Convolution can be understood
as “blending” one function with another, producing a kind of very general “moving
average.” One can illustrate the formation of the convoluted function by setting down
the origin of the first function in every possible position of the second, multiplying
the values of both functions in each position and taking the sum of all these opera-
tions. The convolution operation is commutative and associative.
302 13 Appendix: Mathematical basics

An illustrative example for a typical convolution function is given in Figure 13.5, where
a triangular and a rectangular intensity function are convoluted leading to a function
approximating a Gaussian (see eq. (13.83)). Most functions cannot be convoluted
analytically and the convolution integral needs to be calculated numerically.

2
1

Figure 13.5: Convolution of a triangular (1) with

a rectangular (2) function leading to a function
–0.2 –0.1 0 0.1 0.2 approximating a Gaussian (3).

An alternative way of calculation follows from the Fourier transform of the

convolution function. The Fourier transform of the convolution function can be
calculated by:

Ð∞ Ð
∞ Ð∞
hðx′Þe2πisx′ dx′ = f ðxÞ gðx′ − xÞdx e2πisx′ dx′
−∞ −∞ −∞
Ð∞ ∞
Ð
= f ðxÞ gðx′ − xÞdx e2πisðx′ − x + xÞ dx′. (13:21)
−∞ −∞

This can be rewritten by using the substitution u = x′ − x and therefore du = dx′ as:

Ð∞ Ð∞
= f ðxÞ gðuÞe2πisx dx e2πisu du
−∞ −∞
Ð Ð∞
∞
= f ðxÞe2πisx dx gðuÞe2πisu du = F ðsÞ GðsÞ. (13:22)
−∞ −∞

From this it follows directly that the Fourier transform of the convolution integral is
the product of the Fourier transforms of all functions participating in the convolution:

f  g $ F ðsÞ GðsÞ, (13:23)

 13.8 The delta (δ) function 303

while the back-transformation of a convolution is the product of the back-trans-

formed functions that participate in the convolution:

f ð xÞ g ð xÞ $ F  G (13:24)

As a practical example, a crystal structure can be regarded as a convolution of the

structural motif (atom, group of atoms or molecule) and a three-dimensional lattice.
The scattered amplitude of a crystal is the product of the Fourier transforms of the motif
(density distribution of the unit cell) and the lattice function (three-dimensional sum of
equally spaced δ-functions).
In TOPAS, the operation of convolution can be performed by means of direct
convolution or by Fast Fourier Transform (FFT) convolution.

13.7 Correlation function

If f ðxÞ and gðxÞ are functions with F ðsÞ and GðsÞ being their Fourier transforms, then
the correlation function is defined as:

R∞
hC ðx′Þ = f ðxÞ gðx′ + xÞdx ≡ f −  g
−∞
f −  g $ Fð − sÞ GðsÞ (13:25)

For real functions, F ð − sÞ is the same as the conjugate complex function F * ðsÞ.
A special form of the correlation function is the autocorrelation (Patterson) function,
where a function is correlated with itself:

  R∞
hK x ′ = f ðxÞ f ðx′ + xÞdx = f −  f
−∞
f −  f $ F ð − sÞ F ðsÞ (13:26)

A typical example is the diffracted intensity, which can be considered as the auto-
correlation of the structure factor amplitude.

13.8 The delta (δ) function

The δ-function is defined by its integration properties:

RB
F ðsÞ = δðx − x0 Þdx is 1 if A < x0 < B otherwise 0
A
RB
F ðsÞ = f ðxÞ δðx − x0 Þdx is f ðx0 Þ if A < x0 < B otherwise 0. (13:27)
A
304 13 Appendix: Mathematical basics

In practice, a δ-function can be approximated by a normalized Gaussian function

with a width approaching 0: limðσ ! 0Þ where σ is related to the width via eq. (13.85).
The Fourier transform of a δ-function can be easily calculated:

ð
∞

F ðsÞ = δðx − x0 Þ e2πisx dx = e2πisx0 , (13:28)

−∞

which in case of x0 = 0 equals to 1. The back transformation then is:

ð
∞ ð
∞

δðx − x0 Þ = e 2πisx0
e − 2πisx
ds = e − 2πisðx − x0 Þ ds. (13:29)
−∞ −∞

A series of N equally spaced δ-functions separated by distance a represents a one-

dimensional lattice:

Figure 13.6: Amplitude (top) and squared amplitude (bottom) of the Laue function with N = 3 and a= 2.
 13.9 Matrix algebra 305

X
N
fν ðxÞ = δðx − aνÞ. (13:30)
ν=0

The Fourier transform of this one-dimensional lattice can be directly calculated with
help of eq. (13.12)1 as (Figure 13.6):

ðX
N X
N
SðsÞ = δðx − aνÞe2πisx dx = e2πiaνs
ν=0 ν=0
2πiðN + 1Þas πiðN + 1Þas
e −1 e eπiðN + 1Þas − e − πiðN + 1Þas
= =
−1
e2πias eπias eπias − e − πias
sinðπðN + 1ÞasÞ
= eπiNas , (13:31)
sinðπasÞ

which can easily be expanded to higher-dimensional lattices.

13.9 Matrix algebra

A rectangular array of real numbers in m rows and n columns is called a real (m × n)
matrix A:
0 1
A11 A12  A1n
B A21 A22  A2n C
B C
A = B .. .. .. .. C = ðAik Þ (13:32)
@ . . . . A
Am1 Am2  Amn

The left index i, running from 1 to m, is called the row index, the right index k,
running from 1 to n, is the column index of the matrix. An n × n matrix is called a
square matrix, an (m × 1) matrix a column matrix (or column vector) and a (1 × n)
matrix a row matrix (or row vector). A transposed matrix is obtained by exchanging
rows and columns:

A = ðAkiÞ. (13:33)

A symmetric matrix is called a diagonal matrix if Aik = 0 for i ≠ k . A diagonal matrix

with all elements Aik = 1 is called the unit matrix I. A matrix consisting of zeroes only,

NP−1
1 PN = ex ν = 1 + ex + e2x + ... + eðN − 1Þx PN ex PN = ex + e2x + ... + eNx PN − ex PN = 1 − eNx →PN ð1 − ex Þ =
ν=0
  Nx Nx
1 − eNx → PN = 11−−eex or PN = eex −−11
306 13 Appendix: Mathematical basics

that is, Aik = 0 is called the 0 matrix. The sum of the diagonal terms of a square matrix
is called the trace of the matrix:
X
n
trðAÞ = Ajj . (13:34)
j=1

An (m × n) matrix A is multiplied by a (real or complex) number λ by multiplying each

element with λ:

λA = ðλAik Þ. (13:35)

Two (m × n) matrices A and B can be summed (subtracted) by summing (subtracting)

each pair:

C = ðAik ± Bik Þ. (13:36)

Two matrices A and B can be multiplied if the matrix A has the same number r of
columns as B has rows:

X
r
C = AB with Cik = Aij Bjk , i = 1, ..., m; k = 1, ..., n. (13:37)
j=1

The determinant is a number, which can be calculated for any square matrix. The
formula for a (2 × 2) matrix is:

A A12
detðAÞ = 11 = A11 A22 − A12 A21 (13:38)
A21 A22

and for a (3 × 3) matrix:

A11 A12 A13



detðAÞ = A21 A22 A23


A A32 A (13:39)
31 32

= A11 A22 A33 + A12 A23 A31 + A13 A21 A32

− A11 A23 A32 − A12 A21 A33 − A12 A22 A31 .

If detðAÞ ≠ 0 the matrix A is called regular, otherwise singular. The inverse of a matrix
is defined by:

A − 1A = I (13:40)
 13.10 Crystallographic computing 307

with:
 
A−1 ik
= ðdetðAÞÞ − 1 ð − 1Þi + k Bki , (13:41)

where Bki is the determinant obtained from det(A) by cancelling the kth row and ith
column.
If A = A (Aik = Aki), the matrix is called symmetric. If A − 1 = A, the matrix is called
orthogonal.

13.10 Crystallographic computing

A crystallographic coordinate system D consists of a set of three nonlinear base
vectors spanning in a right handed manner D = fa, b, cg (Figure 13.7).

α c
b
γ b
a a Figure 13.7: Right-handed base D = fa, b, cg in three-
dimensional space.

An arbitrary vector within such a coordinate system can be written as:

0 1
x

r = xa + yb + zc = ðabcÞ@ y A = Ax. (13:42)
z

The scalar product of two vectors r1 and r2 enclosing the angle ϕ is defined as:

r1 r2 = r1 r2 cos ϕ = ðx1 a + y1 b + z1 cÞ  ðx2 a + y2 b + z2 cÞ

0 10 1
aa ab ac x2
B CB C (13:43)
B CB C
= ðx1 y1 z1 ÞB
Bb a b b b  cC B C
CB y2 C = x1 Gx2
@ A@ A
ca cb cc z2

with the metric tensor:

0 1 0 1
aa ab ac a2 a b cos γ a c cos β
G=@b  a b  b b  c A = @ a b cos γ b2 b c cos α A. (13:44)
ca cb cc a c cos β b c cos α c2
308 13 Appendix: Mathematical basics

In the case of a Cartesian coordinate system, the metric tensor simplifies to G = I.

In general, the lengths of a vector r (e.g., the distance between two points) can be
calculated as:
pffiffiffiffiffiffiffiffiffi
r = xGx (13:45)

and, therefore, the angle ϕ between two vectors r1 and r2 is:

x1 Gx2
cos ϕ = pffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffi . (13:46)
x1 Gx1 x2 Gx2

A unit vector br is defined by dividing the vector by its length:

r Ax
br = = pffiffiffiffiffiffiffiffiffi . (13:47)
r xGx

The cross product of two vectors r1 and r2 leads to a vector, which is perpendicular to
r1 and r2, and which completes a right-handed system:

r1 × r2 = ðx1 a + y1 b + z1 cÞ × ðx2 a + y2 b + z2 cÞ =
ðy1 z2 − y2 z1 Þb × c + ðz1 x2 − z2 x1 Þc × a + ðx1 y2 − x2 y1 Þa × b, (13:48)

which in terms of reciprocal lattice parameters (see below) can be written as:
0 1
  y1 z2 − y2 z1
r1 × r2 = a b c @ z1 x2 − z2 x1 A,
* * *
(13:49)
x1 y2 − x2 y1

and, if a, b and c are base vectors of a Cartesian coordinate system, a × b = c , b × c = a

and c × a = b simplify to:
0 1
y1 z2 − y2 z1
r1 × r2 = ðabcÞ@ z1 x2 − z2 x1 A. (13:50)
x1 y2 − x2 y1

The length of the cross product is:

pffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffi
jr1 × r2 j = r1 r2 sin ϕ = x1 Gx1 x2 Gx2 sin ϕ. (13:51)

The scalar triple product of three vectors r1, r2 and r3 is defined as:

0 1
x1 y1 z1
r1  r2 × r3 = Vdet@ x2 y2 z2 A (13:52)
x3 y3 z3
 13.10 Crystallographic computing 309

where V is the volume of the unit cell and can be calculated according to:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
V= detðGÞ. (13:53)

In addition to the “normal” crystal lattice with the lattice parameters a, b, c, α, β, γ,

and volume V of the unit cell, a second lattice with lattice parameters of a*, b*, c*, α*,
β*, γ *, and the volume V* with the same origin can be defined such that:

a  b* = a  c* = b  c* = a*  b = a*  c = b*  c = 0

a  a* = b  b* = c  c* = 1. (13:54)

This is known as the reciprocal lattice,2 which exists in so-called reciprocal space.
The length of the reciprocal base vectors can be defined according to:

a* = xðb × cÞ, (13:55)

where the scale factor x can easily be deduced, using the equations above as:

1
a*  a = xðb × c  aÞ = xV ) x = (13:56)
V

leading to:

1 1 1
a* = ðb × cÞ, b* = ðc × aÞ, c* = ða × bÞ (13:57)
V V V

and vice versa:

1  * * 1   1  
a= *
b × c , b = * c* × a* , c = * a* × b* . (13:58)
V V V

The relationship between the reciprocal and the real lattice parameters expressed
geometrically rather than in the vector formalism above is:

bc sin α
a* = ,
V

ac sin β
b* = ,
V
ab sinγ
c* = ,
V

2 The reciprocal lattice is a commonly used construct in solid state physics, but with a different
normalization: a  a* = 2π.
310 13 Appendix: Mathematical basics

cos β cos γ − cos α

cos α* = ,
sin β sin γ
cos γ cos α − cos β
cos β* = ,
sin γ sin α (13:59)
cos α cos β − cos γ
cos γ* = ,
sin α sin β
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
V = abc 1 + 2 cos α cos β cos γ − cos2 α − cos2 β − cos2 γ,

which is the most general expression for non-orthogonal lattices. The expressions
simplify considerably for higher-symmetry crystal systems.
An arbitrary vector within such a coordinate system can be written as:
0 1
  x
r* = xa* + yb* + zc* = a* b* c* @ y A = A x.
*
(13:60)
z

Accordingly, the reciprocal metric tensor is:

0 1
a*  a* a*  b* a*  c*
* @ *
b*  c* A = G
−1
G = b  a* b*  b* (13:61)
c*  a* c*  b* c c
* *

A vector and its reciprocal counterpart are connected via the metric tensor:

r* = Gr. (13:62)

13.11 Basis transformations

If a matrix T transforms the base vector aD1 in base D1 into the base vector aD2 in base
D2 and the transformation leaves the origin invariant then:

aD2 = TaD1 and aD1 = T − 1 aD2 . (13:63)

Any vector can then be transformed from one base to the other and vice versa by

−1
xD2 = T xD and xD1 = TxD . (13:64)

The metric tensors can be calculated as

−1
GD1 = TGD2 T and GD2 = T GD1 T − 1 . (13:65)
 13.12 Rotations in the Cartesian system 311

13.12 Rotations in the Cartesian system

A counterclockwise active rotation of an angle α around an axis of a Cartesian

coordinate system, which transforms a point P to a point P′ can be written as a (3 × 3)
rotation matrix Raxis ðαÞ (Figure 13.8).

y
P‘ = (x‘,y‘)

r
P = (x,y)

α
φ
x
0 r cos(φ + α) r cos(φ) Figure 13.8: Counterclockwise rotation of a point P
around z-axis in a Cartesian coordinate system.

A rotation α around the z-axis (axis perpendicular to the xy-plane) follows from
Figure 13.8:
   
x r cos ’
= (13:66)
y r sin ’

and:
       
x′ rcosð’ + αÞ r cos ’ cos α − r sin ’ sin α x cos α − y sin α
= = = ,
y′ rsinð’ + αÞ r sin ’ cos α + r cos ’ sin α y cos α + x sin α
(13:67)

giving in three dimensions:

0 1
cos α − sin α 0
Rz ðαÞ = @ sin α cos α 0 A (13:68)
0 0 1

Similarly:
0 1 0 1
1 0 0 1 0 0
Rx ðαÞ = @ 0 cos α − sin α A and Ry ðαÞ = @ 0 cos α − sin α A: (13:69)
0 sin α cos α 0 sin α cos α

For an active rotation, the coordinate system is not rotated but fixed.
312 13 Appendix: Mathematical basics

13.13 Eulerian angles

If the coordinate system is not fixed but rotated, any rotation in three-dimensional
space can be described by a rotation ϕ around z-axis, followed by a rotation around
the new x′-axis θ, followed by a third rotation around the new z′′-axis ψ (Figure 13.9).
The three angles are called Eulerian angles.

z,z‘

y‘‘

z‘‘

y
x‘‘
ψ
φ

x‘

Figure 13.9: Proper Eulerian angles representing rotations about z, x′ and z′′ axes. The xyz (original)
system is shown in blue, the rotated systems are shown in green and red.

The convention used here is only one of several possibilities:

REuler = Rz ðψÞRx ðθÞRz ð’Þ

0 1
cos ψ cos ’ − sin ψ cos θ sin ’ sin ψ cos ’ + cos ψ cos θ sin ’ sin θ sin ’
B C
=B
@ − sin ψ cos θ cos ’ − cos ψ sin ’ − sin ψ sin ’ + cos ψ cos θ cos ’ sin θ cos ’ C
A
sin ψ sin θ − cos ψ sin θ cos θ
(13:70)

13.14 Spherical coordinates

In many cases, it is more convenient to describe a point in three-dimensional space in
spherical instead of Cartesian coordinates. Spherical coordinates consist of the radial
distance r of that point from a fixed origin, its polar angle ϕ measured from a fixed
zenith direction, and the azimuth angle θ of its orthogonal projection onto a refer-
ence plane that passes through the origin and is orthogonal to the zenith, measured
from a fixed reference direction on that plane (Figure 13.10):
 13.15 Triaxial ellipsoid 313

r
Θ
0 y
φ

Figure 13.10: Relation between spherical and Cartesian coordi-

x nates at the point P.

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r= x2 + y2 + z 2

θ = arccos zr (13:71)

’ = arctan xy

The inverse tangent denoted in φ = arctan(y/x) must be suitably defined, taking into
account the correct quadrant of (x, y). The ranges of the spherical coordinates are
usually restricted to, for example, r ∈ [0, ∞], θ ∈ [0, π], φ ∈ [0, 2π]. Then the
Cartesian coordinates may be retrieved from the spherical coordinates by:

x = r sin θ cos ’
y = r sin θ sin ’ (13:72)
z = r cos θ.

13.15 Triaxial ellipsoid

A general orthogonal triaxial ellipsoid is a quadratic surface. In Cartesian space with
the main axes aligned along the base vectors, such an ellipsoid is defined by:

x2 y2 z 2
+ + =1 (13:73)
rx2 ry2 rz2

with semi-axes of lengths rx , ry and rz (Figure 13.11). In spherical coordinates as

defined in Section 13.14 the Cartesian coordinates can be parameterized by

x = rx sin θ cos ’
y = ry sin θ sin ’ (13:74)
z = rz cos θ.
314 13 Appendix: Mathematical basics

rx
ry
rz
Figure 13.11: Definition of the main radii for an orthogonal triaxial
ellipsoid.

The volume of a triaxial ellipsoid is:

4
V= πrx ry rz . (13:75)
3

13.16 Symmetry operator

Crystallographic symmetry operations are special affine mappings leaving the crystal
structure invariant. In three-dimensional space they are represented by a (3 × 3)
rotation matrix and a (3 × 1) translation vector. The symmetry operation that maps a
point:
0 1
x
x=@yA (13:76)
z

onto a symmetry equivalent point:

0 1
x′
x′ = @ y′ A (13:77)
z′

is given by:
0 1 0 1
R11 R12 R13 t1
x = @ R21
′
R22 R23 Ax + @ t2 A = Rx + t = ðRjtÞ. (13:78)
R31 R32 R33 t3

ðRjtÞ is called a symmetry operator in Seitz notation. Some important properties of

the symmetry operator are given in the following:
If a symmetry operation ðUjuÞ is followed by a second symmetry operation, then
the combined symmetry operator is:

ðVjvÞðUjuÞ = ðVUjVu + vÞ. (13:79)

The symmetry operator for the identity operation is:

 13.17 Probability density functions 315

ðIj0Þ = ðUjuÞ − 1 ðUjuÞ, (13:80)

from which the operator for the inverse symmetry operation can be deduced as:
 
ðUjuÞ − 1 = U − 1 j − U − 1 u . (13:81)

If a (3 × 3) matrix P transforms the old to the new base vectors and a vector p translates
the old to the new origin, then the symmetry operator changes according to:
 
ðU′ju′Þ = P − 1 UPju + ðU − IÞp . (13:82)

13.17 Probability density functions

13.17.1 Gaussian distribution

The Gaussian function is a very common continuous probability density function

(normal distribution, Figure 13.12). Physical quantities that are expected to be the
sum of many independent processes (such as measurement errors) often have dis-
tributions that are nearly normal. A Gaussian function normalized to an area of unity
centered at a mean or expectation value x0 is:

1 1 x − x0 2
Gðx − x0 Þ = pffiffiffiffiffi e − 2ð σ Þ (13:83)
σ 2π

Figure 13.12: Series of normalized Gaussians with different fwhmG.

where σ is the standard deviation and σ2 the variance. The pre-factor follows from
normalization:
316 13 Appendix: Mathematical basics

+ð∞

Gðx − x0 Þdx = 1. (13:84)

−∞

The full width at half-maximum (fwhm) is nowadays the most commonly used
measurement in powder diffraction of peak widths. The fwhmG of a Gaussian follows
the simple relation:
pffiffiffiffiffiffiffiffiffi
fwhmG = 2 2ln2σ  2.355σ. (13:85)

Using this, the Gaussian function can be written as:

qffiffiffiffiffi  2
2 ln2
π − 4lnð2Þ
x − x0
Gðx − x0 Þ = e fwhmG
. (13:86)
fwhmG

13.17.2 Lorentz distribution

The Lorentzian (or Cauchy) function is another important continuous probability den-
sity distribution. The normalized Lorentz distribution with an fwhmL (Figure 13.13) is
defined as:

2
πfwhmL
Lðx − x0 Þ =  2 . (13:87)
x−x
1 + 4 fwhm0
L

Figure 13.13: Series of normal-

ized Lorentzians with different
fwhmL.
 13.17 Probability density functions 317

The Lorentz distribution is an example of a distribution that has no mean, variance

or higher moments defined. Its mode and median are well defined and are both
equal to x0 .

13.17.3 The Voigt distribution

The Voigt distribution is the convolution of a Gaussian and a Lorentzian. No analy-

tical solution exists for the convolution integral but it can be expressed by the real
part of the complex error function for which good approximations exist:

Re½wð zÞ x + iy
V ðxÞ = ðG  LÞðxÞ = pffiffiffiffiffi with z = pffiffiffi (13:88)
σ 2π σ 2

where wð zÞ is called Faddeeva-function (also called Kramp-function or relativistic

plasma-dispersion-function), and is a scalable complex conjugated error function
(see Section 13.17.6).
Voigt functions are important profile functions, and typically normalized
Gaussians and Lorentzian functions of different fwhm values, fwhmG and fwhmL are
convoluted to produce them. The total fwhm of such a Voigt function can be approxi-
mated by a polynomial of fifth order (Thompson et al. 1987):

!1
fwhmG5 þ 2:69269 fwhmG4 fwhmL þ 2:42843 fwhmG3 fwhmL2 2
fwhm ¼ (13:89)
þ4:47163 fwhmG2 fwhmL3 þ 0:07842 fwhmG fwhmL4 þ fwhmL5

An important property of both Gaussian and Lorentzian functions is that the con-
volution of a Gaussian with a Gaussian is a Gaussian and of a Lorentzian with a
Lorentzian is a Lorentzian. Thereby, it holds:

fwhmG2 = fwhmG,
2 2
1 + fwhmG, 2 (13:90)

and:

fwhmL = fwhmL, 1 + fwhmL, 2 (13:91)

where fwhmG=L, 1=2 are the fwhms of the two Gaussian or the two Lorentzian functions
to be convoluted, where fwhmG=L are the fwhms of the Gaussian and Lorentzian
function resulting from the convolution.
In combination with eq. (13.88), the convolution properties of two Gaussian and
Lorentzian functions illustrated above imply that the convolution of two Voigt func-
tions V1 and V2 is also a Voigt:
318 13 Appendix: Mathematical basics

V= V1  V2 (13:92)

where the Gaussian and Lorentzian fwhm values fwhmG and fwhmL of V can be
obtained from the corresponding values pertaining to V1 and V2 via eqs. (13.88) and
(13.89). The functions can be centered by introducing the variable X = x − x0 with the
position x0 at the average or maximum of the function.

13.17.4 The pseudo-Voigt distribution

The Voigt function can easily be convoluted, which is convenient when modeling the
profile of a diffraction peak. The requirement for a numerical approximation of the
Voigt function (due to lack of an analytical formulation), however, makes the use of
the true Voigt function computationally expensive. A popular approximation to the
Voigt function is the pseudo-Voigt function PV, which is defined as:

PV ðxÞ = ð1 − ηÞ GðxÞ + η LðxÞ, (13:93)

with G and L usually being a Gaussian and Lorentzian function with equal fwhms, that
is, fwhmG = fwhmL. These fwhm values are equal to the fwhm of the whole PV function. η
is a weighting factor, where η = 0 implies a pure Gaussian (Figure 13.12) and η = 1 a pure
Lorentzian function (Figure 13.13). Typically, the values of η are confined 0 ≤ η ≤ 1.
As illustrated in Figure 13.14, η < 0 implies negative values for some ranges of x, and η
values somewhat above 1 give PV functions which still have unimodal shapes. These
may occasionally agree with experimentally observed shapes, for example, in some
line broadening cases. Such shapes are often referred to super-Lorentz. Even larger
values of η result in multimodal distributions and or distributions with negative values
in certain ranges of x (not shown in Figure 13.14).

Figure 13.14: Normalized pseudo-Voigt functions located at x0= 0 with fwhm = 1 and η=–0.2, 0, .2,
0.5, 0.8, 1.0, 1.2.
 13.17 Probability density functions 319

In spite of its convenience and the fact that it fits experimental data well, there is
little theoretical basis for physically relevant distributions to be of pseudo-Voigt
shape. In particular, a convolution of one pseudo-Voigt function with a second is
not a pseudo-Voigt, except for the trivial cases where η1= η2 = 0 or η1= η2 = 1. It is,
however, possible to approximate a given PV by a Voigt function using eq. (13.89)
and:

η = 13.36603fwhmL =fwhm − 0.47719ð fwhmL =fwhmÞ2 + 0.11116ð fwhmL =fwhmÞ3 .

(13:94)

Then the convolution rules for Voigt functions can be approximated for pseudo-Voigt
functions, at least if 0 ≤ η ≤ 1 holds.

13.17.5 More width parameters

In addition to the fwhm, there are other important parameters quantifying widths of
distributions, each with advantages and disadvantages. We consider these width
parameters on the x scale as above.
The square root of the variance varðxÞ1=2 = σ or the standard deviation
hðx − hxiÞ2 i1=2 of a distribution described by f(x) (which is normalized to an area of
unity) is given as:
 ð∞ 1=2
hðx − hxiÞ2 i1=2 = x2 f ðxÞdx , (13:95)
−∞

R∞
with the average given as hxi = xf ðxÞdx , which agrees (in the case of a symmetric
−∞
unimodal distribution) with the peak maximum x0. The standard deviation of a
Gaussian corresponds to:

fwhmG
hðx − hxiÞ2 i1=2 = pffiffiffiffiffiffiffiffiffi , (13:96)
2 2ln2

whereas the standard deviation of Lorentzian is not defined. The same holds for a
pseudo-Voigt function in the case of η > 0. Voigt functions have a defined standard
deviation only for sufficiently small values of fwhmL =fwhm.
The integral breadth β of a function f(x), which is normalized to an area of unity,
is defined as:

1
β= , (13:97)
f ðx0 Þ

with the position of the peak maximum x0. The integral breadth of a Gaussian is given by:
320 13 Appendix: Mathematical basics

rffiffiffiffiffiffiffiffiffi
π
βG = fwhmG , (13:98)
4ln2

whereas that of a Lorentzian is given as:

π
βL = fwhmL (13:99)
2

Occasionally, a Voigt function is given in terms of its Gaussian and Lorentzian fwhm
values in terms of:

fwhmG = ð1 − ζ ÞE
(13:100)
fwhmL = ζ E.

In this case, E is also some kind of width parameter related to the fwhm of the Voigt
function in terms of:

fwhm = ð1 − ζ Þ5 + 2.69269ð1 − ζ Þ4 ζ + 2.42843ð1 − ζ Þ3 ζ 2 + 4.47163ð1 − ζ Þ2 ζ 3

1
5
+ 0.07842ð1 − ζ Þζ 4 + ζ 5 E (13:101)

Note that fwhm = E holds for ζ= 0 or 1. Note that calculation of η according to eq.
(13.94) does not lead to the pseudo-Voigt approximating the corresponding Voigt to
η= ζ except in the case of ζ= 0 or 1.

13.17.6 Error function

When the results of a series of measurements are described by a normal distribution

with standard deviation σ and expected value 0, then:
 
a
erf pffiffiffi (13:102)
σ 2

is the probability that the error of a single measurement lies between −a and +a, for
positive a, which is plotted in Figure 13.15. The error function is defined as:
ðx
2 2
erf ðxÞ = pffiffiffi et dt (13:103)
π
0

The complementary error function is defined as:

 13.18 The method of least squares 321

Figure 13.15: Plot of the error function given in eq. (13.104).

ð
∞
2 2
erfcðxÞ = 1 − erf ðxÞ = pffiffiffi e − t dt (13:104)
π
x

Using complex-valued arguments z instead of x, one obtains the so-called Faddeeva

function:

ðz !
− z2 − z2 2i t2
wð zÞ = e erfcð − izÞ = e 1 + pffiffiffi e dt  (13:105)
π
0

13.18 The method of least squares

The aim of the least-squares method is to adjust the parameters of a model function to
 
best fit a data set with, for example, n data pairs xj , yj with the independent variable
xj. The dependent variable yj, which is found by observation, can be treated as a
vector:
0 1
y1
B y2 C
B C
y=B . C (13:106)
@ .. A
yn

of dimension n. The model function of the form f ðx, pÞ contains adjustable para-
meters in form of the vector:
322 13 Appendix: Mathematical basics

0 1
p1
B p2 C
B C
p=B . C (13:107)
@ .. A
pm

of dimension m. The residual vector (observed – calc) of dimension n is then given

as:
0 1
y1 − f ðx1 , pÞ
B C
B y2 − f ðx2 , pÞ C
B C
r=B
B ..
C.
C (13:108)
B . C
@ A
yn − f ðxn , pÞ

The vector p must be determined in a way that minimizes the weighted sum S of the
squared residuals between the measured data and the model function at the posi-
tions xj :

Min X 2 X
n n    2
S! rj = yj − f xj , p . (13:109)
j=1 j=1

S is called the objective function. To minimize S, a set of m equations containing

the partial derivatives of S must be calculated with respect to all parameters pk with
k=1, . . ., m:
 
∂S Xn
∂rj Xn
∂f xj , p
=2 rj = −2 rj . (13:110)
∂pk j=1
∂pk j=1
∂pk

^ minimizes S, the partial derivatives must be zero:

If the least squares estimate p

∂S
= 0. (13:111)
∂pk

Linear and nonlinear cases must be distinguished. A regression model is called linear
when the model comprises a linear combination of the parameters, that is:

X
m
f ðx, pÞ = pk Φk ðxÞ (13:112)
k=1

where the function Φj ðxÞ is a function of x. The partial derivatives in this case
are:
 13.18 The method of least squares 323

 
∂f xj , p  
= Φk xj = Xj, k (13:113)
∂pk

or in matrix form
0 1
X1, 1  X1, n
B .. C.
X = @ ... ..
. . A (13:114)
Xm, 1  Xm, n

The partial derivatives of S can then be written as:

!
X
n X
m
−2 yi − ^k Xj, k = 0,
Xj, k p (13:115)
j=1 k=1

which can be rearranged to obtain the m normal equations denoted by index i:

n X
X m X
n
^k =
Xj, i Xj, k p Xj, i yi ði ¼ 1; :::; mÞ. (13:116)
j=1 k=1 k=1

Eq. (13.116) can be conveniently written in matrix notation:

b = Xy.
XXp (13:117)

If there are weights applied to the observations with wj > 0 as the weight of the jth
observation (usually the reciprocal of the variance of the measured value) and:
0 1
W1, 1  0
B .. C
W = @ ... ..
. . A (13:118)
0  Wn, m

is the diagonal matrix of such weights, then the normal equations are written as:
 
XWX pb = XWy. (13:119)

The algebraic solution is:

 −1
b = XWX XWy.
p (13:120)

Many different ways exist for solving eq. (13.120). One possibility is the Cholesky
decomposition where XX is decomposed into RR where R is an upper triangular
matrix.
324 13 Appendix: Mathematical basics

In practice, f ðx, pÞ is usually a nonlinear function. There is no closed-form

solution to a nonlinear least squares problem. Instead, numerical algorithms are
used to find the values of the parameters of p that minimize the objective. Most
algorithms involve choosing initial values for the parameters. Then, the para-
meters are refined iteratively, that is, the values are obtained by successive
approximation:

p  pt + 1 = pt + Δp (13:121)

where t is an iteration number, and the vector of increments Δp is called the shift
vector. In most algorithms, at each iteration the model may be linearized approxi-
mately by a first-order Taylor series expansion (see Section 13.2) around the current
values of the parameters pt ðFigure 13.16Þ :
 
    Xm
∂f xj , pt     X m
f xj , p  f xj , p +
t
pk − ptk = f xj , pt + Jjk Δpk (13:122)
k=1
∂pk k=1

Figure 13.16: Taylor expansion of first

order around the red point in a complex
two-dimensional function.

with the residual vector of dimension n now defined as:

0 1
y1 − f ðx1 , pt Þ
B y2 − f ðx2 , pt Þ C
B C
Δy = B .. C, (13:123)
@ . A
yn − f ðxn , pt Þ
 13.19 Spherical harmonics 325

which in matrix notation can be then written as:

JJΔp = JΔy (13:124)

or in case of weighted data:

 
JWJ Δp = JWΔy. (13:125)

Methods for solving the system of equations comprise the Gauss–Newton algorithm,
Levenberg–Marquardt algorithm, QR decomposition, gradient and direct search
methods.

13.19 Spherical harmonics

The spherical harmonics Ylm ðθ, ’Þ are a complete and orthogonal set of solutions of
the angular part of Laplace’s equation in three dimensions:

1
Ylm ðθ, ’Þ = pffiffiffiffiffi Nlm Plm ðcos θÞeim’ (13:126)
2π

with scaling factors:

rffiffiffiffiffiffiffiffiffiffi
2l + 1 ðl − mÞ!
Nlm = (13:127)
2 ðl + mÞ!

and the corresponding Legendre polynomials:

ð − 1Þm  m=2 dl + m  l
Plm ðxÞ = l
1 − x2 l + m
1 − x2 . (13:128)
2 l! dx

θ and ’ are the coordinates of a spherical surface. They are similar to latitude and
longitude except that θ goes from 0 to π and ’ goes from 0 to 2π. The simplest
spherical harmonic represents a sphere (Figure 13.17):

1
Y0, 0 ðθ, ’Þ = pffiffiffi . (13:129)
2 π

The spherical harmonics functions are orthogonal because if one integrates the
product of any two different harmonics over the surface of the sphere, the result is
zero. The spherical harmonics of second order are defined as:
qffiffi
Y2, 0 ðθ, ’Þ = 41 π5 ð3cos2 θ − 1Þ
326 13 Appendix: Mathematical basics

Figure 13.17: Graphical visualization of the normalized real components of second-order spherical
harmonics Y0, 0 ðθ, ’Þ, Y2, 0 ðθ, ’Þ, Y2, 1 ðθ, ’Þ and Y2, − 2 ðθ, ’Þ.

qffiffiffiffi
Y2, 1 ðθ, ’Þ = − 21 ei’ 2π cos θ sin θ
15
qffiffiffiffi
Y2, − 1 ðθ, ’Þ = 21 e − i’ 2π cos θ sin θ
15
(13:130)
qffiffiffiffi
Y2, 2 ðθ, ’Þ = 1 2i’
4e 2πsin θ
15 2
qffiffiffiffi
Y2, − 2 ðθ, ’Þ = 41 e − 2i’ 2πsin θ
15 2

For powder diffraction the symmetrized and normalized real spherical harmonics of
even order (due to the inversion center introduced by diffraction) are of most
importance. The functions are normalized such that the maximum value of each
component is 1 (Järvinen, 1993). As an example, the normalized real components of
second-order spherical harmonics become:
 References 327

Y2, 0 ðθ, ’Þ = 21 ð3cos2 θ − 1Þ

Y2, 1 ðθ, ’Þ = 2 cos’ cos θ sin θ
Y2, − 1 ðθ, ’Þ = 2 sin’ cos θ sin θ (13:131)
2
Y2, 2 ðθ, ’Þ = cosð2’Þsin θ
Y2, − 2 ðθ, ’Þ = sinð2’Þsin2 θ,

which are visualized in Figure 13.17.

The spherical harmonic functions can be expanded in a series to describe, in
principle, any direction ðθ, ’Þ dependent function.

References

Giacovazzo, C., Monaco, H.L., Artioli, G., Viterbo, D., Milaneso, M., Ferraris, G., Gilli, G., Gilli,
P., Zanotti, G., Catti, M. (2011): Fundamentals of crystallography, Edited by C. Giacovazzo. IUCr
Texts on Crystallography No. 15, 3rd editon, New York, IUCr/Oxford University Press, 842 pages.
Järvinen, M. (1993): Application of symmetrized harmonics expansion to correction of the preferred
orientation effect. J. Appl. Cryst. 26, 525–531.
Müller, U. (2013): Symmetry relationships between crystal structures, Oxford University Press, New
York, IUCr Texts on Crystallography No. 18, 332 pages.
Papula, L. (2014): Mathematik für Ingenieure und Naturwissenschaftler" Band 1, Springer Vieweg,
14th edition, 854 pages. (in German)
Thompson, P., Cox, D.E., Hastings, J.B. (1987): Rietveld refinement of Debye-Scherrer synchrotron
X-ray data from Al2O3. J. Appl. Cryst. 20, 79–83.
Index
Absorption 26, 41, 42, 45, 64, 81, 98 – Z-matrix 162, 171, 172
Absorption edge 69, 70, 71 Correlation function 303
Accidental overlap 5, 13, 16, 103, 228 Correlation matrix 186
Admissible magnetic point groups 223 Cross product 308
Agreement (R-) factor 77 Crystallite (domain) size see domain size
– Bragg 78 Crystallographic point groups 222
– expected 77 Cubic close packing (ccp) 237
– profile 77 Cumulative χ2 86
A matrix 75, 155 Cylindrical absorption correction 45
Ambiguities 227
Amorphous content 142, 146–148 Damping 260, 261
Angle restraint 157 Darwin width 59
Angular dispersive 16, 18, 20, 21, 26, 70, 97 Data collection 80
Anomalous scattering 27, 28, 34 Debye equation 10–12
Aristotype 196 Debye-Scherrer cone (ring) 5, 7, 25, 38, 62
Atomic form factor 3, 27, 29 Debye–Waller factor see displacement
Axial divergence 62, 63, 100, 104, 108 parameter
Axial vector 222 #define 286
Degrees of freedom 165, 172
Background 19, 71, 73, 80, 81 – internal 165, 166, 172
Base vectors 307 – rotational 165, 174
Black and white space group 221 – translational 165, 174
Black and white point groups 223 Delta (δ) function 1, 3, 56, 70, 303
Born-Mayer potentials 152, 153, 160 DIFFaX 243
Box function 55, 70, 103 Diffraction 1, 218
Bragg equation 5, 7, 16, 22 Diffuse scattering 253
Bragg-Brentano geometry 24, 99, 100, 106, 107 Dispersion coefficients 27
Branches 201 Displacement factor 85
Brindley correction 53, 135 Displacement parameter 31, 177
– anisotropic 32, 33, 313
Capillary 25, 45, 81 – isotropic 32
Chebyshev polynomials 71–73 – TLS 177
Child phase 196 Distance restraint 159
Circles function 62 Distortion mode 201
Command line mode 272 Distortion symmetry 200
Complementary error function 321 Divergence slit 67, 68, 81, 101, 104
Complex number 296 – fixed 67
Constraints 151, 153, 154 – variable 68
Continue_after_convegence 290 Domain size 8, 57, 59, 62, 109, 111, 127, 210
Convolution 4, 53, 56, 97, 99, 101, 109, 113, 301, Double salt Mg(H2O)6RbBr3 83, 179, 210
302 Dummy atoms 173
Coordinate system 162, 163, 296, 307, 311 Durbin–Watson statistic 78
– Cartesian 162, 163, 168, 175, 296, 297, 311,
312, 313 Elemental composition 136
– crystallographic/fractional 162, 168, Ellipsoid 313
175, 307 Emission profile 68, 99–102, 107
– spherical 162, 168, 312, 313 Energy dispersive 16, 19, 23, 70

https://doi.org/10.1515/9783110461381-014
 Index 329

Equatorial divergence 104 Instrumental resolution function (IRF) 53, 54,

#endif 286 80, 83, 97, 98, 112, 125, 137
Error function 320 Intercalation 241
Euler formula 299 Internal standard method 134
Euler identity 33 Interstratification 242
Eulerian angles 312 Irreducible representation (irrep) 200
Ewald sphere 4, 5, 7, 37 ISODISTORT 199, 201, 210, 211, 232
Exponential function 63 Isomorphic subgroup 198
External standard method 132, 140 Isotropy subgroup 200
Extinction 47
Jahn–Teller distortion 197, 228
Faddeeva function 317, 321 Joint refinements 266
Fault probability 244, 247, 250
FINDSYMM 208 Kernel 201, 203
Flatten restraint 157 Kinetics 293
for strs 286–291 Klassengleiche 198
for xdds 286
Form factor 27, 229 LaMnO3 228–235
Fourier integral 300 Lattice function 1
Fourier series 300 Laue equations 2, 3
Fourier transform 300–304 Layers 227, 246
Full width at half maximum (fwhm) 56, 59, 61, Le Bail WPPF method 89, 91, 93, 102
62, 80, 97, 111, 116, 126, 319 Least-squares method 74, 321
Fundamental parameters approach 54, 99 Legendre polynomials 325
Lennard-Jones potentials 152, 153, 160
G(r) 257 Linear absorption see Absorption
Gauss function 36, 106, 112, 117, 127, 315 List 271
General Real Space interaction 153 Local keyword 286
Genetic algorithm 208 Local structure 253
Glazer tilt system 228 Lorentz factor 37–39
Global optimizer 183, 184 Lorentz function 58, 107, 108, 112, 117, 126, 316
Gnuplot 294 Lorentz–polarization factor 37, 40, 41, 100
Goodness of fit 77
Gray groups 223
MacLaurin series 298, 299
Group theory 199 Mag_only_for_mag_sites 231
Group–subgroup 197 Magnetic form factor 220
Magnetic interaction vector 219
Hat function 70 Magnetic point groups 222
Height correction 24 Magnetic refinements 218
Height error 24, 55 Magnetic space groups 223, 225
Hettotypes 196 Magnetic structures 218
Hexagonal close packing (hcp) 237 Magnetic symmetry 220
March–Dollase function 51
#ifdef 286 Marquardt constant 76
in situ 269 Mass attenuation coefficient (MAC) 133
Inclusion runs 209 Matrix 305
Index 198 – determinant 306
Infinite thickness criterion 42 – diagonal 305, 323
330 Index

– inverse 306 – damping 260–261

– orthogonal 307 – position 21
– square 305 – profile 53, 97, 319
– symmetric 305, 307 – overlap 13
– trace 306 – shapes 97, 243
– transpose 305 Penalties 154–156
– triangular 323 Penalty functions 151, 152
– unit 305 Phase angle 35
Metric tensor 307, 310 Phase transitions 196
Mg(H2O)6RbBr3 83, 179, 210, Plotting tricks 294
Microstrain 7, 57, 59, 60, 62, 109, 116, 127 Pointer 299
Microstructure 53, 109 Polar vector 221
Miller indices 3, 6, 7, 120, 121 Polarization factor 39, 40
Monochrome/colorless groups 222 PONKCS 131, 134, 149
Multi-k 227 Porosity effect 46
Multiple data sets 269 Power series 297
Multiplicity 36, 37, 85 Preferred orientation 50, 51, 81
Primary extinction 47
Neutron diffraction 218 Primary irrep 202
Neutron scattering length 31, 125 Primed element 222
Newton-Raphson method 76 Pseudo Voigt function 62, 99, 106,
Ni 264–267 126, 318
Non–admissible magnetic point group 223
Non–crystallographic parameters 293 Quantitative phase analysis (QPA) 53, 131
Non-isomorphic subgroup 198 Quantitative Rietveld refinement 134

Objective function 74, 151, 322 R-factor see agreement factor

1/X decay function 67 Radial distribution function 258
Only penalties 156 Real space Rietveld refinement 259
Operando 269 Receiving slit width 55, 99, 100,
Order parameter components 201 103, 109
Order parameter direction (OPD) 201, 203 Reciprocal lattice 3, 309
Order parameters 199 Reciprocal metric tensor 310
Ordering 196 Reciprocal space 300
Overspill correction 48 Reduced pair distribution function 257
Reduced total scattering structure
Pair distribution function (PDF) 253 function 256
Parametric refinement 269, 281 Relative weight fraction 53
Parent 196 Remove_Phase 277
Pawley method 89, 95, 102, 103 Restraints 151, 154–157
Pb3O4 119, 190 – anti bumping 157, 158
PDF 253 – bond angles 157, 159
– area 260 – bond lengths 157, 158
– position 260 – flatness 157, 159
PDF resolution 262 Riding model 161
Pdf_data 264 Rigid bodies 151, 161, 162,
Peak 176, 179
– asymmetry 50, 98, 99, 102, 110 Robust refinement 80
– broadening/line broadening 97 Rotation 311
 Index 331

Rotation vector 167 – identity 314

Rotational rigid body 216 – inverse 315
– Seitz notation 314
Scalar product 307 – symmetry operator 314, 315
Scalar triple product 308 Systematic overlap 5, 13, 16
Scale factor 52, 53, 108, 131, 144, 231
Scattering length 30 Taylor series 74, 76, 297, 324
Scattering vector 2, 3, 219 Temperature calibration 293
Scherrer equation 8, 9, 111 Temperature factor see displacement
Scripting 272 parameter
Secondary irreps 202 Time reversal 222
Sequential analysis 269, 274, 277 TLS matrices 162, 177
SF_smooth 251 TOF (time-of-flight) 16, 23, 38, 39, 41, 61, 64,
Shubnikov groups 221 66, 70, 125
Simulated annealing (SA) 184, 187 Total-scattering structure function 254
– Boltzmann criterion 184 Transformation 310
– cost function 185 Transition probability matrix 244
– occupancy-merge 189 Translationengleiche 198
– temperature scheme 187 Transmission factor 41
Simulating powder patterns 269 Transparency 64, 65, 66, 81
Single peak fitting 89 Triaxial ellipsoid 313
– constrained 90 Trivial magnetic groups 222
– unconstrained 89 Tube tails 69, 70, 104, 106
Size broadening 109, 111 Turbostratic disorder 241, 242
Specimen displacement see height correction Twinning 240
Spherical coordinates 168, 312, 313 Two-colour space group 221
Special position 35, 85
Spherical harmonics function 51, 52, 114, 124, Unit vector 308
325, 327
Spin reversal 222 Variable divergence slits 49
Stacking disorder 237 Variance 319
Stacking faults 237 Voigt function 59, 113, 118, 120, 317
Standard deviation 319 Volume of the unit cell 309
Strain broadening see Microstrain
Structure determination 183–195 Warren peak shape 242, 243
Structure factor 33–35 Weight fraction 131
Structures 227 Weighting 78, 79, 191
Surface roughness 46, 81 Whole powder pattern fitting (WPPF) 89, 90
Symmetry adapted distortion modes 199 WO3 197, 203, 205–210, 282
Symmetry mode labels 201
Symmetry mode refinements 196, 205 Zero error 22
Symmetry modes 232 Z-matrix 171–174
Symmetry operation 314