(Louisette Priester (Auth.) ) Grain Boundaries

Springer Series in Materials Science
Volume 172
Series Editors
Zhiming M. Wang, Fayetteville, AR, USA
Chennupati Jagadish, Canberra, ACT, Australia
Robert Hull, Charlottesville, VA, USA
Richard M. Osgood, New York, NY, USA
Jürgen Parisi, Oldenburg, Germany
For further volumes:

http://www.springer.com/series/856
The Springer Series in Materials Science covers the complete spectrum of
materials physics, including fundamental principles, physical properties, materials
theory and design. Recognizing the increasing importance of materials science in
future device technologies, the book titles in this series reflect the state-of-the-art
in understanding and controlling the structure and properties of all important
classes of materials.
Louisette Priester
Grain Boundaries
From Theory to Engineering
123
Louisette Priester
Université Paris Sud 11
Paris
France
ISSN 0933-033X
ISBN 978-94-007-4968-9 ISBN 978-94-007-4969-6 (eBook)
EDP ISBN 978-2-7598-0769-7
DOI 10.1007/978-94-007-4969-6
Springer Dordrecht Heidelberg New York London
Library of Congress Control Number: 2012952016
Originally published in French as ‘‘Les Joints de Grains’’ by EDP Sciences. Ó EDP Sciences 2006.
A co-publication with EDP Sciences, 17, av. du Hoggar, F-91944 Les Ulis, France
Ó Springer Science+Business Media Dordrecht 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publisher’s location, in its current version, and permission for use must always
be obtained from Springer. Permissions for use may be obtained through RightsLink at the Copyright
Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
Most current studies tend to present the knowledge on interfaces in crystalline

materials by simultaneously considering homophase interfaces or grain bound-
aries, and hetero-interfaces located either between two crystals of the same?
material but with two different structures (such as ferrite/austenite in steels), or
between two different materials (metal A/metal B, metal oxide, metal–semicon-
ductor). In this work, we deliberately choose to limit our presentation to grain
boundaries. Despite the fact that the crystals on both sides of the interface display
the same structure and the same composition, these interfaces are not simple.
Numerous questions about their structures, their defects, and their organizations in
the material still have to be solved. In each chapter, we aim to highlight these
questions after having selected well-established data. In particular, we underline
the difficulties in going from an ideal grain boundary (akin to a perfect crystal) to a
real grain boundary (analogous to a crystal with defects). Subsequently we address
the difficulties in going from an isolated grain boundary in a bicrystal to that
included in a polycrystal grain boundary network, where each grain boundary is
constrained at triple junctions.
Two main ideas prevail in the concept of this work. The first idea is implicitly
contained in the book sub-title From Theory to Engineering: to know in order to
control and even improve. In this perspective, we not only approach the grain
boundaries at the current time, but we also try to look into the future of grain
boundary research and applications. The underlying question is the grain boundary
contribution to the overall material properties, the improvement of which being the
final goal in materials science. Now, at the beginning of the twenty-first century,
grain boundary engineering, the dream of the 1980s, seems to be taking shape on
the horizon. Indeed, with the development of new experimental and computational
techniques, progress has been made which enables us to fill the gap between the
scales and to move backward and forward between the world of atoms and that of
objects. For this reason, the present work is not restricted to a state-of-the-art
report, but it moves toward engineering by considering the exchanges between a
grain boundary and the other crystalline defects and, moreover, by immersing the
boundary in a practical environment, i.e. connected with other grain boundaries.
v
vi Preface
The second idea, strongly associated to the previous one, reveals a constant
effort to overcome the dichotomy between the ‘‘whole’’ and the ‘‘parts’’. It can be
summarized as: from the individual to the collective or from the element to the
whole. And again, the objective is to gain a better understanding and an awareness
of the practical applications.
In engineering, not only the responses of individual grain boundaries to various
stimuli, but also the collective behavior of a grain boundary network must be
known. However, grain boundary properties are not explicitly considered in this
book for the following reasons. The basic mechanisms at grain boundaries are
similar to those occurring in the crystal in several cases such as diffusion and
plastic deformation. The reader may find information in specific books dedicated
to these properties (references are given at the end of the book). Although the
behavior of isolated grain boundaries has been studied extensively, no agreement
exists on the elementary processes in cases of migration, corrosion, and wetting.
Furthermore, very little is known concerning the grain boundary electric and
magnetic properties. Generally, the properties vary with the grain boundary
geometry. However, apart from the coherent twins that display very particular
behaviors, ambiguities remain about the specific behaviors of other grain bound-
aries. Most often, the grain boundary chemistry erases the particularities linked to
the geometry. In numerous experiments, the difficulty to evaluate the solute
content in the intergranular region leads to contradictory results for a same
stimulus, a same material, and a same type of grain boundary. The main reason to
postpone the consideration of grain boundary properties is the lack of data on the
collective grain boundary behaviors in polycrystals. However, the situation could
progressively evolve with the simulation of local grain boundary textures and by
the percolation approaches of the grain boundary ensemble. Finally, although not
treated explicitly in a separate chapter, grain boundary properties are certainly
treated implicitly. The mechanical behavior may be understood on the basis of the
interactions between lattice dislocations and grain boundaries and of the inter-
granular stress relaxation under the effects of temperature and time. Other
approaches of grain boundary network properties are also briefly stipulated.
The examples treated in this book concern different crystalline materials:
metals, ceramics, semiconductors, and superconductors. In earlier times, grain
boundary studies mainly developed in the field of metallurgy. Conceptual
advances were obtained from high-resolution transmission electron microscopy
observations of semiconductor bicrystals. Only the results of experiments and
simulations are given in this volume. For the understanding of the electron
microscopy images and of the calculated grain boundary structures, the reader may
refer to the general references given at the end of the book, which also include
books dedicated to grain boundaries and to certain properties often mentioned.
Specific references are given at the end of each part and may be repeated from one
chapter to the other.
Going From Theory to Engineering, three stages need to be passed, constituting
the three parts of the book.
Preface vii
Part I deals with the concept of a perfect grain boundary, at equilibrium, and
questions the maintenance of its crystalline state. The notions of order and disorder
always raise questions from a philosophical point of view. Beauty is traditionally
linked to order and science cannot escape this esthetic connotation. Working on a
beautiful object is a noble task. For several years, studies in the grain boundary
domain mainly concern grain boundaries that possess symmetry and purity. Noble
tools such as transmission electron microscopy and atomistic simulations were
used to improve understanding of grain boundary order. At the other end of this
hierarchy of the beauty, there are ugly and impure grain boundaries, their pro-
portion in polycrystals generally being high. They play a major role in the material
properties. This first part presents the notion of bicrystallography, followed by the
description of grain boundaries in terms of dislocations and in terms of structural
units of atoms, with a special focus on the limits of these descriptions. The reasons
for which a grain boundary adopts a given structure are also discussed, knowing
that order and energy are not necessarily linked.
Part II brings us to the faulted grain boundary. It attempts to reveal the
influence of the grain boundary structure on its defects, their formation, and their
accommodation. Point, line, and volume defects are considered. Interstitial and
substitutional solutes in excess in a grain boundary, resulting from a segregation
phenomenon, strongly change the grain boundary behaviors. They may lead to pre-
wetting accompanied by an important widening of the grain boundary region that
possibly becomes non-crystalline. In the presence of segregated elements, the
differences due to geometrical parameters may be obscured. Segregation may be
the origin of the preferential formation of a second phase at grain boundaries that
notoriously affects intergranular corrosion, migration, and deformation. Precipi-
tation at grain boundaries and 3D defects well deserve to be analysed in this part.
Finally, the interactions between boundaries and lattice dislocations yield strong
disturbances at grain boundaries. The elementary mechanisms for the entrance of
dislocations in a grain boundary and for the relaxation of the associated inter-
granular stresses are discussed in greater detail in this part. They constitute the
necessary support to allow a good understanding of the mechanical properties of
isolated grain boundaries, and subsequently for grain boundaries included in an
ensemble.
Part III of the book is specifically devoted to these grain boundary ensembles
starting from the triple junction to real grain boundary networks in polycrystals.
To our knowledge, this is the first monograph to sum up the different approaches
that have been developed in recent years in an engineering direction. Despite
reserves, we take risks to select some mesoscopic and macroscopic studies of grain
boundaries that contribute to a better understanding of the grain boundary network
configuration in a material. It is an attempt to combine our knowledge of the part
and of the whole, aware that there is still a huge territory to explore. In particular,
general grain boundaries remain largely unknown. Furthermore, grain boundaries
spatially confined in nanocrystals confer specific properties to the material. They
also seem to escape order. With the development of Chaos theory, these disordered
objects give rise to new interest. Generally, though, the scientific attitude in the
viii Preface
field of grain boundaries remains classical: it deals with the research of an hin-
dered order behind an apparent disorder. The approaches developed in that domain
are beyond the scope of the deterministic disorder.
The book is addressed to graduate students preparing a thesis, to engineers and
to researchers in materials science; it attempts to give basic notions on grain
boundaries and to give rise to research subjects in that field. It also tries to interest
the scientific community in a major component of the material microstructure, the
importance of which increasing as the grain size decreases. This constitutes a real
challenge with the development of nanomaterials.
It is my sincere hope and final goal that the knowledge presented and the ideas
developed in this work may help future researches on grain boundaries.
Acknowledgments
This book would not exist without the unfailing support and encouragement of the
companion of my life since the beginning of my commitment to science:
......My very first thanks go to Pierre
This book is not only the fruit of personal reflection, but also the result of work
with students as part of their theses and/or in collaboration with several colleagues.
I am indebted to all of them for accompanying me on the route of Grain
Boundaries, either part of the way or until the end of the road.
Without being exhaustive, I will quote some of them and, primarily, Daniele
Bouchet, not to follow a chronological order, but because her work was indirectly
responsible for my orientation toward grain boundaries. Indeed, the discovery that
electrochemical behavior of iron–chromium alloys is closely related to the
microstructure linked to the chemistry of intergranular regions encouraged me to
probe further into these areas in order to understand their geography, their history,
and their future. My trip did not disappoint me. Professor R. W. Balluffi, at Cornell
University, opened to me the border of this region. I warmly thank him for his
hospitality in his laboratory and his availability to help me in my initiation.
On my return to the University of Orsay, I had the pleasure of working with two
Ph.D. students who have greatly contributed to the launching of the theme Grain
Boundaries: Sylvie Lartigue-Korinek and Omar Khalfallah. The many issues
raised by them and their enthusiasm to debate have strengthened my resolution to
establish a small research group on this subject. Over time, we always kept a
strong collaboration with the development of original themes: calculations of
elastic image force on a dislocation near the boundary with Omar, grain boundary
structures, and behaviors in rhombohedral alumina with Sylvie. From their work,
the major concern of all my scientific activity was forged: the interaction of
dislocations with grain boundaries. My incursion into the world of ceramics was
limited; Sylvie became the specialist. This intransigent researcher has remained
my most faithful collaborator; we have not ceased to exchange experiences and
ideas and for that I thank her very much.
ix
x Acknowledgments
Then I found again the metal with a multiscale approach, by simultaneously

implementing several techniques of transmission electron microscopy. In this
approach, I was helped by Brigitte Décamps to whom I am particularly grateful.
Together, we supervised two Ph.D. students that I shall not forget in my thanks,
Sophie Poulat and Jean-Philippe Couzinié, because their results allowed, among
other things, a very promising improvement on general grain boundaries. It is
through knowledge of these boundaries that we can hope to move toward a grain
boundary engineering.
Finally, I have always appreciated the collaboration of two French colleagues,
Jany Thibault -Penisson and Olivier Hardouin Duparc, with whom I have always
had fruitful discussions, especially during the writing of this book. I want to thank
them for rapid and documented answers they have always given to my questions. I
add a special recognition for their contributions to our recent work on nickel and
copper. The results obtained by coupling the studies of atomic grain boundary
structure by transmission electron microscopy and those by high-resolution sim-
ulation, well illustrate the first two chapters of this book.
Other colleagues, French and foreign, have also crossed my path by providing
their contributions which have certainly enriched the content of this work and for
which I collectively thank them.
L. Priester
Contents
Part I From Grain Boundary Order to Disorder
1 Geometrical Order of Grain Boundaries . . . . . . . . . . . . . . . . . . . 3

1.1 Grain Boundary Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Crystallographical Parameters . . . . . . . . . . . . . . . . . 4
1.1.2 Equivalent Rotations: Disorientation . . . . . . . . . . . . . 6
1.1.3 Rodrigues Vector and Quaternions . . . . . . . . . . . . . . 7
1.2 Bicrystallography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 General Methodology: From the Dichromatic
Complex to the Grain Boundary. . . . . . . . . . . . . ... 9
1.2.2 Dichromatic Pattern: Coincidence
Site Lattice (CSL) . . . . . . . . . . . . . . . . . . . . . . ... 12
1.2.3 Translation Lattice of the Bicrystal (DSC Lattice) ... 14
1.2.4 Extension of the Coincidence Notion . . . . . . . . . ... 16
1.2.5 Generalization of the Coincidence: O-Lattice
and O2-Lattice . . . . . . . . . . . . . . . . . . . . . . . . . ... 20
1.2.6 Interest and Limit of the Bollmann Approach . . . ... 22
1.3 Different Types of Grain Boundaries: Terminology . . . . . ... 23
1.3.1 Terminology Based on the Macroscopic
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 23
1.3.2 Terminology Based on the Microscopic
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 25
1.3.3 Practical Distinction Between Grain Boundaries. . ... 25
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 27
2 Mechanical Stress Order of Grain Boundaries. . . . . . . . . . . . . . . 29

2.1 Continuous Approach: The Frank-Bilby Equation. . . . . . . . . . 29
2.2 Discrete Approach: The Read and Shockley Model . . . . . . . . 32
2.3 Bollmann’s Discrete Approach: Intrinsic Dislocations . . . . . . . 33
2.3.1 Primary Intrinsic Dislocations . . . . . . . . . . . . . . . . . 33
2.3.2 Secondary Intrinsic Dislocation . . . . . . . . . . . . . . . . 37
xi
xii Contents
2.3.3 Interest and Limit of the Intrinsic

Dislocation Model . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4 Partial Intergranular Dislocations . . . . . . . . . . . . . . . . . . . . . 44
2.5 Stress Fields Associated to Intrinsic Dislocations . . . . . . . . . . 45
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3 Atomic Order of Grain Boundaries. . . . . . . . . . . . . . . . . . . . . . . 49

3.1 Hard Sphere Model: Geometrical Construction. . . . . . . . . . . . 50
3.2 Structural Unit Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2.1 Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2.2 Hierarchy of the Descriptions. . . . . . . . . . . . . . . . . . 55
3.2.3 Multiplicity of the Descriptions . . . . . . . . . . . . . . . . 58
3.2.4 Geometrical Construction of the Grain
Boundary Structure . . . . . . . . . . . . . . . . . . . . . .... 61
3.2.5 Algorithm for the Determination
of the Grain Boundary Structure. . . . . . . . . . . . .... 62
3.2.6 Determination of the Grain Boundary Structure
by the Strip Method . . . . . . . . . . . . . . . . . . . . .... 64
3.3 Interests and Limits of the Structural Unit Model:
Any Model has its Exceptions . . . . . . . . . . . . . . . . . . . .... 66
3.3.1 Three-Dimensional Tilt Grain
Boundaries in Metals . . . . . . . . . . . . . . . . . . . . . . . 68
3.3.2 Asymmetrical Tilt Grain Boundaries in Metals . . . . . 72
3.3.3 Twist Grain Boundaries in Metals . . . . . . . . . . . . . . 76
3.3.4 Grain Boundaries in Covalent Materials . . . . . . . . . . 76
3.3.5 Grain Boundaries in Ionic Materials . . . . . . . . . . . . . 79
3.4 Structural Unit Model/Intrinsic Grain
Boundary Dislocations . . . . . . . . . . . . . . . . . . . . . . . . .... 83
3.4.1 Principle of the SU/GBD Model. . . . . . . . . . . . .... 84
3.4.2 Characterization of the Dislocations Associated
to Structural Units . . . . . . . . . . . . . . . . . . . . . .... 85
3.4.3 Application of the SU/GBD Model
to Tilt Grain Boundaries . . . . . . . . . . . . . . . . . .... 88
3.4.4 Limits of the SU/GBD Model for the Twist
Grain Boundaries . . . . . . . . . . . . . . . . . . . . . . .... 89
3.5 Structural Unit/Disclination Model . . . . . . . . . . . . . . . . .... 91
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 92
4 Grain Boundary Order/Disorder and Energy . . . . . . . . . . . . ... 93

4.1 Grain Boundary Order or Disorder at High Temperature? . ... 93
4.1.1 Solid/Solid Phase Transformations
at Grain Boundaries . . . . . . . . . . . . . . . . . . . . . ... 93
4.1.2 Grain Boundary Pre-melting?. . . . . . . . . . . . . . . ... 95
Contents xiii
4.2 Interfacial Energy: Thermodynamic Aspects

and Energy Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 99
4.3 Macroscopic Degrees of Freedom and Interfacial Energy . . .. 100
4.3.1 Variation of the Interfacial Energy
with the Misorientation Angle . . . . . . . . . . . . . . . .. 101
4.3.2 Variation of the Interfacial Energy with the
Grain Boundary Plane Inclination . . . . . . . . . . . . . .. 106
4.4 Microscopic Degrees of Freedom and Interfacial Energy. . . .. 113
in Plane Rigid Body Translation. . . . . . . . . . . . . . .. 113
Expansion Normal to the Grain Boundary Plane . . .. 115
with Local Individual Atomic Relaxations . . . . . . . . . 117
4.5 Are There any Geometrical Criteria of Minimum Energy? . . . 117
4.5.1 Low Sigma Value Criterion . . . . . . . . . . . . . . . . . . . 118
4.5.2 High Gamma Value Criterion . . . . . . . . . . . . . . . . . 119
4.5.3 High d Value Criterion . . . . . . . . . . . . . . . . . . . . . . 120
4.5.4 High Gamma Value with Constant d Criterion . . . . . . 120
4.6 Energy and Classification of Grain Boundaries: The Limits. . . 122
4.6.1 Classification Directly Based
on the Grain Boundary Energy. . . . . . . . . . . . . . . .. 122
4.6.2 Classification Based on the Grain Boundary
Interplanar Spacing . . . . . . . . . . . . . . . . . . . . . . . .. 123
4.7 Grain Boundary Order or Disorder: What Conclusion? . . . . .. 127
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 132
Part II From Ideal to Real Grain Boundary
5 Defects in the Grain Boundary Structure . . . . . . . . . . . . . . . . . . 135

5.1 Point Defects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.2 Linear Defects: Extrinsic Dislocations. . . . . . . . . . . . . . . . . . 138
5.2.1 Definition of an Extrinsic Dislocation . . . . . . . . . . . . 138
5.2.2 Geometrical Characteristics of an Extrinsic
Dislocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.2.3 Origin of an Extrinsic Dislocation . . . . . . . . . . . . . . 142
5.2.4 Extrinsic Dislocation Core . . . . . . . . . . . . . . . . . . . . 142
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
6 Grain Boundary Segregation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

6.1 Driving Forces Equilibrium Segregation . . . . . . . . . . . . . . . . 148
6.1.1 Elastic Interactions . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.1.2 Electronic Effects . . . . . . . . . . . . . . . . . . . . . . . . . . 151
xiv Contents
6.2 Thermodynamic Approaches of Equilibrium Segregation. .... 151

6.2.1 Gibbs Adsorption Isotherm . . . . . . . . . . . . . . . .... 152
6.2.2 Segregation in Regular Solid Solution
Without Interaction . . . . . . . . . . . . . . . . . . . . . .... 153
6.2.3 Segregation in Regular Solid Solution
with Interactions. . . . . . . . . . . . . . . . . . . . . . . .... 156
6.3 Segregation Models Based on the Statistical Mechanics . .... 159
6.3.1 Model of Regular Solid Solution
with the Bragg-Williams Approximation . . . . . . .... 160
6.3.2 Mean Field Approximation (MFA) Models . . . . .... 161
6.4 Average Segregation at Grain Boundaries . . . . . . . . . . . .... 162
6.4.1 Dependence of the Average Grain Boundary
Segregation on the Temperature and the
Solute Concentration. . . . . . . . . . . . . . . . . . . . .... 162
6.4.2 Influence of the Interaction Term
on the Average Segregation . . . . . . . . . . . . . . . .... 163
6.5 Relation Between Segregation and Grain
Boundary Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 165
6.5.1 Segregation and Grain Boundary
Geometrical Parameters. . . . . . . . . . . . . . . . . . .... 166
6.5.2 Grain Boundary Segregation and Intrinsic
Dislocations . . . . . . . . . . . . . . . . . . . . . . . . . . .... 177
6.5.3 Grain Boundary Segregation and Grain
Boundary Atomic Structure . . . . . . . . . . . . . . . .... 182
6.5.4 Grain Boundary Segregation and Grain
Boundary Electronic Structure . . . . . . . . . . . . . .... 192
6.6 Prewetting Transition Upon Segregation
at Grain Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . .... 200
6.7 Concept of Complexions . . . . . . . . . . . . . . . . . . . . . . . .... 203
6.8 Role of Extrinsic Dislocations in Grain Boundary
Equilibrium Segregation . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6.9 Non-equilibrium Segregation at Grain Boundaries . . . . . . . . . 207
6.10 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
7 Precipitation at Grain Boundaries. . . . . . . . . . . . . . . . . . . . . . . . 217

7.1 Energetic Aspect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
7.2 Different Types of Interfaces and Precipitates . . . . . . . . . . . . 220
7.2.1 Coherent Interface . . . . . . . . . . . . . . . . . . . . . . . . . 220
7.2.2 Semicoherent Interface . . . . . . . . . . . . . . . . . . . . . . 222
7.2.3 Incoherent Interface . . . . . . . . . . . . . . . . . . . . . . . . 223
7.2.4 Different Types of Precipitates . . . . . . . . . . . . . . . . . 223
Contents xv
7.3 Generalised Wulff Construction for Nucleus Equilibrium

at Grain Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 224
7.3.1 Equilibrium Condition for a Nucleus
at a Grain Boundary . . . . . . . . . . . . . . . . . . . . .... 224
7.3.2 Principle of the Generalised Wulff Construction. .... 226
7.3.3 Equilibrium Shape of Two-Dimensional Nuclei
at Grain Boundaries . . . . . . . . . . . . . . . . . . . . .... 227
7.3.4 Influence on the Grain Boundary Plane
on the Nucleus Shape . . . . . . . . . . . . . . . . . . . .... 232
7.3.5 Equilibrium Shape of Three-Dimensional Nuclei
at Grain Boundaries . . . . . . . . . . . . . . . . . . . . . . . . 233
7.3.6 Grain Boundary Puckering Phenomenon . . . . . . . . . . 234
7.4 Grain Boundary Precipitate Growth . . . . . . . . . . . . . . . . . . . 235
7.4.1 Migration of a Curved Incoherent Interface . . . . . . . . 236
7.4.2 Migration of a Planar Interface
(Coherent or Semicoherent) . . . . . . . . . . . . . . . .... 236
7.5 Localisation of Grain Boundary Precipitates
on Extrinsic Dislocations . . . . . . . . . . . . . . . . . . . . . . . .... 237
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 240
8 Interactions Between Dislocations and Grain Boundaries . . . . . . . 241

8.1 Long-Range Elastic Interaction: Image Force. . . . . . . . . . . . . 242
8.2 Dislocation Configurations in the Vicinity
of a Grain Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
8.3 Short-Range Interactions Between Linear
and Planar Defects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
8.3.1 Combination Processes . . . . . . . . . . . . . . . . . . . . . . 249
8.3.2 Decomposition Processes. . . . . . . . . . . . . . . . . . . . . 250
8.3.3 Transmission Processes . . . . . . . . . . . . . . . . . . . . . . 254
8.3.4 Entrance of a Dissociated Dislocation
in a Grain Boundary . . . . . . . . . . . . . . . . . . . . . . . . 257
8.3.5 Simulation of the Interaction of a Lattice Dislocation
with a Grain Boundary . . . . . . . . . . . . . . . . . . . . . . 261
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
9 Intergranular Stress Relaxation . . . . . . . . . . . . . . . . . ........ 269

9.1 Extrinsic Dislocation Accommodation Models . . . . ........ 270
9.1.1 Extrinsic Dislocation Core Delocalisation . ........ 270
9.1.2 Extrinsic Dislocation Decomposition
and Product Reorganisation . . . . . . . . . . . ........ 272
9.1.3 Extrinsic Dislocation Incorporation Within
the Intrinsic Structure . . . . . . . . . . . . . . . ........ 274
xvi Contents
9.2 Evolution of Extrinsic Dislocation Stress Fields . . . . . . . . . . . 277

9.2.1 Random Disordered Dislocation Wall Model . . . . . . . 277
9.2.2 Quasi-Equidistant Grain Boundary Model . . . . . . . . . 279
9.3 Evolution of Extrinsic Dislocation Stress Fields with Time . . . 282
9.4 Experimental Studies of Extrinsic Dislocation
Accommodation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 284
9.4.1 Accommodation in Symmetrical Tilt
Grain Boundaries in Semiconductors . . . . . . . . . . .. 284
9.4.2 Accommodation in Singular, Vicinal
and General Grain Boundaries in Metals . . . . . . . . .. 286
9.4.3 Accommodation Kinetics. . . . . . . . . . . . . . . . . . . .. 297
9.5 Conclusion on the Extrinsic Dislocation
Relaxation Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . .. 300
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 300
Part III From the Free to the Constrained Grain Boundary
10 The Triple Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 305

10.1 Triple Junction Geometry . . . . . . . . . . . . . . . . . . . . . ..... 305
10.1.1 Geometrical Parameters and Triple
Junction Classification. . . . . . . . . . . . . . . . . . . . . . . 305
10.1.2 Tricrystallography. . . . . . . . . . . . . . . . . . . . . . . . . . 307
10.2 Triple Junction Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . 310
10.2.1 Thermodynamic Approach: Limits . . . . . . . . . . . . . . 311
10.2.2 Equilibrium in Terms of Intrinsic Dislocations . . . . . . 317
10.2.3 Equilibrium in Terms of Structural Units. . . . . . . . . . 319
10.3 Triple Junction Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
10.3.1 Calculation of the Triple Junction Energy . . . . . . . . . 322
10.3.2 Experimental Determination
of the Triple Junction Energy. . . . . . . . . . . . . ..... 325
10.4 Triple Junction Defects . . . . . . . . . . . . . . . . . . . . . . . ..... 326
10.4.1 Intrinsic Defects of a Triple Junction:
Geometrical Approach. . . . . . . . . . . . . . . . . . ..... 327
10.4.2 Extrinsic Defects of a Triple Junction:
Mechanical Approach . . . . . . . . . . . . . . . . . . ..... 330
10.5 From Tricrystal to Polycrystal . . . . . . . . . . . . . . . . . . ..... 333
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 336
11 Grain Boundary Network: Grain Boundary Texture . . . . . . . . . . 337

11.1 Criteria for Grain Boundary Distribution . . . . . . . . . . . . . . . . 339
11.1.1 Misorientation Criterion . . . . . . . . . . . . . . . . . . . . . 340
11.1.2 Grain Boundary Plane Criterion . . . . . . . . . . . . . . . . 341
11.1.3 Non Geometrical Criteria . . . . . . . . . . . . . . . . . . . . 343
11.2 Calculation of the Misorientation Distribution . . . . . . . . . . . . 344
Contents xvii
11.2.1Crystal Orientation and Grain Boundary

Misorientation Distribution Functions . . . . . . . . . . .. 345
11.2.2 Theoretical Misorientation Distributions . . . . . . . . .. 347
11.3 Experimental Grain Boundary Misorientation Distributions . .. 353
11.3.1 Different Types of Experimental
Misorientation Distributions . . . . . . . . . . . . . . . . . .. 355
11.3.2 Effects of the Structure and of the Stacking
Fault Energy of the Material . . . . . . . . . . . . . . . . .. 356
11.3.3 Effects of the Material Purity. . . . . . . . . . . . . . . . .. 362
11.3.4 Effects of Thermo-Mechanical Treatments:
Relation with Crystal Texture . . . . . . . . . . . . . . . .. 366
11.4 Grain Boundary Plane Distributions . . . . . . . . . . . . . . . . . .. 383
11.5 Five Grain Boundary Macroscopic Parameters
Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
11.5.1 Theoretical Approach . . . . . . . . . . . . . . . . . . . . . . . 389
11.5.2 Experimental Approach . . . . . . . . . . . . . . . . . . . . . . 390
11.6 Grain Boundary Property Distributions . . . . . . . . . . . . . . . . . 393
11.6.1 Grain Boundary Diffusivity Distribution . . . . . . . . . . 394
11.6.2 Grain Boundary Energy Distribution . . . . . . . . . . . . . 394
11.7 Triple Junction Distributions . . . . . . . . . . . . . . . . . . . . . . . . 397
11.7.1 Limit for Application of the Coincidence
Index Combination Rule . . . . . . . . . . . . . . . . . . . .. 398
11.7.2 Theoretical Approach of the Triple
Junction Distribution . . . . . . . . . . . . . . . . . . . . . . .. 402
11.7.3 Experimental Triple Junction Distributions . . . . . . .. 407
11.7.4 Grain Boundary Energy Distribution Starting
from the Dihedral Angle Distribution
at Triple Junctions . . . . . . . . . . . . . . . . . . . . . . . . . 408
11.8 Local Grain Boundary Texture . . . . . . . . . . . . . . . . . . . . . . . 411
11.8.1 Different Types of Clusters . . . . . . . . . . . . . . . . . . . 411
11.8.2 Observed Cluster Configurations . . . . . . . . . . . . . . . 413
11.8.3 Simulated Cluster Configurations . . . . . . . . . . . . . . . 418
11.9 Percolation Concept Applied to Grain Boundary Networks . . . 424
11.9.1 Infinite Grain Boundary Network . . . . . . . . . . . . . . . 424
11.9.2 Finite Grain Boundary Network . . . . . . . . . . . . . . . . 429
11.9.3 Correlated Percolation . . . . . . . . . . . . . . . . . . . . . . . 429
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
Recommended Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
Acronyms
CAD Coincidence Axis Direction

CGBD Coincidence Grain Boundary Distribution
c.n.i.d. Cell of Non Identical Displacements
CSL Coincidence Site Lattice
b.c.c. Base-centered cubic
f.c.c. Face-centered cubic
DSC Displacement Shift Complete1
Displacement Symmetry Conserving2
EAM Embedded Atom Method
EBSD Electron Back Scattering Diffraction
EGBD Extrinsic Grain Boundary Dislocation
GBCD Grain Boundary Character Distribution
GBD Grain Boundary Dislocation
GBMD Grain Boundary Misorientation3 Distribution
IGBD Intrinsic Grain Boundary Dislocation
ISDF Intercrystalline Structure Distribution Function
LDFT Local Density Functional Theory
MDF Misorientation Distribution Function
SEM Scanning Electron Microscopy
TEM Transmission Electron Microscopy
HRTEM High-resolution Transmission Electron Microscopy
MFA Mean Field Approximation
NNR Nearest Neighbor Relationship
OCF Orientation Coherence Function–Orientation Correlation Function
1
Definition 1 is the original.
2
Definition 2 is more explicit.
3
The term ‘‘Disorientation’’ indicates one of the relative orientation relationships between two
crystals retained by convention as characteristic of the class of all the equivalent rotations
deduced by symmetry operations. The term ‘‘Misorientation’’, generally used in this book is valid
for any one of these relative orientations.
xix
xx Acronyms
ODF Orientation Distribution Function

OFE Oxygen-free Electronic (copper)
OIM Orientation Imaging Microscopy
S.U. Structural Unit
YAG Yttrium Aluminum Garnet
Introduction
Brief History of the Grain
Boundary Order Concept
During the first half of the twentieth century, the idea that the grain boundary
constitutes an amorphous layer between two crystals prevailed, until some vec-
torial properties were discovered in around the 1950s. The strong anisotropy of the
intergranular penetration of some elements in function of the direction in the grain
boundary was found inconsistent with an amorphous grain boundary nature;
besides, it was previously argued that a grain boundary may be formed by regions
of 00 good fit00 separated by regions of 00 bad fit00 . The first results and models on grain
boundary diffusion were obtained at the same time than the first theoretical
descriptions of a grain boundary in terms of dislocations which was proposed by
[1]. This description historically precedes the geometrical approaches that are first
presented in this Part.
The geometrical models and the intrinsic dislocation models of the grain
boundary structure, described in Chaps. 1 and 2 are based, on one hand, on the
preservation of the crystalline symmetry and, on the other hand, on the possible
modes of intergranular stress relaxation. They are not physical models in that sense
that they do not specify the atom arrangement in the grain boundary region.
Chapter 3 deals with the description of the atomic grain boundary structure.
First predicated by a rigid model that takes into account the sizes and the coor-
dination of atoms, the grain boundary structure was then described as an
arrangement of atomic polyhedra; this was deduced from grain boundary energy
calculations and minimizations. The calculated structure is, as far as possible,
compared to the structure observed by high-resolution transmission electron
microscopy. The atom organization in a restricted number of polyhedral clusters is
confirmed and the periodic arrangement of these units is formalized in the
structural unit model. Simultaneously, a strong correlation is established between
the latter model and the grain boundary description in terms of intrinsic disloca-
tions, giving rise to the structural unit / grain boundary dislocation (SU/GBD)
model. But the nature of the grain boundary is more unworkable and cannot be
xxi
xxii Introduction
classified as ordered without restriction. Indeed, even when the structural unit
model supports the presence of an atomic order in a large number of grain
boundaries, it questions this order with its extension to quasi-periodic grain
boundaries. Besides, the experiments reveal the existence of three-dimensional
(3D) grain boundaries, not theoretically predicted. Finally and more recently,
amorphous grain boundaries seem to re-appear. These contradictions justify the
title retained for this chapter: From grain boundary order to disorder.
Chapter 4 is dedicated to the grain boundary order and energy. First, we deal
with the order evolution in function of the temperature; in particular, we are
concerned with the possibility of proper grain boundary phase transformations and
with that of pre-melting in the intergranular region. The effects of solutes on grain
boundary structural transformations are analysed in Part II of the book.
Inherent to the structure calculations, the grain boundary energy appears as a
factor that may influence the grain boundary behavior. Energetic considerations
are the main subject of Chap. 4 with an underlying question: is it possible to
predict the grain boundary energy in function of the macroscopic parameters, the
only ones available for real grain boundaries. In fact, we show that no reciprocal
relationship exists between the grain boundary energy and the grain boundary
geometry. Moreover, energy values do not permit to predict intergranular prop-
erties. Finally, we raise the problems associated to any classification of grain
boundaries based on their energies.
Finally, we attempt to review the fundamental question suggested by the Part I
title: what are the limits of the intergranular order? Could the structure of the real
grain boundaries be disordered? The responses that we may bring to this question
are of great importance for all points developed in the following parts of this book.
Indeed, we cannot approach the defects of a grain boundary without knowing its
perfect structure. We cannot study triple junctions appearing in polycrystals
without knowing the structures of all the abutting grain boundaries. Finally, the
grain boundary properties strongly depend on the grain boundary structures.
Reference
1. W.T. Read, W. Shockley, Phys. Rev. 78, 275 (1950).

Part I
From Grain Boundary
Order to Disorder
Les invariances physiques appuyées sur la structure des
groupes nous paraissent donner une valeur rationnelle aux
principes de permanence à la base des phénoménes physiques
(G. Bachelard)
The physical invariances that rest on the group structure seem

to give a rational value to the principles of permanence at the
basis of the physical phenomena.
(G. Bachelard)
This title want to be provocative insofar as the present knowledge allows us to say
that the grain boundary is much well described as an ordered structure than a
disordered one, as it was the case in the past. Indeed, most of the grain boundaries
until now studied display a periodic structure. However, the description of certain
grain boundaries in terms of quasi crystalline structures, the observation of non-
localized intergranular stresses or the presence, revealed by simulation, of strongly
disordered atomic layers in the core of some grain boundaries question the
existence of amorphous grain boundaries. By another way, the definition of the
amorphous state has greatly evolved since the first grain boundary model in terms
of cement layer between two crystals. The crystalline and quasi-crystalline states
define long range ordered structures; on the contrary, the amorphous state may be
only ordered over very short distances. However, for some properties, it is not
excluded that periodicity could be secondary compared to the organization of the
atomic motif.
In this first part, we attempt to show how a pseudo backward and forward
motion in the conception of the grain boundary structure has occurred: from the
whole amorphous to the strictly periodic, then back to a quasi-crystalline state that
may be considered as intermediate between the two previous states. Efforts must
be undertaken to know general grain boundaries, likely predominant in real
materials, in order to confirm or invalidate the existence of an amorphous state, at
least at high temperature. These preoccupations prevail in the different chapters of
this part and explain their chronology.
Chapter 1
Geometrical Order of Grain Boundaries
The idea at the origin of a grain boundary geometrical order comes from a general law
in physics: between two physical objects (here two crystals) or between two physical
quantities in relation to each other, an invariant always exists. More precisely and
on the contrary to the amorphous layer concept [1], we postulate that two periodic
structures that are interpenetrated must also give rise to a periodic bi-structure. The
existence of order was supported by diffusion experiments [2] and by some physical
descriptions [3, 4]
To establish a geometrical order requires searching for invariants between the
two crystals (I and II) misorientated and/or translated [5, 6]. The description of the
geometrical models that have been developed on the basis of this invariance imposes
first to specify the grain boundary geometrical characteristics, then to determine the
symmetry operations of the bicrystal formed by the two interpenetrated adjacent
crystals: it is the subject of the bicrystallography. Finally, we present a terminology
of grain boundaries that rests on their geometrical characteristics and which will be
used throughout the book.
1.1 Grain Boundary Geometry
A grain boundary separates two crystals of same nature and same structure related
to each other by the general Eq. (1.1):
x II = (A/τ )GI x I (1.1)
x I and x II are two vectors that define two homologous points in crystals I and II,
A is a homogeneous linear transformation, τ is a rigid body translation of crystal (II)
with respect to crystal (I) and GI represents all the symmetry operations of the space
group of crystal I. The transformation A followed by the translation τ is named
interface operation or isometry: α = (A/τ ) [7]. A is a 3 × 3 matrix and τ is a
L. Priester, Grain Boundaries, Springer Series in Materials Science 172, 3

DOI: 10.1007/978-94-007-4969-6_1, © Springer Science+Business Media Dordrecht 2013
4 1 Geometrical Order of Grain Boundaries
column vector. The rigid body translation is unique (modulo a crystal translation
vector) only for the symmorphic space groups with only one atom per primitive unit
cell. On the contrary, a translation associated to a point operation in the space group
(helical axis for example) exists that cannot be suppressed by any translation of the
reference lattice; the operation that relies two homologous points is then given by:
(A/τ )(GI /t) = (AGI /τ + GI t). Thereafter, we consider only the translation τ .
In a crystal, there is a set of equivalent points to a given point x; these points are
obtained by applying to x any symmetry operation of the crystal space group, this set
is called G-orbit of x. Thus, any isometry (A /τ ) that relates a point of the G-orbit
of xI to any point of the G-orbit of xII is equivalent to (A/τ ) (1.1):
(A /τ ) = (A/τ ) · GI = αGI (1.2)
The set of all equivalent isometries (A/τ ) G is called the co-set associated to the
interface.
The inverse operations that relate any point of crystal I to any homologous point
of crystal II are given by GI (A/τ )−1 or (A/τ )−1 GII . A relation of conjugation links
the GI and GII groups:
GII = (A/τ )GI (A/τ )−1 (1.3)
When considering the operations from the right to the left, this relation expresses
that applying to a point xII of crystal II the inverse operation (A/τ )−1 , we obtain an
homologous point xI of crystal I, then by applying GI an equivalent point to xI . The
latter then undergoes a direct operation (A/τ ) that leads to any point of the GII orbit
of crystal II. Substituting (A/τ ) with α:
GII = αGI α−1 and GI = α−1 GII α (1.3 bis)
In the previous equations, A is generally a complex transformation that implies a

rotation R and a deformation D. It can be reduced to a pure rotation, apart from the
case of enantiomorphic crystals, or to a pure deformation in the case on semicoherent
interfaces between two phases. In the rest of the book, that concerns particularly
grain boundaries in cubic materials, the interface operation only implies a rotation
R, accompanied or not by a rigid translation τ . If only rigid translation applies, the
interface is a stacking fault or an anti-phase boundary. However, the grain boundary
description in materials of lower symmetry may require, in addition to the rotation,
a lattice deformation; this will be clarified in the cases of materials with hexagonal
and rhombohedral structures.
1.1.1 Crystallographical Parameters
The parameters that are necessary to characterize a grain boundary are also called
geometrical degrees of freedom of the grain boundary [8]. They can be divided into
1.1 Grain Boundary Geometry 5
Table 1.1 The nine geometrical degrees of freedom of a grain boundary: six for the interface
operation and three for the grain boundary plane orientation and position
Macroscopic parameters Microscopic parameters
Number Type Number Type
2 [uvw] 2 τ 1 and τ 2
Interface operation Rotation axis Translation in the grain
boundary plane
1 θ 1 τ3
Rotation angle Expansion
Grain boundary 2 n 1 d<n
plane
Orientation of the grain Position of the grain
boundary plane boundary plane
Five are macroscopic parameters. The dimension of each of the four microscopic parameters is less
than the grain boundary period
two groups: six are necessary for describing the interface operation (R/τ ) appearing
in Eq. (1.1), with A = R in that case; three are required to define the orientation
and the position of the grain boundary plane. We may also distinguish between the
macroscopic and the microscopic parameters (Table 1.1).
There are five macroscopic degrees of freedom:
• One for the rotation angle θ between the two crystals
• Two for the rotation axis [uvw] defined by its direction cosines
• Two for the orientation of the grain boundary plane defined by its normal n.
There are four microscopic degrees of freedom:
• Three for the translation of one crystal with respect to the other: the rigid body
translation vector τ
• One for the vector d normal to the gain boundary plane that indicates the position
of the plane along its normal; the intensity of d is necessarily less than the unit
vector n.
Thus, a grain boundary possesses nine geometrical degrees of freedom; however
the microscopic parameters result from relaxation processes and are not independent
of the macroscopic parameters. Finally, the five macroscopic degrees of freedom
constitute thermodynamic variables that are sufficient to give a complete thermody-
namic description of the grain boundary.
Other definitions of the geometrical parameters, proposed by Wolf [9], highlight
the importance of the grain boundary plane. Four of them specify the unit vectors
n1 and n2 that define the orientations of the normal to the grain boundary plane
in crystals I and II; the fifth gives the twist rotation angle α between the crystals
around the grain boundary plane normal. For non-cubic structures, establishing a
relationship between two crystals requires to express the five macroscopic parameters
in an adequate coordinate system.
Fig. 1.1 Different stages to form a grain: a interpenetration of two misorientated crystals; b rigid
body translation of one crystal with respect to the other; c introduction of a grain boundary plane
(different positions are possible for a same orientation); d atoms of crystal I and crystal II are
rejected on one side and on the other side, respectively
The construction of a grain boundary is schematized on Fig. 1.1.

The macroscopic geometrical parameters are imposed in the simulations of the
grain boundary structure; the microscopic parameters are then deduced after grain
boundary energy minimization. On the contrary, in experimental studies, the exact
values of the macroscopic parameters are not known because there are uncertainties
in measurements. These parameters are random (unpredictable) for real grain bound-
aries in polycrystals whereas they are more or less controlled in bicrystal elaboration.
Indeed, for most materials apart for semiconductors, the obtained bicrystal grain
boundary always deviates from the intended misorientation and plane.
1.1.2 Equivalent Rotations: Disorientation
We now consider that the point transformation A in the relation (1.2) is reduced
to a rotation R, which is always the case for cubic materials. The misorientation
is the difference in crystallographic orientation between the two crystals; it may be
expressed by different rotations deduced from each other by a symmetry operation S
of the crystal space group G. If we fix a coordinate system in the crystal I, then any
rotation R has an equivalent rotation R = R Si , with i = 1 to n (n is the number of
point group operations of the crystal). The rotation may also be expressed in another
coordinate system equivalent by symmetry: R = S−1 i R Si ; moreover, each of the
lattices may be considered as reference system: R = R−1 .
Finally, the descriptions by R or R are equivalent if:
R = Si R Sj or R = S−1
i R Sj . (1.4)
The N different and equivalent rotations form a class; N = 1152 for a grain
boundary in the cubic materials (24×24×2). For any crystalline system, there exists
a limited number of equivalent geometrical descriptions of a same grain boundary,
for example N = 72 in the case of the rhombohedral symmetry (6 × 6 × 2). It is
always possible to choose a rotation axis [uvw] within the standard stereographic
triangle with u ≥ v ≥ w ≥ 0. The misorientation with the smallest possible rotation
1.1 Grain Boundary Geometry 7
angle out of all symmetrically equivalent rotations that falls within that triangle is
called Disorientation. It characterizes the class of equivalent rotations.
In the cubic structures, the disorientation angle is always less than 62◦ . For
example, the disorientation of a twin is 60◦ around [111], but this grain bound-
ary may also be described by 70.32◦ around a 110 axis. The disorientation has
no specific meaning; it just permits to have a conventional denomination for a grain
boundary. In particular, among all the equivalent descriptions, there is a 180◦ rotation
angle around a uvw axis that indicates the symmetry plane of the bicrystal normal
to this axis. For example, a grain boundary with a disorientation of 38.14◦ around
[110] has two equivalent descriptions with a 180◦ rotation angle, one around 221
and the other around 411. The {221} and {411} planes are the symmetry planes
of that grain boundary [10].
For the cubic materials, the components rij of the rotation matrix R (with i, j =
1, 2, 3) are expressed in function of the rotation angle θ and of the unit vector t
components t1 , t2 et t3 on the [uvw] rotation axis:
2
1 0 0 t 0 −t3 t2
1 t12t2 t1 t3
R = ri j = cos θ 0 1 0 + (1− cos θ )t2 t1 t2 t2 t3 + sin θ t3 0 −t1
0 0 1 t3 t1 t3 t2 t 2 −t 2 t1 0
3
(1.5)
with Det. R = 1
Reciprocally, if we know the rotation matrix, the rotation angle and the rotation
axis indices may be deduced:
trace R − 1
cos θ = (1.6)
2
The components t1 , t2 and t3 of the elementary vector on the rotation axis are deter-
mined by solving a series of equation obtained from Eq. (1.5) knowing R and θ .
The matrices of the equivalent rotations are obtained by change of the sign of
one column or one line and by permutation of the columns and the lines of the
matrix R such as the value of the determinant is always equal to 1. In particular, such
exchanges allow us to align the higher elements of R on the diagonal of a matrix R
that represents the Disorientation.
1.1.3 Rodrigues Vector and Quaternions
Apart from the rotation matrix or the couple formed by the rotation angle and axis,
there are two other ways to mathematically describe a grain boundary.
• The Rodrigues vector R:
R = t tg θ/2 (1.7)
t is the elementary vector on the rotation axis defined in the expression (1.5) by its
components that also are its direction cosines. For example, the couple angle/axis
70.5◦ [110] is represented by a Rodrigues vector 1/2, 1/2, 0. The set of all possible
rotations may be mapped onto 3D space forming the Frank-Rodrigues map.
Initially defined to describe the position of a point in a coordinate system and used
to account for the texture of a material, the Rodrigues representation has proved very
useful to approach the grain boundary disorientation distribution in a polycrystal (see
Chap. 11).
• Quaternions:
The quaternion representation of a grain boundary has been particularly useful
to determine all the equivalent rotations in a cubic crystal. The rotation θ around a
[uvw] axis is given by the unit quaternion:
Q = (e1 , e2 , e3 , e4 ) = (t1 sin θ/2, t2 sin θ/2, t3 sin θ/2, cos θ/2) (1.8)
The four components e1 , e2 , e3 and e4 are related by: e1 2 + e2 2 + e3 2 + e4 2 = 1.

The quaternions will not be used later in the book. We generally describe a grain
boundary by one of the relative orientations between the neighbouring crystals or
misorientation (not necessarily the disorientation) and by the mean grain boundary
plane.
1.2 Bicrystallography
We have already noted that most laws in Physics are conservation laws and that
researching the invariance between two interpenetrated crystals is the basis of the
geometrical approach of grain boundaries. The first invariant is evidently the crystal.
The second invariant has been introduced by Friedel in 1926 [11] with the concept of
coincidence, i.e. the existence of a multiple lattice common to the two misorientated
lattices.
The notion of coincidence, initially applied to twins, has been extended to any
grain boundary for materials with a simple structure by Ranganathan [12], then
by Bollmann [5]. The latter author has also introduced specific bicrystallographic
tools: DSC lattice, 0-lattices … by only considering the translation symmetry. This
approach permits to determine the equivalent nodes of two lattices but presents the
disadvantage to apply only for symmorphic structures with only one atom per unit
cell whereas numerous crystalline materials display a primitive unit cell with several
atoms (i.e. an atomic motif or atomic basis is attached at each node of the cell).
Then, the method requires to be completed by putting the atoms of the motif on the
lattice nodes. This approach proposed by Pond (1989) [13] to describe the interface
symmetry is detailed in the book of Sutton and Balluffi [8].
The research of the bicrystalline symmetry generalized to any grain boundary
in any material whatever its structure (even to any interface between two phases)
1.2 Bicrystallography 9
requires to take into account the crystal space groups that, generally, are not the
direct product of the orientation symmetry by the translation symmetry. We first
present this methodology in order to show the extension of the geometrical approach
to any interface. Then, we enter more in details in the Bollmann formalism based on
the notion of lattices [5, 6] that has been largely used to describe the grain boundary
structures.
1.2.1 General Methodology: From the Dichromatic Complex

to the Grain Boundary
This methodology is built by directly introducing the symmetry of the system formed
by the two interpenetrated crystals named dichromatic complex. The following steps
concern the bicrystal creation obtained by cutting the dichromatic complex by a
plane, then the research of the symmetries that are preserved in the bicrystal. Table 1.2
summarises the different steps to describe the bicrystalline symmetry.
Table 1.2 The different steps DICHROMATIC COMPLEX

to describe the bicrystalline
symmetry Grain boundary plane
IDEAL BICRYSTAL SYMMETRY
Atomic relaxations
REAL BICRYSTAL SYMMETRY
1.2.1.1 Dichromatic Complex
The dichromatic complex is formed by the crystals I and II misorientated and

interpenetrated (dichromatic is used because each crystal is defined by a colour,
white/black for example). This complex displays two types of symmetry elements:
elements of the first type leave the two crystals invariant and are obtained by the
intersections of the two crystal space groups:
H = GI ∩ GII (1.9)
This symmetry named coincident links to each other the sites of the same colour.
The second type of symmetry named anti-symmetry exchanges the crystalline
sites. The anti-symmetry elements link the sites of one colour (one crystal) to the
sites of the other colour (other crystal); they are obtained by the intersection of the
co-set αG of all the operations transforming crystal I in crystal II with the co-set
Gα−1 that corresponds to the inverse transformations:
H = α G ∩ G α−1 (1.10)
Finally, the symmetry group S of the dichromatic complex is given by the union
of the two intersection groups [7]:
S = H ∪ H (1.11)
We show that H = εH with ε, any symmetry element that exchanges the crystals.
The dichromatic complex for a grain boundary in a diamond cubic structure is
represented on Fig. 1.2. The anti-symmetry elements are indicated by the symbol (‘ ).
The consideration of the atomic motif yields a loss of symmetry elements with respect
to the dichromatic pattern (see Sect. 1.2.2) that only takes into account the symmetry
group of the crystalline lattice.
Fig. 1.2 Projections in the (001) plane of the dichromatic pattern and complex of a 36.9◦ [001]
grain boundary; the symbols ◦ and • correspond to the nodes of lattice I and II, respectively.
The different symbol sizes refer to nodes at different positions along the direction of projection.
a Dichromatic pattern (here equivalent to the CSL lattice) formed by two interpenetrating f.c.c.
lattices; b dichromatic complex formed by two interpenetrating diamond cubic crystals (now the
symbols represent atomic sites). The fine mesh corresponds to the DSC lattice in (a) and (b) (see
Sect. 1.2.3)
Pond has shown how the different ways to break the symmetry of the dichromatic
complex allow us to describe all the interfacial defects [13].
The knowledge of the symmetry group of the dichromatic complex brings very
useful physical information on the grain boundary properties [14]. The principle
of symmetry dictated extreme explains in what extent the symmetry may help to
detect special properties. The energy functional displays extrema depending on
the symmetry; these extremes are not necessarily minima, but their identification
allows us to extract the possible values of the physical parameters that may lead to
a minimum. As for any rule, exceptions to this principle may exist; in particular, we
will see later that grain boundary energy minima may be associated to non-symmetric
grain boundaries.
The dichromatic complex serves as a precursor for the atomistic simulation of the
grain boundary structure. The energy minimization leads to the determination of the
rigid body translation and of the local relaxations that allow specifying the atomic
structure adopted by a grain boundary (see Chap. 3).
1.2.1.2 Ideal Bicrystal
A bicrystal is formed by cutting the dichromatic complex by a plane, then by rejecting

the atoms of crystal I on one side, and the atoms of crystal II in the other side of
that plane. The symmetry operations of the ideal (non relaxed) bicrystal are those
of the dichromatic complex that leave invariant the normal to the grain boundary
plane. The grain boundary plane may always be chosen such as it does not contain
any translation vector of the dichromatic complex (random plane), even in the case
where a high 3D symmetry exists.
1.2.1.3 Real Bicrystal: Real Grain Boundary
Practically, three types of relaxation may occur: a rigid body translation τ of one
crystal with respect to the other, a displacement d of the grain boundary plane along
its normal (Table 1.1) and, finally, local individual atomic relaxations. These collective
or individual relaxations depend on the interatomic forces; they yield a decrease of
the grain boundary free enthalpy.
The rigid body translation has two components in the grain boundary plane τ 1
and τ 2 , implying the existence of a 2D periodicity in that plane. The intensities
of these in-plane vectors τ// must be inferior to one period in order to give rise
to the formation of different structures for a same grain boundary. All the possible
in-plane vectors not equivalent by adding a bicrystal translation vector form the cell of
non-identical displacements (c.n.i.d.) (Fig. 1.3) [15]. The c.n.i.d. concept is strongly
linked to the definition of the γ -surface, which represents the energy of the grain
boundary as a function of the in-plane rigid body translations (see Sect. 6.3.1). These
two notions, c.n.i.d. and γ -surface, will take a particular importance for analysing
the stress relaxation processes of intergranular dislocations (see Chap. 9).
The third component τ 3 or τ ⊥ of the rigid body translation is perpendicular to
the grain boundary plane and corresponds to a pure expansion (Table 1.1). In the
Bollmann approach [5, 6], the rigid body translation is considered as null (τ = 0).
Fig. 1.3 Cell of non-identical

displacements (c.n.i.d.) in a
bi-dimensional lattice
The last microscopic degree of freedom, the displacement d of the grain boundary
plane is distinct from the rigid expansion τ ⊥ only for grain boundaries in crystals
with more than one atom by primitive unit cell. In that cases, the grain boundary
plane displacement between two parallel atomic planes leads to the formation of
different “c.n.i.d.” for a given gain boundary.
Finally, local atomic relaxations that vary from one atom to the other may be
necessary to minimize the grain boundary free energy. They occur at the close vicinity
of the grain boundary whereas rigid translations extend far from the interface. The
individual atomic relaxations further reduce the bicrystalline symmetry.
1.2.2 Dichromatic Pattern: Coincidence Site Lattice (CSL)
The dichromatic pattern is obtained by simply replacing the crystal space groups by
the lattice space groups in relations (1.9) to (1.11). We devote a special paragraph
to this pattern at the basis of the most used method for studying grain boundaries in
cubic structures, and even in non-cubic provided approximations.
The Coincidence Site Lattice (CSL) is a particular case of dichromatic pattern
(Fig. 1.2) introduced in the Bollmann approach [5, 6] where only the translation sub-
groups TI and TII of the symmetry groups are taken into account. This 1D, 2D or 3D
lattice corresponds to the intersection, if not empty, of these two sub-groups:
CSL = I = TI ∩ TII (1.12)
The CSL lattice may also be defined as the smallest lattice included in the lattices
of the two crystals. Let us remark that there are no common sites (i.e. atomic sites
even empty) between two lattices but only common nodes, i.e. mathematical points
of the lattices. It would have been wiser to call the CSL coincidence lattice (C.L.).
However, we continue to use in the book the original and devoted terminology CSL.
The CSL exists in 3 dimensions if all the elements of the transformation matrix
A = RD which relates the two crystals are rational, R and D being a rotation matrix
and a deformation matrix, respectively [16]. The CSL is characterised by an integer
named the coincidence index:
Coincidence unit cell volume 1
= = (1.13)
Crystal primitive unit cell volume ρ
ρ is the inverse of the density of common nodes between the two lattices in the CSL
lattice.
For cubic lattices, the coincidence indices are always odd and are related to the
misorientation angle θ and axis uvw by the following expressions [12]:
2
tg(θ/2) = mλ /n and = [n2 + m2 λ]α (1.14)
√
with m and n two prime integers: λ = u 2 + v 2 + w 2 and α generally equal to
1 or 1/2.
Furthermore, the index is linked to the (hkl) grain boundary plane indices:

= n h 2 + k 2 + l 2 with n = 1 or 1/2 (1.15)
The coincidence index varies discontinuously with the misorientation angle θ

around a given axis. Table 1.3 presents an example of these variations for grain
boundaries around 110 in cubic materials; these boundaries will often serve as
structure models in the rest of the book.
Table 1.3 List of certain θ ◦ 110

values of the coincidence 0 1
index for misorientations
26.53 19
less than 70.53◦ around 110
in cubic materials 31.59 27
38.94 9
50.48 11
58.99 33
70.53 3
The evolution of the grain boundary energy of the symmetrical tilt grain boundaries
around 110 versus the misorientation θ clearly shows that there is no reciprocal
relationship between the value and the energy (Fig. 1.4) [17]. The energy of the
= 9 grain boundary is higher than that of = 11; furthermore, for a same
value, the energy depends on the grain boundary plane: very low for = 3 {111},
it takes a high value for = 3 {211}, akin to that of general grain boundaries (for
the terms in italic applied to the grain boundary type, see Sect. 1.3.1).
Fig. 1.4 Evolution of the

grain boundary energy (in
mJ m−2 ) calculated at 0 K for
symmetrical grain boundaries
around 110 in function
of the misorientation θ in
aluminium [17]
It is always possible to define exact coincidence relationships between two crystals

as so far as the rotation between them are expressed by a rational matrix. For cubic
materials, this condition only implies a rational rotation axis. Tables giving all the
coincidence relationships for the cubic systems have been established by Warrington
[18] and Grimmer [19]. A list of these relations until = 101 has been reported by
Mykura [10].
The coincidence lattice may be reduced to one or two dimensions, whatever
the crystal symmetry. Besides, in the structures of lower symmetry than cubic, the
existence of coincidence relationships i.e. the determination of multiple cells com-
mon to the two crystals often requires an additional deformation. In these structures,
the rotation alone cannot give rise to exact coincidences but only to approximate coin-
cidence (or near coincidence). The notions of reduced coincidence and approximate
coincidence will be approached later (see Sect. 1.2.4.3).
1.2.3 Translation Lattice of the Bicrystal (DSC Lattice)
Starting with two interpenetrated lattices in coincidence relationship, let us trans-

late randomly one of them; it is then impossible to find the multiple common cell
characteristic of the original coincidence. However, for certain translation vectors,
the common cell is reproduced with simply a displacement of its origin. The set of
all the translation vectors that possess this property constitutes the DSC lattice for
Displacement Shift Complete, according to the original denomination by Bollmann,
then in a more explicit manner, Displacement Symmetry Conserving according to
Pond and Bollmann [20].
Mathematically, the DSC lattice results from the union of the translation sub-
groups of the two crystalline lattices:
DSC = U = TI ∪ TII (1.16)
It may also be defined as the largest lattice containing the two crystal lattices.
It is formed by the linear combinations of all the unit translation vectors that relate
one node of crystal I to one node of crystal II, the two crystals being in coincidence
relationship.
The DSC vectors are most often expressed on a matrix form, in a consistent
manner such as the cross-product of two of its elements gives the third (1.17):
=3 = 11
−1 2 2 3 −7 2

a/6 2 −1 2 a/22−2 1 6
1 (1.17)
1 −2 −3 −4 −2
↓ ↓ ↓ ↓ ↓ ↓
b1 b2 b3 b1 b2 b3
Fig. 1.5 CSL lattice (full lines) and DSC lattice (dotted lines) for a = 11.50◦ 48 110 grain
boundary in a f.c.c. material
The CSL and the DSC lattices corresponding to = 11 in the materials of cubic
symmetry are given on Fig. 1.5 with their basic vectors projected in the (101) plane:
b1 = a/22 [3 − 2 − 3]I and b3 = a/11 [13 − 1]I . This construction will be used
later for the description of grain boundary linear defects (see Chap. 8).
The intensities of the elemental DSC vectors decrease when the coincidence index
increases and become extremely low when tends to infinity. However, when
the coincidence is reduced to two (or one) dimensions, these vectors may preserve
a non-negligible length when increases. In particular, if the coincidence only
exists in one dimension, two vectors tend to zero but the third one, almost parallel
to the rotation axis, may stay approximately constant even when tends to infinity.
The consideration of DSC vectors (not necessarily elemental) takes all its impor-
tance in the description in terms of dislocations of high-angle grain boundaries near
coincidence misorientations (see Sect. 2.3.2).
The DSC lattices have been determined, simultaneously to the exact and rational
coincidence lattices, for different cubic systems, and for different parametric ratios
in the cases of non-cubic systems [6].
1.2.4 Extension of the Coincidence Notion
1.2.4.1 Planar or Bi-dimensional (2D) Coincidence
The grain boundary being generally a 2D defect, only the periodicity in the grain
boundary plane has a physical meaning. Moreover, when the period perpendicular to
the grain boundary plane largely increases, the coincidence index tends to infinity
and only a bi-dimensional coincidence lattice may exist.
A complementary notion must be introduced: it is the density of common sites
in the grain boundary plane that strongly depends on the plane inclination [21]. The
coincidence planar density is a rational fraction of if the interface is commensu-
rable; it is equal to 1 when any node in the grain boundary plane is a common node
(the grain boundary plane is then a mirror plane or twin plane). For a = 9 grain
boundary, may take the values 1, 1/3, 1/9 and 0 (0 if there is no common node
in the grain boundary plane). The bi-dimensional coincidence is defined by a planar
coincidence index:
σ = 1/ (1.18)
σ is also defined in the grain boundary plane as the ratio of the area of the smallest
unit cell common to both crystals over the area of the crystal primitive unit cell.
At first, we can suggest that the grain boundary plane position that strongly
influences the planar coincidence must have important consequences on the grain
boundary properties. However, the physical relevance of this coincidence for pre-
dicting energy is still controversial. In particular, we must distinguish between the
CSL dense planes (planar density = 1) that most often possess high Miller indices
and the crystal dense planes. Only the twin = 3 in the f.c.c. materials displays a
common {111} plane that is simultaneously the dense plane in the CSL and in the
crystal. Otherwise, the preferential plane configuration for any other grain boundary
is questionable: dense plane of the crystal or dense plane of the CSL? The planar coin-
cidence density is often calculated by assuming that a curved surface may always be
decomposed in planar facets and that a random surface may always be approximated
to the neighbouring surface with low indices.
An exact bi-dimensional coincidence may theoretically exist in cubic systems
if the rotation axis is not rational. It mostly occurs in non-cubic systems that do
not present a 3D rational coincidence, for instance in the case of the (1–102) plane
of the rhombohedral twin in alumina. Most often, in these systems, there are only
approximate bi-dimensional coincidences.
1.2.4.2 One-Dimensional Coincidence (1D) or “Plane Matching” Model
Initially proposed by Pumphrey [22], the “Plane Matching” model is equivalent to a

1D coincidence around an axis in the grain boundary plane, also named Coincident
Axial Direction (CAD) [23]. On either side of the grain boundary, a good fit exists
Fig. 1.6 Schematic drawing explaining the one-dimensional coincidence model or “plane match-
ing”: a misorientated traces of crystals I and II planes of same indices in the grain boundary plane
(fine lines) and the resulting pseudo-Moiré fringes equidistant from d (bold lines); b the spacing, in
the grain boundary plane, of the traces of two plane families in each crystal differs
between two families of planes with the same indices and displaying a relatively high
atomic density. The traces of these planes on the grain boundary plane correspond
to atomic rows slightly inclined relative to each other and/or with very close spacing
(Fig. 1.6). These rows have a high atomic density; therefore, relaxations necessarily
occur depending on the grain boundary periodicity. The periodic stress fields are
observed by transmission electron microscopy as a pseudo-Moiré phenomenon, the
opaque lines corresponding to high atomic disorder regions in the grain boundary
plane. This model is equivalent to that of semi-coherency for interphase interfaces.
Such a phenomenon has been observed in several grain boundaries displaying a
good fit for the {200}, {220} and {111} planes in the f.c.c. materials and only for
the {110} planes in the b.c.c. materials. It concerns grain boundaries deviated from
any tri-dimensional coincidence. Thus, the periodicity of a θ = 42◦ [001] tilt grain
boundary with an additional 4◦ twist angle in gold has been explained by a good fit
of the {200} planes (Fig. 1.7) [24].
On the basis of the CSL approach, the previous model is equivalent to a coin-
cidence around an axis in the grain boundary plane when →∞ and σ →∞,
simultaneously. In that case, the intensities of the b1 and b2 DSC vectors located
in a grain boundary normal to the fundamental rotation axis cancel, but the b3 vec-
tor almost parallel to this rotation axis is independent of the value and preserves
a length close to an interplanar distance. Thus, the physical reality of the model
only depends on the fundamental rotation axis indices. This coincidence has a great
interest in polycrystalline materials where its probability to occur is larger than that
of tri-dimensional coincidence, particularly for textured materials.
1.2.4.3 Exact Coincidence and Approximate Coincidence
Most works on grain boundaries are very cautious in approaching the notion of
approximate coincidence even though numerous materials display a non-cubic
Fig. 1.7 Geometry of a mixed

grain boundary with a major
tilt component θ = 42◦ [001]
and a small twist deviation α
showing a good fit between
the {200} planes of the two
adjacent crystals
structure. To reduce this lack of information and because this type of coincidence
implies specific dislocation descriptions of grain boundaries in hexagonal (zinc,
titanium, cobalt …) and rhombohedral (hematite, alumina …) materials, we now
especially deal in details with the questions raised up by the coincidence notion.
When two hexagonal (rhombohedral) lattices are misorientated and interpen-
etrated, only the rotations around the 6-fold axis (threefold for rhombohedral
symmetry) may give rise to 3D coincidence lattices, whatever the c/a ratio of the
material. These common rotations, relevant for all hexagonal materials, are named
exact coincidence (by analogy with the coincidences for cubic symmetry). Rational
matrices in a coordinate system based on the crystal I (or crystal II) lattice express
these coincidences.
For any other rotation axis, there is no rotation angle that enables to exactly super-
impose the two primitive unit cells of crystals II and I. However, for certain specific
rotations, two multiple cells M1 and M2 may be approximately superimposed. The
obtained quasi-common cell has a multiplicity (or coincidence) index as previously
defined for cubic systems. Two indices 1 and 2 are then required to characterize
the coincidence. So-called rational coincidences or approximate coincidences are
associated to these specific rotations. A rational matrix expresses an exact coinci-
dence, but in the case of a hexagonal material, the rotation matrix depends on the c/a
ratio. Only certain values of this ratio corresponding to an ideal fictitious material may
give rise to an exact coincidence. For a real material whose c/a ratio is close that an
ideal material, first the specific rotations yield the multiple cells in near coincidence,
then a deformation is necessary to bring a perfect matching. The transformation that
relates the M1 and M2 cells consists in a rotation followed by a deformation:
A = R· D (1.19)
The way to derive the CSL lattices corresponding to coincidence grain boundaries
in different crystalline structures has been explained by Warrington, Grimmer and
Bonnet [25, 26]. Coincidence tables have been established for the hexagonal mate-
rials taking into account different c/a ratios [27, 28]. A more complete approach to
determine the coincidence relationships has been developed for hexagonal [29] then
for rhombohedral crystals [30].
Figure 1.8 illustrates the case of an approximate bi-dimensional coincidence of
an asymmetrical grain boundary in alumina, the (0001)I plane being parallel to the
(1011)II plane. Two multiple common cells may be selected [31].
Fig. 1.8 Schematic representation of two possible approximate coincidence cells in the
(0001)I //(1011)II grain boundary plane in alumina. The coincidence indices in the plane are:
σI = 87 σII = 26 for the cell built on u and u (full lines), σI = 87 σII = 26 for the cell built on
v and v (dotted lines) [31]
• The cell M1 built on the vectors u and u corresponds to high coincidence indices:
σI = 87, σII = 26, but the misfits between the basic vectors are small: δ = 0.014
for the uI //uII vectors and δ = 0.002 for the u I //u II vectors.
• The cell M2 built on the vectors v and v displays lower coincidence indices with
respect to the previous ones: σI = 39, σII = 12, but the misfits are larger: δ = 0.069
for the v I //v II vectors and δ = 0.039 for the v I //v II vectors.
The main objective to enter more in details in these complex notions of coincidence
is to point out the fact that the transformation yielding a coincidence relationship
may often be the product of a rotation by a deformation (expansion, contraction,
shear). A same grain boundary may be approximated to different neighbouring exact
coincidences provided different deformations. The deformation may be chosen as
small as possible at the expense of a high value. The question then arises: what
is the closest description of reality, a large coincidence index, but small deformation
or vice versa? Only the determination of the grain boundary structure in terms of
dislocations may answer the questions.
1.2.5 Generalization of the Coincidence: O-Lattice and O2-Lattice
The Bollmann approach is at the basis of the grain boundary structure description
in terms of dislocations. According to this approach, any grain boundary that devi-
ates from an invariant configuration, single crystal or coincidence, has a tendency
to preserve this configuration, either by a crystal lattice translation, or by a DSC
lattice translation. Translation defects then occur to build the grain boundary struc-
ture: they are structural or intrinsic dislocations (see Chap. 2). Their description
requires introducing a generalised coincidence concept, the O-lattice [5, 6]. This
concept is included in the standard crystallography; its essential tool is the group
theory.
1.2.5.1 O-lattice
The CSL lattice only exists for certain grain boundary misorientations. Considering
that the single crystal invariant must be preserved regardless the relative orientation
of the two crystals, Bollmann extends the coincidence concept with the introduction
of the O-lattice [5].
On the contrary of the CSL lattice, the O-lattice is actually a coincidence site
lattice. It is the locus of all points in good fit between the regions in bad fit between
the two neighbouring crystals. The coincidence sites named O-points may have
internal coordinates in the unit cells of crystals I and II. Consequently and unlike
the CSL lattice (Table 1.3), the O-lattice continuously varies with the rotation angle
between the crystals. When the CSL lattice exists, it is a sub-lattice of the O-lattice
(Fig. 1.9). The term O-lattice is that each point O can be considered as origin of the
rotation that relates the two crystals.
Fig. 1.9 O-lattice for a

= 5, 36.87◦ [001] (310)
grain boundary in a diamond
cubic structure. The crystal
cells are drawn in fine lines.
The O-locus is formed of
planes (traces in full lines) and
of lines (dotted lines). The
positions, along the direction
of projection, of these traces
and lines are indicated on the
sides of the CSL cell (ABCD)
A point x belongs to the O-lattice if it remains invariant when the transformation

that relates crystal I to crystal II is applied (Eq. 1.1). Then, the fundamental equation
of the O-lattice is:
x 0 = (A/τ )GI x 0 (1.20)
In the Bollmann formalism (G = T and τ = 0), Eq. 1.20 becomes:
x 0 = (A)GI x 0 = Ax 0 + B m (1.21)
B m is a translation vector of crystal I (the subscript m is for matrix).

Then the fundamental equation of the O-lattice, relevant whatever the symmetry
of the material, takes the form:
(I − A−1 )x O = B m (1.22)
A may be a single rotation or a deformation or the product of a rotation by a deforma-

tion, I is the identity matrix. If A is an expansion or a shear, the O-lattice is constituted
of points or planes, respectively.
For grain boundaries in f.c.c. or b.c.c. materials, Eq. (1.22) becomes:
(I − R−1 )x O = B m (1.23)
R is the rotation matrix.

This is the form that we will use later, and in this case, the O-lattice consists of
O-lines.
The plurality of the descriptions (equivalent rotations) describing the same

relationship between the crystals leads to the existence of several O-lattices. This
ambiguity will be discussed when we describe the grain boundaries in terms of
dislocations.
1.2.5.2 O2-Lattice
Most often a high-angle grain boundary presents an angular deviation θ around

a [u v w ] axis with respect to its closest coincidence misorientation defined by θ
[uvw]. By considering that the second invariant, i.e. the coincidence lattice, tends
to be preserved, Bollmann introduces the O2-lattice [5]. In the same way that the
O-lattice defines the invariant sites between two misorientated translation lattices of
the crystals, the O2-lattice specifies the invariant sites between two DSC translation
lattices of the bicrystal rotated from θ around [u v w ].
In the cubic structures, an equation similar to (1.23) defines the O2-lattice:
(I − D−1 )x O2 = bDSC(I) (1.24)
D is the deviation matrix between the real rotation between the crystals and the
nearest exact coincidence rotation, bDSC is a DSC translation vector, not necessarily
elemental. The O2-points which satisfy this equation are the invariant points between
the DSCI of lattice I, chosen as reference, and the DSCII lattice obtained from DSCI
by a transformation D = Rréel .Rcoin−1 .
The 02-lattice always exists and varies continuously with the angular deviation
θ to the coincidence misorientation. On the contrary of the O-lattice that differs
according to the choice of the relative rotation among all the equivalent rotations, the
O2-lattice is unique in that the matrix D is unique. Indeed, to determine D we always
choose the nearest coincidence rotation from the real rotation and the deviation
between these two rotations is always the same.
For grain boundaries in non-cubic structures, the deviation matrix must be define
with respect to the approximate coincidence; it is then the product of a rotation by a
deformation.
1.2.6 Interest and Limit of the Bollmann Approach
The Bollmann formalism only predicts the preferential orientation relationships

between the crystals but not the preferential grain boundary planes that impose
the grain boundary structures in terms of dislocations and the interfacial energy. A
preferential orientation relationship does not imply a particular grain boundary plane
and reciprocally. Several criteria have been proposed to predict the grain boundary
plane; they are presented later in this chapter along with the intergranular energy.
The O and O2 lattices are used to describe the grain boundaries in terms of intrinsic
dislocations. These dislocations are geometrically necessary, they must be located
midway between the good fit regions predicted by the O locus and the O2 locus.
But the physical dislocations, really present in the grain boundary plane, may or may
not correspond to the geometrical dislocations, according to the local relaxations that
occur to minimize the intergranular energy and that depend on the atomic interactions.
Despite these shortcomings, a good knowledge of the bicrystallography, the
Bollmann model being just a particular aspect, is a necessary platform whatever
the way to approach the grain boundary structure, in terms of dislocations or in
terms of atomic arrangements.
1.3 Different Types of Grain Boundaries: Terminology
We do not aim here a classification of grain boundaries, that will be addressed after
the presentation of all the structural models; we just want to familiarize readers with
a terminology specific to grain boundaries and to raise some inherent ambiguities in
this domain.
In the preface, we have identified the homo-phase interfaces to the grain boundaries.
In fact, interfaces between two identical crystals also include order domain walls and
stacking faults obtained by reducing the interface operation to τ in relation (1.1).
The terminology presented here only deals with the grain boundaries. Let us first
underline that there is no satisfying denomination of grain boundaries because it
depends on the scale they are observed, i.e. of the observation tool. A distinction
between grain boundaries at the mesoscopic or macroscopic scale is based on the
mean geometrical parameters as they are measured by X-ray diffraction, backscat-
tered electron diffraction or by conventional transmission electron microscopy. For
each grain boundary, its macroscopic degrees of freedom are determined. A dis-
tinction at the atomic scale rests on the knowledge of the microscopic geometrical
parameters, determined by high-resolution transmission electron microscopy and/or
obtained by simulation. The observations of the grain boundaries at different scales
result in terminological ambiguities.
1.3.1 Terminology Based on the Macroscopic Parameters
First, we distinguish between low-angle grain boundaries, generally θ < 15◦ (this
limit may slightly differ according to the materials) and the high-angle grain
boundaries. The sub-grain boundaries are low-angle grain boundaries, but this term
is reserved for dislocations walls that form within the grains under the effect of a
work hardening followed by a heat treatment. Otherwise, the term low-angle grain
boundary indicates an interface between two slightly misorientated grains resulting
from the material elaboration (solidification, sintering …).
Among the high-angle grain boundaries, those that correspond to a coincidence

relationship are named coincidence grain boundaries. If they are slightly deviated
from this preferential relation, they are near-coincident grain boundaries; the term
general grain boundary is usually reserved for a grain boundary far from any low
index coincidence. This term, widely used, does not imply the periodicity nor the
energy of the grain boundary; indeed, it is based only on the coincidence notion and
does not take into account the grain boundary plane.
A grain boundary is said to be a tilt grain boundary if the [uvw] rotation axis
is contained within the boundary plane and a twist grain boundary if this axis is
perpendicular to the plane; any other position of the [uvw] axis with respect to the
boundary plane leads to a mixed tilt/twist grain boundary. To better highlight the
degrees of freedom of these different grain boundaries, we refer to the parameter
definitions proposed by D. Wolf (see Sect. 1.1) [9]. A tilt grain boundary is perfectly
defined by the two vectors nI and nII normal to its plane; the rotation axis and the
rotation angle are the zone axis and the angle between the two planes, respectively.
Then, to describe a tilt grain boundary, we generally need four degrees of freedom.
A twist grain boundary possesses three degrees of freedom: two for the normal to its
plane (the rotation axis) and one for the tilt angle θ . Only the mixed grain boundaries
possess five degrees of freedom, they are often assimilated to really general grain
boundaries (see later discussion). The distinction tilt/twist is ambiguous: for instance
in a centro-symmetric structure, any (hkl) symmetrical tilt grain boundary may be
described as a twist grain boundary with a 180◦ rotation angle around an axis normal
to (hkl). An asymmetrical twist grain boundary may be seen as an asymmetrical tilt
grain boundary that underwent an additional rotation around the normal to its plane.
This ambiguity is inherent to any grain boundary geometrical definition.
A tilt grain boundary is said asymmetrical when its plane displays different Miller
indices in the two adjacent crystals {hkl}I = {hkl}II ; in that case, the grain boundary
really possesses four degrees of freedom. If the boundary plane has the same indices
in crystals I and II, {hkl}I ≡ {hkl}II , the grain boundary is said symmetrical, the
number of degrees of freedom required for its description is reduced to two.
We later adopt the following designation: a symmetrical tilt grain boundary is
defined by θ uvw{hkl} or if it is coincident by {hkl}; If it is asymmetrical, the
plane indices in the two crystals are mentioned {hkl}I //{hkl}II . A twist grain bound-
ary is defined by θ uvw or {hkl}.
In a symmetrical tilt gain boundary, the median plane is a mirror planes for each
crystals, provided they are centro-symmetric. The median plane must not be confused
with the mean {hkl} plane, the later being simply rotated by θ/2 with respect to the
mirror plane. In the cases of the symmetrical tilt grain boundaries around [1–10],
the (110) median plane is confused with the mean grain boundary plane only for the
extreme value of = 1 (single crystal). As soon as a θ rotation occurs between the
crystals, one crystal is rotated of +θ/2, the other of −θ/2 with respect to the median
plane. The grain boundary plane is then obtained by superimposing the two planes
of same indices initially located at the (110) position. We obtain a series of grain
boundaries whose the plane indices take the same (hkl) form: (331), (221), (332) …
(111), with increasing θ values; the corresponding coincident grain boundaries are
1.3 Different Types of Grain Boundaries: Terminology 25
= 19, = 9, = 11, = 3. In the cubic crystals, only {100} and {110} may
be median planes; this reduced the possible tilt axis indices for symmetrical grain
boundaries.
Until now, the symmetrical/asymmetrical distinction only concerns the mean grain
boundary plane defined by its macroscopic parameters and does not take into account
the faceting phenomenon that often occurs in real grain boundaries.
A symmetrical grain boundary is said commensurable or incommensurable
depending on whether a periodicity exists or not in the grain boundary plane, in
one direction at least. In the cubic symmetry, a grain boundary is incommensurable
if the ratio (h 21 + k12 + l12 )/(h 22 + k22 + l22 ) = A1 /A2 is irrational; h i ki et li being
the Miller indices of the boundary plane in the crystal i; as an example, a grain
boundary with the (111)1 plane parallel to (331)2 is incommensurable. If the grain
boundary is commensurable, the coincidence index equal to (A1 /A2 )1/2 . In the
materials of lower symmetry, the parameter ratios must be taken into account (c/a
in the hexagonal system).
1.3.2 Terminology Based on the Microscopic Parameters
Although they are called microscopic parameters (see Sect. 1.1.1), the dimensions
of these parameters are inferior to the micrometre, being inferior to a crystalline
lattice period. At this scale, a grain boundary is symmetrical, not only if the grain
boundary plane indices are equivalent in both crystals but, in addition, if no rigid
body translation from one crystal with respect to the other yields a symmetry loss
of the atomic positions on both sides of the grain boundary plane. A grain boundary
considered as symmetrical at the microscopic scale may be asymmetrical at the
atomic scale. Thus, it is important to emphasize the observation scale when speaking
of the grain boundary symmetry.
The commensurability at the atomic scale is related to the cell of non-identical
displacements c.n.i.d. (see Sect. 1.2.1). A commensurable grain boundary presents a
planar periodicity, a c.n.i.d. cell may then be defined, the dimensions of which being
linked to the coincidence site planar density. The area of the c.n.i.d. is reduced to
zero for an incommensurable grain boundary.
1.3.3 Practical Distinction Between Grain Boundaries
It seems useful to introduce here a practical distinction between grain boundaries

depending on the degree of knowledge of the parameters that can be achieved. The
grain boundaries studied by simulation are exact in that their macroscopic parame-
ters are perfectly defined, the simulation aiming, inter alia, to the determination of
their microscopic parameters. We also may consider that the symmetrical tilt grain
boundaries in semiconductor obtained by bicrystal growth are quasi-exact in so far
as the obtained macroscopic parameters very slightly deviate from the expected
ones; this is due in particular to the great purity of the materials. On the contrary, in
cases of metals, the grain boundaries manufactured by various bicrystal elaboration
techniques (unidirectional solidification or diffusion bonding) more or less deviate
from the expected parameters, with an exception for the coherent twin boundary. It is
thus necessary to verify the misorientation and the grain boundary plane orientation
after bicrystal elaboration. Furthermore, these characteristics generally vary with the
position of the observed grain boundary region along the bicrystal. However, even
if the macroscopic parameters of a grain boundary in a bicrystal differ from the
expected ones, they may be accurately determined, as well as its tilt/twist character.
The last type of grain boundaries, likely the most important from a practical
point-of-view, is that of the real grain boundaries in polycrystals. Their geometrical
parameters are uncontrollable and are generally determined with less accuracy than
those of the bicrystals. Furthermore, these grain boundaries are rarely observable by
high-resolution transmission electron microscopy.
A question must now be raised: how to attribute to a grain boundary in a polycrystal
a tilt or twist (or mixed) character? This character is fixed for exact and manufac-
tured grain boundaries, theoretically in one case, experimentally in the other. On the
contrary, the genesis of a real grain boundary in a polycrystal is unknown (apart from
the case of columnar growth on a substrate). Practically, one of the misorientation
relationships among all the equivalent descriptions is experimentally determined;
the disorientation is then deduced. The grain boundary plane may or may not be
indexed depending of the observation mode (this indication is often missing in the
polycrystal studies). Even if the grain boundary plane is known, what misorientation
description to use for concluding that the rotation axis is or not in the grain boundary
plane? In the macroscopic studies, the misorientation and the grain boundary plane
parameters are determined in an experimental coordinate system (Euler angles for
example); Whereas the grain boundary plane is unique, no one of the misorientation
descriptions enables to decide of the grain boundary character (tilt, twist or mixed)
without observing the grain boundary content in terms of dislocations (if no its fine
structure). Indeed, any symmetrical tilt grain boundary may be considered as a twist
grain boundary, the rotation axis being that of the 180◦ equivalent description (see
Sect. 1.2). A = 3 {111} grain boundary may also be a 70.32◦ around 110 tilt grain
boundary or a 60◦ 111 twist grain boundary. For this particular boundary, the choice
of the description has no consequence on its intrinsic structure, the coherent twin
being free of dislocations. Otherwise, the choice of the geometrical description affects
the structure description in terms of dislocations. The following example enables to
clarify this effect: the geometrical parameters of a grain boundary as experimentally
determined are close to 63◦ 211 not far from a = 11 coincident grain boundary,
its plane is close {332} in each crystal; in this description, the grain boundary has
a mixed (tilt/twist) character. If we now consider the equivalent 180◦ 332 descrip-
tion, it is a twist grain boundary; finally, the choice of the 50.48◦ 110 description
confers to that grain boundary a fundamental tilt character. Only the knowledge of
the dislocation structure of this near coincidence grain boundary enables to specify
the real grain boundary character. The grain boundary responses to various stimuli
1.3 Different Types of Grain Boundaries: Terminology 27
(thermal, mechanical and others) will depend on the real structure, only. Thus, we
must be very cautious when establishing the grain boundary character distribution
in a polycrystal (see Chap. 11).
Until now, the knowledge of the grain boundary structures and on the relations
between structures and grain boundary properties mainly concerns symmetrical tilt
grain boundaries that only possess two degrees of freedom. Recently, some twist
grain boundaries (three degrees of freedom) and some asymmetrical and incommen-
surable grain boundaries (four degrees of freedom) have been experimentally stud-
ied by high-resolution transmission electron microscopy; their structures have been
simulated by using approximants with respect to the nearest rational grain bound-
aries. Otherwise, there is no (or very few) data on really general grain boundaries that
require five degrees of freedom to be described. A unique structure corresponding
to these five parameters seems unlike. Such grain boundaries can always be consti-
tuted by different regions with simple characteristics, as suggested by the faceting
phenomenon.
Note that there is another classification of the symmetrical and asymmetrical
grain boundaries; although only geometrical, this description is presented later, as
it requires for understanding the knowledge of the atomic grain boundary structure
(see Sect. 4.6.2).
In the overall conclusion of this chapter, we stress the necessity to know the
bicrystallography before to study grain boundaries, both experimentally and by
simulation. But, is the geometrical long-range order predicted by the crystallography
respected when we approach the grain boundary with other concepts? In particular,
is there an order of the stresses that necessarily develop in the grain boundary region
that is faulted compared to the perfect crystal? This question is at the core of Chap. 2.
References
1. W. Rosenhain, D.J. Ewen, J. Inst. Metals 8, 149 (1912)

2. R.R.L. Couling, R. Smoluchowski, J. Appl. Phys. 25, 1538 (1954)
3. N.F. Mott, Proc. Phys. Soc. 60, 3 (1948)
4. W.T. Read, W. Shockley, Phys. Rev. 78(195), 275 (1950)
5. W. Bollmann, Crystal Defects and Crystalline Interfaces (Springler, Berlin, 1970)
6. W. Bollmann (ed.), Crystal Lattices, Interfaces, Matrices: An Extension of Crystallography
(BoUman, Geneva, 1982)
7. D. Gratias, R. Portier, J. de Phys. C6–43, 15 (1982)
8. A.P. Sutton, R.W. Balluffi, Interfaces in Crystalline Materials (Clarendon Press, Oxford, 1995)
9. D. Wolf, S. Yip, Materials Interfaces (Chapman and Hall, London, 1992)
10. Mykura, in Grain Boundary Structure and Kinetics, ed. by R.W. Balluffi (ASM, Metals Park
Ohio, 1980), p. 445.
11. G. Friedel, Leçons de Cristallographie (Berger-Levrault, Paris, 1926), seconde édition,
(Blanchard, Paris, 1964)
12. S. Ranganathan, Acta Cryst. 21, 197 (1966)
13. R.C. Pond, in Dislocations, in Solids, ed. by F. Nabarro, vol. 8, (North Holland, Amsterdam,
1989), pp. 1–6
14. G. Kalonji, J. Cahn, J. de Phys. C6–43, 25 (1982)
15. V. Vitek, A.P. Sutton, D.A. Smith, R.C. Pond, in Grain Boundary Structure and kinetics, ed.
by R.W. Balluffi (ASM, Metals Park Ohio, 1980), p. 115
16. H. Grimmer, Acta Cryst. A32, 783 (1976)
17. G. Hasson, M. Biscondi, P. Lagarde, J. Levy, C. Goux, in The Nature and Behavior of Grain
Boundaries, ed. by H. Hu (Plenum Press, New York, 1972), p. 3
18. D.H. Warington, P. Bufalini, Scripta Met. 5, 771 (1971)
20. R.C. Pond, W. Bollmann, Phil. Trans. R. Soc. Lond. A292, 449 (1979)
21. D.G. Brandon, B. Ralph, S. Ranganathan, M.S. Wald, Acta Metall. 12, 813 (1964)
22. P.H. Pumphrey, Scripta Met. 6, 107 (1972)
23. L. Priester, Rev. de Phys. Appl. 15, 789 (1980)
24. R. Schindler, J.E. Clemans, R.W. Balluffi, Phys. Stat. Sol. A56, 749 (1979)
25. D.H. Warrington, J. Phys. coll. C-4 36, 87 (1975).
26. R. Bonnet, E. Cousineau, Acta Cryst. A33, 850 (1977)
27. G.L. Bleris, G. Nouet, S. Hagege, P. Delavignette, Acta Cryst. A38, 550 (1982)
28. P. Delavignette, J. Phys. Coll C-6, n◦ 12, tome 43, 1 (1982).
30. H. Grimmer, R. Bonnet, S. Lartigue, L. Priester, Phil. Mag. A61, 493 (1990)
31. S. Lartigue, L. Priester, J. de Phys. 49, C5–451 (1988)
Chapter 2
Mechanical Stress Order of Grain Boundaries
The mechanical aspect is particularly important for understanding the grain boundary
behaviour during plastic deformation or recrystallization, both processes involving
interactions between planar and linear defects.
Two types of mechanical approaches of the grain boundary description exist:
• The continuous Frank [1] and Bilby [2] approach describes any interface as a
surface dislocation characterized by a Burgers vector density or by a second
order surface dislocation tensor. This approach is analogous to the elegant the-
ory developed later by Kröner to account for the stresses, the energies and the
curvatures introduced in a material under internal stress sources [3, 4].
• The discrete approach of Bollmann [5, 6] considers the presence of dislocations
in the centres of the disturbed regions between the good crystalline regions. It
generalizes at any grain boundary the first description of an intergranular order
proposed by Read and Shockley [7]. It rests on the symmetry notions introduced
in Sect. 1.2.
2.1 Continuous Approach: The Frank-Bilby Equation
Following Frank who describes the dislocation distribution in a low-angle grain

boundary [1], Bilby introduces a procedure to deal with the dislocation distribution
in a surface delimiting two Crystals I and II with a distortion discontinuity between
them [2]:
εijk nl βIsj − εijk nl βII

sj = αks (2.1)
with αks the surface dislocation tensor, k and s refer to the Burgers vector and to the
dislocation line, respectively; n is the normal to the grain boundary plane pointing
from grain I to grain II and εijk is the permutation tensor (εiij = 0).

30 2 Mechanical Stress Order of Grain Boundaries
Crystal I Crystal II
SII
Y
X
SI
Fig. 2.1 Diagram explaining how lattices of crystals I and II are generated from a reference system
XYZ by the transformations SI and SII , respectively
For a 2D distribution, this approach is equivalent to the Kröner generalized 3D

approach in which the continuous distribution of dislocations is expressed by a dis-
location density tensor α related to the β distortion by [3, 4]:
curl β = α (2.2)
The expression (2.1) reformulated by Christian in matrix notation leads to the

equation known as the Frank-Bilby equation that gives the Burgers vector density B
necessary to realise the compatibility at the interface between the crystals I and II:
B = (S−1 −1
I − SII ) · X (2.3)
SI and SII are affine transformations that generate the lattices of crystals I and II
starting from a reference system (Fig. 2.1), X is a vector of relatively large intensity
in the grain boundary plane and B is the Burgers vector content of all the dislocations
that are crossed by the vector X. If this equation is satisfied, the grain boundary is
free of long-range stresses.
Equation (2.3) is relevant for any interface but we focus here on grain boundaries.
For grain boundaries in cubic materials, a simple relation links the SI and SII
lattices to the rotation R between the crystals: R = SII .S−1 I . Equation (2.3) then
becomes equivalent to the 0-lattice equation in which I is the identity matrix:
B = (I − R−1 )X (2.4)
In Eqs. (2.3) and (2.4), B is not considered as the sum of discrete Burgers vectors
belonging to the crystal translation lattice or to the DSC lattice. Such a discretization
depends on the relaxation processes and is meaningless when the dislocation density
is very high (Fig. 2.2).
B may be determined by using the Frank circuit drawn around the grain boundary,
equivalent to the Burgers circuit within the crystal (Fig. 2.2). A vector X = X II ,
traced in the grain boundary plane and indexed in lattice II, is obtained by applying
2.1 Continuous Approach: The Frank-Bilby Equation 31
Fig. 2.2 Frank circuit: a

vector X I = X II is drawn in
lattice I, then surrounded by
a closed circuit starting and
finishing at the vector origin.
The same circuit is reproduced
around the grain boundary: it
starts at the extremity of X II
(vector in the grain boundary
plane obtained by the rotation
θ of the vector (X I ) passes
through the origin O and
terminates at the extremity
of X I . This circuit displays
a closure failure B given by
Eq. (2.4) by substituting X for
X II
the rotation R to a vector X I in lattice I:
X I = R−1 X II .
In the cases of symmetrical grain boundaries, if a median lattice between the two
crystal lattices is considered as the reference system, then the I and II lattices are
deduced from this reference by equal and opposite rotations (± θ/2) around a ρ axis;
expression (2.4) takes the well-known form of the Frank formula:
B = 2 sin(θ /2)(X ∧ ρ) (2.5)
And provided that X is perpendicular to the rotation axis:
B = 2 sin(θ/2)|X| (2.6)
The absence of long-range stresses around the interface may be modelled by the
use of two continuous distributions of dislocations with Burgers vector densities of
opposite sign that cancel. One distribution may be regarded as an array of stress-
generator dislocations or coherency dislocations, the other as an array of stress-
annihilator dislocations or anti-coherency dislocations.For a detailed description of
this approach, the reader may refer to the book by Sutton and Balluffi [8] and to the
articles by Olson and Cohen [9] and Bonnet [10, 11].
2.2 Discrete Approach: The Read and Shockley Model
Studies of edge dislocation walls in a crystal, later recognised as a low-angle sym-

metrical tilt grain boundaries, developed about 20 years before the explicit structural
model of these grain boundaries due to Read and Shockley [7]. This model links the
grain boundary misorientation angle θ to the spacing d between the edge disloca-
tions of Burgers vector b that form a periodic network in the grain boundary plane
(Fig. 2.3):
b
d= (2.7)
θ
Fig. 2.3 Scheme showing

a low angle symmetrical tilt
grain boundary according to
Read and Shockley. The tilt
axis is perpendicular to the
plane of the figure
The classical Read and Shockley formula gives the intergranular energy in func-
tion of the misorientation angle, in the low-angle limit:
γθ = γθ θ (A − lnθ ) (2.8)
γ0 = μb/4π(1 − ν) and A = b/2πr0 with μ, the shear modulus, ν the Poisson

coefficient and r0 the dislocation core radius.
Equation (2.8) is easily obtained by summing the elastic energies of the N
edge dislocations (of unit length) contained in a unit area of the grain boundary
(N = 1/d). This calculation may be done provided the dislocations are sufficiently
far from each other (low θ angle) and the extension of the stress fields is equal to the
distance d. This equation, initially established for θ < 15◦ , is fundamental. Indeed,
after having established their model, Read and Shockley predict its relevance for
high-angle grain boundaries formed by dislocations evenly spaced. In 1989, Wolf
proposed an empirical extrapolation of the Read and Shockley formula for high-angle
grain boundaries sharing the same mean plane, by substituting θ by sin θ [12].
In 1947, Lacombe and Beaujard [13] revealed the existence of sub-grain boundaries
in aluminium single crystals owing to the preferential etching of the crystal surface
on the form of etch pit alignments. But no relationship was established between etch
pits and dislocations. In 1950, Read and Shockley suggested a correlation between
2.2 Discrete Approach: The Read and Shockley Model 33
their model and the previous observations. It was only in 1953 that a proper verifica-
tion of the validity of the dislocation model for sub-grain boundaries was obtained
on a germanium bicrystal by Vogel et al. [14].
2.3 Bollmann’s Discrete Approach: Intrinsic Dislocations
The Bollmann approach generalizes the Read and Shockley model of intergranular
dislocations at any grain boundary. It rests on Eqs. (1.23) and (1.24) that define the
O-lattice and the 02-lattice. The term intrinsic, which comes from the Latin intrin-
secus, indicates that these dislocations are inherent in the grain boundary structure
that cannot exist in their absence; they actually are structural dislocations.
2.3.1 Primary Intrinsic Dislocations
In the 0-lattice equation (I − R−1 ) x O = B m , the B m vector may always be con-

sidered as the sum of Burgers vectors bm of crystal I (subscript m for matrix).
Equations (1.23) and (1.22) for non-cubic materials) may be seen as the quantified
form of the Frank-Bilby Eq. (2.4) for grain boundaries in which x 0 is a O-lattice
element. The different O elements are surrounded by Wigner-Seitz cell walls. The
intersections of the grain boundary plane with these walls form the b-net; they define
the lines of the geometrically necessary dislocations called primary intrinsic dislo-
cations (IGBD) localized in the regions of worse fit between the two crystals. The
term primary indicates that these dislocations account for the deviation from the first
invariant of the bicystal, i.e. the single crystal (θ = 0). The term secondary will be
used to explain the deviation from the second invariant, the CSL lattice.
The primary intrinsic dislocations possess a Burgers vector bm of the perfect
crystal dislocations and form a network with the same periodicity than the 0-lattice
(Fig. 2.4). Their spacing decreases when the misorientation angle increases according
to the formula derived from the Frank Eq. (2.3) where |X|= d:
bm
d= (2.9)
2sinθ/2
For low θ values, Eq. (2.9) is identical to the Read and Shockley expression (2.7).
The uniformly spaced distribution of the primary dislocations mathematically
predicted by Eq. (2.9) is generally not respected. Indeed, a spacing between two
dislocations must be physically compatible with the lattice period. Only the average
spacing measured on an arrangement of several dislocations meet this formula.
Let us illustrate the application of Eq. (2.4) to the prediction of the primary dislo-
cation content in a tilt grain boundary, then in a twist grain boundary, both displaying
a small misorientation θ = ε.
Fig. 2.4 Diagram showing the localization of the primary intrinsic dislocations in a low angle tilt
grain boundary. A, B and O are nodes of the O-lattice. PI and PII (QI and QII ) are homologous nodes
assimilated to one node. D represents a region of bad fit where the primary dislocation is localized
• For a tilt grain boundary around [011] with θ = ε, cos ε # 1 and sin ε # ε. The
rotation matrix R between the crystals is:

1 −ε ε

R = ε 1 0 (2.10)
−ε 0 1

0 −ε ε

I − R −1 = ε 0 0 (2.11)
−ε 0 0
The indices of any vector in the (01−1) grain boundary plane are [xyy].
The product (I − R−1 )X in Eq. (2.4) enables to define the three components of the
dislocation density vector B:
Bx = 0 By = εx Bz = − εx (2.12)
B is then perpendicular to the rotation axis and to the grain boundary plane (Fig. 2.5)
Furthermore:
B = nbm and bm = a/2 [01 − 1]. (2.12bis)
The dislocations have an edge character.

There are exact solutions for relations (2.4) only for vectors X = x 0 = [x yy]
such as:
x = ± bm / ε x = ±2 bm / ε x = ± n bm / ε (2.13)
The X vectors cut the intergranular dislocations along [100], the dislocation lines
perpendicular to [100] in the grain boundary plane are parallel to [011]. The low-angle
tilt grain boundary is well described by edge dislocations parallel to the rotation axis
2.3 Bollmann’s Discrete Approach: Intrinsic Dislocations 35
Fig. 2.5 Geometry of the

primary dislocation network
(b = a/2 [01−1]) in a low angle
[011] tilt grain boundary
and distant from d = b/θ, as predicted by Read and Shockley. In this description,
the distance between the dislocations is large as θ = ε is a very small angle.
We may also describe the low angle tilt boundary by using an equivalent rotation
between the crystals: θ = (ε +π ), then cos (ε +π )# −1 and sin(ε +π ) − ε. The
application of the previous procedure leads to predict the occurrence of dislocations
also parallel to the tilt axis but with a Burgers vector b = a [100]. Furthermore,
the distance between the dislocations is now very small ( θ = ε +π being large).
As an example, for a 2◦ misorientation in copper, the dislocations are distant from
about 7 nm according to the first description; they may be observed by conventional
transmission electron microscopy. According to the second description, the distance
is about 0.12 nm and the dislocations are no longer individually distinguishable; the
physical reality of this description in terms of dislocations is questionable. Rather in
this case to use a continuous distribution of the deformation. It seems reasonable to
consider that the dislocation structure predicted by the first description is likely to
be realised. This point will be discussed further.
Figure 2.6 presents a primary dislocation network in a low-angle tilt grain bound-
ary (2◦ around 111) in a iron–molybdenum alloy with a b.c.c. structure [15].
The same approach for a low-angle twist grain boundary with a {001} plane results
in a structure composed of two screw dislocation networks perpendicular to each
other and with a Burgers vector b = a/2 110; the distances between the dislocations
Fig. 2.6 Electronic micro-

graph of a 2◦ 111 tilt grain
boundary in a Fe-1 % Mo
showing a primary edge dislo-
cation network [15]
Fig. 2.7 Electronic micrograph of a low-angle twist grain boundary θ# 4◦ around 001, in gold
showing two primary screw dislocation networks aligned along the 110 directions (at 45◦ of g);
the spacing between the dislocations is d# 40 nm [16]. Bubbles are artefacts (bad welded bicrystal
regions) resulting from the bicrystal manufacturing; disturbed parts of the network (as EF) indicate
the presence of extrinsic dislocations (see Sect. 5.2)
agree with the experiments (Fig. 2.7) [16]. As previously, other descriptions exist but
their physical meaning is questionable.
It must be stressed that, whatever the chosen description of a given grain boundary,
the dislocation content totally accounts for the rotation between the crystals: these
dislocations are geometrically necessary. But the adopted grain boundary structure
depends on the relaxations, and only the observation enables to specify what descrip-
tion is physically meaningful.
Besides, the distance between the dislocations decreases as the misorientation
increases and becomes so small that it is no longer compatible with the crystal lattice
periodicity; the dislocation stress fields then overlap. However, the O-lattice being a
continuous function of the misorientation, the primary dislocations must exist from
a strict geometrical point of view. This continuity is confirmed by the observation
of an electron (or X-ray synchrotron) diffraction pattern of a grain boundary plane
[17]. Indeed, the periodic distribution of deformations in the plane constitutes an
optic grid; by selecting a diffraction spot, an electron microscopy image of the grid
may be obtained. The distance between the rows of this rebuilt network well satisfies
Eq. (2.9).
In summary, the equation giving the total dislocation content B in a grain boundary
is mathematically valid whatever the misorientation θ ; but, for high θ values, it covers
the physical concept of dislocation density. Another discrete description of the grain
boundary structure is then proposed which refers to the second invariant of the
bicrystal, i.e. the CSL lattice.
2.3.2 Secondary Intrinsic Dislocation
As postulated by Bollmann, any grain boundary adopts as much as possible the struc-
ture of the neighboured coincidence grain boundary, provided relaxations on the form
of secondary intrinsic dislocations. These dislocations are localized in the regions
of bad fit between the two DSC lattices of the coincidence grain boundary θc rotated
one with respect to other by an angle equal to the angular deviation between the real
and the coincidence grain boundaries. These lines correspond to the intersections of
the grain boundary plane with the walls of the O2-lattice predicted by Eq. (1.24).
The term secondary comes from the fact that these defects break the periodicity
of the primary dislocations. They form a sub-grain boundary in the grain boundary.
Then, coincidence grain boundary regions delimited by secondary dislocations con-
stitute the grain boundary (Fig. 2.8).
Fig. 2.8 Schematic representation of the periodic arrangement of secondary dislocations (⊥) in
the primary network (⊥)
Conversely to the plurality of the primary dislocation structures for a given grain
boundary, only one secondary dislocation structure is predicted by Eq. (1.24). Indeed,
whatever the equivalent rotations chosen for the real and the coincidence grain
boundaries, the angular deviation matrix D is identical. However, different struc-
tures may exist for non-mathematical reasons, linked to the atomic relaxations in the
boundary.
The characteristics of the secondary dislocations depend on the tilt/twist character
of the deviation to the coincidence. In a tilt grain boundary, a pure tilt deviation may
exist either around the fundamental rotation axis or around another axis; otherwise
the deviation may have a twist or mixed character. Depending on the case, the grain
boundary structure is constituted of only one secondary edge dislocation network
or of two screw dislocation networks or even several mixed dislocation networks.
Similarly, a twist grain boundary may present a twist, tilt or mixed angular deviation
to the coincidence. In any case, the dislocation lines correspond to the intersections
of the grain boundary plane with the O2-lattice walls.
The Burgers vectors of the secondary dislocations belong to the DSC lattice, but
are not necessarily the elemental vectors of this lattice. In order to determine the
Burgers vector of a secondary dislocation, a closed circuit drawn around the defect
in the grain boundary must be reproduced in the DSC lattice of the nearest coincident
grain boundary. A construction very useful for analysing the high-resolution trans-
mission electron microscopy images enables to determine the dislocation Burgers
vector and its associated step (Fig. 2.9) [18]. A secondary dislocation provokes a
translation of the CSL lattice. A step vector s may be defined, that relates an initial
coincidence node to the coincidence node obtained by translation. If the crystal I is
fixed and the crystal II translated, then the step sI is defined in the coordinate system
of crystal I; and reciprocally for sII if the crystal II is fixed. There are several choices
for the step vectors, related to each other by a DSC vector; the shortest one is usually
chosen. The height of the step associated to the dislocation is h I = sI n in crystal I
and hII = sII n in crystal II. It follows that:
b = sI − sII
h = n(sI + sII )/2 (2.14)
where n is the normal to the grain boundary plane (pointed from crystal I to crystal II).
The Burgers vectors of the secondary dislocations must be necessarily indexed in
the reference system of the bicrystal translation lattice (the DSC lattice). Otherwise,
if each vector si is indexed in the reference system of crystal i, we obtain the primary
Burgers vector of the defect.
Fig. 2.9 Construction in the DSC lattice of a = 5 grain boundary of three possible steps s, s and
s associated to a dislocation with vector ± b1 . The empty and full symbols correspond to crystal
II and crystal I, respectively; the triangular symbols represent sites located at ± a/2 [100] along the
projection axis; a CSL in its original position; b Displacement of crystal II relative to crystal I by
a vector + b1 ; c displacement of crystal I relative to crystal II by a vector − b1 [18]
The average distance between the secondary dislocations is given by a formula

similar to (2.9)
|bDSC |
d= (2.15)
2sinθ/2
This distance decreases when the coincidence index increases (|bDSC | decreases)
and/or when the angular deviation θ to the coincidence increases. As a result, the
dislocations cannot be experimentally distinguished beyond a certain value or
beyond a certain θ deviation from the neighbouring coincidence relationship. Fur-
thermore, with the diminution of |bDSC |, the contrasts associated to dislocations in
transmission electron microscopy decreases. Thus, while preserving an almost con-
stant spacing, the dislocations are less and less visible when increases. Figure 2.10
illustrates this effect for a series of 001 twist grain boundaries in gold [19].
Fig. 2.10 Intrinsic secondary dislocation networks in 001 twist grain boundaries in gold showing
the diminution of their contrast when increases: a = 1, b = 0,289 nm; b = 5, b = 0,129 nm;
c = 13, b = 0,080 nm; d = 17, b = 0,070 nm [19]
Figure 2.11 presents a network of secondary dislocations in a tilt grain boundary

in nickel that totally account for the deviation θ = 0, 09◦ 112 of this boundary
with respect to the = 3 {111} coincidence grain boundary [20]. Dislocations A
are parallel to the 112 direction, their Burgers vector is the DSC vector of = 3
normal to the grain boundary plane, b = a/3 111, and their distance (≈ 130 nm) is
in good agreement with the calculated distance.
Fig. 2.11 Electronic micrograph in dark field (crystal C2 is in extinction condition) showing a
network of intrinsic secondary dislocations (as A) in a near = 3 {111} tilt grain boundary in
nickel (B is an extrinsic dislocation that will be analysed in part II, Chap. 5) [20]
The near one-dimensional coincidence grain boundaries displays a network of

secondary dislocations, the lines of which being parallel to the intersections of the
crystal planes in good fit with the grain boundary plane.
In an approximate coincidence grain boundary, the structure in terms of dis-
location must account for the rotation and the deformation necessary to bring in
coincidence two multiple cells of the crystals. According to the choice of the rational
value of the (c/a)2 ratio that approximates the real value, these two components
(rotation and deformation) differ. Only the observed intergranular structure enables
to specify the relaxation mode chosen by the grain boundary. Let us consider a grain
boundary near the rhombohedral twin, often observed in alumina or hematite poly-
crystals. When the grain boundary plane is strictly parallel to the {01-12} twin plane,
the grain boundary is described by an exact bi-dimensional coincidence. But, most
often, the observed grain boundary deviates from {01-12}. Then, the grain boundary
is described in terms of tri-dimensional approximate coincidence, by reference to
specific rotations θ around (02-21) associated to rational values of (c/a)2 as close
as possible of the real value, (2.730)2 for alumina. The specific rotations with their
corresponding deformations ε for alumina are (Table 2.1) [21]:
Table 2.1 Specific rotations

c/a θ◦ ε
and associated deformations
giving rise to rational 7 2.739 85.90 0.003
coincidences for rational 29 2.728 86.05 0.001
values of the c/a ratio [21] 36 2.730 86.02 0
The deformations decrease when increases, this is also true for hematite
(c/a = 2,733). Among the three previous descriptions, which is the one adopted by
the grain boundary: high coincidence index ( = 36) and zero deformation or small
coincidence index ( = 7) at the expense of a deformation? Only the experiment
enables to answer the question.
The DSC vectors of the different descriptions are such that the intensities of two
of them, b2 and b3 , are almost identical whereas the third vector b1 decreases as
increases. A secondary dislocation network may be observed in a grain boundary,
the misorientation of which being close to that of the rhombohedral twin in alumina
(θ = 85◦ 5); the grain boundary plane is deviated by 8◦ from the rhombohedral plane
(Fig. 2.12). The Burgers vector of the intrinsic dislocations have been identified as
a linear combination of the two almost identical vectors (b2 and b3 ), whatever the
grain boundary description. In that case, no approximate coincidence description
seems to be preferred. However, in a grain boundary with a plane very close {01-12}
in hematite, the secondary dislocations, observed by high-resolution transmission
electron microscopy, possess a Burgers vector that is the smallest DSC vector cor-
responding to a = 7 coincidence relationship [22].
Fig. 2.12 Network of intrin-

sic secondary dislocations C
in a grain boundary near the
rhombohedral twin in alumina
[21] (Extrinsic dislocations B
will be described in part II)
In the θ = 90◦ (10-10) grain boundaries that are numerous in tungsten carbide of
hexagonal symmetry, the secondary dislocations have Burgers vectors of the = 2
coincidence relationship with a deformation equal to 0.024 whereas another neigh-
boured description with a = 39 coincidence index only requires a 0.0016 defor-
mation [23]. Similar conclusions have been obtained for the (01-12) twin in zinc of
hexagonal structure. To decide what approximate coincidence is better adapted to
the grain boundary structure, the dislocation lines are determined; they correspond
to the intersections of the grain boundary plane with the O2-lattice associated to
the smallest coincidence index, = 13 in this case. But the description requires a
non-negligible deformation, about 0.5 % [24].
Thus, in non-cubic system, the structure of a grain boundary close to different
approximate coincidences seems to adopt that of a grain boundary with a specific
rotation corresponding to the rational coincidence with the smallest index at the
expense of a larger deformation of the coincidence cell.
2.3.3 Interest and Limit of the Intrinsic Dislocation Model
To sum up, intrinsic primary dislocations with Burgers vectors of the crystals describe
the low-angle grain boundary structures. Similarly, a high angle coincidence grain
boundary only contains primary dislocations, but the close spacing of these disloca-
tions does not give them any physical reality. Finally, a high-angle non-coincident
grain boundary contains primary dislocations accounting for the nearest coincidence
structure and secondary dislocations that accommodate the deviation between the real
and the coincidence grain boundaries. Most often, only the secondary dislocations
are visualized as the distances between the primary dislocations are too small.
Figure 2.13 shows the evolutions of the expected dislocation spacing in function
of the misorientation angle θ and in function of the deviation θ from different coin-
cidences for 001 symmetrical tilt grain boundaries in gold [25]. The points reported
on the curves indicate the spacing experimentally determined on the conventional
transmission electron microscopy images. The smaller the coincidence index, the
larger the deviation θ authorized to visualize secondary dislocations.
Fig. 2.13 Evolution of the spacing d (in nm) in function of the misorientation θ of the dislocations in
001 symmetrical tilt grain boundaries in gold. The full line and the dotted lines give the spacing of
the primary and secondary dislocations, respectively. The points indicate in which grain boundaries
periodic dislocations have been observed [25]
The possibilities to simultaneously observe primary and secondary dislocations

are limited to few experiments; it was the case in gold for a [001] tilt grain boundary
deviated from about 1◦ from the nearest coincidence = 13 {320} (Fig. 2.14).
The primary dislocations with Burgers vector a/2 [110] are distant from 0.77 nm
(calculated distance 0.75 nm) and the secondary edge dislocations with DSC Burgers
vector a/13 [320] are distant from 7 nm, in agreement with the theoretical distance
of 7.1 nm [25].
The Bollmann formalism does not claim to describe the physical reality of a
grain boundary, but constitutes a possible approach to knowledge. Bollmann himself
insists that the use of the 0-lattices only permits a discrete selection of the possible
Fig. 2.14 Electronic micrograph of a symmetrical [001] tilt grain boundary near = 13 {320} in
gold (g = 200). The fine periodic contrasts are due to primary dislocations with Burgers vector a/2
[110], the large contrasts (arrows) are attributed to secondary dislocations with Burgers vector a/13
[320] [25]
displacement vectors B, characteristic of the continuous displacement field of the

transformation A. It is a first approximate model of linear relaxation in the grain
boundary plane where the fact that it is linear is pure convenience.
Moreover, whatever the expression used to describe the grain boundary structure
in terms of dislocations, Burgers vector density according to Frank-Bilby or discrete
dislocations according to Bollmann, there are a large number of transformations S
(or R) equivalent because of the symmetry operations of the crystals, i.e. depending
on the choice of the unit cell in the reference coordinate lattice. An example has been
given for the low-angle [011] tilt grain boundary (see Sect. 2.3.1). The equivalent
matrices are obtained by multiplying SI , SII or R by a uni-modular matrix U corre-
sponding to an operation of the point group symmetry of lattices I or II. Thus, there
are a very large number of possible descriptions of the structure of a given grain
boundary. Bollmann favours a particular description called NNR for Nearest Neigh-
bour Relationship that corresponds to the minimal dislocation density by arguing
that it must be associated to an energy minimum (E proportional to b2 ) [26]. This
argument is based on the consideration of the long-range elastic strain fields, but
the local distortions due to intrinsic dislocations cancel at large distance. A grain
boundary at equilibrium only generates short-range strains; the strain distribution
depends on the intergranular relaxation modes and cannot be predicted by a geo-
metrical model. The Frank-Bilby and Bollmann equations are disconnected from the
grain boundary energy. Finally, a grain boundary is a unique physical object with a
precise structure and a precise energy. Among all the possible structures, none can be
selected a priori. This question will be discussed more thoroughly in the description
of grain boundaries constrained at triple junctions (see Sect. 10.2.2) where it takes a
particular importance.
Despite its geometrical nature, the Bollmann model has been successfully applied
to identify intrinsic dislocations in numerous grain boundaries in different types of
materials. It is especially remarkable that periodic dislocations have been observed,
by high-resolution transmission electron microscopy, in symmetrical tilt grain
boundaries with high coincidence indices, as = 337 in silicon; in that case,
the Burgers vectors of the intrinsic secondary dislocations do not correspond to the
elemental DSC vectors [27].
A dislocation network always reveals the existence of a preferred ordered state of
the grain boundary, whatever its description. The reason why a preferred state exists
is a subject of questioning in solid-state physics.
However, the high-coincidence index grain boundaries displaying periodic dislo-
cations are not true general grain boundaries; they generally display misorientations
around axes with simple indices and possess symmetrical grain boundary planes.
What happens for grain boundaries with random misorientations and asymmetrical
high index planes? These grain boundaries do not present dislocation networks at
the scale of the conventional electron microscopy and are not observable by high-
resolution electron microscopy. Do they obey a discrete or continuous description? A
partial answer to this question has been advanced when we deal with the grain bound-
ary structure at high temperature (See Sect. 4.1), then with relaxation phenomena of
non-equilibrated grain boundaries (See Chap. 9).
In summary, the dislocations predicted by Eqs. (1.23) and (1.24) are geometrically
necessary, but the actual dislocation content depends on the local intergranular stress
relaxations. The structure is governed by the energy minimization and any forecast
based on geometry alone cannot reflect reality. It is physics that determines the
real mode of relaxation and a physical description requires an atomistic approach
(Chap. 3).
2.4 Partial Intergranular Dislocations
The grain boundaries may also contain partial dislocations, generally associated to
rigid body translations τ , thus not predicted by the geometrical formalism. Although
non-periodic, these dislocations belong to the grain boundary structure [28]. Their
Burgers vectors are determined by using the circuit mapping method proposed by
Pond [29]. This circuit is analogous to a Frank circuit but is drawn in the dichromatic
complex; it takes into account all the symmetry operations of the space group and
not only the translations. Partial dislocations may also result from the existence of
grain boundary structure variants. If regions of different structures along the grain
boundary are related by a symmetry operation, they are energetically degenerate.
Partial dislocations separating such regions have been observed in germanium [30]
and in a copper–silicon alloy [31]. Partial dislocations are often located at the junction
between two boundary facets, but may be also present in a straight part of the grain
boundary. They may result of the decomposition of the DSC intrinsic dislocations, as
observed in a = 3 {211} grain boundary where a perfect dislocation with Burgers
vector a/3 111 dissociates into two products with Burgers vectors a/9 111 and
2a/9 111 [28].
2.5 Stress Fields Associated to Intrinsic Dislocations 45
2.5 Stress Fields Associated to Intrinsic Dislocations
The stress field of a low-angle grain boundary has been calculated by Hirth and
Lothe, using isotropic linear elasticity and by adding the stress fields of the periodic
dislocations [32]. The dislocations are mixed; three components of their Burgers
vector (bx by bz ) and thus three periodic networks are considered (Fig. 2.15).
Fig. 2.15 The three periodic networks of dislocations that are considered to calculate the elastic
stress fields of a low angle grain boundary. The Burgers vector of the mixed dislocations is decom-
posed in two edge components along Ox (a), Oy (b) and one screw component along Oy (c) [32]
For a symmetrical tilt grain boundary constituted of egde dislocations with

Burgers vector bx, parallel to the z axis and distant from D along Oy (Fig. 2.15a),
the stress components (in regions such as x D/2π ) exponentially decrease with
the distance x from the grain boundary ; on the contrary the stress field of an isolated
dislocation is inversely proportional to the distance from its line. The expressions of
the shear component σx y for the two types of defects may be compared:
• for a dislocation
μb x(x 2 − y 2 )
σx y = (2.16)
2π(1 − ν) (x 2 + y2)
• for a tilt grain boundary
μb 2π x −2π x 2π y
σx y = exp cos (2.17)
(1 − ν) D D D
where, μ is the shear modulus and ν the Poisson coefficient.

The expression of σxx for the grain boundary is similar to that of σxy with
sin2π y/D instead of cos2π y/D. The value of x = 2D (y = 0), the shear compo-
nent of σxy due to the periodic arrangement, is about 1 % of its value for an isolated
dislocation. In the grain boundary plane (x = 0) and for distances y < D/2π of a given
dislocation in the network, more than 90 % of the grain boundary stress σxx is due to
the dislocation at the origin. Midway between the two dislocations, the compression
stress of one dislocation is annihilated by the tension stress of the other. Then, there
is no long-range stress field induced by a symmetrical tilt grain boundary. If now
we consider the stress field due to a network of edge dislocations with a Burgers
vector by (Fig. 2.15b), the component σyy does not cancel at long distance, in contra-
diction with the equilibrium condition given by Eqs. (2.3) or (2.4). The equilibrium of
the grain boundary imposes that there is no total Burgers vector parallel to the grain
boundary plane associated to edge dislocations. For the last component of the Burgers
vector bz (Fig. 2.15c), the dislocations have a twist character. For a uni-dimensional
arrangement of dislocations parallel to Oz, the shear component σyz exponentially
tends to zero with the distance to the grain boundary, but the stress component σxz
preserves non-negligible values at large distance. A simple arrangement of screw
dislocations is then incompatible with the equilibrium condition (2.4). Equilibrium
in a pure twist grain boundary requires the formation of two screw dislocation net-
works perpendicular to each other that result in a rotation around the normal x; the
stress fields of these networks annihilate.
More generally, three dislocation networks with non-coplanar Burgers vectors
account for the misorientation of a random grain boundary. For a grain boundary
that satisfies the Frank-Bilby formula, the elastic stress field analysis shows that the
combination of the three stress fields yields their annihilation, even if one of the
network generates long-range stresses. In the calculation, the elastic constants of the
crystals are used, whereas the elastic constants of the grain boundary may differ from
those of the crystals as found by simulation [33]. Although the elastic constant tensor
of a boundary is not known, these constants may strongly vary in the grain boundary
plane [34].
Another model for calculating the stress fields is based on the grain boundary
description in terms of disclinations [35]. On the contrary to the dislocation model,
it takes into account the disclination core energy and may apply to high-angle grain
boundaries (see Sect. 3.5).
More complex situations than those previously described by Hirth and Lothe
have been addressed, in particular by considering no strictly periodic networks and
by studying the evolution of their stress fields during return to periodicity. These
approaches will be detailed in part II, Chap. 9 when dealing with the return to equi-
librium of a non-equilibrated grain boundary.
To conclude, the knowledge of the elastic contribution is insufficient to account
for the intergranular energy, we must know the contribution of the grain boundary
core. This constitutes the subject of the next session focussing on the grain boundary
atomic order.
References 47
References
1. F.C. Frank, Symposium on the Plastic Deformation of Crystalline Solids (Office of Naval
Research Pittsburgh, Pennsylvania, 1950), p. 150
2. B.A. Bilby, Report on the Conference on Defects in Crystalline Solids (The Physical Society,
London, 1955), p. 123
3. E. Kröner, in Statistical Continuum Mechanics (Springer, Berlin, 1972)
4. E. Kröner, in Physics of Defects, ed. by R. Balian, M. Kleman et J Poirier (North Holland,
Amsterdam, 1981), p. 219
5. W. Bollmann, Crystal Defects and Crystalline Interfaces (Springer, Berlin, 1970)
6. Bollmann, in Crystal Lattices, Interfaces, Matrices: An Extension of Crystallography ed. by
W. Bollmann (1982)
7. W.T. Read, W. Shockley, Phys. Rev 78(195), 275 (1950)
9. G.B. Olson, M Cohen. Acta Metall. 27, 1907 (1979)
10. R. Bonnet, J. Phys. 43, C6–215 (1982)
11. R. Bonnet, Phil. Mag. A51, 51 (1985)
12. D. Wolf, Scripta Metall. 23, 1713 (1989)
13. P. Lacombe, L. Beaujard, J. Inst. Metals 74, 1 (1947)
14. H.E. Vogel, E.E. Thomas, Phys. Rev. 90, 489 (1953)
15. S. Lartigue, communication personnelle (1980)
16. L. Priester, R.W. Balluffi, J. Microsc. Spectrosc. Electr. 4, 615 (1979)
17. S.L. Sass, T.Y. Tan, R.W. Balluffi, Phil. Mag. 31, 559 (1975)
18. A.H. King, D.A. Smith, Acta Cryst. 36, 335 (1980)
19. T. Schober, R.W. Balluffi, Phil. Mag. 21, 109 (1970)
20. S. Poulat, B. Décamps, L. Priester, Phil. Mag. A 77, 1381 (1998)
21. H. Grimmer, R. Bonnet, S. Lartigue, L. Priester L, Phil. Mag. A61, 493 (1990)
22. L. A Bursill, R.L. Withers, Phil. Mag. A 40, 213 (1979)
23. J. Vicens, E. Laurent-Pinson, J.L. Chermant, G. Nouet, J. Phys. 49, C5–271 (1988)
24. A.H. King, F.R. Chen, Phil. Mag. A 57, 431 (1988)
25. E.P. Kwan, R.W. Balluffi, Phil. Mag. A 56, 137 (1987)
26. W. Bollmann, Mat. Sci. Eng. A113, 129 (1989)
27. J. Thibault, J.L. Putaux, A. Jacques, A. George, H.M.Michaud a et X. Baillin, Mat. Sci. Eng.
1164, 93 (1993)
28. R.C. Pond, Proc. R. Soc. Lond. A 357, 471 (1977)
29. R.C. Pond in Dislocations in Solids, vol. 8, ed. by F. Nabarro (North Holland, Amsterdam,
1989), pp. 1–66
30. J.J. Bacmann, G. Sylvestre, M. Petit, W. Bollmann, Phil. Mag. A 43, 189 (1981)
31. C.T. Forwood, L.M. Clarebrough, Phil. Mag. 47, 135 (1983)
32. J.P. Hirth, I. Lotte, Theory of Dislocations, 2nd edn. (McGrawHill, New York 1968), p. 671
33. D. Wolf, J.F. Lutsko, J. Mater. Res. 6, 1427 (1989)
34. A. Aber, J.C. Bassani, M. Khanta, V. Vitek, G.J. Wang, Phil. Trans. Roy. Soc. A339, 555 (1992)
35. V.Y. Gertsman, A.A. Nazarov, A.E. Romanov, R.Z. Valiev, V.I. Vladimirov, Phil. Mag. A59,
1113 (1989)
Chapter 3
Atomic Order of Grain Boundaries
The model of the grain boundary structure in terms of intrinsic dislocation networks
does not enable to satisfactorily account for grain boundary energy and functional
property differences. It does not take into account the five microscopic degrees of
freedom and the local relaxations that lead to more stable structures. Furthermore,
it does not consider the different types of bonding between atoms depending on
the metallic, ionic or covalent crystal character. According to this model, the grain
boundaries in materials displaying the same Bravais lattice would have the same
structure; while, for instance, the = 9 tilt grain boundary has a low energy in
silicon and a high energy in f.c.c. metals. To describe the intergranular structure, it
appears necessary to consider the atomic interactions in the grain boundary core, i.e.
to explicitly calculate the core energy of the intrinsic dislocations.
In this approach, the progression of the idea of grain boundary order is particu-
larly interesting. First, searching for a local arrangement of the atoms, the concept
of polyhedral unit does not concern any long-range order. The latter appears fur-
ther, then develop owing to the simulations of the intergranular structure and the
so-called structural unit model. This model predicts the structures of a series of
grain boundaries, providing that their geometrical parameters are rational and that
only one parameter varies in the series. The extension of the model to irrational grain
boundaries leads to the notion of quasi-periodic boundaries that no longer display
a translational symmetry but preserve a long-range order. Finally, recent simula-
tions reveal that an amorphous layer may occur in twist grain boundary cores in
semiconductors.
Note that from the outset and in their more recent development, the grain boundary
atomic structure models are closely associated with the intrinsic dislocation models.
For clarity reasons, we first present the atomic models without reference to disloca-
tions although the interpretations of the structures imply local relaxations. Then we
develop the elegant model that links structural units and intrinsic dislocations named
SU/GBD for Structural Unit/Grain Boundary Dislocation model. Finally we just
evoke a model based on the association between structural units and disclinations.

50 3 Atomic Order of Grain Boundaries
3.1 Hard Sphere Model: Geometrical Construction
In 1959, Hornstra conceived the idea that, in semiconductors because of the tetra-
hedral atomic coordination, a well-defined atomic configuration (polyhedral unit or
structural unit) is attached to the core of a grain boundary dislocation. By using
a model of balls and sticks, he built the core structures of several 110 tilt grain
boundaries [1].
The first models of grain boundary atomic structure postulate that, for all materials,
the grain boundary core consists of an arrangement of polyhedral atomic units built
by optimizing the number of local coordinations and by avoiding atomic repulsions
[2, 3]. These polyhedra take the shape of deltahedra, i.e. all their faces are equilateral
triangles. There are a limited number of polyhedra (Fig. 3.1) [4], five of them are well
known as being the units in the hard sphere model of the liquid [5]. This suggests
a back and forth movement between the notion of a grain boundary as a crystalline
structure and as an amorphous layer. This model is more concerned by describing
the best local arrangement of atoms in a polyhedron than the repetition of an atomic
motif that would result in a periodicity. It does not deal with the long-range grain
boundary structure but only considers local relaxations.
Fig. 3.1 Different polyhedra that may constitute grain boundary structural units [4]
On the basis of this description, the excess volumes of symmetrical [110], [100]
and [111] tilt grain boundaries and the possible segregation sites, substitutional or
interstitial, within the polyhedra have been calculated (Fig. 3.2). The excess volumes
3.1 Hard Sphere Model: Geometrical Construction 51
Fig. 3.2 Description in the hard sphere model of a symmetrical = 17, 28.07◦ [100] tilt grain
boundary with a {053} plane. The grain boundary is composed of capped trigonal prisms, each
displaying an interstice of size 0.82, separated by channels with five sides containing two other
interstitial sites. a The larger substitutional site has a size of 1.12. b The grain boundary excess
volume is 0.44 [6]
vary between 0 for the coherent twin until about 0.5 for tilt [111] grain boundaries.
Almost all the grain boundaries contain an interstitial site of size 0.6 or larger
(we recall that in an f.c.c. crystal, the maximum interstitial site size is 0.414).
The substitutional site sizes do not exceed 1.3 in grain boundaries with non-dense
planes [6].
This model has also been used to explain the first grain boundary structures
of semiconductors observed by high-resolution transmission electron microscopy.
However, it proved to be limited for the understanding of grain boundaries in metals,
and moreover it has no predictive character. The structural unit model tries to meet
the needs for predicting the structure of any grain boundary and to serve as potential
guide for the experimental investigations and the simulations.
3.2 Structural Unit Model
The model is based on simulations that enable to associate an intergranular structure

to a minimum energy, on the contrary of the hard sphere model which is essentially
geometric. It should be noted however that a criteria of energy minimization was
not absent in the polyhedra consideration but no control of that energy would be
warranted. The great contribution of this model, compared to the previous one, is its
predictability although there are some restrictions to this aspect.
3.2.1 Principle
Akin to the hard sphere model, this model starts by considering a characteristic
configuration of a small group of atoms called structural unit, but now it considers
the ordered arrangement of these polyhedra in the whole bi-dimensional space of
the grain boundary. It describes the systematic changes of the intergranular structure
when a macroscopic degree of freedom, misorientation or grain boundary plane
inclination, continuously varies. The basic idea of this model is attributed to Bishop
and Chalmers [7]. It is presented here with its extension and its current formalism
due to Sutton and Vitek [8]. It results from energy calculations for a great number
of tilt grain boundaries, symmetrical or not, and for certain twist grain boundaries in
f.c.c. and b.c.c. structures.
The principle of the structural unit model may be formulated as follows:
Any long-period grain boundary with misorientation θ [uvw] may be described
by a sequential arrangement of structural units associated to two short-period grain
boundaries θi and θ j located on each side of θ and possessing the same [uvw] rotation
axis and the same (hkl)m median plane (Fig. 3.3).
A first description of the grain boundary atomic structure uses as references the so-
called favoured grain boundaries, constituted from only one type of structural units
that are indivisible in smaller units; they are represented by A A A. . . or B B B. . .
(Fig. 3.3). A more restrictive formulation of the previous principle may be proposed:
an intermediary grain boundary between two favoured grain boundaries is composed
of a well defined mixture of two structural units, one of them being a structural unit
of a favoured boundary, at least. The favoured grain boundaries are not necessarily
associated to small values of the coincidence index and do not necessarily possess a
low minimum energy. The term favoured only refers to the grain boundary structure.
Besides, a favoured grain boundary in a given metal may be not favoured in another
one and, a fortiori, in an ionic or covalent material where the atomic bonds differ.
The choice of a favoured unit, its structure and its ability to describe neighboured
grain boundaries depend on the interatomic potentials chosen in the simulation.
Choosing the possible structural units may rest on the dichromatic complex, but
the final shapes are dictated by the authorized displacements during relaxation. No
geometrical criterion enables to predict favoured grain boundaries, only properties
Fig. 3.3 Diagram explaining the atomic description of any grain boundary of misorientation θ in
terms of structural units (A and B) of short-period grain boundaries, the misorientations of which
surrounding θ
3.2 Structural Unit Model 53
that strongly depend on the core structure, such as boundary self-diffusion, may serve
as guides to select these reference grain boundaries.
The only favoured grain boundaries in materials with the same Bravais lattice are
the atomic planes of the single crystals ( = 1). Then, the grain boundaries with
misorientations 0◦ or 180◦ around 110, with {001} or {110} planes, may be chosen
as favoured to describe the series of symmetrical 110 tilt grain boundaries in cubic
materials. The coherent twin = 3, 70.53◦ 110 {111} is also a favoured grain
boundary in f.c.c. structures, then any tilt grain boundary may be located between
the two favoured = 1 and = 3 {111} grain boundaries.
A low index coincident grain boundary is often composed by a simple sequence of
two structural units, such as AB for the θ3 grain boundary (Fig. 3.3). The intermediary
grain boundaries with longer period, such as θ4 and θ5 , are considered as general.
A grain boundary close to the misorientation θ1 contains a majority of A units, and
vice-versa when its misorientation is higher than θ3 .
The period of the grain boundary perpendicular to the tilt axis may be represented
by delineating the units that compose its structure by vertical lines, such as |A A AB|
for the θ4 misorientation. If the period is decomposed into two identical half-period,
relatively displaced one with respect to the other along the tilt axis, a dot is placed
between the two identical sets of units. Then, a favoured grain boundary is not always
represented by only one unit, but may be represented by |A.A|; similarly, the period
of a general grain boundary may be |A A AB.A A AB|. Moreover, an exponent such as
A+ A− permits to distinguish two identical units differently orientated with respect
to the grain boundary plane.
The size of the structural unit in the grain boundary plane is determined by the grain
boundary translation symmetry. We may choose different shapes for the structural
unit in a bi-dimensional lattice provided that its translation generates the lattice; the
preferred shape is simply for convenience. In a tilt gain boundary, the convenient
unit shape is a rectangle, the sides of which being the period vectors of the boundary,
parallel and perpendicular to the tilt axis. It is the section by the grain boundary
Fig. 3.4 Relaxed structure of a symmetrical = 11 {113} grain boundary (50.48◦ [110]). Two
capped trigonal prisms, one being translated with respect to the other along the tilt axis, compose
the boundary period. The unit vectors in the grain boundary plane (normal to the figure) are the
CSL basic vectors: 1/2 [1-10] and 1/4 [33-2] (see Fig. 1.5) [8]
plane of the coincidence cell corresponding to the index. The C unit describing
the symmetrical = 11 [1-10] (113) tilt grain boundary may be delimited by the
unit vectors 1/2 [1-10] and 1/4 [33-2] (Fig. 3.4), which well correspond to the unit
vectors of the CSL cell (Fig. 1.5).
The dimension normal to the boundary plane is arbitrary; the smallest size is
usually retained. In the graphical representation of the structural units of a tilt grain
boundary, the atomic positions are projected parallel to the tilt axis. The atoms on
successive planes normal to the projection axis are distinguished by different symbols
(crosses, circles, triangles. . .). Figure 3.4 shows the relaxed structure of the = 11
(113) grain boundary with the atomic positions on the successive (2-20) planes; the
grain boundary period, composed of two identical C units displaced one with respect
to the other along the tilt axis, is represented by |C.C|. The structural unit is a capped
trigonal prism.
Table 3.1 reproduces part of the first simulation results obtained on symmetrical
[1-10] tilt grain boundaries in aluminium that well illustrates the principle at the basis
of the structural unit model [8]. On this series, the = 11 (113) and = 27 (115)
grain boundaries are favoured. The period vector of a grain boundary is obtained by a
simple vector composition: p = mu B + nνC in which u B and νC are the dimensions
of the B and C structural units, respectively. The period of the = 89 grain boundary
in Table 1.4 is: pBBBC = 3u B + 1vC or 1/2 [99-4] = 3/4 [55-2] + 1/4 [33-2]. The
letters attributed to the units in Table 1.4 are those adopted by Rittner and Seidman
[9] in an exhaustive study of tilt grain boundaries in metals; we refer especially to
this study later in this chapter.
Table 3.1 Periods and structures in terms of structural units of a series of symmetrical [110] tilt
grain boundaries in aluminium [8]
θ◦ Grain bound- Period vector Structure
ary plane
34.59 27 (1 1 5) 1/2 [55-2] |B.B|
33.79 107 (3 3 14) [77-3] |BBBBBC|
34.89 89 (2 2 9) 1/2 [99-4] |BBBC|
37.22 491 (5 5 21) 1/2 [21, 21, -10] |BCBCB.BCBCB|
38.94 9 (1 1 4) [22-1] |BC|
40.83 411 (5 5 19) 1/2 [19, 19, -10] |CBCBC.CBCBC|
42.18 139 (3 3 11) 1/2 [11, 11, -6] |CCB.CCB|
44.00 57 (2 2 7) 1/2 [77-4] |CCCB|
50.48 11 (1 1 3) 1/2 [33-2] |C.C|
The boundary plane has (hhl) indices in crystal I and (-h-hl) in crystal II. The period vector is
parallel to [uu-w] in crystal I and to [uuw] in crystal II (2hu = -lw)
The period of a given grain boundary is the same for all the materials with the same
Bravais lattice, only the shape and the atomic motif of the structural unit change.
Such similarities are evidenced in the case of the = 11 {332}, 129.52◦ 110
grain boundary in silicon (cubic diamond structure) and in nickel (f.c.c. structure).
Fig. 3.5 The periods of the atomic structures of the = 11 {332} grain boundaries in nickel (a)
and in silicon (b) are similar but the shapes of the structural units differ [10, 11]
The period of this grain boundary is represented by |M − T M + T | in silicon [10] and

by |E − D E + D| in nickel [11]. In each case, the period is composed of two units of
the = 3 {111}, 109.47◦ 110 and two units of the = 9 {221}, 141.06◦ 110
grain boundaries (Fig. 3.5).
The structural unit model rests on the principle that the boundary structure changes
as continuously as possible between two adjacent misorientations. In order to satisfy
this principle, the minority units must be distributed as regularly as possible between
the majority units and two minority units never appear together; a period constituted
by three A units and two B units is represented by |A AB AB| and not |A A AB B|.
The rigid body translations of two favoured grain boundaries that enter a period
must be compatible to avoid long-range stresses; the majority units may force the
formation of metastable minority units to fulfill this condition.
The structural units of a non-favoured grain boundary are inevitably distorted,
relaxations occur to minimize these distortions. When an intermediary grain bound-
ary is too far from a favoured one, the unit of the latter is deformed such as its
identification may be difficult. It is likely the case of the B units in the grain bound-
ary with the θ4 misorientation (Fig. 3.3). The use of another reference boundary
co-called delimiting grain boundary enables to improve the description but yields a
hierarchy of the descriptions.
Moreover, the structural unit of a favoured grain boundary can have multiple
configurations leading to similar values of the interfacial energies. The existence of
different reference structures results in a multiplicity of the stable structures of the
intermediary grain boundaries.
3.2.2 Hierarchy of the Descriptions
The θ3 grain boundary, composed of a simple sequence of units |AB AB AB|. . .

(Fig. 3.3) may also be described by using only one multiple unit C = |AB|. Then,
it constitutes a delimiting grain boundary for all the structures between θ3 and θ1 or
Fig. 3.6 Hierarchy principle of the grain boundary atomic description in terms of structural units.
The units A and B are elementary units (not divisible), the units C and D are multiple units
θ3 and θ2 . The θ4 boundary may be composed by an alternation of C units of the

delimiting θ3 boundary and two A units associated to the θ1 isorientation. The first
description belongs to the level 1 and the second to the level 2 of the hierarchy. We
can thus increase the level of description (Fig. 3.6).
An example of hierarchy of the atomic descriptions of the symmetrical =
73 001 (830) tilt grain boundary in gold is given on Fig. 3.7 [12]. At the first level,
the boundary is described by using as references the = 1 boundaries with (100)
and (110) planes but a strong distortion of the crystal units appears. At the second
level, the delimiting boundaries are = 1 (100) and = 5 (210), the minority
units (100) are still deformed. Finally, when the = 73 boundary is considered as
surrounded by = 5 (210) and = 5 (310), the shape of each component unit is
preserved.
Fig. 3.7 Hierarchy of the descriptions in terms of structural units of the symmetrical = 73 001
(830) tilt grain boundary in gold; a minority units of = 1 (110) and majority units of = 1
(100); b minority units of = 1 (110) and majority units of = 5 (210); c minority units of
= 5 (210) and majority units of = 5 (310). The dislocation distributions on the right of each
drawing are explained later (see Sect. 3.4.1) [12]
A selection rule of the delimiting boundaries specifies that one of them, at least,
must be centred in order that all intermediary grain boundaries may be decomposed
into structural units of these boundaries; this is the case for 110 tilt boundaries
with odd plane indices (Table 3.1). The structural unit of a delimiting grain boundary
repeats in each half-period of this boundary (i.e. |B.B|), but the two units are
displaced with respect to each other along the tilt axis. Thus, the choice of the
delimiting boundaries does not rest only on their energy. The use of delimiting grain
boundaries with no simple units enables to reduce distortions [9]. Larger is the num-
ber of simple units combined to form a multiple unit smaller is the angular range
of the structures that may be built using this unit. Simultaneously, the description
becomes more accurate as the constituent units are less distorted. Whatever the choice
of the description (for θ4 : three A units and one elemental B unit, or one C multiple
unit and two A units, or one D unit and one A unit), the necessary configuration to
characterise the boundary always constitutes the grain boundary period. This period
must be compatible with the boundary translational symmetry defined by the cell of
non-identical displacements (c.n.i.d.).
Fig. 3.8 Structures of the symmetrical 110 tilt grain boundaries obtained by simulation using a
EAM potential for metals with a low stacking fault energy [9]. A detailed configuration of the D
unit of = 3 shows that it is composed of two A units of = 1 rotated by 109.47◦ (dotted lines
in m); the E unit of = 9 may also be decomposed into two A/A units strongly deformed (dotted
lines in p)
The = 9 {221} (θ = 141◦ 06) grain boundary in metals is chosen as delimiting

for the description of the 110 tilt boundaries with {110} as median plane, the
misorientations varying between those of the favoured = 3 {111} (θ = 109◦ 47)
and = 1 {110} (θ = 180◦ ) boundaries (Fig. 3.8) [9]. These structures have been
obtained by simulation for f.c.c. metals with a low stacking fault energy. The E unit
of = 9 may be seen as the junction of one A unit of = 1 {100} (θ = 0◦ ) and
one A unit of = 1 {110} (θ = 180◦ ), both strongly deformed. The units A and
A are morphologically identical, only their orientations differ from 90◦ ; as soon as
they are rotated and/or distorted in a grain boundary, we may use either one or the
other. Alternately, the = 9 {221} boundary may be described as composed of one
deformed unit of = 1 and one D unit of = 3. Note that the D unit is formed
by the junction of two A (or A ) units that are rotated compared to their orientation
in = 1 [11]. The structures observed in most intermediary grain boundaries agree
with the structural unit model with = 9 as delimiting boundary, the A and A units
are not identified due to their strong distortions.
Atomic structures of symmetrical 110 tilt grain boundaries in nickel, obtained
by quenched molecular dynamic simulations, are found identical to those of Fig. 3.8
although the calculations use a relatively high stacking fault energy, close to that
experimentally determined [13]. The |E − D E + D| structure of the = 11 {332}
in nickel has been observed by high-resolution transmission electron microscopy
(Fig. 3.9) [11].
The boundaries of the series between = 9 and = 1 (110) are described using
E and A units, but large distortions of the E unit occur, even for a small deviation
with respect to = 9 (7.35◦ for = 27 {552}). Besides, the E unit is rotated by
180◦ giving rise to the E unit in the = 73 (661); a priori, this is incomprehensible
in the context of the structural unit model. Akin to the previous remark, the rotation
in = 33 (441) of the A unit, normally favoured for the boundary misorientations
close to θ = 180◦ , cannot be explained. Such deviations to the continuity of the
structural units model are discussed later (see Sect. 3.3).
The distinction between a non-breakable unit (favoured boundary) and a multiple
unit (delimiting boundary) is not easy; it depends on the distortion level of the unit
reasonably acceptable and thus of the nature of the material. It is generally easier
in a covalent material than in a metal, the elasticity of the latter authorizing larger
distortions. For example, the structural unit of the = 9 {221} delimiting boundary
in silicon and germanium cannot be decomposed into units of = 1 and = 3 as
it is the case for metals.
3.2.3 Multiplicity of the Descriptions
A favoured grain boundary is formed by only one type of structural units; but, different
configurations of the unit exist that result from different rigid body translations or
local relaxations leading to almost equal energies. Generally, the multiplicity of the
favoured grain boundary structures increases with the size of the c.n.i.d., i.e. the
Fig. 3.9 Atomic structure of the = 11 {332} grain boundary in nickel: a observed by HRTEM;
b obtained by molecular dynamic simulation [11]
possibilities of rigid body translations. Then, different combinations of the favoured

boundary units may describe an intermediary boundary (Fig. 3.10).
If the favoured boundaries 1 and 2 display several stable structures (i and j,
respectively), then the boundary composed of n units of 1 and m units of 2 has a
number N of configurations given by [14]:
Fig. 3.10 Principle of the multiplicity of descriptions of a same grain boundary (see text)
N = i n jm (3.1)
If the favoured = 5 [001] (210) boundary may be formed by units B or B

leading to almost the same energy E B /E B = 1.04, then the = 17 [001] (530)
boundary, composed of two = 5 units and one = 1 (110) unit, displays four
possible structures (N = 22 11 ) (Fig. 3.11); the energy ratios of the different structures
are 1.07, 1.09 and 1 for E B B A , E AB B et E AB B = E AB B , respectively.
Fig. 3.11 B and B structures of the favoured = 5 [001] (210) grain boundary (E B /E B = 1.04)
and of the intermediary = 17 [001] (530) boundary, located between = 5 and the single crystal
= 1 (110). The energy ratios E(a)/E(b)/E(c, d) of the structures a |B B A|, b |B B A|, c |B B A|
and d |B B A| follow the order 1.07/1.09/1 [14]
All the descriptions are not stable; they must be compared to the local stress dis-
tribution (see Sect. 3.4.1) and to the images obtained by high-resolution transmission
electron microscopy. For silicon and germanium, a metastable unit P may describe
a high-energy = 3 grain boundary; although this unit is not stable at low tem-
perature in the favoured boundary, it may exist in stable structures of intermediary
boundaries.
3.2.4 Geometrical Construction of the Grain Boundary

Structure
A vector is associated to each structural unit, the intensity of which is equal to the
dimension of the non-distorted unit in the direction normal to the tilt axis:
1 0
uA = vB = (3.2)
0 1
A boundary constituted by m A units with vector u A and nB units with vector v B ,

A and B being the units of the delimiting boundaries, has a period:
p = mu A + nv B (3.3)
A decomposition lattice is built with basic vectors u A and v B , the angle between
these vectors being equal to half the value of the angle between the θ A and θ B
misorientations (indeed, a symmetrical grain boundary is generated by two equal and
opposite rotations around the tilt axis). A vector attached to each node P of this lattice
corresponds to one (or one half-period for centred boundary) of an intermediary
boundary between θ A and θ B . The angles between the OP vector and the basic vectors
of the decomposition lattice are (θ − θ A )/2 and (θ B − θ )/2, respectively (Fig. 3.12).
The length of the period of a boundary represented by the point P (m,n) is:
p = mu A cos((θ − θ A )/2) + nν B cos((θ B − θ )/2) (3.4)
with m = 3 and n = 2 for the point P (3,2) on Fig. 3.12.

Compared to (3.3), the Eq. (3.4) expresses that the structural units are distorted in
the intermediary boundary and that their lengths differ from those they have in the
delimiting boundaries.
By expressing the length TS = m u A sin((θ − θ A )/2) = n ν B sin((θ B − θ ))/2
(Fig. 3.12), we may deduce the ratio of the numbers of structural units or level
rule [15]:
m ν B sin((θ B − θ )/2)
= (3.5)
n u A sin((θ − θ A )/2)
At that point, only the repartition of these m A units and n B units is still to
be determined. By taking into account the continuity principle, the structure of the
boundary defined by P (3,2) in the decomposition lattice is described by |AABAB|.
3.2.5 Algorithm for the Determination of the Grain Boundary

Structure
An algorithm to determine the sequence of structural units rests on the principle that
the grain boundary structure must change as continuously as possible between two
delimiting boundaries. In particular, not more than one minority unit may break the
sequence of majority units [16].
Let us consider a grain boundary X formed by m A units and n B units (n < m) and
search for the sequence of structural units in one period. The ratio n/m lies between
two neighbouring rational ratios i.e. similar to place the grain boundary between two
delimiting boundaries:
1 n 1
< < (3.6)
p+1 m p
⇓ ⇓ ⇓
B· (p + 1) A X B·pA
Let us call C the period of the first delimiting boundary: C = B·( p + 1)A and
D that of the second D = B· p A. The period of the boundary X, first described by
p = m A·n B, is then represented by p = rC·sD with r and s easily deduced from the
previous inequalities:
r = m − np s = n(p + 1) − m (3.7)
The ratio r/s is again situated between the two neighbouring rational ratios:
1 r 1
< < (3.8)
q+1 s q
⇓ ⇓ ⇓
C. (q + 1) D X C.qD
Fig. 3.12 Decomposition lattice for a series of symmetrical tilt grain boundaries. Each boundary
is represented by a point P, the coordinates (X, Y ) of which giving the numbers of A and B units in
one period. The basic vectors attached to the units are u A = (1, 0) et v B = (0, 1). The OP vector
is inclined with respect to the basic vectors by equal angles (θ − θ A )/2 and (θ B − θ)/2
And so until the number of minority unit is equal to 1.

To illustrate the procedure, we consider the symmetrical = 1881 tilt grain
boundary with a misorientation of 38.06◦ around [1−10] and a (10,10,41) plane.
This boundary belongs to the series of Table 3.1; it is composed of 11 units A and
9 units B of the = 27 and = 11, respectively. To homogenize the presentation,
we use the same letters as previously to identify the units, A and B, even though they
are named B and C in Table 3.1. The period X of the boundary is located between
those of the delimiting boundaries represented by |AAB| and |AB|; then the multiples
units C = |AAB| and D = |AB| are considered:
1 9
< < 1 (3.9)
2 11
⇓ ⇓ ⇓
| A AB| X | AB|
⇓ ⇓
C D
The period of the boundary composed by the delimiting boundary periods C and D
is now of r units C and s units D with r and s deduced from Eq. (3.7): r = 2 et s = 7.
1 2 1
< < (3.10)
4 7 3
⇓ ⇓ ⇓
| C D D D D| X | C D D D|
⇓ ⇓
E F
At the next step, the use of equations similar to (3.7) leads to an equal number of
E and F units in the = 1881 grain boundary period, then the structure may be
represented by:
|E F| = |C D D D DC D D D| = |A AB AB AB AB AB A AB AB AB AB|
This structure well obeys the principle of the most regular distribution of the B
minority units within the A majority units.
If the boundary is irrational, the ratio of the unit numbers may still be written on
the form:
nB n
R= = +λ (3.11)
mA m
λ is an irrational number, which, by simplifying the problem, is considered as
quadratic:
λ2 − λ − 1 = 0 (3.12)
The sequence of structural units may be obtained by substituting recursively a

sequence of units to one A or B unit (self-similarity) [16]:
Iteration number Structural unit sequence R

0 |AB| 1/1
1 |ABA| 1/2
2 |ABAAB| 2/3
3 |ABAABABA| 3/5
4 |ABAABABAABAAB| 5/8
↓
∞ Quasi-periodic 1/τ
With τ the golden number.

The general quasi-periodic grain boundaries [17] appear as the limits of the coin-
cidence boundaries when the period increases. The following method, developed for
quasi-crystals, enables to elegantly determine the exact composition of a period in
any grain boundary.
3.2.6 Determination of the Grain Boundary Structure

by the Strip Method
Another method to determine the structural unit sequence is based on the principle
of cut and projection used to describe the quasi-crystal structures, so-called the strip
method [18]. We built a bi-dimensional lattice with basic vectors u A and ν B forming
a unit square (or rectangle) K. We trace a line E with a slope equal to m/n, the ratio
Fig. 3.13 Determination of the sequence of structural units in a grain boundary by the strip method
(see text)
of the favoured units in a boundary being known but not their sequence. A strip is
then obtained by translating the unit square K along the line E; this strip contains a
unique broken line passing through the bi-dimensional lattice nodes. The projection
of the edges of the broken line on the line L determines a sequence of two types of
tiles that represent the A and B units (Fig. 3.13). If the grain boundary has a rational
axis, the slope of the line E is rational; then, the distribution of the A and B units
is periodic. Applying the strip method to the symmetrical = 1881 boundary, the
structure of which having been obtained by the algorithm method, we well find again
a sequence composed by: |A AB AB AB AB A AB AB AB AB AB|.
This elegant construction enables to extend the structural unit model to irrational
boundaries. If we consider an irrational slope of the line E (m/n irrational), then
the structure obtained by projection is quasi-periodic. However this construction is
restricted to symmetrical tilt grain boundaries around a simple axis with an irra-
tional plane, provided that the median plane is common to the series of boundaries
composed of the A and B favoured units.
The structure, observed by high-resolution transmission electron microscopy,
of an incommensurable {111}I //{331}II grain boundary in nickel has been found
in good agreement with that obtained by simulation using an approximant [13].
The one-dimensional quasi-periodic structure of a 90◦ [001], (100)I //(110)II tilt
grain boundary in gold has been studied in details [19]. Starting with the image
obtained by high-resolution transmission electron microscopy, a numerically calcu-
lated diffraction pattern displays elongated peaks on the form of sticks perpendicular
to the grain boundary plane and characteristic of this plane. The direct image and the
filtered image, obtained by using the diffraction sticks, reveal that this grain boundary
is not periodic (Fig. 3.14). The boundary structure has been simulated
√ by using the
ratios 12/17 and 29/41 that approximate the irrational ratio 1/ 2; with the rational
Fig. 3.14 a High-resolution transmission electron microscopy micrograph of a non-periodic grain

boundary of misorientation 90◦ around [001] with a (100)I //(110)II plane in gold; b fourier image
by using the diffraction spots proper to the grain boundary plane [19]
Fig. 3.15 Structure of the (100)√ I //(110)II grain boundary simulated by using an approximant of
the irrational parametric ratio 1/ 2 [19]
ratio 29/41, the parametric misfit is only 3 × 10−4 (Fig. 3.15). The simulated image,
taking into account the atomic positions deduced from the simulation, well fits with
the experimental image (Fig. 3.16).
3.3 Interests and Limits of the Structural Unit Model:

Any Model has its Exceptions
The major interest of the structural units lies in its possibility to predict the grain
boundary structures, even though there are several restrictions to this predictability.
In its principle, this model applies only to series of grain boundaries in which only
one degree of freedom varies. Most simulations of the atomic structure are focused
on grain boundaries with a simple rotation axis and preserving a common median
plane. Among them, those of the symmetrical tilt boundaries are by far the most
numerous and most systematic. The changes linked to the misorientation alone have
also been studied for pure twist grain boundaries, asymmetrical tilt and, in extreme
cases, for boundaries with irrational planes. Although much less studied, the change
of the median plane orientation for a fixed grain boundary misorientation has also
been investigated; the results are often used to understand the faceting phenomenon.
Structures of different asymmetrical boundaries with the same misorientation have
been described in terms of structural units, owing to the use in the simulation of anti-
periodic conditions, with free surfaces in the direction perpendicular to the grain
boundary plane [20].
Generally, whatever the interatomic potentials used in the simulation, energy
minima occur for the same misorientations and correspond to the same grain bound-
ary structures; only the energy values differ according to the retained potential.
However, erroneous structures may be obtained if the potential does not reproduce
3.3 Interests and Limits of the Structural Unit Model: Any Model has its Exceptions 67
Fig. 3.16 a Experimental image of the (100)I //(110)II grain boundary; b simulated image starting
with the atomic positions of the simulated structure [19]
important properties of the material, in particular its stacking fault energy in the case
of metals.
The structural unit model is always applicable from a theoretical point of view and
by considering only one degree of freedom; however it has proved really useful, i.e.
with a predictive power, only for a limited number of grain boundaries. The structure
and, in certain cases, a property as the self-diffusion have been predicted only for
pure tilt boundaries and, in a more restrictive way, for pure twist boundaries around
simple axis 100, 110 or 112 in cubic crystals [21]. There are few systematic
experimental investigations of the evolution of the grain boundary atomic structures
with the misorientation. The most detailed studies deal with the symmetrical tilt
boundaries around 100 and 110 in semi-conductors [22, 23]. Less common are
the simulations for system of lower symmetry and even fewer are the observations
of grain boundaries in these systems.
Even in the cases of symmetrical tilt grain boundaries around a simple axis,
restrictions in the structural unit model appear. Indeed, when the unit distortions
become too important, the local stresses can no longer relax and the boundary may
then adopt another relaxation mode: change of rigid body translation, additional unit
not predicted, formation of facets, boundary dissociation (3D structure), steps. . . We
may formally distinguish geometrical restrictions from those linked to the material;
but, whatever its origin, any restriction is generated by an incompatibility between
structural units; it avoids the development of long-range stresses.
Among the restrictions depending on the boundary type, the limitation of the
model to rotation axes with indices equal and, a fortiori, superior to 221 or when
the boundary has a mixed (tilt/twist) character is explained by the requirement of an
important number of delimiting grain boundaries; the necessary data then exceed the
obtained information. The model may still be applied, but the content of the predic-
tions is reduced [21]. Another geometrical condition for applying the model is the
compatibility of the rigid body translations of the delimiting boundaries. Besides, the
prediction of the asymmetrical boundary structures comes up against the formation of
facets. In the twist grain boundaries, additional units so-called fillers, non-predicted
by the model, are required to fulfill the voids between the predicted units.
An important restriction depending on the material is linked to the stacking fault
energy that yields a delocalization of the relaxations associated to the distortions
and a dissociation of the grain boundary, as it was observed in metals. We must also
evoke the specific relaxations occurring in ionic materials to avoid two ions of same
sign to be neighbours and the resulting normal expansion. In covalent materials, the
formation of the covalent bonding determines the structure; the possibility or not to
have dangling bonds is then questionable.
To these restrictions for applying the structural unit model, we must add those
linked to the simulation methods: choice of the potential, limit conditions, entropy
term often neglected and questions about obtaining the true energy minimum. . .
Complex grain boundary structures, that do not obey or only partially the structural
unit model, are hereafter presented. The examples are deliberately taken from studies
of different materials (metal, oxide, semi-conductor) and the reasons for the deviation
from the model cover the various restrictions set out above.
3.3.1 Three-Dimensional Tilt Grain Boundaries in Metals
The structures of the symmetrical 110 tilt grain boundaries in aluminium (Table 3.1),
in complete agreement with the predictions of the structural unit model, concern a
metal with high stacking fault energy. As this energy decreases, the calculated struc-
tures become more complex. The comparison of Fig. 3.17a and b shows the effect of
the stacking fault energy on the evolutions of the grain boundary period and atomic
structure for symmetrical [1-10] tilt boundaries located between θ = 0◦ ( = 1)
and θ = 50.48◦ ( = 11), the symmetry median plane being (110) [9]. The = 1
(001) and = 11 (113) are favoured in both cases; the = 27 (115) boundary
is also favoured for the high stacking fault energy metals as aluminium; but it is
not favoured in the low stacking fault energy metals as copper or silver where two
additional units, A of = 1 and C of = 11, are present in the structure. The
intermediary boundaries, as = 33 (118) and = 19 (116), widen out in low
stacking fault energy metals and display an arbitrary structure with respect to the
neighbouring favoured boundaries.
The description continues for low stacking fault energy metals until θ = 93.37◦
with the occurrence of dissociated grain boundaries (Fig. 3.18) [9]. The authors
include nickel in the previous series but nickel possesses a non-negligible stacking
fault energy (of the order of 130 mJ·m−2 ) compared to that of copper (45 mJ·m−2 ).
However, more recent simulations, using a low stacking fault energy value close
to that experimentally determined for nickel, find structures identical to those of
Fig. 3.18 [13]. The units of the favoured = 11 (113), θ = 50.48◦ and
= 3 (111), θ = 109.47◦ boundaries (the latter is described in Fig. 3.8) well
appear in the nickel intermediary boundaries, but the structures of these boundaries
are unpredictable because of stacking faults that are generated in the adjacent crystals.
The boundary dissociation phenomenon is observed and analyzed in the case of
an incoherent = 3 (112), θ = 70.53◦ twin in copper [24]; it is also explained
for three simulated boundaries = 33 (225), = 3 (112) and = 17 (223)
in gold and nickel [25]. In each case, the initial grain boundary is dissociated into
Fig. 3.17 Atomic structures of symmetrical 110 tilt grain boundaries, between θ = 0◦ and
θ = 50.48◦ , simulated using an EAM potential: a for high stacking fault energy metals (as alu-
minium), from = 33 to = 11; b for low stacking fault energy metals (as copper, silver), from
= 1 to = 57. The structural units are drawn in full lines. Only the reference structures of
= 1 (001) and = 11 (113) are identical [9]
Fig. 3.18 Atomic structures of symmetrical 110 tilt grain boundaries between θ = 50.48◦ and
θ = 93.37◦ simulated with a potential based on a low stacking fault energy. The structural units
are drawn in full lines [9]. The dissociations of the = 33 (225), = 3 (112) and = 17 (223)
boundaries are underlined by dotted lines [24]. For the structure of = 3 (111) θ = 109.47◦
boundary, see Fig. 3.8
two boundaries separated by a region with a width of about 1 nm. The C units of
= 11 (113) are found on the left boundary, whereas the D units of = 3 (111),
θ = 109.47◦ are distributed along the right boundary (Fig. 3.18). The shape of the
D units of the = 3 (111) boundary is explained (see = 33 and = 3 figs)
as composed of two A (or A ) units of = 1 [71]. The dissociation occurs by
emission of stacking faults bordered by Shockley partial dislocations, the cores of
which being associated to the structural unit of = 3 (111). The faults are all the
more closely spaced than the misorientation of the intermediary boundary is far from
that of = 11 (113). Their distance corresponds to three C units of = 11 in the
= 33 (225) boundary; they are inserted between two C units in the = 3 (112)
boundary and are not separated in the = 17 (223) boundary. In the region between
the two boundaries, the (111) planes are curved [25]. In the case of the = 3 (112)
boundary in copper, this region has been considered as a slab of metastable 9R phase,
i.e. a polytype of f.c.c. structure, with a stacking fault for every three {111} planes
and described by ABCBCACAB (Fig. 3.19) [24].
Generally, the dissociated region width observed by high-resolution transmis-
sion electron microscopy is somewhat smaller than that predicted by simulation
(Fig. 3.20) [25].
Two questions must be raised: what is the limit value of the stacking fault energy
yielding or not grain boundary dissociation? Are there other factors that possibly
influence this dissociation? The consideration of the relative stacking fault energy,
compared to the shear modulus better reflects the differences between metals. Indeed,
Fig. 3.19 Relaxed structure of a {112} symmetrical till grain boundary in copper. A slab of 9R
phase extends from the initial boundary (indicated by large triangles) and the low angle boundary
resulting from by dissociation (dotted lines). The stacking faults, bordered by partial dislocations,
are indicated in full lines every three {111} planes inside the 9R phase [24]
Fig. 3.20 a Simulated image obtained on the basis of the calculated structure of the = 33{225}
grain boundary in gold; b High-resolution electron microscopy image showing a good qualitative
agreement with the simulated image, but the dissociation width is smaller. The partial dislocations
being delocalized, it is difficult to determine the exact position of the second boundary [25]
the order of the ratios (γSFE /μ)M is 10/12−1.5/2−1 with M = Al, Ni and Cu, respec-
tively; the two mentioned values account for the lack of accuracy of the µ values for
nickel and copper, due to their anisotropy.
These dissociations that also occur in asymmetrical grain boundaries are
energetically favoured [24]. The dissociation phenomenon may take a special impor-
tance for real metals where the presence of solutes or impurities generally decreases
the stacking fault energy; this must affects the grain boundary mechanical properties.
3.3.2 Asymmetrical Tilt Grain Boundaries in Metals
Before to approach the asymmetrical grain boundaries, we recall that the structural
unit model is applicable if only one degree of freedom varies; until now, it was
the coincidence index that depends on θ , while a constant median plane was
maintained. The parameter ξ defines a system in which all the grain boundaries have
the same median plane; it varies with the plane inclination. The parameters and
ξ , that are independent from each other, totally define a grain boundary, for a given
tilt axis. When ξ is constant and varies, the relaxation occurs by introduction of
intrinsic secondary dislocations. This point will be explained in Sect. 3.4, where the
structural unit (SU) model is closely linked to the intrinsic dislocations (GBD) model
to form a unique model SU/GBD. If is constant and ξ varies, the relaxation occurs
by faceting.
The structural unit model has proved to be relevant to describe the structures of
the asymmetrical [1−10] tilt grain boundaries belonging to a same system ξ = 1
with (111) as median plane; the investigated misorientations are situated between
109.47◦ and 141.06◦ corresponding to the favoured = 3(001)I //(221)II and
= 9(115)I //(111)II boundaries. The basic units are more complex than those
of the symmetrical boundaries, but the principle of structure continuity applies to
intermediary boundaries [8].
If the median plane varies for a given misorientation, the simulations that often use
periodic boundary conditions do not allow determining the intergranular structure but
may account for the faceting phenomenon [8]. The decomposition of an asymmetrical
grain boundary in symmetrical parts with same misorientation has been first predicted
for a = 5(36◦ 87 100) boundary in aluminium. The boundary displaying an
asymmetry angle of about 8◦ is formed of symmetrical {210} facets alternating with
symmetrical {310} facets. However, the decomposition no longer occurs when the
asymmetry degree exceeds a certain value [26]; beyond more complex structures
appear.
Merkle et al. have studied grain boundaries with fixed misorientations θ = 38◦ 9
( = 9) and θ = 50◦ 5 ( = 11), the boundary plane being free to adopt any inclina-
tion ϕ, symmetrical or asymmetrical [27]. All the observed boundaries present facets
with well-defined inclinations, apart from the {113} boundary that stays straight; this
is the only favoured boundary in the = 11 series. According to the calculations for
gold and copper, the asymmetrical facets = 11{225}I //{441}II and {557}I //{771}II
have energies inferior to those of the symmetrical {332} boundary, but there are no
descriptions of these boundaries in terms of structural unit [27]. For a given metal, the
energy order may be questioned if the potentials chosen for the calculations differ.
Molecular dynamic simulations, using limit conditions that maintain free the two
surfaces normal to the boundary plane, enable to propose atomic structures of some
asymmetrical coincidence or near coincidence grain boundaries [13]. Figure 3.21
shows the structure of an asymmetrical {225}I //{441}II boundary in nickel, obtained
by using a potential that well reproduces the relatively high stacking fault energy of
this metal (O. Hardouin Duparc, personal communication (2003–2005)). The bound-
Fig. 3.21 Structure of the asymmetrical {225}I //{441}II grain boundary in nickel, obtained by
molecular dynamic simulation (O. Hardouin Duparc, personal communication (2003–2005)). The
boundary is periodically formed by the units of the symmetrical = 33 {441} boundary and those
of the = 33 {225} boundary, with occurrence on the {225} side of a stacking fault similar to that
of the symmetrical {225} boundaries in low stacking fault energy metals [9]
ary is formed by the |AE A| units (half-period) of the symmetrical = 33 {441}

boundary (Fig. 3.8) and by the |C D| units of the = 33 {225} boundary (Fig. 3.18).
In particular, the stacking fault that appears on the {225} side in the crystal is similar
to that obtained by simulation of the structure of the {225} boundary in low stacking
fault energy metals [9].
From a practical point of view, a bicrystal with a high-energy symmetrical tilt
boundary is not easy to manufacture; most often, the mean plane deviates from
the symmetrical position. This is the case when we attempt to manufacture a
= 11 {332} in nickel by uniaxial growth; apart from the region of the ingot very
near the seed, the bicrystal presents various facets; the boundary plane is often a dense
{111} plane in one crystal. The {111}I //{331}II facet seems to be favoured although
it does not belong to the = 11 boundary family, its misorientation being 131.47◦ .
This incommensurable facet is most likely described by a quasi-periodic arrange-
ment of structural units. The commensurable {111}I //{13, 13, 5}II facet, very near the
incommensurable one, belongs to the = 11 family (θ = 129◦ 52). The energies of
these two asymmetrical boundaries and those of the symmetrical boundaries between
= 3 et = 9 (among them {13, 13, 5} et {331}), are calculated; their structures
are simulated in order to understand the development of the {111}I //{331}II facet
(Table 3.2) [11, 13]. To simulate this incommensurable boundary, an approximant
5/2 is used provided a very small lattice mismatch δ = 0.0066.
Although the structures of the two asymmetrical grain boundaries are complex,
sequential arrangements may be determined; they are composed of the A and E units
of the symmetrical boundaries of the same series (see Fig. 3.8). The choice of the
74
Table 3.2 Description, in terms of structural units, of symmetrical and asymmetrical 110 tilt grain boundaries with their respective energies in nickel. To
simplify, the orientation (±) of the units has been omitted [11, 13]
Grain boundary plane θ◦ Period Energy (mJ·m−2 )
{331}I //{331}II 19 153.4 |E AE A| 1095
{13, 13, 5}I //{13, 13, 5}II 363 149.47 |E A A AE A A AE A.E A A AE A A AE A| 1161
|E E 1 AE E 1 AE A.E E 1 AE E 1 AE A|
{552}I //{552}II 27 148.41 |E A A AE A A A| 1149
|E E 1 AE E 1 A|
{221}I //{221}II 9 141.06 |E E| 1146
{332}I //{332}II 11 129.52 |E D E D| 970
{111}I //{111}II 3 109.47 |D D| 42
{111}I //{331}II Incom 131.47 |E A A AE A A A A| 1074
|E A A AE AE 1 A|
{111}I //{13, 13, 5}II 11 129.52 |E A A A A A A A A A A.E A A A A A A A A A A| 1094
|E A A A A AE 1 A A A.E A A A A AE 1 A A A|
3 Atomic Order of Grain Boundaries
Fig. 3.22 Simulated structures of asymmetrical tilt grain boundaries: a = 11

{111}I //{13, 13, 5}II boundary; b near = 11 {111}I //{331}II boundary(two hatched A units
constitute a distorted E unit called E 1 in Table 3.5); c experimental image of a {111}I //{331}II facet,
the structure of which is in agreement with (b) [13]
descriptions in terms of structural units of the grain boundaries in metals may be

arbitrary. Table 3.2, established on the basis of simulations, presents for certain
boundaries two descriptions with a different hierarchy order, a distorted E unit
named E 1 being equivalent to two deformed A (or A ) units. The description using
E 1 is more consistent with the principle of continuity of the structural unit model;
indeed, this principle implies that the proportion of E units increases from 1/2 in
= 19 until 1 in = 19. For example, the period of the symmetrical {13,
13, 5} boundary is better described by |E E 1 AE E 1 AE A.E E 1 AE E 1 AE A| than
|E A A AE A A AE A.E A A AE A A AE A| because the percentage of E units is higher
in the first description and in agreement with the proximity of the = 363 boundary
from = 9. The following boundaries in Table 3.2 may also be described using the
E 1 unit. The period of the = 27 {552} boundary represented by |E E 1 AE E 1 A|
is then similar to that of the same boundary in silicon [LLCLLC]. But, whatever
the chosen description, the symmetrical {13, 13, 5} boundary is well composed of
two periods of the = 27 {552} boundary and one period of the = 19 {331}
boundary: 27E A·27E A.
The energies of the two asymmetrical grain boundaries are close to each other
and similar to that of the symmetrical {331} boundary; they are also significantly
more elevated that the symmetrical = 11 {332} energy, but inferior to the other
symmetrical boundary energies. The period of the {111}//{311} boundary, much
smaller than that of the {111}//{13, 13, 5} boundary, turns out to be the observed
configuration; indeed, this configuration reported on Fig. 3.22c is identical to the
description implying the E 1 unit (Fig. 3.22b). A detailed analysis of the calculated
and experimental images enables to better describe the structural unit arrangement on
the asymmetrical {hhl} // {111} boundaries. On the {111} side of the boundary, and
whatever the opposite plane, a regular arrangement of A units is clearly evidenced.
Remember that the A unit may be seen as half a D unit of the symmetrical = 3 {111}
boundary. The sequences proposed in Table 3.2 appear on the simulated images on
the {331} and {13, 13, 5}, respectively (Fig. 3.22). The higher number of A units on
the {hhl} side, compared to their number in the symmetrical {hhl} boundary, may
result from the necessary additions of A units as second halves of the D units of the
opposite {111} plane [11].
The asymmetrical boundaries seem to reproduce the dissociations that appear in
the corresponding symmetrical ones, as it is the case for {225}//{441}. A 9R slab
has been observed in several asymmetrical boundaries near = 3 (112) in silver,
but in that case the 9R structure is faulted [27].
In conclusion, the simulations and the observations of the asymmetrical tilt grain
boundaries seem to converge to describe them either as composed by alternating
symmetrical facets, or by using the structural units that appear in the symmetrical
tilt boundaries.
3.3.3 Twist Grain Boundaries in Metals
From the outset of the structural unit model, it has been applied to twist grain bound-
aries [28]. These boundaries being hardly observable by high-resolution microscopy,
information on their structures often comes from simulations. The pavement of the
boundary plane by the structural units of the delimiting boundaries is never perfect
and requires additional unit called fillers to fulfil the voids between the basic units
(Fig. 3.23) [29].
The structure of a filler unit cannot be determined by simulation; in contrast, its
size may be known as it depends on the A and B unit sizes. The structural unit model
cannot totally predict the structure of a twist boundary (and mixed a fortiori). Larger
are the filler units less predictive is the model.
3.3.4 Grain Boundaries in Covalent Materials
In covalent materials, the atomic structure of a grain boundary is strongly linked to

its electronic structure. The formation of the covalent bond determines the atomic
structure. However, we cannot exclude the occurrence of dangling bonds associated
with electronic states in the fundamental energy gap. The tilt grain boundaries in
semiconductors, particularly silicon and germanium, were the first to be studied by
transmission electron microscopy; this is due to their relatively large parameters and,
probably also, because they can be obtained with a high degree of purity, leaving hope
for a good fit between the manufactured boundary and the intended one. In any case,
the atoms have been found tetra-coordinated: a reconstruction of the boundary occurs
such as there are no dangling bonds. There are no intrinsic electronic states localized
at the tilt boundaries. However, the structural units going by pair in semiconductors,
the continuity principle cannot always be strictly applied.
Table 3.3 shows the evolution of the structures for a series of symmetrical 110
tilt grain boundaries in semiconductors [23]. It is partly similar to that established for
metals (Table 3.2) with an additional low energy favoured boundary = 9 {221},
the unit of which being not divisible in one unit of = 1 {110} and one unit of
{111}. We recall that in metals, = 9 with a non-negligible energy may constitute
a delimiting boundary.
The C unit of the = 1 {110} boundary is composed of two six-atom rings;
two units, one being translated with respect to the other along the tilt axis by
τ = a/4110, are necessary to describe the period |C.C|. The period of the
= 3 {111} is formed by two T units, each of them is constituted by a boat-
shaped six-atom ring; there is also a translation of one unit relative to the other,
thus the boundary is represented by |T.T |. The = 9 {221} boundary has two
stable structures, the periods of which are constituted by two L units or two M units
symmetrically orientated with respect to the boundary plane: |L + L − | or |M + M − |.
Each unit consists of a five-atom ring and a seven-atom ring, but the connection of
the cycles depends on the type of unit L or M (Fig. 3.24) [23]. The L and M units occur
in the boundaries with misorientations superior or inferior to 129.52◦ , respectively
(Table 1.6). The presence of only one C or T unit within a sequence results in a
translation along the tilt axis; the period of the corresponding grain boundary is then
doubled.
The intermediary boundaries between = 1 and = 9 perfectly obey the
structural unit model. On the contrary, an anomaly appears in the = 11B {332}
Fig. 3.23 a Relaxed structure of a = 85, 8.8◦ 001 twist grain boundary in copper. The vectors
of the CSL lattice (1/2 760) delimit the period. 25 A units of = 1 predominantly compose the
period, in agreement with the low misorientation angle. At the corner of the period cell there is a B
unit of = 5. The units called fillers are localized between two B units along the square formed
by the A units [29]
Fig. 3.24 Shapes of the structural units of the favoured = 1 (a) = 9 (b) and = 3
(c) boundaries in silicon and germanium [23]
structure and in the two following boundaries of the series with the occurrence of an
additional P unit non-predicted by the model. This unit is also formed by five and
seven-atom rings, but differently connected comparatively to the M and L units. It has
been found in a high-energy = 3 boundary. The A structure is associated to glide-
mirror plane whereas the B structure is found in a pure mirror plane (Fig. 3.25) [10].
In the latter case, the additional P unit accounts for the incompatibility between the
rigid body translations of the two favoured units M (τ = 0) and (τ = 1/4 110).
The = 11A structure is stable in silicon at low temperature while the B structure
may appear at high temperature. The situation is inversed in germanium in which the
two structures have been observed. This finding is hardly explainable by the energy
differences of the two structures at low temperature (in mJ·m−2 ): 467 for = 11A,
478 for = 11B in silicon and 373 for = 11A, 366 for = 11B in germanium,
respectively [30].
By considering that P is a possible = 3 unit, then all the series on Table 3.3
verifies the structural unit model; but the model does not predict certain structures
deduced from the calculations and observed.
The prediction by simulation of the structural units with tetra-coordinated atoms
in twist grain boundaries of semiconductors is a difficult task. These boundaries are
relatively disordered; their energies are higher by a factor of two to four, than the
energies of the tilt boundaries.
Table 3.3 Periods and structures in terms of structural units of a series of symmetrical 110 tilt
boundaries in silicon and germanium (The period of the = 123 boundary may also be composed
by two identical sequences of units with inverse signs) [23]
θ◦ Grain boundary plane Structure
180 1 {110} |C.C|
163.9 51 {551} |L + CC L − CC|
153.47 19 {331} |LC.LC|
148.41 27 {552} |L + L − C L − L + C|
141.06 9 {221} |L + L − | or |M + M − |
129.52 11 A {332} |M + T M − T |
11 B |M + T M − P + M − T M + P − |
126.41 123 {775} |M − P + P − M + T.M − P + P − M + T |
124.12 41 {443} |M + P − P + M + M − T T |
109.47 3 {111} |T.T |
Fig. 3.25 Simulated and observed structures of the = 11A and = 11B grain boundaries in
silicon and germanium [10]
3.3.5 Grain Boundaries in Ionic Materials
The experimental studies and by simulation of the grain boundaries in ionic materials
are relatively less numerous than those developed for metallic and covalent materials,
for the following main reasons:
• The crystallographic structures of these materials are generally more complex
(lower symmetry and/or a great number of atoms by unit cell)
• The strong repulsive force between ions of the same sign leads to relaxations
and reconstructions to avoid ions of like sign being nearest neighbours. This
results in a non-negligible expansion and grain boundary structures more open
than those occurring in metals. The relaxations differ according to the bound-
ary plane character: checkerboard (alternating anions and cations in the plane)
or double-layer (planes uniquely composed of anions alternating with planes of
cations). The long-range interactions make the simulations more complex com-
pared to those in metals and semiconductors where the interactions are felt at
relatively short distance.
• The boundary core may acquire a charge that must be compensated by a space
charge extending at large distances from both sides of the boundary. The ionic
charge may also not be constant in the boundary.
These restrictions result in rare agreements between the simulated and the
observed structures, compared with the relatively good fit in metals. They explain
why the majority of the studies have been performed on simple ionic oxides (MgO
or NiO) and, apart from some exceptions, on structures with a small elemental unit
cell. The simulated structures of the symmetrical [001] tilt boundaries in NiO [31]
are often more open than those observed by high-resolution electron microscopy, as
illustrated for = 5 (210) on Fig. 3.26. This dichotomy has been explained on the
basis of electronic images by the presence of vacant sites along the atomic columns
parallel to the tilt axis [32]. The proportion of vacant sites, that may reach 25 % in
certain boundaries, is not considered in the simulations. We must also remark that
the observed structures are obtained from bicrystals that have been quenched from
Fig. 3.26 Structures of symmetrical = 5 [001], (210) tilt grain boundary in NiO: a simulated
image (atomic columns in black) starting from the calculations of Duffy and Tasker [31]; b image
in transmission electron microscopy showing a denser arrangement of the atoms in the boundary
core than that deduced from the simulations [32]
the high temperatures while the simulated structures were determined by energy
minimization at 0 K.
The other 001 tilt boundaries, symmetrical or not, are generally faceted with
a {100} facet in a grain, at least [32]. This facet is the densest plane whatever the
misorientation.
Finally, Duffy and Tasker show that the expansion occurring at the core of a
double-layer boundary produces a positive charge on one side of the boundary and a
negative charge on the other side; An electric potential difference is then created that
may reach about 2 V for the symmetrical {311} in NiO [33]. The charged defects,
present in the crystals, attempt to reduce this difference by forming regions with a
space charge on each boundary side. The double-layer plane plays the role of sinks
and sources for point defects.
When the oxide has a more complex structure, the energy and atomic force
calculations, using ab initio band structure method based on density functional theory
(DFT) in the local density approximation (LDA) proved to be quantitatively more
accurate than atomic simulations. For the rhombohedral twin in alumina, these cal-
culations predict that the grain boundary plane of lower energy is situated on the
octahedral interstitial vacancies of the lattice. Two metastable structures of this type
are determined with a clear preference for the structure of symmetry n (twofold
symmetry axis with glide) or screw twin that does not present any expansion and
possesses a very low energy; comparatively, the glide-twin has a non-zero expansion
and a high energy (Fig. 3.27) [34]. In the n symmetry structure, the oxygen ions form
a configuration very similar to that they have in the crystal.
These predictions are in good agreement with the interpretations of the electron
microscopy images following an approach based on the symmetry groups [35]. The
Fig. 3.27 a Projection along [12-10] of the initial structure, considered in the simulations, of the
rhombohedral twin with a vacancy plane in alumina; b, c projections along [10-11] of the twin after
relaxation for two expansion states τ ⊥ : τ ⊥ = 0 (b) and τ ⊥ = 0.03 aalumina (c). The aluminium
atoms are represented by small black full circles, the oxygen atoms by large grey full circles with
decreasing sizes along the projection axis [34]
phase image obtained by a technique of reconstruction of the complex electron exit

wave function from a focal series of images directly shows that the rhombohedral
twin is well situated on the aluminium vacancies (Fig. 3.28) [36]. On the image,
small dots correspond to oxygen columns and large dots to aluminium columns; the
distance between Al and O columns is 85 pm.
The use of the Z contrast imaging in a scanning electron microscope, combined
with a technique of electron energy loss spectroscopy, allows Browning et al. to
Fig. 3.28 Phase image of the electron exit wave showing atomic resolution of the rhombohedral
twin in alumina (see text). The boundary plane is well situated on the aluminium vacancies. An
extrinsic dislocation (see part 2) associated to a step may be also visualized [36]
Fig. 3.29 Structural units of the delimiting boundaries that serve to build the structures of all the
[001] tilt boundaries in SrTiO3 . The E and E ∗ units correspond to misorientations 0◦ and 90◦ of
the (100) and (110) planes of the single crystal, respectively. The F and F ∗ units form the = 17
boundary with a misorientation 28.07◦ and a (410) plane; the G and H units are attached to the
= 5, 36.87◦ (310) and 53.13◦ (210), respectively. (The hatched circles indicate columns partially
occupied by atoms) [37]
determine the structures of a large number of symmetrical [001] tilt grain boundaries
in the oxide SrTiO3 [37]. Six elemental units of favoured boundaries enable to build
the structures of all the boundaries, with a complete agreement with the continuity
principle of the structural unit model (Fig. 3.29). The sites of certain columns are
only partially fulfilled to avoid the repulsive interactions between the ions of like
sign in the boundary core, as it was already suggested for NiO [32]. The structures
of the = 13 (67.38◦ ) and = 25 (73.74◦ ) display the periods |E ∗ H | and
|E ∗ H E ∗ | constituted by the somewhat distorted units of the = 5 and = 1
delimiting boundaries. A structural refinement based on valence bond model calcu-
lations shows that all the grain boundaries analysed are neutral and stoichiometric,
provided small displacements (<0.02 nm) of the metallic ion columns; an example
is given on Fig. 3.30.
Fig. 3.30 a Structure of the = 13, 67◦ [001] {320} grain boundary in SrTiO3 : a derived from
the Z contrast images; b after refinement of the structure by valence bond calculations (identical
symbols than on Fig. 3.29) [37]
To conclude, although various intergranular structures presented in this section

do not strictly obey the structural unit model, the latter remains as the reference to
which it is useful to refer to describe grain boundaries, otherwise predict them.
3.4 Structural Unit Model/Intrinsic Grain

Boundary Dislocations
For clarity reasons, we choose to present separately the Structural Unit/ Grain
Boundary Dislocation (SU/GBD) model that links the structural units to the intrinsic
dislocations. However, as early as 1959, Hornstra associates a dislocation to a struc-
tural unit [1]. This link is also present in the original paper by Sutton and Vitek
[8]. These authors not only calculate the interfacial energy leading to the determina-
tion of grain boundary structure, but also describe the hydrostatic stress fields that
reveal a strong relationship between the minority units and the intrinsic secondary
dislocations.
3.4.1 Principle of the SU/GBD Model
The principle of the SU/GBD model may be expressed as follows:

In a long period grain boundary such as |A A A AB. . .|, each minority unit B is
the core of an intrinsic secondary dislocation, the Burgers vector of which belonging
to the DSC lattice of the delimiting boundary composed by the majority units A
(Fig. 3.31).
Fig. 3.31 Schematic drawing
of the relationship between
the descriptions of a grain
boundary in terms of structural
units and in terms of intrinsic
dislocations. The secondary
dislocation network attached
to the B units constitutes a
sub-boundary in the favoured
grain boundary composed of
A units
The Burgers vector b A of a secondary dislocation may be expressed with respect

to the length of the minority structural unit ν B and to the misorientation angles θ A and
θ B of the delimiting boundaries; by referring to Fig. 3.12, to expressions (3.4) and
(3.5) and by applying the Frank formula (2.6), we obtain for one boundary period:
b A = 2νB sin((θ B − θ A )/2) (3.13)
This equation implies a choice of the delimiting boundary A. This may be related to
the interest, explained in Sect. 3.2.2, to surround the intermediary grain boundary by
two delimiting boundaries with misorientations as close as possible from each other,
in order to reduce the distortions of the units. The values of the Burgers vector of a
secondary dislocation associated to a structural unit in various symmetrical tilt grain
boundaries in silicon will be presented later (See Sect. 8.3.2).
The intrinsic secondary dislocations localize the angular deviation between the
intermediary boundary and the delimiting boundary composed by the majority units.
Better is the localization of the dislocations less the structural units are distorted.
If this localization is not physically realised in an intermediary boundary, then the
SU/GBD model is only a geometrical description without any relation with the atomic
relaxations in the boundary.
The favoured boundary is free from secondary dislocations. It may even be con-
sidered as free of any dislocation if it can be described as an invariant plane under
3.4 Structural Unit Model/Intrinsic Grain Boundary Dislocations 85
a simple shear. Alternately, a favoured grain boundary may be described as an

arrangement of primary dislocations evenly spaced and very close to each other, thus
hardly observable by transmission electron microscopy. The = 3 {111} boundary,
described by a rotation of 70.53◦ around 110 (or 60◦ 111) is also obtained by
a shear parallel to the {111} planes. Although described as free of dislocations, a
favoured boundary is then not free of short-range stresses.
In the same way that it exists a hierarchy of the grain boundary descriptions
in terms of structural units, there is also a hierarchy of their description in terms
of dislocations. On Fig. 3.7 (see Sect. 3.2.2), the dislocation structure hierarchy is
associated to the structural unit hierarchy for the tilt = 73 (001) boundary [12]. The
dislocations in the second or third order description are considered as disturbances in
the arrangement corresponding to the first order. The a/2 [110] Burgers vectors of the
secondary dislocations that occur at the first order are associated to the majority units
of the (001) plane of the single crystal; at the second order, the Burgers vectors are a/5
[210] and at the third order they become equal to a/10 [310]. Only the observation
of the dislocations may specify what structure the grain boundary adopts.
3.4.2 Characterization of the Dislocations Associated

to Structural Units
Expressions (2.14) give the Burgers vector b and the associated step height h of an
intergranular dislocation: b = sI − sII and h = n(sI + sII )/2 [38].
An algebraic method had been developed to attribute a Burgers vector to each
structural unit. A couple of vectors s I and s I I is attached to each unit; this couple
does not depend on the grain boundary in which the unit is situated. The vector length
is that of the structural unit period. Thus a unit may be described by a (2 × 2) matrix,
the components of which are those of the s I and s I I vectors in the coordinate systems
of crystal I and crystal II, respectively [10]:
s I = m 1 a1 + n 1 b1
s I I = m 2 a2 + n 2 b2 (3.14)
The corresponding structural unit is defined:

m1 m2

SU = (3.15)
n1 n2
For pure tilt boundaries, the basic units may be chosen such as their sum is equal
to the period of the = 1 boundary (the crystal). For example, a1 = 1/4 [2-11]I ,
b1 = 1/4 [21-1]I , a2 = 1/4 [2-11]II , b2 = 1/4 [21-1]II with a + b = [100]. This
method has been applied to symmetrical tilt boundaries in silicon and germanium.
Figure 3.32 shows the couple of vectors (sI , sII ) for the T units of = 3 and M +
Fig. 3.32 Determination of the couple of vectors (sI sII ) for the structural units T and M + in silicon
and germanium [10]
Table 3.4 Matrix notation of the structural units in some delimiting boundaries in the series of the
110 tilt boundaries with {110} as median plane [10]
C+ C− L+ M+ L− M− P+ P− T
00 11 01 12 −1 0 12 01
11 00 21 10 2 1 0 −1 10
Table 3.5 Matrix notation of the periods of several symmetrical 110 tilt boundaries with {110}
as median plane
=1 = 19∗ = 27 =9 = 337 = 59 = 11A = 11B∗∗ = 3∗
11 24 37 13 5 19 14 15 2 10 02
11 42 73 31 19 5 41 51 10 2 20
The symbol * indicates that the period is constituted by two identical units that are translated from
one to another; the symbol ** indicates that the period is doubled compared to that of the CSL
lattice (see Table 3.3) [10]
units of = 9; Table 3.4 gives the different matrices associated to the structural
units of the delimiting boundaries for the series of 110 tilt boundaries in these
semiconductors [10].
This matrix notation enables operations on the structural units that will take a
special interest for introducing defects in the grain boundary structure (Part II). The
matrix linked to the sum of the structural units is the sum of the matrices of each
unit, in agreement with the conservation rule of the Burgers vectors and step heights.
Then for a boundary with period |A A AB|, we have in each grain i: s i |A A AB| =
s i |A| + s i |A AB|. A symmetrical boundary is defined by a period in which the
structural unit (or the unit group) is represented by a symmetrical matrix:

pq

SU = (3.16)
q p
Table 3.5 gives the matrices associated to the periods of several symmetrical tilt grain
boundaries. When the components p and q are even, the resulting period is made of
two groups with a same content in structural units or is doubled with respect to the
CSL lattice period.
A boundary with a mirror plane is made of two groups of structural units related
by a mirror symmetry, its period is such that:

p q m 1 m 2 n2 n1
= + (3.17)
q p n1 n2 m 2 m 1
This is the case of the = 9 boundary made of two M + M − units or two L + L −

units.
Using now the vectors sI and sII in Eq. (2.14), we obtain the Burgers vector b of a
dislocation associated to a structural unit. If sI and sII are indexed in their respective
coordinate system and that RI is the rotation that relates crystal I to crystal II, then
the primary Burgers vector (in the crystal I coordinate) is:
b = sI − sII = sI − R−1
r sI (3.18)
In expression (3.18), b being a perfect Burgers vector of the crystal, sI must belong
to the CSL lattice and thus the rotation Rr is that of a reference lattice (subscript r
for reference). Note however that even if the structural unit does not correspond to a
dislocation array, a primary pseudo Burgers vector can be found.
To get the Burgers vector bS of a secondary dislocation, we must define the vectors
s in the bicrystal, i.e. sII must be replaced by Re sII (Re is the rotation between the
crystals for the studied boundary near the reference), then we have:
bs = sI − Re sII (3.19)
If Re = Rr in Eq. (3.19), then bs = 0; this is normal as no secondary dislocation

is associated to a period of the reference boundary; all the symmetrical boundaries
in Table 1.8 are such as bs = 0. Otherwise, if the studied boundary differs from the
reference (Re = Rr ) and if we substitute sII in (3.19) by its expression used in (3.18),
we obtain an equation analogous to that defining the 0-lattice (1.23):
bs = sI − Re Rr−1 sI = (I − D−1 )sI . (3.20)
3.4.3 Application of the SU/GBD Model to Tilt Grain

Boundaries
The localization of the secondary dislocations in a tilt boundary is deduced from its
hydrostatic stress field. Figures 3.33 and 3.34 give two examples of structures and
stress distributions for [1-10] tilt boundaries in aluminium: the favoured = 11
(113) and the general = 89 (229) [8]. On these maps, an arrow centred on each
atomic site represents the hydrostatic stress at this site; arrows pointing to the right
denote sites in compression and those to the left sites in tension. The length of an arrow
Fig. 3.33 Hydrostatic stress field map for the [1-10] = 11 (113) tilt boundary in aluminium.
(The boundary structure is given on Fig. 3.4) [8]
indicates the relative stress magnitude. For a favoured boundary with period |C.C|,
each unit C bounded by the vectors 1/2 [1-10] and 1/4 [33-2] in the boundary plane
may be formally associated to a primary dislocation, the Burgers vector of which
being equal to a translation vector of the crystal. But the absence of significant over-
concentration of stresses in the boundary and the small spacing (d = a/4 [33-2])
of the primary dislocations make preferable to consider the boundary as resulting
from a shear (as indicated by the arrows in Fig. 3.33). In the = 89 (229) boundary,
there is a minority of C units, the distance between them is equal to one boundary
period d = a/2 [99−4]; their localization, at the extremity of the {115} plane of the
two grains clearly reveal their association to edge dislocations with a Burgers vector
equal to the double of the {115} plane spacing: d = 2a/27 [115]1 (Fig. 3.34a). The
magnitude of this vector is also the double of that of the elemental DSC vector of the
= 27 boundary. The location of the secondary dislocation cores is highlighted by
the local concentration of hydrostatic stresses (Fig. 3.34b).
In semiconductors, the = 27 {552} boundary with the |L + L − C L − L + C|
period (Table 3.3) is formed of C and L units of = 1 and = 9 favoured
boundaries, but also of one period |LC| of = 19 and one period L of = 9. The
Fig. 3.34 a Atomic structure and b hydrostatic stress field map for the [1−10] = 89 (229) tilt
boundary in aluminium [8]
= 19 boundary may be considered as a low angle grain boundary, constituted of

Lomer edge dislocations a/2 110 characterized by a core L.
The observation, in high-resolution transmission electron microscopy, of the local-
ization of an grain boundary dislocation and its association to a structural unit is
relatively easy in symmetrical tilt boundaries but it becomes problematic in asym-
metrical and general boundaries, the latter being described as quasi-periodic.
3.4.4 Limits of the SU/GBD Model for the Twist Grain

Boundaries
The shear stress map gives the localization of the screw dislocations in twist grain
boundaries. Figure 3.35 shows the stress distribution in a = 85 twist boundary
with a low angle (θ = 8.8◦ ) around [001] that must be compared to the structure
of this boundary (Fig. 3.23) [29]. The stress distribution reveals that the filler units
correspond to the core of screw dislocations of Burgers vector b = a/2 110 and
the minority units are located at the intersections of these dislocations. The SU/GBD
model cannot totally predict the atomic structures, but the localization of the screw
dislocations are determined owing to the fact that they bound the regions formed by
the majority units and that they cross each others over the minority units.
The low angle 001 boundary in silicon presents a grid of screw dislocations with
Burgers vector a/2 110, their intersections may correspond to several non-tetra-
coordinated structures with similar energies suggesting that these regions are close
to an amorphous state. On the contrary, in the 90◦ 001 boundary, the structural
units are localized at the intersections of the screw dislocations, the atoms being
coordinated to four neighbours [39]. Finally, in the twist boundaries of high energy
(close to 1.4 J·m−2 for = 29 (100) in silicon and 1 J·m−2 for = 11 (110)
Fig. 3.35 Shear stress map of

the = 85 (001) (θ = 8.8◦ )
twist grain boundary (see
Fig. 3.23) [29]
in palladium) equilibrated at high temperature then slowly cooled, the molecular

dynamic simulations reveal the presence of a confined amorphous layer of width
inferior to 1 nm [40]. The twist boundaries with lower energies (about 640 mJ·m−2 in
silicon) are crystalline. The introduction of an amorphous layer provokes a widening
of the boundary core and a decrease of the grain boundary energy compared to
the characteristics of the same boundary at 0 K. Furthermore, in the presence of
this amorphous layer, the boundary does not evolve as a crystalline solid when
the temperature increases (see Sect. 4.1). The existence of a disordered structure
in a = 5 (100) twist boundary in germanium, similar to that of the amorphous
germanium bulk, has also been determined by calculations entirely based on quantic
mechanics [41].
Most of the observations in electron microscopy of the 001 twist boundaries
show that the interface is relaxed with well-ordered arrangements of screw dislo-
cations. Very often, the absence of experimental information on the exact atomic
arrangement does not allow to deny or to confirm the local disorder in some regions.
However, in certain (100) twist boundaries in gold, a disorder has been observed in
the close vicinity of the boundary plane by high-resolution microscopy. We espe-
cially mentioned the contradictory results on the twist boundary structures as they
well illustrate the underlying question in the chapter title: From grain boundary
order to disorder. Indeed, if we admit that simple (001) twist boundaries in the cubic
symmetry may be disordered, what about the general grain boundaries with a mixed
(tilt/twist) character, the energies of which reaching elevated values (γ > 1 J·m−2 )
(see Sect. 4.7).
Finally, the usefulness of the SU/GBD model to describe mixed grain boundaries
is very limited; the cores of the secondary dislocations are too large and the energy
cusp for each delimiting boundary is too weak to be detected. This is also true
for pure tilt and twist boundaries when the rotation axis indices increase. There-
fore, the SU/GBD model presents the same restrictions than the structural unit
model: it is only relevant for pure tilt or twist boundaries with 100, 110 or
111 rotation axis and in majority for cubic materials; the close relationship
between a structural unit and a dislocation with its Burgers vector will be more
thoroughly approached when introducing a linear defect in the boundary structure
(Part II).
3.5 Structural Unit/Disclination Model
This model combines the description of the boundary in terms of structural units and
its mesoscopic approach in terms of disclinations [42]. Although presented in the
cases of symmetrical tilt boundaries, it may be extended to general boundaries. In a
tilt boundary, the junction between two structural units, each of them being charac-
teristic of a delimiting boundary dislocation, is a disclination. Thus, a disclination
dipole may be associated to each structural element and the boundary is described
as disclination dipole wall (Fig. 3.36).
3.5 Structural Unit/Disclination Model 91
Fig. 3.36 Diagram

showing the boundary
structure in terms of
disclinations localized
at the junction between two
different units. Between these
defects, the arrangements of
identical units are disclination
dipoles and the boundary is a
dipole wall [43]
The boundary period H is equal to the sum of the lengths u A , v B of all the
dipoles contained in H, corresponding to the regions of misorientations θ A and θ B ,
respectively. H is geometrically deduced by taking into account the distortions of the
structural units in the boundary with a mean misorientation θ [43]:
H sin(θ/2) = mu A sin(θA /2) + nνB sin(θB /2) (3.21)
The mean misorientation is:

muA sin(θA /2) + nνB sin(θB /2)sin θ/2
sin θ/2 = (3.22)
muA + nνB
By combining the grain boundary descriptions at the mesoscopic and at the atomic
scales, the structural unit/Disclination model proves its interest to derive the grain
boundary energy (see Sect. 4.3.1).
References
1. J. Hornstra, Physica 25, 409 (1959)

2. M.F. Ashby, F. Spaepen, S. Williams, Acta Metall. 26, 1647 (1978)
3. R.C. Pond, D.A. Smith, V. Vitek, Scr. Metall. 12, 699 (1978)
4. V. Vitek, A.P. Sutton, D.A. Smith, R.C. Pond, in Grain Boundary Structure and Kinetics, ed.
by R.W. Balluffi (ASM, Metals Park, 1980), p. 115
5. J.D. Bernal, Proc. Roy. Soc. Lond. A 280, 299 (1964)
6. H.J. Frost, M.F. Ashby, F. Spaepen, in Grain Boundary Structure and Kinetic, ed. by R.W.
Balluffi (ASM pub., Metals Park, 1980), p. 149
7. G.H. Bishop, B. Chalmers, Scr. Metall. 2, 133 (1968)
8. A.P. Sutton, V.Vitek, Phil. Trans. R. Soc. Lond. A 309, 1, 37, 55 (1983)
9. J.D. Rittner, D.N. Seidman, Phys. Rev. B 54, 6999 (1996)
10. J. Thibault, J.L. Putaux, A. Jacques, A. George, H.M. Michaud, X. Baillin, Mat. Sci. Eng.
1164, 93 (1993)
11. O. Hardouin Duparc, S. Poulat, A. Larere, J. Thibault, L. Priester, Phil. Mag. A 80, 853 (2000)
12. E.P. Kwan, R.W. Balluffi, Phil. Mag. A 56, 137 (1987)
13. O. Hardouin Duparc, A.Larère, S. Poulat, L. Priester, J. Thibault, Interf. Controlled Mater. 9,
231 (2000)
14. G.J. Wang, A.P. Sutton, V. Vitek, Acta Metall. 32, 1093 (1984)
15. A.A. Nazarov, A.E. Romanov, R.Z. Valiev, Phys. Stat. Sol. (a) 122, 495 (1990)
16. A.P.Sutton, Acta Metall. 36, 1291 (1988)
17. D. Gratias, A. Thallal, Phil. Mag. Lett. 57, 63 (1988)
18. A.P. Sutton, Prog. Mat. Sci. 36, 167 (1992)
19. J.M. Pénisson, F. Lançon, U. Dahmen, Mat. Sci. Forum Trans. Tech. Pub. 294–296, 27 (1999)
20. O. Hardouin Duparc, M. Torrent, Interf. Sci. 2, 7 (1994)
21. A.P. Sutton, Phil. Mag. Lett. 59, 53 (1989)
22. A. Bourret, J.L. Rouvière puis, in Polycrystalline Semiconductor Grain Boundaries and Inter-
faces, ed. by H.J. Möller, H.P. Strunk, J.H. Werner (Springer, Berlin, 1989), pp. 8–24
23. J. Thibault, J.L.Rouvière, A. Bourret, in Handbook of Semiconductor Technology: Electronic
Structure and Properties of Semiconductors, vol. 1, ed. by K.A. Jackson, W. Schröter (Wiley
VCH Pub., Weinheim, 2000), pp. 377–451
24. U. Wolf, F. Ernst, T. Muschik, M.W. Finnis, H.F. Fishmeister, Phil. Mag. A 66, 991 (1992)
25. J.D. Rittner, D.N. Seidman, K.L. Merkle, Phys. Rev. B 53, 4241 (1996)
26. M. Biscondi, J. Phys. Colloq. France 36, C4-58 (1975)
27. K.L. Merkle, D. Wolf, Phil. Mag. A 65, 513 (1992)
28. A.P. Sutton, Phil. Mag. A 46, 171 (1982)
29. D. Schwartz, V. Vitek, A.P. Sutton, Phil. Mag. A 51, 499 (1985)
30. J. Wilder, H. Teicher, Phil. Mag. Lett. 76, 83 (1997)
31. D.M. Duffy, R.W. Tasker, Phil. Mag. A 47, 817 (1983)
32. K.L. Merkle, Interf. Sci. 2, 311 (1995)
33. D.M. Duffy, R.W. Tasker, J. Appl. Phys. 56, 971 (1984)
34. A.G. Marinopoulos, C. Elsässer, Acta Metall. 48, 4375 (2000)
35. S. Lartigue Korinek, S. Hagège, Mater. Sci. Forum 294–296, 281 (1999)
36. S. Lartigue-Korinek, S. Hagege, C. Kisielowski, A. Serra, Phil. Mag. 88, 1569 (2008)
37. N.D. Browning, S.J. Pennycook, M.F. Chisholm, M.M. McGibbon, A.K. McGibbon, Interf.
Sci. 2, 397 (1995)
39. A.Y. Belov, K. Scheerschmidt, K. Gösele, Phys. Stat. Sol. (a) 171, 1 (1998)
40. P. Keblinsky, D. Wolf, S.R. Phillpot, H. Gleiter, Phys. Rev. Lett. 77, 2965 (1996)
41. E. Tarnow, P. Dallot, P.D. Bristowe, Phys. Rev. B 42, 3644 (1990)
42. V.Y. Gertsman, A.A. Nazarov, A.E. Romanov, R.Z. Valiev, V.I. Vladimirov, Phil. Mag. A 59,
1113 (1989)
43. A.A. Nazarov, A.E. Romanov, Phil. Mag. Lett. 60, 187 (1989)
Chapter 4
Grain Boundary Order/Disorder and Energy
4.1 Grain Boundary Order or Disorder at High Temperature?
A grain boundary is described as a 2-dimensional phase distinct of the 3D phase of

the adjacent crystals [1]. We are now concerned with the question of the existence of
structural changes specific to this 2D phase and the possibility of loss of intergranular
order with increasing temperature. Phase transformations at grain boundaries may
also result from chemical composition variations; this effect is not considered in
this section but will be discussed when dealing with the segregation phenomenon
(Part II).
Phase transitions at grain boundaries may be separated into two groups [2]:
• The congruent phase transitions imply no change of the geometric parameters,

but changes in the grain boundary core structure.
• The non-congruent phase transitions are accompanied by changes of the misori-
entation and/or the grain boundary plane.
From a practical point of view, we distinguish between the transformation of the

stable solid boundary phase in another solid phase at medium temperature and its
transformation in a liquid phase at high temperature. The adsorption and emission
of vacancies by a boundary is also considered as local structural transformations;
indeed, a grain boundary disorder may result from the cooperative production of
vacancies. These phenomena are similar to those occurring in the order/disorder
transformations or in the melting of a 3D crystalline phase [1].
4.1.1 Solid/Solid Phase Transformations at Grain Boundaries
In terms of structural units, the solid/solid congruent phase transformations are linked
to the multiplicity of the grain boundary structures (see Sect. 3.2.3). The existence
of such transformations has been first proposed on the basis of pure thermodynamic

94 4 Grain Boundary Order/Disorder and Energy
arguments [1, 2]; it depends on the availability of various delimiting boundary struc-
tures and on their energy differences [3]. At any temperature, the vacancy mecha-
nisms induce the replacement in the boundary of certain fractions of the structural
units found at 0 K by alternative units of low energy. An order/disorder transforma-
tion may then occur at a critical temperature such that there is an equality of the
free energies of the structures corresponding to the ordered and to the disordered
distributions of the structural units. The solid/solid transformation may also occur
by a shear resulting from the vacancy coalescence in certain regions of the boundary.
A phase transformation of a = 11 {332} grain boundary is revealed by high-
resolution transmission electron microscopy in silicon and germanium bicrystals
(Fig. 4.1) [4]. Two structures A and B may exist in the boundary coming from different
rigid body translations parallel to the boundary plane, equal to a/44 [311] for τ A
and null for τ B (a is the parameter of the crystal unit cell). The components of
the translation normal to the boundary plane are very small in silicon. The atomic
structures observed in three = 11 {332} boundaries in germanium depend on
the thermal treatment underwent by the bicrystal: as elaborated, annealed at 850 ◦ C
then quenched, annealed at the same temperature followed by a slow cooling. As
predicted by Vitek et al. [3], the two first boundaries are not totally composed by
the two units deduced from the simulations, certain regions are described by a new
structure = 11 E. The units of the A structure appear after quenching from 850 ◦ C,
then A is the stable structure at high temperature. A transformation of the phase A
into the phase B, stable at low temperature (B and B correspond to the sequences
|TM+ P− M+ | and |TM− P+ M− |) occurs during the slow cooling from 850 ◦ C to
the room temperature. We have already mentioned that the situation is inversed in
silicon: the A structure being stable at low temperature. In both cases, the phase
transformation from A to B occurs by a rigid body translation equal to a/44 [311].
Also belonging to the congruent transitions, we may connect the transition occur-
ring in a delimiting boundary of misorientation θ , described by units B. For misori-
entations lower than θ but still very close to it, the boundaries are described by an
arrangement of majority B units and few A units. For misorientations close to θ but
higher, the majority units are still B but the minority units are now of the type C.
The discontinuity in energy, corresponding to the minimum in the curve E = fn(θ )
(Fig. 1.4), reveals a phase transition.
Non-congruent transformations also occur, as the formation of facets or the dis-
sociation of the boundary plane. The identification of a faceting phenomenon with
a phase transformation has been clearly explained by Cahn who established a phase
rule for grain boundaries; this imposes to consider a close analogy between the orien-
tation of the normal to the boundary plane and the composition of a system with three
constituents [2]. This type of phase transition may occur in the form of a roughening
transition induced by a temperature rise. A symmetrical = 3 111 tilt boundary
in aluminium, initially formed by a series of facets at equilibrium, becomes corru-
gated or rough when annealed at 230 ◦ C [5]. This transition may progressively yield
a disordered boundary structure.
4.1 Grain Boundary Order or Disorder at High Temperature? 95
Fig. 4.1 High-resolution

transmission electron
microscopy images showing
the structure transition of the
= 11{332} boundary in ger-
manium: a structure obtained
after bicrystal manufacturing;
b structure after annealing
the bicrystal at 850 ◦ C then
quenching; c structure after
annealing the bicrystal at
850 ◦ C then slow cooling [4]
4.1.2 Grain Boundary Pre-melting?
Given the lower atomic density in the intergranular region, the question is the possible
transformation of the solid grain boundary into liquid before the normal melting point
of the material. This phenomenon called grain boundary pre-melting belongs to the
category of congruent transitions. It can only occur when the grain boundary energy is
more than twice the energy of the solid/liquid interface; this condition is more easily
achieved in general high-energy grain boundaries. The short period tilt boundaries do
not seem to undergo such transformation; however, atomic jumps may occur under
heating that may induce structural changes. The pre-melting question remains totally
open for general high-angle boundaries that display a less dense structure.
In a first series of work, simulations of the grain boundary thermodynamic

properties focus on the calculations of the excess quantities compared to those of the
crystal, particularly, the Gibbs energy [6, 7]. For several symmetrical tilt boundaries,
discontinuities appear in the evolutions with temperature of the free enthalpy and
of the excess volume, in favour of a first-order transition at a temperature below
Tm (Tm , the crystal melting temperature). In this transition, the crystalline boundary
is substituted by a strongly disordered boundary akin to a liquid layer, but preserving
a good coherency between the neighbouring crystals [6]. The transition temperature
is inversely proportional to the boundary excess volume.
The experimental studies of the boundary structure evolutions at the vicinity of the
material melting point are difficult and their interpretations ambiguous. We may men-
tion the observation of dislocation networks in boundaries annealed at temperatures
close the melting point [5], that is inconsistent with an interpretation of the disorder
as a widening and an overlap of the intrinsic dislocation cores [8]. Studies of grain
boundary sliding [9, 10] and grain boundary migration [11] reveal the existence of a
transition temperature Tc that separates two types of boundary behaviours. In partic-
ular, the activation energy of grain boundary sliding for T < Tc is comparable to that
of grain boundary self-diffusion, whereas it becomes close to that of bulk diffusion
for T > Tc [10]. But, in both cases, we cannot exclude the effects of impurities that,
even in very small amounts, are known to strongly influence sliding and migration.
In the absence of experimental proofs, the molecular dynamic simulation appears
as a good tool to follow the evolution of the grain boundary structures with
temperature. Such study of several boundaries in palladium shows that there is a tran-
sition of the grain boundary diffusion energies between the low temperature regime
and the high temperature regime [12]. Above a critical temperature Tc , higher when
the boundary energy is lower, a notable acceleration of the diffusion occurs (Fig. 4.2).
In order to emphasise the controlling diffusion mechanism above TC , the tem-
porary evolution of the configuration of the atoms diffusing in the boundary plane
is followed at different simulation times (Fig. 4.3): liquid-type clusters, with high
Fig. 4.2 Positions of the atoms moving at least one distance between neighbouring sites during a
given simulation time, projected in a plane normal to (110) of a = 11 twist grain boundary (
is the planar coincidence index): a T = 1000 K; b T = 1400 K. The horizontal lines indicate the
boundary structural width. The density of mobile atoms gives a view of the diffusion width [12]
4.1 Grain Boundary Order or Disorder at High Temperature? 97
Fig. 4.3 Instantaneous views of the atoms mobile at 1000 K in the (110) plane of the = 11
twist boundary, recorded with 100.000 simulation steps between the views (a) and (b). The squares
delimit two identical regions at two different times, illustrating the formation and the disappearance
of very mobile clusters [12]
mobility, are continuously created before to disappear while others appear in other
regions of the boundary.
When the temperature increases, the clusters coalesce to form a confined liquid
layer at Tm . However, the boundary width remains finite until Tm, in agreement with
the conclusion of non-existence of a boundary pre-melting [13].
It is important to note that the temperature TC of the transition between a solid-
type diffusion mechanism and a liquid-type mechanism is not reached for low energy
boundaries that preserve a short-range order until Tm .
In a review paper of the simulation studies of the grain boundary structures at
temperature above 0.4 Tm [14], an agreement emerges on the existence of a transi-
tion from a periodic structure to a disordered structure of several symmetrical tilt
boundaries at 0.5 Tm , the width of the disordered region increasing with the tempera-
ture rise. But the results do not agree on the existence of a pre-melting, the differences
may result from the data taken into account in the simulation and the boundary condi-
tions used. Indeed, the simulations generally use periodic limit conditions implying
the presence of two boundaries in the simulation box. These conditions are not
adapted to the studies of phenomena occurring at high temperature (diffusion, phase
transformation…) because a migration of the two boundaries may lead to their anni-
hilation. In that case, the atomic configuration near the thermodynamic melting point
cannot be described.
Anti-periodic limit conditions so-called Möbius conditions have been proposed
[15], that enable to introduce only one boundary on the simulation box, thus limiting
the artefacts. The use of Möbius conditions allows, among other things, to highlight
pre-melting phenomena in the particular case of the = 25 {710} 001 boundary in
silicon [16]. Note that the thermodynamic melting temperature of the silicon simu-
lated in this study is about 1700 K, very close that of the real silicon, i.e. 1687 K; but,
the number of atoms involved in the simulation being limited, and in the absence of
surfaces, the simulated system only melts at temperatures superior to Tm . A dynamic
disorder appears at around 1100–1200 K (about 30 % lower that Tm ). From 1600 K,
and more clearly at 1700 K (below the melting temperature of the simulated system)
Fig. 4.4 Views of the atom

trajectories during 90 ps et
1700 K for the = 25
(710) [001] (16.26◦ ) bound-
ary in silicon: a view along
[001] showing the trajec-
tories of all the atoms;
b view along [1-70] revealing
the trajectory of only one atom
and the character quasi-liquid
of the boundary at the struc-
tural unit level. The simulation
box contains 16 (001) planes
with 920 atoms [16]
we observe a spontaneous diffusion, of the liquid-type, along a cylinder parallel to

the tilt axis (Fig. 4.4). The diffusion remains limited in the directions perpendicular
to this axis.
In this simulation at 1700 K, pre-melting remains localized in the boundary core
region. Beyond 1800–1900 K, this region extends to the volume: it is no longer
pre-melting but melting that starts in the more fragile zone.
This very interesting example, is however, isolated compared to what happens in
the other tilt boundaries in silicon simulated with the same limit conditions [16]. The
pre-melting existence would not be a question that may be answered universally: it
depends on the considered boundaries and probably on their purity.
Finally, given the overall results, whatever the crystalline state of a boundary, we
may conclude that the boundary fills up with vacancies and local disorders when
the temperature increases, but the concerned layer width stays very small until the
melting point. The boundary generally maintains a solid character until a macroscopic
width is reached.
The grain boundary behaviour at high temperature will be again discussed at the
end of this part (see Sect. 4.7) when we will come back on a debate that always
accompanies the state of knowledge on the boundary structure:
• Are all boundaries crystalline? Like those which have been studied to date by
simulation and experiments?
• Could they be quasi-crystalline like those predicted by mathematical considera-
tions (algorithm, geometrical construction)?
• Finally, are there really amorphous boundaries, as considered a century ago by
Rosenhain [17], the idea of which has been reintroduced, after a long decline,
according to the simulation results for high-energy grain boundaries?
4.2 Interfacial Energy: Thermodynamic Aspects and Energy Factors 99
4.2 Interfacial Energy: Thermodynamic Aspects

and Energy Factors
The presence of a grain boundary increases the energy of a system initially on the
form of a single crystal. According to the thermodynamic variables that are used,
we define the interfacial energy γ (per unit area) as the increase of an energetic term
associated to the bicrystal, due to the presence of the grain boundary. By choosing the
two variables, temperature T and pressure P (the usual thermodynamic variables),
the increase of the Gibbs enthalpy is given by:

C
dG = −SdT + V d P + μI dNi + γ d A (4.1)
i=1
With Ni and μi are the quantity and the chemical potential of the constituent i; C is
the number of constituents, S the entropy of the system and V the volume of the
bicrystal.
Then, the free energy of the boundary is defined:
∂G
γ = (4.2)
∂ A T,P,Ni
If the area of the boundary plane increases, the total energy variation (dG = γ d A+
Adγ ) is equal to the work made by a force F (per unit length) to provoke an
augmentation of the area, i .e. F · d A. Then we have:
F = γ + Adγ /d A (4.3)
If the interfacial energy is independent on the boundary plane area (dγ /d A = 0),
then the free energy γ (in J·m−2 ) exerts an interfacial tension equal to γ (in N·m−1 ).
Although this independence of the energy with the area is not evident in the case of
solid interfaces, except close the melting point, we often consider that tension ≡
energy. Note that the definition of γ is totally similar to that of the surface tension
σ used in the studies of liquids. To avoid any ambiguity with a stress, we will use
the Greek letter γ - and not σ - to represent the interfacial tension, equivalent in
certain cases to the interfacial energy. We consider the thermodynamic equilibrium
of the system constituted by the crystals and the boundary that may exchange matter
and energy. The bicrystal is composed by two homogeneous regions (the crystals)
forming a single phase and a narrow region separating them, and considered as second
phase [1]. By using generalised thermodynamic quantities and by extrapolating the
application of Gibbs-Duhem relation to the interfacial region, we obtain two coupled
equations that rule the equilibrium of the system:

− S I dT + V I d P − NiI dμi −dγ = 0
i=1

− SdT + V d P − Ni dμi −dγ = 0 (4.4)
i=1
The exponent I indicates quantities in the interface (grain boundary plane).

For grain boundaries in a system with only one constituent (i = 1), dμ1 may be
eliminated between the two equations, then:
N1I NI
dγ = −S I − SdT + V I − 1 V d P (4.5)
N1 N1
and:
∂γ ∂γ
= V and = −S (4.6)
∂ PT ∂ TP
V and S are the excess quantities in the grain boundary region compared to those
in a region of the crystal containing the same number of atoms. They are generally
positive for boundaries in compact materials. The interfacial energy tends to diminish
when the temperature increases.
But the interfacial free energy is not only function of the variables T and P; indeed,
the boundary is in a deformed state compared to the crystals, moreover, its structure
(and thus its energy) is a function of the five macroscopic degrees of freedom. Finally,
γ is a complex function that may be written as:
γ = γ (T, P, μi , εij , ρ1 , ρ2 , θ, n I , n II ) (4.7)
With εij the grain boundary deformations, ρ1 et ρ2 the direction cosines of the rotation
axis, θ the rotation angle and n I , n II the directions of the normal to the boundary plane
in the two crystals.
Other variations of the energy are beyond the classical thermodynamics; they
result from the variations of the microscopic geometrical parameters and from local
atomic relaxations.
4.3 Macroscopic Degrees of Freedom and Interfacial Energy
The evolutions of the interfacial energy γ with the grain boundary macroscopic
geometrical parameters are often studied in two extreme cases:
• Only the misorientation θ varies around a given axis, the boundary plane being
fixed for the twist boundaries, whereas the median plane is fixed for the tilt
boundaries. In the latter, the boundary plane also changes with the misorienta-
tion, but it is determined by the rotation and by the median plane. In the series of
4.3 Macroscopic Degrees of Freedom and Interfacial Energy 101
symmetrical [1-10] tilt boundaries (Table 3.1), the misorientation angle imposes
the boundary plane in each crystal, the median plane being (001).
• The misorientation (axis and angle) is fixed, the boundary plane takes any inclina-
tion, the two variable parameters are the direction cosines of the normals, nI and
nII , to the boundary plane in each crystal.
The results reported below enable to give orders of magnitude of the energies and
to compare their evolution according to the materials and the grain boundary types.

with the Misorientation Angle
We recall that Read and Shockley derived the first formula (2.8) giving the energy
variation of the low-angle tilt grain boundaries with their misorientation; this formula
only takes into account the elastic contribution of the intrinsic primary dislocations.
Then, the energy evolutions have been approached on the basis of the different grain
boundary structure models.
The Read and Shockley formula remains at the basis of the energetic considerations
in terms of dislocations for any grain boundary. Its application requires that the dis-
locations coming from the crystals (b = bcrystal ) are periodically arranged in the
boundary. Simultaneously, the spacing between the dislocations must correspond to
a crystal period. This condition may be fulfilled only for certain rational rotations
θ = 1/m (for rotations around 001 in cubic crystals). For intermediary angles
θ + δ θ , we consider that a sub-boundary is superimposed to the nearest rational
boundary. We find the model of intrinsic primary dislocations, the periodicity of
which is broken by the secondary dislocations of Burgers vector |b/m| that form the
sub-boundary. Generally, the boundary energy is composed of two terms γ and δ γ .
The energy excess linked to the sub-boundary may be written [18] by deriving the
Read and Shockley equation (2.8):
γ0 δ θ
δγ = ln δ θ (4.8)
m
when δ θ → 0, the slope of the curve giving γ in function of θ becomes infinite
and the rational angles θ = 1/m correspond to pronounced minima on the curve
representing the simple Read and Shockley equation (Fig. 4.5). These minima are
said primary. Secondary minima can be considered if tertiary dislocations come
to break the periodicity of the secondary dislocations and so on…The existence of
secondary minima is controverted, in particular the Read and Shockley model ignores
the entropy contribution to the free energy that results in the delocalization of the
secondary dislocation cores and in the diminution of their elastic fields.
Although without physical meaning, the empirical extrapolation of the Read and
Shockley formula, in which sin θ substitutes for θ , applies pretty well to high-angle
Fig. 4.5 Curve giving the interfacial energy in function of the misorientation θ, plotted by taking
into account the δ θ term necessary to respect the crystal periodicity (see text). The primary minima
on the curve γ + δ γ (full line) are located on the curve deduced from the classical Read and Shockley
formula (dotted line) (2.8). The secondary minima between two primary minima (see detail) are
controverted [18]
tilt and twist grain boundaries, possessing the same median plane [19]:
γθ = γ0 sin θ (A − ln(sin θ )) (4.9)
In the previous considerations, the contribution of the core energy of the intrinsic
dislocations, i.e. the boundary core energy, is neglected. This contribution may be
determined on the basis of the structural unit model. For a boundary X constituted of
m A units and n B units (n < m), the core energy depends on the interactions between
the units that are of two types A − A and A − B − A, the minority unit B being always
situated between two majority units. The energy associated to the A − A interaction
is considered as equal to the energy γA (per unit area) of the delimiting A boundary
multiplied by the length of this unit in the boundary X, i.e. u A cos((θ − θA )/2) (see
Sect. 3.2.4). The energy associated to the A − B − A interaction is considered as
equal to the energy γAB of the AB boundary multiplied by the length of the AB
unit in the boundary X, i.e. wAB = u A cos((θ − θA )/2) + vB cos((θB − θ )/2). The
boundary X contains n links A − B − A and (m − n) links A − A, and the length of
its period is given by the formula (3.4), then the expression of the core energy is:
(m − n) u A cos((θ − θA )/2)γ A + n[u A cos((θ − θA )/2) + v B cos((θB − θ )/2)]γ AB

γc =
m u A cos((θ − θA )/2) + nv B cos((θB − θ )/2)
(4.10)
The use of the formulae (3.4) and (3.5) enables to simplify expression (4.10):
2sin ((θ − θA ) /2) w AB

γc = γ A + (γ AB − γ A ) (4.11)
bA vB
bA is the magnitude of the Burgers vector of the secondary dislocations associated to

B units. When the misorientation of the boundary X is very near that of the delimiting
boundary A, expression (4.11) becomes:
θ − θ A w AB
γc = γ A + (γ AB − γ A ) (4.12)
bA vB
The core energy varies linearly between γ A and γ AB .

An elastic energy derived from (2.8) must be added to γc ; then, the boundary total
energy is:
1 w AB μb A bA
γ = γ A + (θ − θ A ) (γ AB − γ A ) + − ln(θ − θ A ) (4.13)
bA v B 4π(1 − v) 2πr0
The core radius of a grain boundary dislocation r0 is not known; it is often

approximated to w AB /2 [20]. Besides, the elastic energy being necessarily posi-
tive, a limit value of the angular deviation (θ − θ A ) appears; the generally low value
of this limit requires consideration of a large number of delimiting boundaries to use
the expression (4.13).
The evolution of the interfacial energy with the misorientation has also been
approached for tilt boundaries on the basis of the disclination/structural unit descrip-
tion [21]. As previously, a grain boundary is constituted by m A units and n B units,
the length of which in the boundary being d A and d B , respectively; the reference
boundary energies are γA and γB . The elastic and core energies are those of a discli-
nation dipole wall, each dipole is formed by the minority unit B (Fig. 3.36). The core
energy of a disclination dipole with a force θ = θ B − θ A is equal to:
α μb2 (θ )2
γc = (4.14)
2 π3 (1 − v)
α is roughly equal to 1.
The elastic energy of disclination dipole wall with period H is given by:
μ (θ )2 H
γe = f n dB , n, H and yi (4.15)
32 π (1 − v)
3
yi locates the position of the dipole along the boundary from an origin that coincides
with a period extremity (Fig. 3.36). In the formula (4.15), fn expresses a complex
function, introduced by Shih and Li [22], that takes a minimum value when the
dipoles are uniformly spaced from H/n. But this distance is rarely strictly achieved
as it must be a multiple of d A ; a distribution as close as possible from the regularity
is only realised;
Finally, the boundary total energy is:
γ = m d A γ A + n d B γ B + nγC /H + γe (4.16)
The two first terms are the surface energies of the two boundaries, constituted of
A units and B units, respectively; they are functions of the misorientation θ as d A =
u A cos((θ − θ A )/2) and d B = v B cos((θ B − θ )/2).
This approach enables to study the energy evolution of symmetrical [001] tilt
boundaries in aluminium by considering four favoured boundaries = 1 (110), =
5 (210), = 5 (310) and = 1(1-10), i.e. by cutting up the angular interval 0–90◦
in three parts: 0–36.9◦ , 36.9◦ –53.1◦ and 53.1◦ –90◦ . The resulting curve (Fig. 4.6) is
in good agreement with the experimental values obtained before [23]. It presents a
large number of shallow minima apart few deeper cusps corresponding to favoured
boundaries.
Fig. 4.6 Curve giving the
energy evolution for symmet-
rical [001] tilt boundaries in
aluminium versus the misori-
entation θ, on the basis of a
description of the boundary
as a disclination dipole wall
[21]. Experimental values are
reported by circles [23]
The simulations provide the knowledge of the relaxed structure of a grain boundary
with its associated energy. It is remarkable that the curves giving γ in function of
θ derived from the simulations, as well as the experimental curves, are in good
agreement with the curves predicted in the previous approaches.
Figures 4.7, 4.8 and 4.9 [24] present the curves of the energy evolution in function
of the misorientation angle θ simulated for pure tilt and twist boundaries, sym-
metrical or not, in f.c.c. (copper, gold) and b.c.c. (iron, molybdenum) metals. The
energy minima are generally shallow for twist boundaries (Fig. 4.8) by compari-
son to the minima appearing in tilt boundaries (Fig. 4.7). Generally, the energies of
the asymmetrical boundaries are more elevated than those of the symmetrical ones
with the same misorientation (Fig. 4.9), but numerous exceptions exist. Whatever the
boundary type, tilt or twist, the lowest energies are not necessarily associated to the
smallest coincidence index. An energy plateau occurs in the curves γ = fn(θ ) for
symmetrical twist boundaries. For tilt grain boundaries, the energy values strongly
depend on the boundary plane; a striking example concerns the = 3 twins: the
energy of the incoherent = 3 {112} twin is 840 mJ·m−2 compared to 22 mJ·m−2
for the coherent = 3 {111} twin in copper [25]. Other important difference appears
for the = 11 {332} and = 11 {113} boundaries: the energy of the former (970
mJ·m−2 ) is almost twice that of the second (448 mJ·m−2 ) in nickel [24].
Although the interfacial energy is not a criterion to choose the favoured grain
boundaries, as it does not depend only of the core structure, the energy minima are
often associated to such boundaries: = 3 {111} and = 11 {113} for symmetrical
110 tilt boundaries (Fig. 1.4).
Fig. 4.7 Curves giving

the energy evolution for
symmetrical 110 tilt
boundaries: a for copper and
gold, both of f.c.c. symmetry:
b for iron and molybde-
num, both of b.c.c. symme-
try. The boundary planes
corresponding to energy
minima are indicated at the
top of the figure [24]
By reporting on a 3D diagram the interfacial energy values calculated for different

tilt or twist angles, D. Wolf draws by interpolation an energy surface that could
predict the energy of the mixed grain boundaries (Fig. 4.10) [26]. The deep valleys
on the 3D surface show the importance of the dense planes and of the neighbouring
orientations (vicinal boundaries). This representation clearly highlights that twist
boundaries with θ = 180◦ are identical to symmetrical 110 tilt boundaries. Then,
the twist boundaries with 0◦ <θ <180◦ may be considered as pure twist or as mixed
boundaries having two components: θtilt and θtwist .
Other elastic approaches of the relation between energy and misorientation will
be evoked when we deal with the return to equilibrium of a grain boundary disturbed
by linear defects (See Chap. 9).
Fig. 4.8 Curves giving the energy evolution for symmetrical twist boundaries: a around [111] for
copper and gold (f.c.c.); b around [001] for iron and molybdenum (b.c.c.). To compare these curves,
the misorientation scales have been multiplied by three for (111) and by two for (001) boundaries.
The symbols (•) and () correspond to the values obtained by using the extended Read and Shockley
formula (4.9) and by simulation, respectively [24]
Fig. 4.9 Curves giving the energy evolution for symmetrical (112) and asymmetrical (112)// (552)
twist boundaries in b.c.c. iron. Only the symmetrical boundaries θ = 0◦ (single crystal) and θ =
180◦ corresponding to = 3 tilt boundary have a low energy [24]
4.3.2 Variation of the Interfacial Energy with the Grain Boundary

Plane Inclination
A convenient way to present the variation of the interfacial energy γ with the interface
inclination ϕ (or its normal orientation) is the Wulff construction that traduces the
anisotropy of the surface energies. We trace a set of vectors from an origin; each
vector has an orientation associated with a normal to a grain boundary plane and
a length proportional to the energy of that plane. The surface that passes through
the vector extremities is a polar representation of the energy γ in function of the
inclination ϕ. Figure 4.11 shows a 2D section of the 3D surface by a random plane;
in order to simplify, the chosen plane is (1-10). The cusps that appear correspond
to energy minima for certain inclinations of the boundary plane. The Wulff diagram
enables to predict the shape of a crystal embedded in a matrix if we know the energies
Fig. 4.10 a Calculated energies for grain boundaries with various tilt and/or twist angles. The
curves represent the energy evolutions with θtwist for five pure twist {110}, {111}, {112}, {113}
and {100} boundaries. For θtwist = 180◦ , the twist boundaries are identical to symmetrical tilt
boundaries; b curved surface drawn by in interpolation of the curves given in (a) allowing to predict
the energies of mixed boundaries [26]
of the interfaces between that crystal and the matrix. Let define A1 , A2 . . ., the areas
of the interfaces surrounding the nucleus and γ1 , γ2 . . ., their respective energies, the
equilibrium shape of the nucleus is such that the total interfacial energy ΣAi γi is
minimal. For each point P of the polar diagram, we trace the plane perpendicular
to OP, the so-called Wulff plane; the inner envelop of the Wulff planes gives the
equilibrium shape of the nucleus.
Fig. 4.11 (1-10) section of a

possible polar Wulff diagram
for a f.c.c. crystal. The length
OA represents the free energy
of a plane perpendicular
to it in A (Wulff plane); so
OB = γ(001) and OC = γ(111) .
The inner envelop of the Wulff
planes gives the equilibrium
shape of the nucleus
If a net energy minimum exists for a particular orientation of the interface, the
latter develop preferentially leading to a nucleus on the form of plate or disc with a
very small thickness as illustrated on Fig. 4.12. In that case, the lateral shape of the
Fig. 4.12 Section of a polar Wulff diagram in the case where a boundary plane orientation is clearly
favoured (very low energy) and equilibrium shape of the corresponding section of the nucleus
nucleus (plate or disk) can be known only by considering other sections of the 3D
Wulff diagram.
When a grain boundary is in an instable or metastable position (as it is often the
case for boundaries in polycrystal), tangential and normal forces act on each region
of the boundary plane. The tangential force FT tends to reduce the extension of the
boundary; this force always exists as a boundary, that is a defect, has a tendency to
decrease its area. The boundary plane tends to rotate under the effect of the normal
force FN ; this rotation yields a decrease of the interfacial energy when it varies with
the inclination of the boundary plane.
Consider a part of a boundary of unit width and of length OP = 1. Forces equal and
opposite to FT and FN must act in O and P to maintain the boundary at equilibrium
(Fig. 4.13a).
The tangential force FT is the interfacial tension equal to γ (see Sect. 5.1). The
normal force FN may be calculated by considering that the point O is fixed and the
point P moves on a small distance δy (Fig. 4.13b); the work done for this displace-
ment is equal to FN · δy and must balance the energy rise linked to the change of
orientation δϕ:
dγ
FN · δ y. = l δϕ (4.17)
dϕ
Fig. 4.13 Schematic drawing to define the tangential and normal forces acting on a grain boundary
segment
as δy = 1δϕ, we have:
dγ
FN = (4.18)
dϕ
Relation (4.18) implies that if the boundary energy is dependent of the plane
orientation, a force dγ /dϕ must be applied to the extremities of the boundary to
avoid its rotation towards a position of lower energy; thus dγ /dϕ is a torque term.
A boundary linked to others at a triple junction exerts a force equal and opposite to
FN on the junction. No couple acts on the boundary if its energy is at a minimum
on the curve γ = fn(ϕ). However, the boundary resists to a normal force only if
this force is inferior to dγ /dϕ. If the boundary energy is independent of the plane
orientation, the torque term is null.
The relative boundary energy may be determined by measuring the angles between
the boundary plane and the intercepted free surface. The method is based on the
measurements of the thermal etching grooves that develop at a metal free surface
previously polished (Fig. 4.14) [27].
Fig. 4.14 Equilibrium

between the surface tensions
and the interfacial tension at
the intersection of a boundary
with a free surface
The equilibrium condition is:
γJ = 2 γSV · cos ψ/2 (4.19)
with ψ the opening angle of the groove.

When a grain boundary has an orientation very close that corresponding to an
energy minimum, it tends to preserve as well as possible the structure of low energy
by formation of facets. Such profiles are often observed in alumina with long facets
parallel to the basal (0001) plane in the two crystals or in one of them, at least [28]
(Fig. 4.15).
An interface far from any energetically preferred orientation may also present
faceting, each facet having a very low energy to compensate the increase of the
total boundary area. The observed facets are most often those predicted by the Wulff
construction. The total energy of the faceted boundary is less than that of the straight
boundary, even by taking into account an energetic term associated to the junction
Fig. 4.15 a A straight boundary with a mean basal plane in alumina; b at high magnification, the
boundary displays asymmetrical facets with a (0001) plane alternately located in one and in the
other crystal [28]
between facets. Steps may also exist at the atomic scale; a vicinal grain boundary
close to = 11 in gold presents symmetrical {113} facets that alternate with asym-
metrical {111}//{001} facets (Fig. 4.16) [29].
Fig. 4.16 Observation at the atomic scale of faceted boundaries in gold: a vicinal boundary close
= 11 displaying symmetrical {113} facets alternating with shorter asymmetrical {111} // {001}
facets; b asymmetrical = 33 {113} // {771} boundary composed of micro-facets alternately
symmetrical {111} and asymmetrical {111} // {001} [29]
Another way to reduce the interfacial energy is the dissociation of one boundary
in three boundaries surrounding a crystal with an orientation that differs from those
Fig. 4.17 a Diagram show-

ing the dissociation of an
asymmetrical {111} // {115}
boundary in two coherent
and an incoherent twins;
b image in high resolu-
tion transmission electron
microscopy of a = 9
bicrystal in gold obtained by
deposing a crystal on a sub-
strate with {111} orientation:
the grain boundary plane on
the left is {114}//{114},
on the right{11,11,1} //
{111}; at the centre the
{111}//{115} boundary is
dissociated according to the
scheme (a) [32]
of the initial crystals [30]. An asymmetrical = 9 {111}//{115} boundary with

high energy may dissociate into three boundaries of lower energy, as observed in a
copper-6 % silicon alloy [31] and in gold [32] (Fig. 4.17).
Only the relative energies compared to the surface energies may be experimentally
obtained; the determination of the absolute energy values requires simulating the
grain boundary structures. Figure 4.18 presents the evolution of the calculated energy
in function of the boundary plane inclination for a = 9 tilt boundary in gold and
copper, using EAM and LJ potentials, respectively [32]. The energy variations are
qualitatively similar for the two metals, with a common relative low energy for
the asymmetrical {111} // {115} boundary. However, despite its favourable energy,
this boundary dissociates in agreement with the scheme and the image of Fig. 4.17.
We note that the use of a N-body potential [33] to simulate copper leads to values
notably different that those reported on Fig. 4.18; the energy of the symmetrical
{221} boundary is 670 mJ·m−2 instead of 756 mJ·m−2 ; moreover, the asymmetrical
{111} // {11, 11, 1} boundary is clearly favoured in the series with an energy of
337 mJ·m−2 (O. Hardouin Duparc, Communications personnelles (2003–2005)).
Another example of the energy variation with the plane inclination angle ϕ is
obtained for a series of = 3 [01−1] boundaries in copper [34]. The authors follow
the evolution of the ratio between, on one hand, the calculated boundary energy and
the surface energy and, on the other hand, the measured energy and the surface energy;
the ratio is measured using the thermal grooving technique. The calculated and
measured relative values are close to each other for the coherent = 3 {111}(ϕ = 0)
that possesses a very low energy (Fig. 4.19). A second minimum appears in the curve
for ϕ = 82◦ , that corresponds to a mean asymmetrical {322}//{11 4 4} plane. This
Fig. 4.18 Calculated energies

in function of the inclination
angle ϕ of the grain boundary
plane for (110 > = 9 tilt
boundaries in copper (◦) and
gold (). The potentials used
in the simulations are of the
EAM type for gold and LJ
for copper. The corresponding
boundary planes are reported
at the top of the figure. The
full symbols indicate the
energies of the dissociated
{111}//{115} boundary [32]

ratio between the interfacial
energy and the surface energy,
calculated and measured in
function of the boundary
plane inclination angle ϕ for
= 3 [011] boundaries
in copper. The curve in full
line represents the theoretical
variation of this ratio. γS0 is
the calculated surface energy
for the {110} plane [34]
minimum is due to the dissociation of the boundary with formation of a 9R phase

that improves the atomic fit in the interface as explained in the case of the incoherent
twin (see Sect. 3.3.1).
Finally, a systematic study of the variation of the energy in function of the
misorientation or the boundary plane inclination enables to detect the structures
that escape to the structural unit model: faceting, dissociation. . . However, the
calculated values may strongly differ according to the potential retained for the
simulation. The values obtained by ab initio simulations, in principle the closest
from the real values, also depend on the parameters used to model the material (in
particular, the compressibility modulus for the defects). The calculated energy may
be said absolute only by comparison to the relative energy (function of the free
surface energy) obtained by experiments. Moreover, the energy is a function of the
microscopic parameters; we must explore the energy values resulting from differ-
ent rigid body translations leading to different structures, sometimes no explicitly
reproduced.
4.4 Microscopic Degrees of Freedom and Interfacial Energy
The microscopic parameters are implicitly taken into account in the previous simu-
lations, as the calculated energies are associated to the atomic grain boundary struc-
tures. They are necessary as soon as the rotation θ angles exceeds a critical value
such that the intrinsic dislocation cores overlap and the core energy dominates the
elastic energy. Although the disclination model takes into account the core energy,
it is a rigid model ignoring the atomic relaxations.
The atomic interactions differ according to the type of material: metallic, ionic
or covalent. The core structures have been more systematically modelled for grain
boundaries in metals and elementary semiconductors. Only, the simulation enables
to reveal the effects of the microscopic parameters on the boundary core energy.
4.4.1 Variation of the Interfacial Energy with the in Plane Rigid

Body Translation
The variation of the interfacial energy with the τ// component of the rigid body
translation is directly connected to the size and the shape of the cell of non-identical
displacements (c.n.i.d.) (see Sect. 1.2.1) that depend on the periodicity, 1D or 2D, in
the boundary plane. Generally, the c.n.i.d. belongs to a sub-lattice of the DSC lattice
in the plane. The c.n.i.d. size is null when the boundary is incommensurable; then,
the interfacial energy is independent of τ // . If the boundary is commensurable, then
the energy is a periodic function of τ // . A simple physical argument predicts that the
boundary energy should not change much if its translation state is slightly modified.
Otherwise, if the c.n.i.d. size is large, important variations of the boundary energy
with τ // are possible; the boundary may explore a large range of translations and
reach energy minima for particular values of the latter, leading to a multiplicity of
grain boundary structures.
To identify the stable structures, non-identical by translation, we use a procedure
based on the γ -surface technique [35]. The evolution of the interfacial energy in
function of the translations parallel to the boundary plane is determined inside the
c.n.i.d.; the energy is calculated by considering either the total relaxation of the
simulation box, or the relaxation in a direction perpendicular to the boundary plane. In
the latter case, the boundary being partially relaxed, its energy is generally larger than
that of the totally relaxed boundary, but it is more rapidly calculated with a pretty good
accuracy. The energy is a continuous function of τ // that may be represented in a 3D
Fig. 4.20 γ -surfaces for four symmetrical [001] tilt grain boundaries in NiAl: a formal = 1
(110) boundary; b = 5 (310), c = 5 (210) and d = 13 (320) boundaries. The distances are
measured in units a (a = unit cell parameter). The local minima and maxima are indicated by full
and empty symbols, respectively [36]
space by a surface or, in a 2D space, by contour lines with well-defined maxima and
minima. This construction so-called γ -surface is illustrated for four [001] tilt grain
boundaries in the intermetallic compound NiAl: a formal = 1 (110) boundary, the
= 5 (310), = 5 (210) and 13 (320) boundaries (Fig. 4.20) [36]. The various
contour lines correspond to energy variations going from 0.04 to 0.1 J·m−2 according
to the surface complexity. The translations along τ x and τ y in the boundary plane
(τ y parallel to the tilt axis) are measured in units a (with a the unit cell parameter).
The translations leading to minima or maxima are represented by full and empty
symbols, respectively; the corresponding values are indicated next to the γ -surface.
For most grain boundaries, minima and maxima are obtained by translations along
the tilt axis equal to zero or to 1/2 [001] = a/2. Only, few maxima appear for
intermediary translations (Fig. 4.20d). It results that all the metastable states of the
[001] grain boundaries may be determined on γ -surface sections for τ y = 0 and
τy = a/2 (Fig. 4.21).
Taking into account the similarity of the γ -surfaces for various boundaries around
a same axis, i.e. energy minima associated to translations parallel to the tilt axis, we
most often explore a restricted number of rigid body translations. For example, two
structures, with or without rigid body translations, have been calculated for the sym-
metrical = 11 {332} boundary in nickel. The stable structure, experimentally
observed (Fig. 3.9), displays a mirror-glide symmetry. The rigid translation deter-
mined on the calculated structure is equal to 0.07 nm; it corresponds to 2/3 of the
length of the a/11[113] DSC vector [37]. The energy of this structure is 970 mJ·m−2
4.4 Microscopic Degrees of Freedom and Interfacial Energy 115
Fig. 4.21 Sections of the γ -surface of a = 13 (320) boundary for (a) τ y = 0 and (b) τ y = a/2.
The energy values of the totally relaxed boundary (o) are reported for comparison. The arrows
indicate the absolute minima of the interfacial energy [36]
while the energy of the structure without translation is 1020 mJ·m−2 [32]. Thus, an
energy difference of the order of 5 % appears sufficient to stabilize a grain boundary.
Finally, we must remark that the deposition on a substrate of metal balls free to
take any orientation leads to preferential misorientations between the balls and the
substrate that possess large c.n.i.d.
4.4.2 Variation of the Interfacial Energy with the Expansion

Normal to the Grain Boundary Plane
To determine the effect of a rigid expansion τ ⊥ on the interfacial energy, we consider

twist grain boundaries at the incommensurability limit (i.e. the energy in function of
the misorientation has its maximum value), the energy of which being independent
of τ // . The expansion that minimizes the energy is calculated, on the basis of an
analytical model, for any pairs of planes with interplanar distances dII and dI in
crystals II and I, respectively [38]. The expansion increases and, simultaneously, the
cleavage energy decreases when the interplanar distance is reduced. The calculated
expression takes a value = 0 only for distances equal to 0.6420 a; remember that the
maximal distance between the {111} planes in a f.c.c. material is equal to 0. 5774
a. An expansion is thus always predicted in the twist grain boundaries. Figure 4.22
shows the continuous evolutions of the expansion and of the bonding function (that
Fig. 4.22 a Calculated equilibrium expansion () and found by atomic relaxation (◦) in function
of the interplanar spacing dhkl given in units a (a = unit cell parameter) for twist grain boundaries
in the incommensurability limit; b bonding function versus dhkl for the same boundaries [38]
determines the cleavage energy) with the interplanar spacing for twist boundaries in
the incommensurability limit.
The continuous evolutions of the expansion and of the cleavage energy are
problematic. Indeed, for a boundary with a {hkl} plane of low indices and of inter-
planar distance dhkl , the closest plane has high indices compared to {hkl} and its
interplanar distance is thus much lower than dhkl . According to the analytical model,
its expansion and its cleavage energy must strongly differ from those of {111}. This
discontinuity is due to the fact that, in the model, the planes may only displace rigidly
in a direction perpendicular to the boundary plane, whereas planes may locally bend.
For mixed grain boundaries for which dI differs from dII , the variables are the
mean distance d and the misfit δ = (dI −dII, )/(dI +dII, ). For a fixed mean distance,
the expansion increases monotonously and the cleavage energy diminishes when the
misfit decreases from 1 to 0. The boundary with dI = dII always presents the largest
expansion.
These predictions are easily explained if we consider the expansion as determined
by a balance between attractive and repulsive forces in the boundary. The larger the
overlap of the atoms before relaxation, the larger the expansion necessary to reach
equilibrium and the weaker the residual attractive force through the boundary. The
associated cleavage energy differs from the interfacial energy an, in the latter, the
energies of the free surfaces of the two boundary planes are taken into account.
4.4 Microscopic Degrees of Freedom and Interfacial Energy 117
4.4.3 Variation of the Interfacial Energy with Local Individual

Atomic Relaxations
The individual atomic relaxations take a special importance when the mean interpla-
nar spacing decreases. In a twist grain boundary, there are displacements parallel to
the plane inducing a small local change of the misorientation angle compensated by a
local screw dislocation network. These local relaxations are energetically favourable
only when the period of the reference boundary (or the c.n.i.d. size) diminishes and
becomes inferior to a critical value.
The displacements are perpendicular to the plane of the tilt boundary; they induce
a deviation from the tilt angle of the reference boundary compensated by a local
edge dislocation network. They are concentrated in the first parallel planes close
the boundary plane when the interplanar spacing is relatively large. Otherwise, they
provoke oscillations that exponentially decrease with the distance from the boundary
plane. The cohesive forces that act across the interface and that decrease when τ ⊥
increases, are at the origin of these local relaxations. The latter eliminate the dis-
continuities in the expansion and in the cleavage energy predicted by the analytical
model. They may reduce the expansion from 25 % at least in the boundaries with a
large interplanar spacing.
To sum up, the calculated boundary energies reported in function of the macro-
scopic parameters (misorientation and grain boundary plane) are minimal energies;
to determine these minimal energies implies to know the boundary translation state.
The microscopic parameters control the interfacial energy and thus ultimately decide
to the (or more) stable structure of a grain boundary.
4.5 Are There any Geometrical Criteria of Minimum Energy?
First we must ask the physical meaning of minimum energy? Why to search for energy
minima when we know that they do not permit to select favoured grain boundaries
in the simple series of symmetrical tilt boundaries, and thus a fortiori in any other
series. Furthermore, strictly periodic boundaries may have high energies. What is the
more relevant criterion for a given boundary property: energy, periodicity or local
structure?
A boundary selected as special because its low energy has not necessarily a special
structure and special properties. However, without satisfying answer to the previous
questioning, a first selection of grain boundaries, based on their energy, enables to
constitute a pool in which one can think retrieve boundaries with specific properties.
This hypothesis seems reasonable for the properties depending of the boundary core;
but it must be limited when dealing with the mechanical properties depending on the
elastic parameters of the intergranular region.
The presentation of the geometrical criteria for minimal energy and the discussion
that follows are largely drawn from the review made by Sutton and Balluffi [30]. Four
criteria are generally proposed to predict the grain boundary energy, by chronological
order:
• A high volume density ρ = 1/ of coincidence sites, thus a low value [39],
• A high planar density of coincidence sites [40],
• A high value of the interplanar spacing dhkl , {hkl} being the grain boundary plane
indices [41],
• A high value of with constant interplanar spacing dhkl [41].
A fifth criterion is also mentioned, the existence in the interface of a high atomic
density row, common to the two crystals [42]. It has been principally used in the case
of an interface between two phases and will not be discussed in this book dedicated
to grain boundaries. It is based on arguments quite similar to those advanced for the
criterion.
4.5.1 Low Σ Value Criterion
The question of a maximal value Σmax physically significant is still the subject
of numerous debates. By physically significant, we mean that for any coincidence
index inferior to Σmax , a more or less pronounced minimum appears on the curve
giving the interfacial energy in function of the misorientation around a give axis.
However, we have seen that there is no reciprocal relationship between the value
and the interfacial energy (see Sect. 1.2.2). The most obvious example is given by the
coherent = 3 {111} twin, the energy of which being extremely low compared to
that of the incoherent 3 {112} twin; the latter is not linked to an energy minimum
on the curve γ = fn(θ ). Besides, two boundaries with 1 < 2 may have energies
such that γ1 > γ2 , as it is the case for = 9 {221} compared to = 11 {113} in
metals (Fig. 1.4).
However, despite severe restrictions, the low value criterion continues to be
used in macroscopic studies, where grain boundary character distributions in poly-
crystals are established, mainly by electron backscattered diffraction (EBSD). The
consideration of requires an associated criterion: a critical value θmax of the
angular deviation with respect to the exact coincidence misorientation such that if
θ < θmax the boundary may be considered as special. The term special refers here
to the possibility for this boundary to possess a periodicity that rests on the possible
visualization of intrinsic dislocation networks by transmission electron microscopy;
this is obviously unsatisfying.
The θmax deviation is given by:
θmax = b/ p (4.20)
with b the magnitude of the possible Burgers vector of the intrinsic dislocations and
p the boundary period.
b = bDSC = −1/n and p = −1/m (4.21)

4.5 Are There any Geometrical Criteria of Minimum Energy? 119
Different values have been attributed to m and n leading to more or less restrictive
criteria on the form:
θmax = θo −p (4.22)
with θo the limit misorientation for the low-angle grain boundaries (8 or 15◦
depending on the authors) and p varying between 1/2 and 1. Although it is the more
permissive (see Sect. 11.1.1), the Brandon criterion is most often used [43]:
θmax = 15◦ −1/2 (4.23)
If we may admit that a selection of the boundaries with low and θ value enables
to establish a first distinction between the boundary arrangements in polycrystals, we
must never forget that a low value criterion has no physical meaning as it neglects
the boundary plane. Its use raises a crucial question to progress in the grain boundary
knowledge: how to conciliate the approaches at different scales? On one hand, the
studies at the mesoscopic scales are necessary to go towards the real grain boundaries
in the materials and their properties; on the other hand, a better knowledge of the
boundary structure and of the intergranular phenomena rests on studies at the atomic
scale that concern only a very restricted number of grain boundaries.
The only physical meaning of the 3D CSL lattice is that a boundary with a plane
parallel to a rational plane of the CSL has a periodic structure. This remark leads to
consider what happens in the grain boundary plane.
4.5.2 High Γ Value Criterion
The coincidence site planar density Γ is inversely proportional to the CSL unit cell
area in the boundary plane. The existence of a low energy boundary associated to a
high Γ value is based on two hypotheses [40]:
• The greater the number of atoms on sites shared between the crystals, the weaker
the boundary core energy
• The higher the Γ value the smaller is the boundary period, less extended are the
stress fields that almost respect this period and, finally, the weaker is the elastic
boundary energy.
The first hypothesis is easily refutable as it ignores the necessary rigid body
translation that allows the other atoms (not shared) in the boundary plane to not
overlap. Besides, saying that the stress fields are not extended means that there is
a short distance between the dislocations, and then their cores may overlap; the
consideration of the period is then meaningless. On its restricted form, this criterion
may apply only to boundaries sharing a same plane.
4.5.3 High d Value Criterion
This criterion put the focus on the atomic density in the grain boundary plane, much
greater than this plane is formed by two dense planes of the crystal structure, thus with
a high interplanar spacing d. In the case of a non-relaxed boundary, the possibility of a
low interfacial energy for a high d value may be explained by the fact that d represents
the closest distance between two atoms on one side and on the other side of the
boundary plane. The atomic relaxations parallel and perpendicular to the boundary
plane impede to establish a quantitative relationship. However, the symmetrical 110
tilt boundaries giving rise to minima in the γ = fn(θ ) curve are those displaying
the largest d values. A quantitative correlation has been established between the
interfacial energy and the interplanar spacing (Fig. 4.23) [41]. Note that the grain
boundaries with the more spaced planes correspond to the favoured boundaries in the
structural unit model. This criterion has been extended to asymmetrical boundaries
by defining an effective distance [41].
d(hkl)I + d(hkl)II
deff = (4.24)
2
with d(hkl)i the distance between planes of crystal i parallel to the boundary plane.
4.5.4 High Γ Value with Constant d Criterion
The criterion of high values for all the twist boundaries with a same plane (con-
stant d) is identical to the first criterion (low value) and comes back to select
the twist boundaries with large c.n.i.d. size. This criterion is respected for all the
boundaries in metallic and ionic systems [30].
For the symmetrical 110 tilt boundaries where there is one crystal site for each
coincident node in the plane:
(hkl) = 1/A(hkl) = d(hkl) /Ω (4.25)
with A(hkl), the (hkl) plane area associated to a crystal site and Ω the atomic volume.
For these boundaries, there is no distinction between the high interplanar spacing
and the high criteria.
On the contrary, the two previous criteria totally differ for the non-periodic
boundaries. The incommensurable {111} // {100} boundaries have a null planar
density of coincidence sites although they have a relatively high effective interplanar
spacing. Such boundaries, as those with {110} // {111} and {110} // {100} planes,
do not seem to be preferred in the grain boundary formation by deposition of metallic
balls on a substrate.
To conclude, the measurement of the interfacial energies, the analyses of the
boundaries selected by the nature or resulting from different manufacturing modes
4.5 Are There any Geometrical Criteria of Minimum Energy? 121
Fig. 4.23 a Evolution of the energy of 110 tilt boundaries in copper with the misorientation
angle 2θ; b evolution of the distances of the planes parallel to the boundary plane in function of
θ. the comparison of the two curves suggests a correlation between low boundary energy and high
interplanar spacing [41]
and/or thermal treatments, finally the observations of periodic dislocation networks

were faced to the different low energy criteria. Generally, the results provide no
support to the use of the low value, nor to the high value and nor to the large
interplanar spacing criteria to predict the interfacial energy. Each of the criteria is in
agreement as often it is in disagreement with experiments. However, is seems that
the high interplanar spacing criterion is a little best in adequacy with the results,
as revealed by the observations and the energy calculations for symmetrical and
asymmetrical tilt boundaries around 011 in gold [32] and around 001 in NiO [29].
In both materials, asymmetrical incommensurable boundaries with low index planes
in the two crystals are currently observed by high-resolution transmission electron
microscopy. Furthermore, in certain cases, this criterion seems to work for predicting
a low sensitivity of a boundary to solute segregation (see Sect. 6.5.1). Finally, recent
simulations of several twist boundary structures in silicon show that the = 31(111)
boundary displays a lower energy than the (110) and (311) boundaries, and it is the
only one to not contain amorphous phase [44].
4.6 Energy and Classification of Grain Boundaries:

The Limits
4.6.1 Classification Directly Based on the Grain

Boundary Energy
The classification generally adopted [18] is similar to that of the surfaces; it

characterizes a grain boundary according to its free energy with respect to one macro-
scopic degree of freedom, i.e. by referring to a curve γ = fn(η); η represents the
misorientation angle θ or the inclination angle ϕ of the boundary plane or even the
orientation of the misorientation axis. Until now, the most used variable has been the
misorientation angle. We distinguish between:
• The singular grain boundary, the free energy of which is located to a local minimum
on one of the curves γ = fn(η), at least.
• The vicinal grain boundary, the free energy of which is near a local minimum
with respect to at least one of the macroscopic degree of freedom η. The structure
of such a boundary then consists in the structure of the neighbouring singular
boundary to which linear defects (dislocations and/or steps) are superimposed.
• The general grain boundary, the free energy of which is at or near a local maximum
with respect to one or more macroscopic degrees of freedom.
This classification established with respect to one macroscopic parameter at least
immediately presents its limits, as a general boundary with respect to one parameter
may be singular or vicinal when considering another parameters. On Fig. 1.4, the
= 3 {111} and = 11 {311} boundaries are singular, those with a small
deviation from the two previous are vicinal and the = 11 {332} is general.
The latter is general if we choose the misorientation angle as variable, but may
be considered as singular if we plot the energy evolution in function of the plane
orientation ϕ for a fixed misorientation 50.5◦ 110 (Fig. 4.24). The energy of the
incommensurable asymmetrical {111} // {331} boundary, at 1.57◦ of {111} // {13,
13, 5}, is also reported on the curve γ = fn(ϕ), although this boundary (θ = 48.53◦ )
4.6 Energy and Classification of Grain Boundaries: The Limits 123
Fig. 4.24 Evolution of the energy of the = 11, 50.5◦ 110 grain boundary in nickel with the
orientation of the boundary plane. The origin of the ϕ angles corresponds to the symmetrical {311}
orientation. The curve has been extrapolated near ϕ = 90◦ on the basis of the calculated energies
of some boundaries in the series (•). The energy of the {331} // {111} boundary is also reported
(x) although this boundary is deviated from the exact = 11 coincidence (see text) [37]
does not belong to the = 11 family [37]. However, the value of its energy appears
reasonably situated between those of the ϕ = 0 and ϕ = 80.05◦ boundaries. As well,
on the curve (Fig. 4.18), the asymmetrical (111) // (115) = 9 boundary is singular
in copper and gold where it has been observed [32].
Furthermore, what is the limit of the angular deviation from a particular η value
of a singular boundary that enables to decide if a neighbouring boundary is vicinal
or general?
Finally, the evolution of the interfacial energy with the orientation of the rotation
axis has not been studied; in the cubic symmetry, we generally consider rational
110, 100 and 111 axes.
This classification partially overlaps that proposed by the structural unit model
insofar as the favoured and delimiting grain boundaries are considered as singular
and the intermediary boundaries are general. It mostly applies to well-controlled
boundaries when only one degree of freedom varies; but its use to classify real grain
boundaries is questionable.
4.6.2 Classification Based on the Grain Boundary Interplanar

Spacing
Paidar proposes a classification of the symmetrical tilt grain boundaries in the f.c.c.
and b.c.c. systems on the form of a geometrical construction that expresses the rela-
tionship between the interplanar spacing and the orientation of the normal to the
Fig. 4.25 Dependence of the interplanar spacing d (in units a/2) with θ around [-101] for symmet-
rical tilt grain boundaries. The pyramids of the classification are given by dashed lines for all the
boundary planes reported on the figure. The points corresponding to other planes points are under
the zigzag line (full line) [45]
boundary plane; the latter totally characterizes the misorientation for any symmetri-
cal tilt boundary [45]. Although geometrical, the principle of this construction, and
the information that may be drawn, cannot be understood without knowledge of the
structural unit model (existence of favoured boundaries) and without a preliminary
discussion on the low energy criteria. That is why we have chosen to postpone the pre-
sentation of this construction here, by paying particular attention to the symmetrical
[-101] tilt boundaries in f.c.c. structure (Fig. 4.25).
At the first level of the classification, we choose a densest plane of the series, (111)
for the f.c.c. structure, and we put it between two median planes corresponding to the
single crystal, i.e. (010) and (101), by respecting its misorientation. All the planes
located between the previous ones on a same zone are defined on a stereographic
projection. Each of them is situated such that its abscissa corresponds to the angle
formed with (010) or (101). The ordinate√ of (111) corresponds to the distance between
the planes of the same family d = a 3/3 The ordinates of the delimiting planes take
into account the structure factors; the reported distances are those of (020) and (202)
planes. A first pyramid is built by joining the abscissa of the delimiting boundaries
for θ = 0◦ and θ = 180◦ to the point representing (111). This pyramid is sub-divided
into two others, the vertices of them being obtained by the intersection of the first
pyramid sides with the lines joining the abscissa of (111) to the (101) and (010)
points on the vertical axes. The so-determined planes have indices that are the sums
of the indices of the forming boundaries, i.e. (131) = (111) + (020). The symmetrical
(131) and (313) planes constitute the second level of the classification. And so on. . .
For example (121) is at the vertex of a pyramid, one side of which passing through
the abscissa of (111) and the (131) point and the other through the abscissa of (131)
and the (111) point. We then obtain boundaries at the third level, then at the fourth
level of the classification (Table 4.1).
This classification differs from that uniquely based on the planar atomic site
density, estimated from the interplanar spacing only; it is based on the relationship
between this spacing and the misorientation that is a discontinuous function. It high-
lights the importance of the grain boundary plane. The proposed classification is
purely geometric and cannot predict the grain boundary energies and the structures.
However, we note that the low energy favoured boundaries determined by simulation
of the atomic structures correspond to the boundaries situated at the levels 1 and 2 of
the present classification. This similarity may be understood as the principles used
for the construction respecting the rules that control the distribution in a general
boundary of the structural units of the favoured boundaries [46]. Furthermore, the
boundaries at the levels 1 and 2 seem to present special properties as the diffusivity,
even though they are not favoured in the structural unit model.
This classification has been extended, by the same author, to asymmetrical tilt
boundaries by taking into account the misorientation θ around a given axis and the
orientation ϕ of the boundary plane that determines its indices in each crystal [47]. An
example of the position in the space θ/ϕ of the asymmetrical [-110] tilt boundaries,
possessing a dense plane in each crystal (dense plane of the 1 and 2 previous levels)
is given on Fig. 4.26. The orientations ϕ of five planes, (001), (113), (111), (331) and
(110), are reported in abscissa and the misorientations around [−110] are reported in
ordinate. Each point of this diagram represents a boundary defined by its parameters
θ a1nd ϕ. the points on the θ axis for ϕ = 0◦ and ϕ = 90◦ correspond to the
symmetrical boundaries. The trajectory of a given atomic plane, the indices of which
being indicated on the horizontal axis, has two branches forming a parallelogram. If
we displace along these branches, we find all the possible boundaries possessing that
boundary plane in one of the crystals. The intersection of trajectories relative to two
different planes indicates the misorientation and the orientation of the asymmetrical
plane built with these two planes. Taking into account the symmetries, we must
only consider the intersections in the half-space 0 ≤ ϕ ≤ = 45◦ , those of the other
half-space are equivalent. As an example, a boundary with a (111) // (113) plane is
represented by the point P (θ = 29.5◦ , ϕ ∼ = 40◦ ) or by the point P corresponding
∼ ◦
to (111) // (11−3) (θ = 80 , ϕ = 14.75 ). ◦
Fig. 4.26 Trajectories of the

planes of the [-110]CFC tilt
boundaries in the space θ/ϕ
of the misorientations and the
boundary plane inclinations
(see text) [47]
Table 4.1 Classification of the symmetrical 110 tilt grain boundaries according to the interplanar
spacing of the planes parallel to the boundary plane in the f.c.c. system [45]
1st level 2nd level 3rd level 4th level
(101) = 1
(515) = 51 (717) = 99
(414) = 33
(313) = 19
(212) = 9 (737) = 107
(535) = 59
(111) = 3
(121) = 3 (353) = 43
(373) = 67
(131) = 11
(151) = 27 (141) = 9
(171) = 51
(010) = 1
This construction enables to define all the asymmetrical grain boundaries that we
may generate by selecting as boundary plane the dense planes until the 2nd level of
the symmetrical boundary classification. Table 4.2 summarises, for the f.c.c system,
the parameters θ and ϕ of the asymmetrical [−110] tilt grain boundaries.
The asymmetrical grain boundaries with dense planes may be considered as candi-
dates for particular properties. However, their CSL lattice is often reduced to 1D (the
coincidence nodes are on the rotation axis). Moreover, the asymmetrical boundaries
Table 4.2 List of the asymmetrical tilt grain boundaries in f.c.c. materials, obtained by selecting
atomic dense planes of the 1st and 2nd levels of the classification of the symmetrical boundaries
(see Table 4.1) [47]
Grain boundary plane θ◦ ϕ◦
(001)//(110) 90 45
(110) //(-1 -11) 144.74 17.63
(001)//(111) 54.74 27.37
(111)//(113) 29.5 39.99
(111)//(-1 -13) 79.9 14.75
(-1 -11)//(331) 131.4 11.00
(-1 -11)//(33 -1) 158.00 24.26
(110)//(-1 -13) 115.24 32.62
(001)//(113) 25.2 12.62
(110)//(-3 -31) 166.74 6.63
(001)//(331) 76.74 38.37
(-1 -31)//(331) 101.98 25.75
(-1 -31)//(33 -1) 128.5 39.01
with (010) // (101) as well as numerous other boundaries in Table 4.2 such as (111)
// (313) are incommensurable. The prediction of the asymmetrical boundaries in the
space θ/ϕ does not permit to specify what are the short period boundaries in this
space? We may think that they have rational misorientations. But, numerous bound-
aries in the θ/ϕ space display irrational misorientations and are not periodic, even
when they possess large interplanar spacings. The possible particularity of these
boundaries is an open question that deserves to be experimentally and theoretically
explored.
4.7 Grain Boundary Order or Disorder: What Conclusion?
The singular and vicinal grain boundaries, for which the dislocations and the atomic
structures are known, most often constitute a small percentage of all the boundaries
in a polycrystal. Few exceptions exist as in the f.c.c. metals with a low stacking fault
energy where a relatively high number of = 3 twins and = 3n boundaries
may appear during recrystallization. Certain manufacturing processes, as directional
growth, favour the formation of particular boundaries with a same low-index rota-
tion axis. But the general boundaries, with a mixed (tilt/twist) character and a
random boundary plane, are the real boundaries that probably contribute more to
the behaviours of the boundary ensemble in a polycrystal (crystal growth, plastic
deformation. . .). However, the knowledge of these high-energy boundaries remains
very limited.
One of the crucial questions that arises is the relaxation mode adopted by a general
grain boundary:
• Does it decompose into facets with known atomic structures and lower energies,
separated by junctions where the two structures co-exist [2] and where linear
defects are localized? In that case, the knowledge of the elementary structures
may permit to approach the general boundary; with difficulty because everything
is never the sum of its parts.
• Does it preserve a mean plane with formation of a very thin amorphous layer
(<1 nm)? Then, the periodic structures are only interesting physical objects but no
relevant for the understanding of the real material properties.
Numerous observations of general symmetrical and asymmetrical boundaries, in
different materials and at different scales, support the first hypothesis. Faceting may
occur from the mesoscopic scale to the nanoscopic scale for a same grain boundary.
Figure 4.27 shows a general 47◦ 211 boundary with a mean {757} // {535} plane
in nickel; it is decomposed in {767} //{212} facets with an average length between
40 and 50 nm, each of the planes being alternately located on one boundary side and
on the other. These facets decompose again into new asymmetrical {331} // {111}
facets of length about 4 nm separated by symmetrical nano-facets of length 1 nm;
the latter correspond to 1/2 period of the symmetrical {332} boundary (Fig. 3.9), the
misorientation of which being at 5◦ from the observed asymmetrical boundary [37].
Fig. 4.27 a Observation in transmission electron microscopy of the decomposition of an

asymmetrical grain boundary with an average {757} // {535} plane into {767} // {212} facets,
themselves decompose into nano-facets; b high-resolution observation of the structure of a {767}
// {212} facet showing that it is composed of asymmetrical {331} // {111} facets alternating with
symmetrical {332} facets with a |E− DE+ D| period [37]
The formation and the structure of the {331} // {111} boundary have been previously
discussed (see Sect. 3.3.2).
A high-energy 90◦ 110 twist boundary in gold decomposes into facets of length
about 1 nm with a {111} plane in one crystal and a {200} plane in the other. Each
facet presents a pretty good coherency, two {111} planes agreeing with three {200}
planes across the boundary plane (Fig. 4.28) [48].
The facet formation is well a usual phenomenon in general grain boundaries; but
it does not appear as the only mode of relaxation; indeed, simulations of the struc-
tures of several high-energy grain boundaries in silicon bicrystals and nanocrystals
reveal the presence of an amorphous phase. Figure 4.29 shows the simulated posi-
tions of the atoms after relaxation in a = 29(100) twist boundary, the energy
of which being 1464 J·m−2 at 0 K [44]. After relaxation at low temperature, a well-
defined planar structure appears. The same boundary equilibrated at high temperature
(growth starting from the melted material or annealing at high temperature) presents
a confined amorphous phase and a widening of the boundary core. The comparison
4.7 Grain Boundary Order or Disorder: What Conclusion? 129
Fig. 4.28 A 90◦ [110] twist

boundary in gold is no longer
planar but is reconstructed at
the atomic scale into facets
with {111} plane in one grain
and {200} plane in the other
grain [48]
Fig. 4.29 Structures of a twist = 943.60◦ [100] grain boundary in silicon: a crystalline structure
resulting from a relaxation at 0 K; b amorphous structure after relaxation at high temperature;
c evolution, plane by plane, of the mean excess energy per atom in function of the boundary plane
equidistance for each structures (the dotted line gives the energy for the amorphous bulk silicon) [44]
of the excess energies per atom (excess with respect to the cohesion energy in the
perfect crystal) in function of the interplanar spacing indicates that the energy of
the boundary containing the amorphous phase is lower than that of the crystalline
boundary; however, this energy is superior to that of the amorphous bulk silicon.
Simulations of the migration and the diffusion for tilt (310) and twist (110) and
(113) boundaries in palladium also reveal indirectly the presence of an amorphous
layer [49].
As underlined by Cahn [50], the possible existence of amorphous grain boundaries
does not constitute a new event. The idea has been proposed at the beginning of the
twentieth century by Rosenhain [17], one among the scientists at the origin of the
physical metallurgy; but the idea was faced to a great scepticism. It had been totally
abandoned in the 1950s, following the Read and Shockley dislocation model, then the
Bollmann formalism and by taking into account the transmission electron microscopy
observations. The simulations have first supported the grain boundary periodicity
owing to the structural unit model. Their recent developments, allowing the studies
of general boundaries whose structures cannot be determined by experiments, seem
to rehabilitate Rosenhain; to what extent?
Certain boundaries maintain their crystalline structure although their elevated
energies weakly differ from those of the amorphous boundaries; this is the case of a
twist = 31(111) boundary in silicon and of the symmetrical (123) tilt boundary
in palladium. A very striking result obtained by simulation is the existence at high
temperature of a transition towards a confined liquid layer in the two boundary
types, crystalline or basically amorphous at low temperature. The understanding of
this strange similarity rests on a better distinction between long-range order and
short-range order.
The long-range order defines the degree of crystallinity, but says nothing about
energy. It may be expressed by the square of the planar structure factor S(k α) with
α = 1 and 2 for crystals I and II, respectively. This factor if function of the radial
distribution of the atoms. At 0 K, for a perfect crystal I |S(k1) |2 = 1 et |S(k2) |2 = 0
and vice versa for the perfect crystal II. On the contrary, for a liquid or an amorphous,
|S(k1) |2 and |S(k2) |2 fluctuate around zero at any temperature. The combination of
the squares of the two structure factors brings a quantitative measurement of the
degree of crystallinity of each plane (Fig. 4.30). The centre of the = 11 (110)
twist boundary in palladium is strongly disordered as |S(k) |2 is almost equal to 0.25
for each plane (in crystal I and in crystal II) neighbouring the grain boundary plane.
By comparison, the symmetrical (123) tilt boundary preserves 95 % of crystalline
state in its centre, although its energy is weakly inferior to that of the (110) twist
boundary [49].

square of the planar structure
factor |S(k α) |2 in function of
the distance from the boundary
plane in the two crystals for: a
the twist 11 (110) boundary;
b the symmetrical (123) tilt
boundary in palladium [49]
4.7 Grain Boundary Order or Disorder: What Conclusion? 131
The energy value at 0 K is a measure of the degree of short-distance disorder; it

predicts the boundary properties, such as diffusion, better than the planar structure
factor does. The higher the boundary energy at 0 K, the higher the temperature of
transition of a confined solid in a confined liquid. A certain degree of local disorder
at low temperature is required to provoke a transition towards a liquid type structure
at high temperature. This local disorder has two origins: a static disorder existing at
0 K and a dynamic origin due to the thermal disorder. Under the effect of a thermal
treatment from 0 K until the melting point, the high-energy tilt and twist boundaries,
at least in the nanocrystals, undergo a reversible transition from a solid structure
at low temperature to a confined amorphous state with activation energy almost
identical for all boundaries. When the boundary energy decreases, the temperature
at which the transition occurs increases until to reach the melting temperature of the
crystal. The low-energy grain boundaries never reach the critical degree of short-
range disorder that induces a transition below Tm . These remarks only concern the
high-angle general boundaries, the cores of which being overlapping, and thus with
a more or less homogeneous distribution of the local disorders along the boundary
plane.
To sum up and to answer the question of a grain boundary order or disorder,
we must first dispel any confusion between order and energy. The energy in not
controlled by the long-range order but by the local organisation of atoms or atomic
motif. Long-standing experiments coupled to simulations have shown the existence
of strictly periodic grain boundaries with low coincidence index as = 9 in metals,
even with high coincident index as = 351 in silicon, both displaying a high energy.
Reciprocally, simulations seem to reveal the occurrence of disordered boundaries
with energies lower than those of the ordered boundaries with the same geometrical
parameters. After the whole amorphous of the beginning of the twentieth century
to the whole crystalline of the second half of the same century, it seems possible
at the dawn of the twenty-first century to consider that the two types of boundaries
exist. This is the time to develop works on really general grain boundaries in order
to confirm or infirm this proposal.
Whatever the grain boundary descriptions, detailed in this first part of the book,
they concern structures at equilibrium; but the real grain boundaries, like the crystals,
are never perfect. To raise the question of order is also to consider the disorders
induces by the defects; this is the purpose of part II of this book. Finally, the struc-
tures proposed until now describe boundaries considered as infinite; may they rep-
resent a finite boundary, constrained at its two extremities? This question will be
approached in part III together with an attempt to deal with boundary networks
in polycrystals. These preoccupations constitute pre-requisites to go towards grain
References
1. E.W. Hart, in The Nature and Behavior of Grain Boundaries, ed. by H. Hu (Plenum Press, New
York, 1972), p. 155
2. J.W. Cahn, J. de Physique 43, C6–199 (1982)
3. V. Vitek, Y. Minonishi, G.W. Wang, J. Physique 46, C4–171 (1985)
4. M. Elkajbaji, J. Thibault, H.O.K. Kirchner, Philos. Mag. Lett. 73, 5 (1996)
5. T.E. Hsieh, R.W. Balluffi, Acta Metall. 37, 1637 (1989)
6. R. Kikuchi, J.W. Cahn, Phys. Rev. B 21, 1893 (1980)
7. P. Deymier, G. Kalonji, in Proceedings of JIMIS on Grain Boundary Structure and Related
Phenomena, vol. 4 (1986), p. 171
8. M.E. Glicksman, C.L. Vold, Surf. Sci. 31, 50 (1972)
9. M. Biscondi, J. Phys. 43, C6–293 (1982)
10. T. Watanabe, S.I. Kimura, S. Karashima, Phil. Mag. 49, 845 (1984)
11. J.C.M. Hwang, R.W. Balluffi, Scr. Metall. 12, 709 (1978)
12. P. Keblinsky, D. Wolf, S.R. Phillipot, H. Gleiter, Phil. Mag. A 79, 2735 (1999)
13. G. Ciccoti, M. Guillope, V. Pontikis, Phys. Rev. B 27, 5576 (1983)
14. V. Pontikis, J. de Physique 49, C5–327 (1988)
15. O. Hardouin Duparc, M. Torrent, Interface Sci. 2, 7 (1994)
16. O.B.M. Hardouin Duparc, M. Torrent, Mater. Sci. Forum 207–209, 221 (1996)
17. W. Rosenhain, D.J. Ewen, J. Inst. Met. 8, 149 (1912)
19. D. Wolf, Scr. Metall. 23, 1713 (1989)
20. G.J. Wang, V. Vitek, Acta Metall. 34, 951 (1986)
21. V.Y. Gertsman, A.A. Nazarov, A.E. Romanov, R.Z. Valiev, V.I. Vladimirov, Philos. Mag. A
59, 1113 (1989)
22. K.K. Shih, J.C.M. Li, Surf. Sci. 50, 109 (1975)
23. G. Hasson, J.Y. Boos, I. Herbeuval, M. Biscondi, C. Goux, Surf. Sci. 31, 115 (1972)
24. D. Wolf, S. Yip, Materials Interfaces (Chapman and Hall, London, 1992)
25. A.P. Sutton, Philos. Mag. Lett. 59, 53 (1989)
26. D. Wolf, J. Mater. Res. 5, 1708 (1990)
27. W.W. Mullins, J. Appl. Phys. 6, 333 (1957)
28. S. Lartigue, L. Priester, J. Physique 46, C4–101 (1985)
29. K L. Merkle, Interface Sci. 2, 311 (1995)
30. A.P. Sutton, R.W. Balluffi, Acta Metall. 35, 2177 (1987)
31. C.T. Forwood, L.M. Clarebrough, Acta Metall. 32, 757 (1984)
32. K.L. Merkle, D. Wolf, Philos. Mag. A 65, 513 (1992)
33. O. Hardouin Duparc, A. Larère, S. Poulat, L. Priester, J. Thibault, Interface Controlled Mater.
9, 231 (2000)
34. U. Wolf, F. Ernst, T. Muschik, M.W. Finnis, H.F. Fishmeister, Philos. Mag. A 66, 991 (1992)
35. V. Vitek, Phil. Mag 73, 773 (1968)
36. Y. Mishin, D. Farkas, Philos. Mag. A 78, 29 (1998)
37. O. Hardouin Duparc, S. Poulat, A. Larere, J. Thibault, L. Priester, Philos. Mag. A 80, 853
(2000)
38. A.P. Sutton, Prog. Mat. Sci. 36, 167 (1992)
39. M.L. Kronberg, F.H. Wilson, Trans. Am. Inst. Min. Eng. 215, 820 (1959)
40. D.G. Brandon, B. Ralph, S. Ranganathan, M.S. Wald, Acta Metall. 12, 813 (1964)
41. D. Wolf, J. Physique 46, C4–197 (1985)
42. H.J. Fecht, H. Gleiter, Acta Metall. 33, 557 (1985)
43. D.G. Brandon, Acta Metall. 14, 1479 (1966)
44. P. Keblinsky, D. Wolf, S.R. Phillpot, H. Gleiter, Phys. Rev. Lett. 77, 2965 (1996)
45. V. Paidar, Acta Metall. 35, 2035 (1987)
46. A.P. Sutton, V.Vitek, Philos. Trans. R. Soc. Lond. A 309, parts I, II and III (1983)
47. V. Paidar, Philos. Mag. A 66, 41 (1992)
48. K.L. Merkle, L.J. Thomson, Phys. Rev. Lett. 83, 556 (1999)
49. D. Wolf, Curr. Opin. Solid State Mater. Sci. 5, 435 (2001)
50. R.W. Cahn, Nature 390, 344 (1997)
Part II
From Ideal to Real Grain Boundary
It is the defects which bring the crystal to life. Without
the defects the crystal would be dead just like the universe
if it were empty.
(E. Kröner, in ‘‘Continuum Theory of defects’’)
By ‘‘ideal’’ grain boundary, we mean singular, vicinal or general grain boundary,

the energy of which only contains configuration terms. This ideal state
theoretically exists only at 0K. The intrinsic or structural dislocations, with their
cores associated to polyhedral clusters of atoms, that describe the grain boundary
atomic structure (see Sect. 3.4), are not grain boundary defects, but only defects
with respect to the crystal structure.
At any temperature differing from 0K, a ‘‘real’’ grain boundary contains
crystalline defects that introduce ruptures of periodicity within the intergranular
structure. These defects are responsible for most grain boundary properties. The
density of point defects, vacancies, interstitial or substitutional atoms (solutes or
impurities), is generally higher in the grain boundaries than in the bulk, apart for
the coherent twin R = 3 {111}. The linear defects are dislocations that are not
necessary to accommodate the transformation relating the two neighbouring
crystals. These dislocations so-called ‘‘extrinsic’’ superimpose themselves to the
intrinsic network and disturb it. The planar defects are pure steps or steps
associated to extrinsic dislocations (also called disconnections). Other planar
defects formed by the dissociation of a grain boundary, that give rise to a 3D grain
boundary structure, have been presented in Part I (see Sect. 3.3.1). Indeed, this 3D
structure is a particular defect with respect to the crystal, but it is the equilibrated
structure of the grain boundary calculated at 0K.
In Chap. 5, we describe the intergranular point and linear defects, how they are
introduce in the grain boundary and the disturbances they provoke.
Then, we deal with the presence of solutes resulting from intergranular
segregation; not only the structure and the energy of the grain boundary, but also
its defects and its behaviours are strongly affected by the presence of solutes. Finally,
the chemistry of the grain boundary controls its properties and may erase differences
due to the geometry. Thus, the grain boundary segregation phenomenon deserves a
long way round; Chap. 6 is devoted to this task.
Closely related to the segregation, and sometimes difficult to distinguish from it,
at least at its first stage, the preferential precipitation of a second phase at grain
134 Part II: From Ideal to Real Grain Boundary
boundaries is analysed in Chap. 7; precipitation also plays an important role in the

grain boundary properties.
Chapter 8 focuses on linear defects or extrinsic dislocations that most often
result from the interactions between lattice dislocations and grain boundaries. The
possibility for a dislocation to enter a grain boundary depends first on a long-range
elastic interaction, attractive or repulsive, governed by the image force. This
effect, very often neglected, is described in the first section of this chapter. Then,
we present the short-range interaction between the cores of the two defects, linear
and planar, that is always attractive. We analyse the different mechanisms that
may occur in a grain boundary yielding the formation of extrinsic dislocations.
In the last chapter (Chap. 9), we pay particular attention to the mechanisms
allowing a disturbed grain boundary to return to an equilibrium state; the extrinsic
dislocation accommodation processes occur via numerous dislocation displace-
ments and reactions. These processes are described as well as the calculations of
the interfacial energy evolution during the relaxation phenomenon. All the grain
boundary reactions, since the formation of an extrinsic dislocation until the com-
plete relaxation of its elastic stress field, are modified in presence of intergranular
segregation or precipitation; this again justifies the place we give to these two
phenomena.
Chapter 5
Defects in the Grain Boundary Structure
5.1 Point Defects
As there exists an equilibrium concentration of vacancies or self-interstitials in a

crystal at any temperature, causing an increase in internal energy, there is also an
equilibrium concentration of these defects in a grain boundary. No technique enables
to directly reveal these defects owing to the low concentration of atoms in the grain
boundary core. However, simulations of grain boundaries in which point defects have
been introduced lead to relaxed intergranular structures containing such defects. The
energy attached to each site and to each type of defect (vacancy, interstitial) differs
from one situation to the other, resulting in a large number of possible configurations
[1–3].
The equilibrium concentration of a defect d on a site i of the grain boundary is
given by a relation similar to that used for the crystal:
Cd i eq. = exp(Sd i f /k)· exp(−E d i f /kT ) (5.1)
where, Sd i f and E d i f are the vibrational entropy and the internal energy of the defect
formation, respectively. For example, the internal energy of the vacancy formation
in a symmetrical = 5 [001] {310} tilt grain boundary in iron varies between 90
and 128 kJ · mol−1 depending on the intergranular site, while it is 130 kJ · mol−1 for a
crystal site. For an interstitial and for the same sites, these values are generally higher,
between 224 and 319 kJ · mol−1 while it is 458 kJ · mol−1 for a crystal site. The most
favourable site for a vacancy differs from that for an interstitial; the latter depends on
steric stresses and on the bonding forces between atoms. The displacements around
the defects remain weak although higher, for a given defect, to those involved in the
crystal (Fig. 5.1) [2].
Interactions between point defects, vacancies and self-interstitial atoms, with
several tilt grain boundaries around 110 and 111 in copper and aluminium have
been simulated using embedded atom (EAM) potentials [3]. The lower formation
energies are correlated to the grain boundary energies for the two metals. For the

136 5 Defects in the Grain Boundary Structure
Fig. 5.1 Atomic

displacement field around
an interstitial atom in a sym-
metrical = 5 {310} tilt
grain boundary in b.c.c. iron
obtained by static molecular
calculations; a view along the
001 tilt axis of the relaxed
structure before insertion of
the interstitial on the encir-
cled site; b atomic relaxations
around the interstitial; the
displacement of each atom
is represented by a vector
projected on the plane of the
figure [2]
twin boundary, the energies are quasi-identical to the defect formation energies in
the bulk. For the others, they differ from one boundary to another and, for a same
boundary, from one site to another; summary tables give the calculated values for
the two types of point defects in copper and aluminium [3].
The vacancy is usually very localized, with small atom relaxations towards the
empty site. However, in some cases, relaxations are stronger, the vacancy becomes
delocalized, leading to a local change of the atomic configuration and of the associated
free volume (Fig. 5.2a). Instability of the vacancy may even occur on certain boundary
sites (Fig. 5.2b).
The self-interstitial may exist on three structural forms in the boundaries of the
two metals: localized in a relatively open inter-atomic position, delocalized in a
pretty large area or divided on the form of a dumbbell perpendicular to the tilt axis
(Fig. 5.2c). The latter configuration requires higher formation energy than the first
two. These results, common to several boundaries in copper and aluminium, may
highly likely be extended to other f.c.c. metals; they are explained by the existence
of internal stresses and of regions alternately in tension and in compression in the
boundary core.
5.1 Point Defects 137
Fig. 5.2 a Delocalization of a vacancy on the site 2 of a = 11 {332} boundary in copper: the
atom 3 moves midway between sites 2 and 3; b Instability of a vacancy on the site 5 of a = 7
{415} boundary in copper: the atom initially at site 9 fulfills the empty site while the atom 9 moves
midway between the sites 9 and 9 ; c Interstitials on a dumbbell configuration at sites K in the
= 9 {411} and = 13 {527} boundaries in copper [3]
The point defects may concentrate in boundaries on the form of complexes. The
simulations show that a = 13; 32.2◦ [0001] grain boundary in zinc oxide acts
as efficient sink for zinc vacancies, for interstitial oxygen atoms and for substi-
tutional bismuth atoms present as impurities. The defects tend to accumulate on
the sub-coordinated sites in the boundary core and prefer to form clusters. This
segregation influences the electric characteristics of the oxide [4]. The gathering of
various solutes or complexes vacancies/solutes will be discussed in more detail in
the approach of segregation phenomena.
The capability of grain boundaries to act as sources and sinks of point defects plays
an important role in several behaviours: recovery, creep, irradiation damaging. . .
Point defects account for the fast grain boundary diffusion that controls numerous
processes in materials at high temperature.
The point defect concentration given by expression (5.1) corresponds to a
thermal equilibrium, but a boundary may be brought out of equilibrium by an
over-concentration in point defects and/or by the presence of higher order defects,
particularly linear defects.
5.2 Linear Defects: Extrinsic Dislocations
When lattice dislocations interact with a grain boundary, isolated linear defects (in
that sense that they are not arranged in periodic networks) are formed within the
boundary. These defects, called extrinsic dislocations, most often result from reac-
tions that occur in two practical situations:
• In the first stages of plastic deformation, when mobile lattice dislocations enter
grain boundaries considered as immobile
• During recrystallization when mobile grain boundaries sweep across deformed
crystal regions containing dislocations considered as immobile; the latter are incor-
porated in the boundaries.
For the boundaries, the extrinsic dislocations are really equivalent to the dis-
locations in the crystals; they explain the response of the boundary to any plastic
deformation coming from the neighbouring crystals or directly under an applied
stress. In this chapter, we describe extrinsic dislocations by reference to structural
models developed in Part I. Long-range stresses generated around them and their
relaxation is discussed later (see Chaps. 8 and 9).
5.2.1 Definition of an Extrinsic Dislocation
From the Latin, the term extrinsecus (meaning from the exterior) applied to grain
boundary dislocation may be interpreted in two ways:
• Either we put the focus on the origin of the dislocation that comes from the exterior
of the boundary, most often from the adjacent crystals; the boundary plays the role
of sink for lattice dislocations. In some materials where the Peierls stress required
to generate dislocation sources in the crystals is elevated (alumina for example),
intergranular sources may be directly activated.
• Or, we mainly consider the faulted grain boundary: extrinsic then means out of
the equilibrium grain boundary structure, whatever the dislocation origin.
In elastic terms, an extrinsic dislocation breaks the periodic arrangement of the
intrinsic dislocations (Fig. 5.3 a, b) and induces long-range elastic stresses. Its Burgers
vector does not enter the Burgers vector density B given by the Frank–Bilby equa-
tion (2.3) for a fixed relationship between the crystals:
B + be = B (5.2)
B does not account for the total misorientation between the crystals; only an
extremely localized change occurs just at the place of the dislocation.
In terms of dislocation core, an extrinsic dislocation corresponds to a break in the
periodic arrangement of structural units, caused by a lack or an excess of one (or
more) of these units.
5.2 Linear Defects: Extrinsic Dislocations 139
Fig. 5.3 a An extrinsic dis-

location D introduces a break
of periodicity in the grain
boundary structure described
in terms of primary (⊥) and
secondary (⊥) intrinsic dislo-
cations or in terms of structural
units A and B; the dislocation
D corresponds to one unit B
more or one unit A less. b
The extrinsic dislocation AB
induces a displacement of the
intrinsic dislocation network
parallel to g = 110 in a low-
angle twist grain boundary
in gold. c Atomic structure
of a = 11 {332} tilt grain
boundary in nickel: the period
is constituted by two D units
of the = 3 {111} boundary
and two E units of the = 9
{221} boundary. The absence
of a D unit at two places
along the boundary (see white
arrows) indicates the presence
of an extrinsic dislocation, the
Burgers vector of which being
equal to a DSC vector of the
= 11 boundary [5]
Figure 5.3a specifies the strong link between the description of an extrinsic dislo-
cation by reference to the intrinsic dislocation network and its description in terms of
structural units; an example is given for a symmetrical = 11 {332} tilt boundary
in nickel (Fig. 5.3c) [5].
5.2.2 Geometrical Characteristics of an Extrinsic Dislocation
The Burgers vectors of the perfect extrinsic dislocations are translation vectors of the
bicrystal. For a given grain boundary, they are equal to DSC vectors of the nearest
coincidence grain boundary: be = bDSC or be = bDSC (in particular bDSC = bm ,
the Burgers vector of the lattice dislocations).
The Frank circuit, drawn in the DSC lattice, enables to reveal the presence of an
extrinsic dislocation (Fig. 5.4). This dislocation is often associated with a step in the
grain boundary plane; this is an important characteristic of the defect that must be
taken into account in most grain boundary behaviours [6].
Fig. 5.4 Construction of a Frank circuit in the DSC lattice of a = 3 {111} twin boundary in an
f.c.c material; a grain boundary without defect; b grain boundary containing an extrinsic dislocation
of Burgers vector b = a/3 111; a step in the boundary plane is associated with the dislocation [6]
A step associated with an extrinsic dislocation results from the displacement of

atoms on one side and on the other side of the grain boundary plane. If the dislocation
is perfect, the grain boundary structure remains unchanged on both sides of the step
that corresponds to a translation vector s of the DSC lattice origin (Fig. 5.5) [7].
The step height h is an essential parameter. Expressions (2.14) giving h and b for
a secondary intrinsic dislocation are relevant for an extrinsic dislocation. So, the
prediction of the minimal heights of steps associated to two extrinsic dislocations
in a = 11 (113) grain boundary in an f.c.c. material is explained in Fig. 5.5 [7].
An extrinsic dislocation of vector bI = a/11[113] provokes a translation of a crystal
lattice with respect to the other along AB. The step vector may be defined in crystal
I and crystal II: sI = [111]I and sII = [002]II .
By indexing all the Burgers vectors in the DSC lattice (see Sect. 2.3.2), we obtain:
√
bI = sII − sI and hI = sI nI = 5a/ 11 (5.3)
also may determine h II for a dislocation of vector bII = a/22 [332]: h II =

We √
2a/ 11.
Fig. 5.5 The translation of a

crystal lattice with respect to
the other due to an extrinsic
dislocation is expressed in one
and the other coordinate sys-
tem. The Burgers vector of the
dislocation and the height of
its associated step are deduced
from this construction (see
text) [7]
Equal height steps correspond to equal and opposite Burgers vectors: the com-
bination of two dislocations that annihilate leads to the formation of a pure step of
height 2h.
A pure step in the grain boundary plane does not display a stress character. Apart
from the previous formation, it may result from the elaboration of a bicrystal by
diffusion bonding or from thermodynamic considerations (diminution of the surface
energy). Its plane possesses a high density of coincidence sites and passes through
all of them (Fig. 5.6).
Fig. 5.6 Pure step in a = 3

twin grain boundary. The
boundary plane is entirely
composed of coincidence sites
Steps associated with extrinsic dislocations or structural steps, also called dis-
connections must be taken into account when dealing with the thermodynamic and
kinetic aspects of the dislocation accommodation. Indeed, the decomposition of a
grain boundary dislocation may be favoured if we only consider the elastic energy
gain of the reaction while it does not occur if the steps associated with the products
lead to an increase of the surface energy. These steps impede grain boundary slid-
ing, but may be favourable sites for emission and absorption of atoms during grain
boundary migration.
5.2.3 Origin of an Extrinsic Dislocation
An extrinsic dislocation may result from the transfer of a lattice dislocation from
one crystal to the other leaving a product in the interface. It may also come from the
decomposition of a lattice dislocation trapped in a grain boundary into dislocations
with Burgers vectors equal DSC vectors (Fig. 5.7).
Fig. 5.7 Formation of extrin-

sic dislocations by reactions
of a lattice dislocation with a
grain boundary. a transmis-
sion of a lattice dislocation
from one crystal to the other
with a residual product in the
boundary. b decomposition
of a lattice dislocation into
several products
The Burgers vector bm of the crystal may be preserved (bm = bDSC ); it is then
a fresh dislocation introduced into the boundary at moderate temperature and whose
line is the intersection of its slip plane in the original crystal with the grain boundary
plane
When the stresses to activate a dislocation source in one crystal are too elevated,
extrinsic dislocations may be generated from an intergranular source under the effect
of an applied stress (Fig. 5.8) [8].
5.2.4 Extrinsic Dislocation Core
Models (analogous to those used for crystals) that describe the core region of an inter-
granular dislocation by taking into account the friction forces have only a qualitative
interest. However, they give information on the degree of localization of the dis-
location core that controls the dislocation reactions and motions within the grain
boundary. First, we present, very briefly, the Peierls–Nabarro model for an isolated
edge dislocation [9, 10] corresponding to the case of glissile extrinsic dislocations
in a tilt boundary. Then, few results obtained by high-resolutions transmission elec-
tron microscopy are evoked; at this scale, the core structure of a grain boundary
dislocation is resolved, but it is difficult to distinguish between its intrinsic or extrin-
sic character. Finally, we discuss about the role of the grain boundary microscopic
Fig. 5.8 Extrinsic dislocation

source within a grain boundary
in alumina [8]
parameters in the degree of localization of the dislocation core; this is a crucial point
to approach the accommodation processes of extrinsic dislocations (see Chap. 9).
5.2.4.1 The Peierls–Nabarro Model
The Peierls–Nabarro model considers the grain boundary as an inelastic layer with
thickness d separating two semi-infinite continuous elastic media. An edge disloca-
tion with a Burgers vector b equal to one crystal period is introduced in the boundary
(Fig. 5.9).
Fig. 5.9 Configuration of a dislocation in a grain boundary used in the Peierls–Nabarro model.
Two semi-infinite crystals are modeled as continuous elastic media separated by an inelastic layer
of thickness d (the grain boundary). A perfect lattice dislocation is introduced in the inelastic layer
[9, 10]
Using an approximation of homogeneous shear, the expression of the shear

stress is:
μb
τxy = − sin [4π u/b] (5.4)
2π d
with u the displacement at a point (x,y).
By substituting the dislocation by a continuous distribution of infinitesimal dislo-
cations (procedure often used in that type of calculations) we derive the displacements
at the dislocation core (y = 0):
b −1
u(x) = − tg (x/ζ ) (5.5)
2π
with
d
ζ = (5.6)
2(1 − ν)
2ζ is the core width of the dislocation and x the distance along the grain boundary.
5.2.4.2 Degree of Localization of a Grain Boundary Dislocation Core
We recall that the core of an extrinsic dislocation may be built by the addition or
the subtraction of a structural unit, in both cases compatible with the grain bound-
ary atomic structure (see Sect. 5.2.1). The evident question then is the possible
equivalence of the intrinsic and extrinsic dislocation cores, as both are associated
with the same structural units. A study by high-resolution electron microscopy of
a series of low-angle tilt grain boundaries around [011] with a {112} mean plane
shows that the dissociation width of the dislocations that constitute these boundaries
in silicon depends on the misorientation angle [11]. This result suggests that the core
of an intrinsic dislocation differs from that of an extrinsic dislocation. Similarly, in
high-angle boundaries, the core of an isolated dislocation is not exactly that of the
structural unit in the delimiting grain boundary. The comparison between disloca-
tion cores is difficult in metals as the basic units are always strongly deformed in the
intermediary boundaries [12].
Any dislocation has a tendency to delocalize its core in order to reduce its energy.
This is more likely for intergranular dislocation that, generally, the friction forces in a
boundary are lower than those in a crystal. But, simultaneously, the disturbed region
in the boundary and the surface energy increase. Thus, a balance between these two
opposite energetic terms governs the delocalization. This idea is at the basis of an
intergranular stress relaxation mode that will be detailed further.
The widening of the dislocation core is equivalent to the spreading of the dis-
tribution of its Burgers vector density and consists of very small translations less
than the boundary period. The relevant parameter of the boundary is thus its rigid
body translation (see Sect. 1.2.1). The component of this translation parallel to the
boundary plane τ// must be taken into account to understand the dislocation core
widening leading to a local shear of this plane. The component perpendicular to

the boundary plane τ ⊥ or expansion is determined by the cohesive forces across the
interface; these forces generally resist to any delocalization provided the temperature
is not too elevated and in the absence of embrittling segregation. On the contrary,
possible delocalization parallel to the boundary plane strongly depends on the grain
boundary structure, periodic, quasi-periodic (even disordered) and requires referring
to the γ -surface concept and to the cell on non-identical displacements (c.n.i.d.)
[13] (Sect. 1.2.1). The shear resistance in the boundary plane is given by the slope of
the γ -surface that represents the dependence of the interfacial energy with the rigid
body translation parallel to the interface. When the boundary is periodic, the c.n.i.d.
has finite dimensions, then the γ = fn(τ // ) curve presents maxima and minima
(Fig. 5.10a); as a result there is a resistance to any change in the translational state
of a boundary starting from the stable state. In that case, the intergranular disloca-
tion cores tend to stay localized in the boundary plane. The degree of localization
depends on the size and on the shape of the c.n.i.d., strongly connected to the grain
boundary periodicity (1D or 2D). On the contrary, the c.n.i.d. size is reduced to zero
when the grain boundary is quasi-periodic; γ is then independent of τ // (Fig. 5.10b),
the translational state may change without resistance. A dislocation introduced into
such a boundary may undergo complete delocalization; it results in the annihilation
of the strain and stress fields associated with the dislocation Burgers vector parallel
to the boundary plane. In fact, only the average grain boundary resistance is null but
there are local resistances to be overcome, then the delocalization only occurs under
thermal activation (See Chap. 9).
Finally, if the consideration of the rigid body translation enables to predict the
degree of localization of an extrinsic dislocation core, it is the kinetics of the
delocalization processes that governs the real extension of the dislocation core. This
extension depends on the temperature, but also strongly on the grain boundary chem-
istry; we will now explore the consequences of the chemistry on the intergranular
structure, defects and energy before to deal with the extrinsic dislocation behaviours.
Fig. 5.10 Change of the interfacial energy γ in function of the rigid body translation parallel to
the grain boundary plane τ// ; a in a periodic boundary; b in a quasi-periodic boundary
References
1. A. Brockman, P.D. Bristowe, R.W. Balluffi, J. Appl. Phys. 52, 6116 (1981)
2. A. P. Sutton, R. W. Balluffi, Interfaces in Crystalline Material (Oxford Science Publications,
1995)
3. A. Suzuki, Y. Mishin, Interface Science 11, 425 (2003)
4. H.S. Domingos, J.M. Carlsson, P.D. Bristowe, B. Hellsing, Interface Science 12, 227 (2004)
5. S. Poulat, J. Thibault, L. Priester, Interface Science 8, 5 (2000)
6. J.P. Hirth, R.W. Balluffi, Acta Metall. 21, 929 (1973)
7. C.M.F. Rae, D.A. Smith, Phil. Mag. A41, 477 (1980)
8. S. Lartigue, L. Priester, J. Am. Ceram. Soc, 71(6), 430 (1988)
9. R.E. Peierls, Proc. Phys. Soc. 52, 23 (1940)
10. F.R.N. Nabarro, Proc. Phy. Soc. 57, 256 (1947)
11. A. Bourret, J. Desseaux, Phil. Mag. A 39, 419 (1979)
12. O. Hardouin Duparc, S. Poulat, A. Larere, J. Thibault, L. Priester, Phil. Mag. 80, 853 (2000)
13. V. Vitek, Phil. Mag. 73, 773 (1968)
Chapter 6
Grain Boundary Segregation
The phenomenon by which a local enrichment in solutes (controlled additions or

impurities) occurs in a material is named segregation; it results from a redistribution
of the solutes between the crystals and certain faulted sites, like dislocations, sur-
faces, interfaces, during a thermal or thermo-mechanical treatment. The interactions
between point defects and linear or planar defects lead to a decrease of the total free
energy of the system and a relaxation of the defects. Concerning grain boundaries,
the segregation on intrinsic dislocations deserves to be discussed, even though the
latter only possess short-range stress fields. Furthermore, real grain boundaries most
often contain extrinsic dislocations that, akin to lattice dislocations, are likely to be
surrounded by solute clouds or Cottrell atmospheres.
Grain boundary segregation strongly modifies the grain boundary behaviours;
the solute content in boundaries may be very high even though the elements are in
very small amount in the crystals. Form an historical point-of-view, the segregation
phenomenon has been revealed and studied in correlation with its remarkable effects
on the metallurgical properties and processes. The role of phosphorus, arsenic and
sulphur on the grain boundary brittleness of iron has been reported in details in
1894 [1]. The consequences of grain boundary segregation on the structural and
functional properties of any material are considerable: weakening or strengthening,
grain boundary corrosion, grain boundary sliding, change in electrical properties…
Two types of grain boundary segregation are considered depending on the driving
forces brought into play:
• The equilibrium segregation corresponds to a solute enrichment at a grain boundary
governed by the parameters of the system (boundary/crystals) at equilibrium and
not by its history. It is a reversible adsorption phenomenon, the driving force of
which being the decrease of the system energy. Only the boundary core is affected
by the segregation (1–2 interatomic distances)
• The non-equilibrium segregation is generally due to the vacancy concentration
gradients appearing when the temperature is abruptly changed. Quenching involves
transport of vacancy–solute pairs from crystals to grain boundaries; the latter then
play the role of sinks for point defects. In binary alloys, we may observe an inverse

148 6 Grain Boundary Segregation
Kirkendall effect. Another non-equilibrium phenomenon is the solute atom drag by

mobile grain boundaries during solidification, recrystallization and grain growth.
Non-equilibrium segregation may also result from neutron irradiation effects or
from high-rate deformation. This type of segregation affects a relatively large
region from each side of the boundary plane, from the order of several micrometers.
In the following, we mainly consider the equilibrium segregation that strongly
depends on the grain boundary structure. It is also much well known although more
difficult to observe. We first present the fundamental reasons for this segregation,
i.e. the physical factors that control the phenomenon; then, we recall the thermo-
dynamic and the mechanical statistic approaches that quantify segregation (excess
quantity in the boundary, segregation enthalpy…). These preliminary considerations
concerning the average segregation, without taking into account the diversity of the
grain boundaries, have been largely detailed in numerous works (books, review arti-
cles) [2–6]. That is the reason why we have chosen to describe only partially the
various theoretical models of intergranular segregation and the results on average
segregation; we put the focus here on the equilibrium segregation anisotropy from
one boundary to the other and, moreover, from one site to the other within a same
boundary. We try to explain the link between structure and chemistry, not only in
metallic alloys (often only considered) but also in ionic and covalent solids. A special
section is devoted to the pre-wetting phenomenon as a result of an advanced state of
segregation. Then, we introduce the new concept of complexions for describing the
different grain boundary structures resulting from segregation and strongly linked to
the different boundary behaviours. Finally we address the role of the grain boundary
defects in the intergranular segregation. We do not deal with the segregation kinetics;
this aspect is well approached in the book of Saindrenan et al. [6].
The non-equilibrium segregation is more briefly described, although it is an impor-
tant practical phenomenon; however, its theoretical approach is less advanced and
it seems to not depend (or few) on the grain boundary structure. We just report on
the major features that accompany this segregation; these features depend on the
particular treatments that provoke segregation and on the properties that are affected.
Whatever the type of segregation, most results concern metallic systems where the
electronic structure effects may be neglected. The information are less numerous for
ionic systems, where effects of charge clouds are associated to the solute distribution,
and for covalent solids where the local electronic structure play an important role in
segregation.
Note that solute atom depletion can also occur to grain boundaries, but it will not
be taken into account in the following.
6.1 Driving Forces Equilibrium Segregation
Grain boundary segregation can be seen as a rejection of solute from crystals. Much
of the effects thus are explained on the basis of the bulk solubility limits given by the
Hume-Rothery rules [7] and controlled by two types of factors, steric and electronic
6.1 Driving Forces Equilibrium Segregation 149
(valence, electronic concentration, electronegativity difference). Once the solute is

in the grain boundary, coupled atomic and electronic relaxations occur in order to
decrease the energy. In principle, at a fundamental level, all the effects come from
an electronic origin; but it is convenient to separate the elastic from the electronic
effects; most often the consideration of the first ones is sufficient to understand
the interactions between solutes and grain boundaries and to obtain the segregation
energies with a good accuracy.
6.1.1 Elastic Interactions
Two types of effects must be taken into account: the first are linked to size differences
and the second to elastic modulus differences [8].
6.1.1.1 Size Effects
In this approach, the solute atom is considered as an inclusion with a parametric

mismatch with the solvent atom. The size effect results from the first Hume-Rothery
rule [7]. When the size difference between two elements increases (>15 %), their
mutual solubility decreases; the intergranular stress associated to their misfit may be
then relieved, at least partially. The interaction energy for the segregation equal to
the elastic strain energy decrease resulting from the fact that an atom too large (or
too small) for a crystal site finds a more adapted site in the grain boundary.
Let us consider r0 , the site radius (in the crystal or in the grain boundary) and
r = r0 (1 + ε) the radius of the solute atom before its incorporation in that site; ε is
positive or negative according to the solute atom size, larger or smaller than the site
size, respectively. The total strain energy has been derived by Sokolnikoff [9]:
6 π K sr03 ε2
W = (6.1)
1 + 3K s /4μ
K s is the compression modulus of the solute and μ the shear modulus of the matrix.
The segregation energy is defined as the difference between the strain energy of
the solute in an intergranular site W GB and that of the solute in a crystal site W:
E seg = W GB − W (6.2)
If the solute atom is totally elastically relaxed at grain boundary, then:
E seg = −W (6.3)
The estimated value of Eseg is of the order of several tenths of an electron volt, in
agreement with the experimental values deduced from adsorption isotherms.
When a solute atom with a size different from that of the matrix is put in a
hydrostatic pressure field P generated by a grain boundary, we must consider an
additional interaction term:
E seg = PV = 4 π K sr03 ε P(fn.(K s , K , μ)) (6.4)
This term may be neglected in most cases, as long as the pressure P remains well
below the shear modulus.
6.1.1.2 Elastic Modulus Effects
The solute atom is now considered as an elastic inhomogeneity. The elastic modulus
effects are generally less pronounced than those linked to the size; they may have
three origins depending on the reason of differences in elastic constants:
• Difference between the solute and the matrix
Solute and matrix generally display different elastic constants; then an interaction
is linked to the elastic energy stored in the inserted solute that depends on the grain
boundary stress field. This energy may be decomposed into two terms: one is due to
the dilatation Wd , the other to the shear Ws :
Wd = Ke2 /2 and Ws = μs 2 /2 (6.5)
With e = V /V, K the compression modulus of the matrix and s the shear modulus
at the boundary.
Depending on whether the solute is more or less hard than the matrix, it is attracted
or repelled from the grain boundary, respectively.
• Difference between the matrix and the grain boundary
The grain boundary is generally considered as a soft medium compared to the

adjacent crystals [10, 11]; the strain energy associated to a solute atom thus decreases
when this atom is localized in a grain boundary. The attractive energy is inversely
proportional to d 4 with d the distance between the solute and the boundary.
• Difference between the two crystals on each side of the grain boundary in an
anisotropic material
Finally, in the case of a strongly anisotropic matrix, the elastic constants differ
from on crystal to the other due to their misorientation. An image force is then created
that attracts or repels the solute according to its localization, on one side or on the
other, in the vicinity of the grain boundary. This force is analogous to that acting on
a dislocation near a grain boundary. The interaction energy is inversely proportional
to d 3 .
6.1 Driving Forces Equilibrium Segregation 151
These three effects linked to the elasticity modulus are all function of the grain
size d, the energy of which being proportional to d −n (n ≥ 1). The size effect is
generally the most important as it is inversely proportional to d.
Whatever its origin, the segregation involves a diminution of the solute content
in the vicinity of the grain boundary that generates a concentration gradient over a
large distance called depleted zone.
Although the driving force for segregation is a long-range elastic interaction, once
in the grain boundary, the solute atom may find a more or less favourable site to its
localization. The choice of this site (favourable bonding configuration) is dictated by
the short-range electronic interaction.
6.1.2 Electronic Effects
The second Hume-Rothery rule deals with the differences of electronegativity

between the solute and the solvent; if these differences are large, they lead to the
precipitation of intermediary compounds and thus limit the solubility in the matrix.
In a metal, this phenomenon may heterogeneously occur within the grain boundary.
The third rule, less general, indicates that a high valence metal is more soluble in a
solvent of lower valence than vice-versa, suggesting a greater tendency to segregation
of the element of lower valence.
The last Hume-Rothery rule concerns the electronic concentration (the number
of valence electrons per atom: e/a); it may have a special importance when the first
two factors are not favourable to segregation. Particular values of the electronic
concentration lead to the formation of compounds or intermediate solid solutions
that differ from each other by their electronic state densities. However, the local
densities of states in the interface and in the volume differ because different atomic
environments. A lower electronic energy can then be obtained if the solute is in the
boundary rather than in the volume.
The atomic relaxations may be weak or strong according to the electronegativity
differences, the e/a ratio and the angular character of the valence orbitals. They are
weak when the solute occupies a boundary site similar to a crystal site, as it is the
case for the segregation of doping elements in semiconductors. They are strong when
new atomic bonds are formed; the boundary site differs from the crystal site leading
to a new density of electronic states in the interface.
6.2 Thermodynamic Approaches of Equilibrium

Segregation
The phenomenology of interfacial segregation is the same whatever the type of

the interface; differences appear only in the thermodynamic parameters linked to
the different structures and chemical bonds in the interface. In particular, the con-
cepts and the laws of surface segregation are generally used to study intergranular
segregation [12]. The approaches of the equilibrium segregation presented here are
not exhaustive; they just trace the outline of the history of the phenomenon. For a
thorough understanding of the various equations that have been proposed depending
on the systems and on the hypotheses, the reader may refer to review papers and
books [2–6, 12].
We first deal with the models derived from the classical thermodynamic which
have been validated by experimental results, essentially obtained by Auger electron
spectroscopy. These approaches initially developed to describe the average segrega-
tion at random grain boundaries in a polycrystal have been extended to the segregation
at a well-characterized boundary in a bicrystal. We then turn to models based on the
statistical mechanics by noting that, under certain assumptions, we find the equa-
tions derived from classical thermodynamic. The distinction adopted here between
the two types of approach of segregation is quite formal; in fact, in the models related
to the classical thermodynamic, the concepts of favourable sites and mean field, at
the heart of statistical mechanics, are already taken implicitly into account [2]. The
different approaches attempt to predict the segregation level in function of the solute
concentration in the bulk and of the temperature; while some are mainly used to
exploit the experimental results, the others are the basis of simulations that aim to
link segregation to the atomic structure of grain boundary.
6.2.1 Gibbs Adsorption Isotherm
The Gibbs adsorption isotherm has been established for an ideal solution sufficiently
dilute to obey the Henri law: aS = kS X S , with aS , the activity of the solute S in the
solid solution and X S , its atom fraction (or mole fraction). The activity coefficient
γS is here called kS to avoid any confusion with the interfacial energy γ .
In such a system, the solute excess in a grain boundary obeys the classical Gibbs
equation:
1
ΓS = − (∂γ /∂ ln X S )T,V (6.6)
RT
Expression (6.6) relates ΓS , the excess of the S species per unit grain boundary
area to the variation of the interfacial energy γ (in J · m−2 ) in function of the atom
fraction of solute X S in the matrix. R is the ideal gas constant. The slope of the curve
(∂γ /∂ ln X S ) for X = 0 indicates the tendency of the solute atoms S to segregate at
grain boundaries; it gives the interfacial activity (in J · m−2 /atom %) by analogy to
the surface activity, often used in surface chemistry.
Despite its simplicity, expression (6.6) only finds limited applications, because
specific interfacial segregation measurements are extremely difficult to carry out
for various values of the matrix solute content X S and the temperature T. Therefore,
several other approaches have been developed to directly connect the grain boundary
concentration to X S and T.
6.2 Thermodynamic Approaches of Equilibrium Segregation 153
6.2.2 Segregation in Regular Solid Solution Without Interaction
Generally, for a regular solid solution and with the assumption that the solute and
the solvent occupy the same partial molar volume, the equation describing grain
boundary segregation takes the following form:
X SGB XS
= exp(−ΔG S /RT ) (6.7)
(X 0
GB
− X SGB ) (1 − X S )
GB
X SGB is the molar fraction of solute S in the boundary; X 0 is this molar fraction
at saturation. The free segregation enthalpy ΔG S of the solute S must be negative
in order that the S atoms segregate to grain boundaries; this energetic term has two
components: the difference ΔG 0S between the standard molar Gibbs free energies of
the two phases, considered separately, and the free enthalpy excess ΔG ES resulting
from the segregation of the solute S. Equation (6.7) can account for the segrega-
tion of substitutional and interstitial solute atoms in the intergranular structure. It
was declined in different forms depending on the assumptions concerning mainly
the maximum coverage (at saturation) of the grain boundary by the solute and the
expression of the excess energy.
6.2.2.1 Langmuir–McLean’s Model
While preserving the treatment of classical thermodynamic, McLean introduces the

statistical mechanics formalism with the concept of distortion energy of the crys-
tal lattice in the vicinity of the solute atom [13]. The grain boundary structure is
considered to be composed of a finite set of distorted sites. The free energy of a
solute atom in an interfacial distorted site differs from that in a crystal site; the
difference defines the segregation energy associated to the boundary site. The total
energy decrease depends on the proportion of distorted sites filled with solute atoms;
a saturated state of segregation is reached when all the sites are occupied.
When the segregation only occurs on substitutional sites starting from a regular
(very diluted) solid solution, the free enthalpy excess is null and the saturation is
reached for a monolayer of solute at the grain boundary (all the solvent atoms are
replaced by solute atoms). Expression (6.7) can then be written on the simplified
form of the classical segregation isotherm of Langmuir–McLean [13]:
X SGB XS
= exp(−ΔG 0S /RT ) (6.8)
(1 − X SGB ) (1 − X S )
This expression is formally equivalent to the Langmuir isotherm for the adsorption
of an element on a free surface.
6.2.2.2 Seah and Hondros model
To extend the previous approach to interstitial-solute segregation at grain boundaries

(even if the elements are in substitution in the matrix), it is necessary to start from
Eq. (6.7) in which the boundary saturation level is reached for a percentage of occu-
GB
pied sites X 0 different from unity. Indeed, the solvent atoms always occupy the
substitutional sites n Jsubst., and the maximum fraction at saturation is:
GB
Nint.
GB
X0 = GB + n GB
(6.9)
Nint. subst
The solute atoms do not necessarily fill all the interstitial sites; a coverage ratio
θSGB is defined [14]:
X SGB
θ GB
S = GB
(6.10)
X0
The isotherm equation of segregation then takes the form:
θ GB XS
S
= exp(−ΔG 0S /RT ) (6.11)
(1 − θ GB
S )
(1 − X S )
Other reasons than the interstitial segregation in the boundary may limit a mono-
layer formation at saturation, like the existence of an ordered interfacial structure
in alloys that has been discussed in [15]. The boundary is considered as formed by
ordered regions in which no segregation may occur and disordered regions that obey
the Langmuir–McLean formalism.
Finally, Seah and Hondros [14] introduce a solid solubility limit of the solute S
in the matrix X S ∗ ; the segregation equation becomes:
X SGB XS
= exp(−ΔG S /RT ) (6.12)
(X 0GB − X SGB ) X S∗
GB
For low enrichment levels X SGB X 0 , expression (6.12) is reduced to:
X SGB 1
= exp(−ΔG S /RT ) (6.13)
GB
X0 X S X S∗
The first member of this equation is named solute enrichment factor β. Like the
interfacial activity (∂γ /∂ ln X S ), it represents the propensity of an element to
segregate at grain boundaries. When the atom fraction of the grain boundary plane
GB
covered by the solute at saturation is considered equal to a monolayer (X 0 = 1),
then β is simply the ratio between the solute concentration in the boundary and its
concentration in the matrix:
X GB
S
β= (6.14)
XS
GB
Whatever the X 0 value, experiments proved that ΔG S is often small and displays
similar values for different systems, the solute enrichment factor of a solute S at a
grain boundary is then considered as inversely proportional to the solubility limit of
S in the matrix:
K
β= (6.15)
X S∗
With 1.8 ≤ K ≤ 10.8.

This simple relation has been used to predict the enrichment factor of an element S
knowing its solubility limit in the matrix X S ∗ . The results giving β in function of X S ∗
are reported in logarithmic coordinates for various metallic systems on Fig. 6.1 [14].
The ratio R = Lnβ/LnX S∗ is almost equal to −1 for a large number of metal-
lic alloys [14]; however, it may vary with certain physical parameters such as the
thickness and the excess volume of a boundary that influence segregation [16]. The
boundary width modifies the ratio R only if it is less than 0.4 nm; beyond, R tends
towards a constant equal to −1; thus it does not play on R, the usually admitted width
being equal to 0.5 nm. On the other hand, there is a strong dependence of R with the
grain boundary atomic density, that is reflected in the ratio between a grain boundary
elastic constant and the corresponding matrix constant: h11 = C11 /C11 (similar for
h44 = C44 /C44 ) (Fig. 6.2). Indeed, C11 and C44 in a grain boundary are usually less
Fig. 6.1 Curves showing

(in logarithmic coordinates)
the relationship between the
enrichment factor β measured
for different solutes S in
various systems (Fe, Cu, Ni)
and the solubility limit of the
solute in the matrix X S∗ [14]
Fig. 6.2 Variation of R (in

logarithmic coordinates)
with the ratio h11 of the
elastic constants of the
boundary and the crystal for a
boundary of width 0.5 nm in
aluminium [16]
than to C11 and C44 [11], and the more so that the boundary is less dense [16]. The
two extreme cases h11 = 1 and h11 = 0 traduce an absence of segregation of a very
diluted solute in a perfect crystal and its very strong segregation to a free surface,
respectively. By considering an atomic density in a high-angle grain boundary infe-
rior from 5 to 15 % to that in the crystal (i.e. a ratio h11 between 0.3 and 0.4), the
R ration varies between −0.8 and −1.1 (Fig. 6.2), in agreement with the determined
empirical value close −1 [14].
The value of the K parameter (relation 6.15) decreases when the temperature
and/or the solute bulk concentration increase. It also depends on the presence of a
third element in the solid solution that, not only affects the solubility limit of the first
solute, but also modifies the interaction between atoms, favouring or not segregation.
However, the enrichment factors being known with an accuracy of only one order of
magnitude, their dependence with the solubility limit remains valuable for a ternary
solution.
The previous models allow a satisfying approach of the segregation in a large
number of binary or multiple systems; but, they do not account for the segregation
dependence with the temperature in several systems of practical interest (sulphur,
phosphorus or boron in iron and alloys, bismuth in copper …), where strong inter-
actions between solutes and solvent or between different solutes occur in the matrix
and in the grain boundary.
6.2.3 Segregation in Regular Solid Solution with Interactions
In the cases where the interactions between atoms cannot be neglected, we must
refer to the general Eq. (6.7) of the segregation isotherm by using an appropriate
approximation of the excess enthalpy ΔG ES .
6.2.3.1 Fowler and Guggenheim’s Model
In the Fowler and Guggenheim model [17], the interaction between the solute atoms
GB
is given by a term so-called Fuller term equal to 2Zωij X SGB / X 0 with Z the number
of nearest neighbours in the boundary and ωij the pair interaction energy. For a binary
solid solution where the solute may be localised on the site i or j of the boundary:
ωij = A[εij − 1/2(εii + εjj )] (6.16)
A is the Avogadro number εij , εii and εjj are the pair potentials between atoms in
the sites i and j. The segregation equation takes the form:
X SGB XS
exp[−(ΔG ◦S + 2Z ωij X SGB / X 0 )/RT ]
GB
= (6.17)
(X 0GB − X SGB ) (1 − X S )
In this expression, the sign convention is such that a positive value of ω indicates
a repulsive interaction between solute atoms in the grain boundary. The total segre-
gation energy ΔG S is increased (absolute value reduced); it results in a segregation
level lower than in the absence of interaction and in its progressive decrease with
X SGB increase.
6.2.3.2 McLean and Guttmann’s Model
An important step was taken in the segregation approach with the models of multiple
segregation of McLean and Guttmann that take into account the mutual interactions
between atoms in a multi-component system [4, 18]. Different cases of segregation
are discussed:
• The first case concerns a regular substitutional solid solution with site competition.
It is supposed that solutes and solvent are randomly distributed on equivalent
sites, in the matrix as well as in the boundary, and the pair interaction energy
between close neighbours is constant. The molar interaction energy αij between
two elements i and j is equal to Z ω where Z is the coordination number and ω
is given by expression (6.16). On the contrary to the previous model, the sites i
and j may be filled with solute or solvent atoms. In a ternary system constituted

by the matrix M and two solutes S and S , the effective interaction coefficient αSS
between solutes is equal to the interaction energy between solutes in the matrix
minus the sum of the interaction terms of each solute with the matrix:

αSS = αSS − αSM − αS M (6.18)
The segregation equation for a solute S takes the form:
X SGB XS
GB
= exp(−ΔG S /RT ) (6.19)
XM XM
ΔG S = ΔG ◦S − 2αSM (X SGB − X S ) + αSS

(X S − X S )
GB
S = S, M (6.20)
And, similarly for the solute S .

Expression 6.20 neglects the fact that the interaction coefficients in the matrix and
in the grain boundaries must differ, owing to the differences between the coordination
numbers and the interatomic distances; it considers that αij = αijGB (the indices
i and j representing the matrix as well as the solutes S and S ). More complex
expressions may be found in Refs. [4, 6]. Positive αSS ,α
SM and αS M interaction
terms indicate a repulsive interaction between elements. Thus, the Gibbs energy of
segregation becomes more negative if the solute atoms S (or S ) repel solvent atoms
(αSM or αS M > 0) or if the solute atoms S and S attract each other in the boundary,
leading to an increase of their segregation levels in the boundary. A strong attraction
between the solutes (αSS
0) intensifies their segregation, even if the activity of
one of them in the binary system is null. In the case of a strong repulsion (αSS 0),

the segregation of one solute may promote the desegregation of the other, even if the
elements are weakly active in the solvent.
• The second case of segregation concerns regular solid solutions with two types
of solutes, substitutional S and interstitial S , without site competition. The site
lattice is subdivided into two sub-lattices: one is completely filled with the solvent
and the substitutional S atoms, the other is filled with the S atoms in insertion in
the boundary. The proportions of the two types of sites are given by a (substitution)
and b (insertion) with +b = 1. The atomic fractions in the subset are YSGB for the
substitutional solute and YSGB for the interstitial solute. The interactions occurring
between the sub-lattices, the α coefficients are replaced by φ coefficients, the
meaning of which being emphasized in [4]. We use the symbol φ instead of β
in order that no confusion can be made between this ternary coefficient and the
enrichment factor previously introduced [14]. Three coefficients are defined: φSM
and φVS indicate the interactions between the solute S and the matrix M in the
substitutional sub-lattice and the interactions between the solute S and the vacan-
cies V (non-occupied sites) in the interstitial sub-lattice, respectively; the last term
expresses the interaction between the two segregated solutes. The segregation
φSS
equation takes the form (6.7) giving the ratio YX J /(1 − YJ ) in function of ΔG
X X

with x = S or S and Y 0 GB
= 1. By neglecting the interactions between the solute
atoms and the solvent atoms or the vacancies, we obtain the expressions of ΔG:

φSS
ΔG S = ΔG ◦S + (YSGB
− YS ) (6.21)
b

φSS
ΔG S = G ◦S + (YSGb − YS ) (6.22)
a
When the φSS coefficient is negative, the segregation energy of each of the
element is lowered be the presence of the other; the interaction is attractive. This
synergistic phenomenon of equilibrium co-segregation is particularly found in sys-

tems where M or S is a transition element and S a metalloid element [4].
• Quasimolecular behaviours with and without site competition constitute the two
following cases studied by McLean and Guttmann. They take into account the for-
mation within the boundary of stable clusters of compounds Sx S y . In the absence
of site competition, the clusters of quasi-molecules may occupy substitutional
sites; interstitial sites are filled either by free atoms of the element S or by S atoms
combined with those of S or with vacancies; moreover, the clusters do not inter-
act. On the contrary, in the presence of site competition, the clusters of the Sx S y
compound can interact. Finally, strong interactions may lead to the formation in
the boundary region of a bi-dimensional film, the composition and the structure
of which prefiguring the three-dimensional compound [4]. Like in the matrix, the
compound may appear as islands. In certain cases, the atomic models predict the
appearance in the boundary of a local environment of the solute atom analogous
to that adopted in a 3D precipitate in the matrix.
In fact, the segregation enthalpy varies from a grain boundary to the other and
from one site to the other in the same boundary. This means that, strictly speaking, the
above equations to determine ΔG ◦S are valid for the segregation of a given element at
a given site of a grain boundary in a bicrystal and assuming a low boundary coverage
by the solute. However, these equations have been most often used to determine
the dependence with the temperature of the average solute concentration at grain
boundaries in polycrystals.
It is thus important to study grain boundaries with well-characterized geometrical
parameters in order to determine significant values of HS◦ and SS◦ and to analyse
their specific behaviours to segregation. In particular, a question is raised up: is
there an appropriate geometrical criterion to predict segregation? At a finer scale, the
segregation is linked to the atomic grain boundary structure and the favoured sites for
a given solute within a structural unit are selected. These are mostly simulation studies
that have advanced in this direction, the experimental evidence remains difficult.
6.3 Segregation Models Based on the Statistical

Mechanics
These models enable to predict the equilibrium distribution, site-by-site, of the solute
and solvent atoms in the grain boundary region for given temperature and pressure.
They generally consider that, for a binary system M–S, all the boundary sites are
occupied by atoms M or S; they limit the approach to substitutional solutions. The
site positions are not fixed; the number of atoms M and S is not defined while the
chemical potentials of atoms M and S are. Furthermore, we assume that there are
no vacant sites. The distribution of atoms on the sites of the interface is actually
not constant, fluctuations exist over time. We then consider the average position and
occupancy of each site over a large period of time to obtain values that converge to
equilibrium.
Let us consider pi the probability of occupation of a site i at a given time:
pi = 1 if the site i is occupied by an atom S

pi = 0 if the site i is occupied by an atom M
The average value over time of pi is

pi = ci ; it is the average probability of
occupancy of a site at equilibrium with 0 ≤ ci ≤ 1.
If Ri is the position of a site at time t, we define ri =
Ri .
In the grand canonical ensemble (the numbers of atoms M and S being not fixed),
the grand potential Ω is expressed by:
Ω = F({ci }, {ri }) − TS c − μM NM − μS NS (6.23)
F if the Helmholtz free energy:
F({ci }, {ri }) = E({ci }, {ri }) − T Sv ({ci }, {ri }) (6.24)
Sc is the configuration entropy, Sv the vibration entropy and μi the chemical potential
of the solute i in the solution.
The equilibrium state is obtained by minimizing Ω by using different approxi-
mations [2].
6.3.1 Model of Regular Solid Solution with the Bragg-Williams

Approximation
In this model, we consider that the configuration entropy is that of an ideal solution
(without interaction) of atoms M and S.
The minimization of yields the following relation for the occupancy of a site k:
ck c0
= exp [−(1/kT )(∂ F/∂ck − ∂ F/∂c0 )] (6.25)
(1 − ck ) (1 − c0 )
For a disordered solid solution, the average occupancy of the sites in the matrix c0
is simply the concentration of solute S in the solution; while, for an ordered solution,
we must only consider the non-equivalent sites in the unit cell. If all the sites in the
boundary are equivalent but different from those in the matrix and if ∂ F/∂ck does
not vary with the site occupancy, then we find again the McLean expression (6.8).
This approach has been extended to a ternary system, S and S being two solutes
in the matrix M; Eq. (6.25) becomes:
ciS c0S
= exp[−(1/kT )(∂ F/∂ciS − ∂ F/∂c0S )] (6.26)
(1 − ciS − ciS ) (1 − c0S − c0S )
and, a similar expression for the solute S .

6.3 Segregation Models Based on the Statistical Mechanics 161
6.3.2 Mean Field Approximation (MFA) Models
The Mean Field Approximation (MFA) models take into account the interactions
between the atoms M and S in the interface, first described by pair potentials then
by Finnis-Sinclair N-body potentials. A hybrid atom, the character M or S of which
varying with the local atomic environment in a self-consistent manner, replaces each
M or S solute atoms. These models enable to analyse the link between the segregation
and the local grain boundary atomic structure. The interaction energies are described
by pair potentials εijMM , εijSS and εijMS that vary with r = (Ri − Rj ), Ri and Rj locate
the positions of sites i and j. The interaction energy between two atoms M and S,
one on a site i and the other on a site j, is unchanged by site inversion: εijMS = εjiMS .
By minimizing the value of the grand potential, we obtain an isotherm linking the
occupancy ck of a site k to the volume concentration of the solute and the temperature:
ck c0
= exp −(1/kT )(γk − γ0 ) (6.27)
1 − ck 1 − c0
γk is the local field at the site k that depends here on the occupancy of the neighbouring
sites of k; γ0 is the local field of a site in the matrix, the occupancy being c0 . The local
field is the energy difference associated to the chemical potential difference between
an atom S and an atom M. The difference (γk − γ0 ) is the energy to exchange an
atom M at the site k of the grain boundary with an atom S of a site in the crystal.
A similarity between the present models and the thermodynamic models for
a binary solution exists if we consider that only certain well-separated sites are
favourable to segregation, each of them displays the same local field and the segre-
gation energy f seg = γk − γ0 is independent of the site k and of the segregation
level.
Equation (6.27) then takes the form of the McLean isotherm (6.8). If we take
into account the local coordination number z, then the expression of the segregation
energy contains a term equivalent to the Fowler term in Eqs. (6.17) and (6.23) and
in equivalent expressions. The link between the statistical mean field approach and
the multiple segregation models is not straightforward, as the latter use an average
segregation energy obtained by considering several possible segregation sites. It is
then useful to introduce the concept of site density in a local field γk . Furthermore,
models derived from the mean field approximation are limited as far as they ignore
the correlations between the different site occupancies. A significant improvement
was made to the segregation approach by taking into account the correlations at
the same site or autocorrelation approximation. In any case, vacancies are seen in
thermal equilibrium (μV = 0), i.e. we neglect the relaxation energy of each vacancy
that affects the equilibrium of the neighbouring sites occupied by atoms.
For a more complete treatment of statistical approaches, the reader can refer to
the book by Sutton and Balluffi [2].
6.4 Average Segregation at Grain Boundaries
Whatever the form of the segregation equation, the average solute concentration at
grain boundaries depends on its matrix concentration, on the temperature at which the
solute transport from the crystals to the boundaries occurs and on the free segregation
enthalpy.
6.4.1 Dependence of the Average Grain Boundary Segregation

on the Temperature and the Solute Concentration
The evolutions of the coverage ratio θ of a solute at grain boundaries in a polycrystal

in function of the temperature, for a fixed concentration X S , and in function of the
material composition, for a fixed temperature, are schematically shown on Fig. 6.3,
by considering a constant segregation free enthalpy without interaction between
matrix and solute. Very often, the segregation isotherms determined on the basis of
experimental results differ from the previous, as shown on the following examples
for pure iron and an iron alloy. The influence of the temperature on the phosphorus
segregation at grain boundaries in iron is given on Fig. 6.4a [19]. The schematic
shape on Fig. 6.3b is truncated; segregation starts as soon as the metal contains
trace amounts of phosphorus. The data of the five previous curve are reproduced
on a unique curve by referring to the overall concentration X S normalised to the
solubility limit X S ∗ in the matrix at each temperature (Fig. 6.4b) [20]. The phosphorus
concentration at grain boundaries is about 80–90 %; this level is reached for matrix
contents much lower than the solubility limit. This result may due to the fact that
the coordination numbers at grain boundaries are not so different than those in the
crystal when numerous boundary sites are rapidly occupied; alternately, it could be
the manifestation of a repulsive force between the phosphorus atoms segregated to
grain boundaries.
The segregation of antimony in an iron–nickel alloy strongly differs from that
of phosphorus in iron, the segregation level is clearly lower and the segregation
(a) (b)
Fig. 6.3 Schematic evolutions of the solute segregation in a given metal (ΔG S constant): a with
the temperature at constant X S [19]; b with the concentration at constant temperature
6.4 Average Segregation at Grain Boundaries 163
(a)
(b)
Fig. 6.4 a Evolution, for five temperatures, of the concentration X PGB (in atom %) of phosphorus at
grain boundaries in iron in function of its concentration X P in the matrix [19]; b same data reported
on a unique curve with X PGB in function of the matrix concentration normalized with respect to the
phosphorus solubility limit X P∗ at the segregation temperature [20]
Fig. 6.5 Antimony segrega-

tion in grain boundaries in an
iron–nickel alloy in function
of the matrix concentration
normalised with respect to the
∗ [20]
solubility limit X Sb / X Sb
continues to increase beyond the solubility limit (Fig. 6.5). The driving force for
antimony segregation is thus less than that for phosphorus and the segregation is
not simply related to the solubility limit. Furthermore, it is likely that sites remain
available for antimony in the boundary even if the content in this element in the
matrix exceeds its solubility limit [20].
6.4.2 Influence of the Interaction Term on the Average

Segregation
In numerous cases of grain boundary segregation, it is necessary to take into

account an interaction term between solutes to describe the total segregation enthalpy
G S . The segregation isotherms calculated from experimental results by using the
Fig. 6.6 Calculated grain boundary segregation isotherms drawn according to the Fowler-
Guggenheim equation (or McLean when ω → 0) starting with the experimental results obtained
on Phosphorus [19], tellurium and selenium [21] and a solute A [22] on grain boundaries of b.c.c.
iron. The percentage X of occupied sites in the boundary is given on the ordinate. On the abscissa,
the term c0 /(1 − c0 ) exp(−βΔf ) is such that β = 1/kT and Δf is the segregation energy without
interaction (∼
=ΔG 0 ). The value of the interaction parameter −2Z ωβ is indicated for each solute
on the corresponding curve. The S shape of the segregation curve (dashed curve) for a very high
value of this parameter (very strong interaction) indicates a metastable state where two phases of
different compositions coexist in the boundary [22]
Fowler-Guggenheim equation (6.17) (reduced to the McLean equation (6.18) when

the interaction ω tends to 0) are reported on Fig. 6.6. The systems considered are
Fe–S where S is a solute, by increasing order of the interaction term: phosphorus
[19], tellurium and selenium [21] and a solute A with high interaction energy in
the boundary [22]. When the interaction energy between solute atoms segregated to
grain boundaries becomes strongly negative (strong interaction), the curve takes an
S shape, indicating a metastable state where two phases with different compositions
coexist in the boundary; selenium displays a behaviour at the transition towards a
phase separation.
The use of the interaction term αSM defined by McLean and Guttmann, for a
substitutional solute, leads to the same effects on the equilibrium curves and on the
segregation isotherms in a binary system. In particular, beyond a certain temperature,
the appearance of a cusp point on the equilibrium curve indicates that three degrees
of overlap may be reached by isothermal treatment (identical to the three intersection
points of a vertical with the dashed curve on Fig. 6.6) corresponding to three solutions
of the segregation equation. Only the energy value that the grain boundary takes
according to its segregation level enables to discriminate the solutions.
Thus, a model of segregation curve in function of the solubility limit does not apply
to all elements likely to segregate at grain boundaries, even in the case of a common
matrix. The chemical bonds created by each element within a grain boundary govern
6.4 Average Segregation at Grain Boundaries 165
its segregation; these bonds strongly depend on the local structure and then vary from
a grain boundary to the other and from one region to the other in a same boundary.
The segregation deviates from the conventional isotherms when the average grain
size becomes less than a critical size that depends on the solid solution nature and
composition and on the temperature. It then displays a clear dependence of this size.
It is the case for the segregation of calcium in titanium oxide (TiO2 ) polycrystals
when the grain size varies between 150 and 350 nm [23]. In this regime, the ratio
between the area occupied by the boundaries and the material volume is an important
factor for the grain boundary coverage by a solute. The grain boundaries are saturated
with calcium when the coverage ratio becomes equivalent to half a monolayer. An
increase of the total equilibrium solubility of the solute due to segregation is also
observed. This size effect on the grain boundary segregation may have important
consequences on the manufacturing, the microstructure and the properties of ultra-
fine grained materials.
6.5 Relation Between Segregation and Grain Boundary

Structure
For a given material, the quantity and the distribution of a solute in a grain boundary
strongly depend on the macroscopic and microscopic parameters that character-
ize this boundary and determine its structure. Reciprocally, a grain boundary may
undergo a change of its structure resulting from solute equilibrium segregation. The
study of the reciprocal relationship between structure and segregation is fundamen-
tal as this is the structure/chemistry couple that confers to the grain boundaries
their properties and thus their contribution to the overall behaviour of polycrystals.
Numerous experiments show the notorious influence of a solute on the grain boundary
answers to various stimuli: corrosion, diffusion, grain boundary sliding and fracture.
They reveal that this influence varies from a grain boundary to the other and thus
indirectly support the existence of a strong link between structure and segregation.
But, very few direct experiments exist that attempt to analyse this relation; this is
mainly due to the spatial resolution required in these analyses, the grain bound-
ary width being of the order of 0.5 nm. Very promising results combining atomic
structure and nanoanalysis were obtained for a few grain boundaries and in some
solvent–solute systems; but until now, a systematic experimental investigation that
tries to directly establish a relationship between grain boundary structure and chem-
istry does not exist. However, many theoretical and simulation approaches have been
developed; They enable, among other things, to determine the segregation energies
for well-characterized boundaries, to predict the structural units favourable for the
segregation of a given solute and the changes of the grain boundary structure likely
to occur in presence of segregated elements.
The results obtained by simulations and by experiments converge on the exis-
tence of a large intergranular segregation anisotropy, on one hand between different
boundaries and, on the other hand, between different sites within a same boundary.
However, this anisotropy progressively vanishes when the temperature increases.
Investigations reveal the existence of grain boundary structural transformations
induced by the presence of solute and belonging to the phase transition groups defined
in part I (see Sect. 4.1.1) [24].
• The congruent phase transitions are such that the five macroscopic geometrical
parameters stay unchanged. Under the effect of segregation and depending on the
observation scale, we note changes in the arrangement of intrinsic dislocations
or the appearance of a structural unit different from that characterising the pure
boundary, even the formation of an ordered solvent–solute structure. Segregation
may also lead to a wetting transition; this aspect will be discussed separately (see
Sect. 6.8) because the formation of a liquid film is somewhat beyond the classical
segregation phenomenon and, in a certain manner, may be considered close the
precipitation (phase transformation) phenomenon.
• The non-congruent phase transitions, for which at least one of the geometrical
parameters changes, manifest themselves by grain boundary faceting (inclination
variation) or dissociation (changes of inclination and misorientation).
We choose to present here some typical results obtained at different investigation

scales, going from the simple consideration of the geometry of the grain boundary to
its description in terms of intrinsic dislocations and finally to its atomic and electronic
structures.
6.5.1 Segregation and Grain Boundary Geometrical

Parameters
We first deal with the different segregation levels experimentally observed and
measured in function of the misorientation and/or the grain boundary plane ori-
entation. Then, the segregation anisotropy is quantitatively described by associated
thermodynamic quantities that have been calculated for different well-geometrically
characterized boundaries. The evolution of the anisotropic character of the segre-
gation with the temperature is discussed as well as the practical consequences it
involves. Finally, in the last session, we account for the changes of some geometrical
parameters under the effect of a solute indicating a phase transformation at grain
boundaries.
6.5.1.1 Observed and Measured Segregation Anisotropy
The existence of a segregation anisotropy is indirectly revealed by the large dis-

persion of the results obtained on polycrystals of a given alloy by Auger electron
spectroscopy. For example, the phosphorus content in steels may vary from 45 %
6.5 Relation Between Segregation and Grain Boundary Structure 167
from a grain boundary to the other; comparatively, the phosphorus variation along a
given boundary appears relatively small [25].
The differences between the segregation levels have been first attributed to the
differences between the boundary misorientations. Knowing that equilibrium seg-
regation is localised at the grain boundary core (about two interplanar distances)
and observing solute variations along a same boundary, consideration of the grain
boundary plane quickly becomes necessary. Auger electron spectroscopy analyses,
after fracture of various grain boundaries in iron, show a high phosphorus concen-
tration when the boundary planes have high indices in both crystals, while the level
is low for low index planes [26, 27]. Moreover, different phosphorus quantities are
detected on the two fracture surfaces of an asymmetrical boundary according to the
corresponding face indices [27].
A detailed approach of the role of the grain boundary plane in segregation was
performed in the case of the nickel–sulphur system [28, 29]. A preferential elec-
trochemical attack of grain boundary regions enriched in sulphur in a nickel poly-
crystal enables to semi-quantitatively estimate the sulphur content in different grain
boundaries and in different regions along a same grain boundary; the answer of a
boundary strongly varies with the plane that it locally adopts (Fig. 6.7). The results
lead to propose a geometrical criterion of segregation based on the average spacing
dm of the planes parallel to the boundary plane. In the case of an asymmetrical grain
boundary, (khl)I = (khl)II , dm is replaced by an effective spacing deff defined by the
relation (4.24) [30]. The reduced value dm /a (a being the unit cell parameter) must
exceed a critical value (dm /a)c ≈ 0.150 in order that the propensity to segregation
at the grain boundary is null (or very low) [29]. The value of the average spacing dm
(or deff ) gives an estimate of the planar atomic density [30].
(a) (b)
Fig. 6.7 Observation in transmission microscopy of the attack grooves of grain boundaries in
nickel: a variation from a grain boundary to the other around a triple junction; b variation along a
same = 3 boundary (the average values d/a are reported in italics) [29]
For the symmetrical

100,
110 and
111 tilt grain boundaries in face-centered
cubic (f.c.c.) materials, the density of coincident sites in the volume being equal to
the density of atomic sites in the boundary plane (σ = 1), the planes that meet the cri-
terion dm /a > 0.150 are: {111}, {100}, {110}, {311}, {210}, {211}, {511}, {531},
{221} et {310}. All belong to the three first levels of the grain boundary classification
proposed by Paidar (Fig. 4.25) [31]. We recall that the dm /a criterion has also been
used to distinguish grain boundaries on the bases of their free energy [30] and their
cleavage energy [32]. Like any geometrical criterion, it must be used with caution: if
it matches numerous segregation cases, there are many contradictory examples [33].
Besides, the critical value dm /ac ≈ 0.150 is too permissive; nevertheless, a high
interplanar spacing is often associated to a low segregation level for most tilt grain
boundaries. This is no longer true for twist grain boundaries. It is finally the same as
for energy, there is no universal criterion to predict segregation; no simple rule may
link a grain boundary geometrical parameter to the segregation level.
The studies of the nickel–sulphur system also allow us to reveal the reciprocal
aspect of the segregation–structure relationship, i.e. the effect of the segregation
on the geometrical boundary parameters. Thermal treatments of sulphurization and
desulphurization of a same thin foil containing a grain boundary deviated from the
= 3 coincidence lead to rotations of the boundary plane such that the average value
of dm /a varies proportionally to the inverse of the sulphur content in the boundary
(Table 6.1). On the contrary, the angular deviation θ to the exact coincidence mis-
orientation continuously decreases with the thermal treatments, independently of the
sulphur content [34].
Table 6.1 Evolutions of the misorientation and of the interplanar spacing of the boundary plane in
a same grain boundary of nickel in function of its sulphur content; originally, the sulphur content
in the matrix is 16 ppm [34]
Thermal treatment Sulphur segregation level in the grain boundary θ ◦ dm /a
Anneal at 625◦ + quenching Medium 9.5 0.190
+Desulphurization Low 4.5 0.335
+Sulphurization High 2.5 0.105
Similar results than those presented above for metals have been obtained in
ceramics. Nanoanalyses performed on several grain boundaries in alumina show
that, apart from the basal twin = 3 (0001), all the boundaries are more-or-less
sensitive to solute segregation. The segregation appears almost independent of the
misorientation, whereas not only the quantity but also the nature of the segregated
species differ from a grain boundary to the other according to its plane. In poly-
crystalline alumina doped with yttrium and containing silicon as impurity, the latter
element is preferentially located at grain boundaries whose one plane, at least, is a
basal (0001) plane, while yttrium is found in boundary with a rhombohedral (01-12)
plane in one grain, at least (Fig. 6.8) [35]. Same results have been found in com-
mercial alumina doped with titanium: for a given misorientation, there is a clear
predominance of titanium in the region of the boundary with a rhombohedral plane
and of silicon in the region with a basal plane (Fig. 6.9) [36].
A selective segregation of yttrium is also mentioned for twin boundaries of bicrys-
tals obtained by diffusion bonding [37], but, contrary to previous results, yttrium is
detected in the basal (0001) twin and not in the rhombohedral (10-12) twin. This
difference may be due to different impurity content from one alumina to the other,
synergetic or rejection phenomenon being not excluded. In particular, a site com-
petition between solutes can exist, one of them occupying the intergranular sites
Fig. 6.8 Concentration profiles of silicon and yttrium obtained by nanoanalysis across various grain
boundaries in alumina, one boundary plane (at least) being either a basal plane or a rhombohedral
plane [35]
Fig. 6.9 a Image in transmission electron microscopy of a curved grain boundary in alumina whose
the plane (in the dark crystal) progressively changes from the basal (0001) to the rhombohedral
(01-12) orientation; b variations of the titanium and silicon contents in the boundary in function of
the plane orientation [36]
occupied, in its absence, by the other; this phenomenon depends on the grain bound-
ary structure. It is therefore conceivable that a solute can be preferentially attracted
to (or repelled from) a type of grain boundary in a polycrystalline ensemble.
Besides, the grain boundaries that are observed in polycrystals are general, with
almost the same energy, and are only differentiated by their planes [35, 36]; while
the boundaries analysed in bicystals ( = 3, 7, 13) are symmetrical but display very
different energies. For a given boundary concentration in yttrium, the calculated
segregation energies follow the order E g GB ( = 3) < E g GB ( = 7) < E g GB
( = 13), in agreement with the differences of interfacial energies [37]. The values
of E YJ keep the same order for other solutes like lanthanum and scandium, but they
increase with the solute size. This result suggests that the decrease of the elastic
energy accompanying the transfer of an atom from a crystal site to a boundary site
constitutes the major contribution to the segregation energy, not only in metals but
also in ionic oxides.
An interpretation of the selective segregation of a particular solute to a given
boundary likely rests on the atom distribution in the planes that have been considered
in alumina and on the possible adequacy with a silicon, yttrium or titanium oxide
structure. Note that such a selectivity has never been mentioned in metals, although
not excluded. Indeed, a bi-dimensional compound that prefigures the 3D compound
may appear at a grain boundary; such a formation is strongly correlated with the
available sites in the structural units of the boundary. For a given solute, these sites
differ with the atomic boundary structure, controlled by the boundary plane.
Finally, at the mesoscopic and microscopic scales, regardless of the material, it
is the reciprocal relationship between the grain boundary plane and the segregation
that seems to predominate.
6.5.1.2 Segregation Anisotropy Quantified by Thermodynamic

Variables
A thermodynamic approach of the segregation anisotropy has been developed on the

basis of the Seah and Hondros model with two new considerations: the segregation
enthalpy depends on the grain boundary structure through a parameter Φ, and the
solid solution does not obey the Henry rule but an empirical law aS = (X S )ν (a is the
activity of the solute in the solution). Then, the segregation enthalpy of the solute S
in the matrix linearly depends on two independent terms, one reflects its dependence
from the grain boundary type, the other its variation with the solubility limit X S ∗ at
the temperature T [38]:
H ◦ (Φ, X S ∗ ) = H ∗ (Φ) + ν R[T ln(X S ∗ )] (6.28)
H ∗ (Φ) varies with the structure and corresponds to the extrapolation of the
segregation enthalpy for a unlimited solubility in the matrix (X S ∗ = 1). The product
[T ln(X S ∗ )] is almost constant with the temperature and thus can be deduced from
the data on the solubility maximum [T ln(X S ∗ )max ]. The parameter ν that relies the
activity to the mole fraction is considered as a matrix characteristic; it differs little
from one boundary to the other and for different solutes. So, the average value of ν
for the segregation of phosphorus and carbon in α-iron, obtained from experiments
on several bicrystals, is equal to 0.77 with a standard deviation of 0.06.
Fig. 6.10 Variations of the

segregation enthalpies of
silicon, phosphorus and car-
bon to symmetrical [100]
tilt grain boundaries in iron
in function of the product
[T ln(X S ∗ )max ] and of the
boundary misorientation θ
[38]
On the basis of Eq. (6.28), ternary diagrams of grain boundary segregation have
been built for α-iron. They represent the variation of the segregation enthalpy with
the product [T ln(X S ∗ )max ] and with one of the geometrical parameters of the
boundary that influences its structure. The curves on Fig. 6.10 have been estab-
lished for symmetrical [100] tilt grain boundaries, the variable is the misorientation
angle θ , the median plane being fixed. Figure 6.11 concerns boundaries with a fixed
misorientation (36.9◦ [100]) but the boundary planes can adopt different orientations,
symmetrical or not. In that case, the geometrical variable is the angle ϕ between the
symmetrical orientation (031) and any random orientation of the boundary plane. We
remark that the anisotropy of the segregation enthalpy is qualitatively similar for the
three solutes silicon, phosphorus and carbon. Furthermore, the enthalpy dependence
with [T ln(X S∗ )max ] is linear with a constant slope ν R that differs according to the
grain boundary, i.e. according to the angle θ or ϕ Fig. 6.11).
The extrapolation of the ternary diagrams for the value of [T ln(X S∗ )max ] = 0
enables to determine H ∗ (Φ); indeed in that case (X S ∗ )max = 1, the system dis-
play a complete mutual solubility. In the iron system and taking an average value of
ν̄ = 0.77, H ∗ (Φ) varies from −8 kJ·mol−1 for a general boundary to +8 kJ·mol−1
for certain boundaries so-called special. The latter terminology is ambiguous, it is
better to say that the grain boundaries for which H ∗ (Φ) > 0 are impervious (or
very few pervious) to the segregation of carbon, phosphorus and silicon. The curve on
Fig. 6.12 presents the evolution of H ∗ in function of θ that brings out the extreme
values (±8 kJ·mol−1 ) and reveals that H ∗ takes positive values for the symmetrical
[001] tilt grain boundaries = 13 {015}, = 5 {013} and {012}. The dashed curve
extrapolated for vicinal and general boundaries close the previous singular bound-
aries has no physical meaning because an interaction exists between these boundaries
Fig. 6.11 Variations of the

segregation enthalpies of
silicon, phosphorus and car-
bon to = 5 36◦ 9 [100] grain
boundaries in iron in function
of the solubility maximum
(X S∗ )max (in mol. %) and of
the inclination angle ϕ of the
boundary plane [38]
Fig. 6.12 Evolution of

H ∗ in function of θ for
symmetrical tilt grain bound-
aries in iron (o). The curve in
full line represents the depen-
dence of HS0 in function of
θ for a system characterised
by X S∗ = 1 [38]
and a solute. We then consider that in the regions close the = 13 and = 5 coinci-
dences such that H ∗ > 0, the segregation enthalpy H ◦ (θ, XS ∗ = 1) is null. The
existence of vicinal boundaries close the previous coincidence boundaries presenting
few or no segregation seems to be confirmed by experiments [39].
By fixing the value of H ∗ (Φ) = −8 kJ/mol., we can predict the values of the
segregation enthalpy HS0 in general grain boundaries in α-iron for different solutes
S, knowing the corresponding values of the term [T ln(X S ∗ )max ] (Fig. 6.13). The
experimental values reported on the figure are in good agreement with the predictions
with a precision of 5 kJ·mol−1 .
The segregation diagrams are established for a given solvent because the solubility
of a given solute in different matrices varies. The structure and the energy of a grain
Fig. 6.13 Predicted seg-

regation enthalpies for
different solutes in iron grain
boundaries (o) in function of
the product [T ln(X S ∗ )max ].
Some experimental values (•)
are reported for comparison
[38]
boundary with a fixed geometry may also vary according to the base element in the
alloy.
The anisotropy is often represented by the enthalpy dependence with one of the
grain boundary geometrical parameters, but equivalent evolutions of the entropy are
also observed. Indeed, a linear relationship (compensation rule) exists between HS◦
and SS◦ . It has been first determined by simulation of the segregation of nickel to
different grain boundaries and of palladium to different sites of a same boundary [40].
Then, it has been experimentally observed for the segregation of silicon, phosphorus
and carbon to several grain boundaries in α-iron [41]. The results of experiments
performed on grain boundaries of bicrystals and polycrystals in iron alloys and for
different solutes are grouped in two sets giving rise to two lines that reflect this linear
relationship (Fig. 6.14) [42]:
HS◦
SS◦ = +σ (6.29)
τ
The term τ is equivalent to a compensation temperature and the term (− τ σ) has the
dimension of an energy.
The branch noted D corresponds to positive segregation entropy, the branch O
to negative entropy. By considering that entropy is a measurement of the system
disorder, positive entropy suggests that the solute segregation yields a disorder within
the boundary. This is actually the case for phosphorus, antimony, tin and sulphur
giving rise to the line D (for Disorder); these solutes accumulate at grain boundaries
on the form of disordered solid solutions. On the contrary, silicon and aluminium
form ordered solid solution in iron and their segregation leads to an ordered (O)
boundary; indeed, an ordered DO3 compound has been observed in an Fe–Si alloy
concentrated in silicon [15]. This interpretation of the existence of two branches for
the linear relationship between HS◦ and SS◦ must be cautiously considered because
the values of enthalpy and entropy are strictly valid only for very dilute solutions and
may traduce only a tendency to grain boundary order. The segregation behaviour of a

segregation entropy SS◦ in
function of the enthalpy HS◦
measured for different grain
boundaries of bicrystals (full
symbols) and of polycrystals
(empty symbols) in various
b.c.c. iron alloys [42]
grain boundary is thus characterized by two coupled thermodynamic quantities, i.e.

we may plot similar curves to those of Figs. 6.10 and 6.11, with the entropy in ordinate.
The values of HS◦ and SS◦ , calculated for different [100] tilt grain boundaries in
iron, the solute being silicon, phosphorus or carbon [42], always vary in the same
direction either with the misorientation, or with the boundary plane orientation. The
authors distinguish two groups of grain boundary segregation behaviours: elevated
absolute enthalpy and entropy values, indicating a strong segregation, characterise
general boundaries, and inversely for the so-called special boundaries. We have
not previously retained this general/special distinction when it rests on geometrical
criteria. But, in so far as it traduces here two different answers to a stimulus, we may
admit its relevance; however, we must keep in mind that really special boundaries, i.e.
from the point of view of their properties, only constitute a very small group among
all the boundaries in polycrystals, essentially twins and few singular boundaries.
Investigations of transitions between singular and vicinal grain boundary behaviours
constitute one of the main challenges to go towards grain boundary engineering.
From the coupling between HS◦ and SS◦ , it results that the dependence with
the temperature of the segregation free enthalpy G ◦S = HS◦ − T SS◦ differs from
one boundary to another. There is thus an intersection between the lines giving the
evolution of G ◦S with the temperature for different solutes S (S = Si, P and C) in
iron; it is remarkable that the intersection points occur at very similar temperatures,
close 930 K (Fig. 6.15) [42]. The parameter τ represents the average value of all the
temperatures for which the free energies of segregation are equal by boundary pair.
There are several practical consequences to the existence of τ. At low temper-
ature, the general grain boundaries have a free enthalpy inferior to those of the
so-called special boundaries, but this anisotropy disappears with increasing temper-
ature. For temperatures close τ, the solute concentration is the same in different grain
boundaries. Beyond τ, the relation segregation/structure appears inversed. The cal-
culations using Eq. (6.8) of the phosphorus concentration in different symmetrical
[100] tilt grain boundaries in iron show that the differences decrease with the tem-
perature and vanish at about 900 K; a slight tendency to inversion is also detected at
Fig. 6.15 Variations of the calculated standard free enthalpy of segregation (G ◦S = HS◦ −T SS◦ )
with the temperature in α-iron and for the solutes: a silicon; b phosphorus; c carbon [42]
1000 K for = 5, θ = 36.9◦ (Fig. 6.16) [42]. An inversion phenomenon permits to

understand the silicon concentration maxima found at about 923 K in symmetrical
{012}, {013} and {023} boundaries in a stainless steel [43].
Fig. 6.16 Dependence of

the phosphorus concentration
with the misorientation around
[100] for symmetrical tilt
grain boundaries in α-iron at
four temperatures and for a
matrix concentration equal to
0.01 at % P [42]
At the temperature τ, not only the concentration of an element is independent

of the grain boundary geometrical parameters but, for a series of elements giving
rise to a relation SS◦ = fn(HS◦ ) (line D or O), the free enthalpy values is also
identical whatever the boundary. For example, G ◦S = −σ τ = −53 kJ · mol−1 for
phosphorus and carbon in iron (Fig. 6.15b, c).
6.5.1.3 Changes of the Geometrical Parameters Under the Effect

of Segregation
The faceting phenomenon, which belongs to the non-congruent phase transformation

group [24], may result from a change of the grain boundary composition. One of the
most striking results is given by the transition of a planar boundary in pure copper
into a boundary with facets in presence of bismuth, the phenomenon being reversible
(Fig. 6.17) [44]. The boundary in pure copper displays random misorientation and
plane, the facets of the impure boundary are alternately close to a {111} plane in
one crystal or the other. This effect is associated to the strong size misfit between the
copper and the bismuth atoms, the latter being much larger. On the effect of tempera-
ture, the facets gradually fade and disappear at about 225 ◦ C (Fig. 6.17b), suggesting
that bismuth is prone to diffuse out of the boundary at relatively low temperature.
Fig. 6.17 a Presence of facets in a grain boundary of copper containing bismuth; b the same
boundary without facets after removal of bismuth under the effect of temperature; c the facets
appear again after re-introduction of bismuth [44]
A so-called rough transition is also observed in grain boundaries of alumina

under the effect of various doping elements (Fig. 6.18) [45]. The boundaries formed
between a (0001) surface of a sapphire single crystal and small grains of alumina
Fig. 6.18 Observations in scanning electron microscopy of grain boundaries between a (0001)
sapphire single crystal and small alumina grains: a rough boundaries appear in the case of pure
polycrystalline alumina; b the boundaries become planar when alumina contains 100 ppm SiO2 +
50 ppm CaO; c they take again a rough profile when 600 ppm MgO are added to the previous
impure alumina [45]
are rough. In presence of additions of SiO2 (100 ppm) and CaO (50 ppm), these
boundaries become planar over a pretty large distance, even across the triple junc-
tions. Addition of 600 ppm of MgO to alumina, already containing SiO2 and CaO,
transforms the planar boundaries into curved boundaries as in the pure state. The
parallelism of the grain boundary plane with the basal plane in one case, and its
curvature in the two other cases are verified at the nanoscopic scale (Fig. 6.19). Same
remarks are made for boundaries between a crystal that underwent an abnormal grain
growth along the basal direction and the other polycrystal boundaries. According to
the authors, this transition is likely to concern other singular grain boundaries than
the (0001) ones.
To conclude, the observed or calculated segregation in function of the geometrical
grain boundary parameters shows a strong dependence of the boundary plane and,
reciprocally, the boundary plane changes under the effect of segregation. This indi-
cates that, in fine, the grain boundary structure determines the possibilities for a
grain boundary to more or less accept foreign elements. Therefore, the next steps to
understand the grain boundary segregation anisotropy go through the analysis of the
interaction between a solute and the boundary structure, described either in terms of
intrinsic dislocations or in terms on structural units.
6.5.2 Grain Boundary Segregation and Intrinsic

Dislocations
Soon after the description of low-angle grain boundaries in terms of dislocations,

Friedel evokes the idea that segregation results from the elastic interaction between
solute atoms and the strain fields of intrinsic dislocations [46]. These effects can be
understood as long as the misorientation angle θ remains small; indeed, the distance
between the primary dislocations and thus the extension of their elastic fields (equal
Fig. 6.19 a Aspects of the grain boundary planes between a (0001) sapphire single crystal and
small alumina grains observed by high-resolution transmission electron microscopy: b and c the
crystal planes are parallel to the basal sapphire plane in a polycrystal containing 100 ppm SiO2 +
50 ppm CaO; The micrographs above b and c show with a lower magnification the regions A and
B on which the observations have been made, respectively; d adding 600 ppm MgO yields a curved
boundary (region C) even at the nanoscopic scale (from Park et al. [45])
to the boundary period) rapidly decreases with θ. Then, the driving force for segrega-
tion can no longer result from the elastic attraction of the solutes by the dislocations.
This is also true for high-angle grain boundaries that do not possess long-range
strain fields, the distance between the secondary dislocations and the Burgers vec-
tors of these dislocations (bDSC ) being generally small. There are few experimental
studies of the attraction exerted by intrinsic dislocations on solute atoms and of the
influence of the latter on the boundary structure; only simulations enable to specify
this reciprocal relationship.
A change in the dislocation structure of a low-angle grain boundary under the
effect of segregation has been observed by transmission electron microscopy in the
iron–gold system [47]. This transition is of the congruent type. In a cubic material,
a twist low-angle grain boundary around [001] displays a grid of screw dislocations
parallel to the
110 directions. In the case of pure body-centered cubic (b.c.c.) iron,
the screw component of the Burgers vector b = a/2
111 account for the dislocation
grid. Adding 0.18 at.% of gold to an iron bicrystal of the same geometry leads to the
formation in some boundary regions of an arrangement of screw dislocations with
Burgers vector b = a
100 (Fig. 6.20). The gold content in these regions is double
than that where the normal structure is preserved. The gold atoms are supposed to
decorate the dislocation cores.
Fig. 6.20 Electronic micrographs showing grids of screw dislocations in a low-angle

(θ=1.5◦ [001]) twist grain boundary in iron, seen at normal incidence: a pure iron b Fe-0.18 % at.
Au: the structure differs from that of the same boundary in pure iron [47]
A change of structure of a low-angle twist grain boundary also occurs in magnesia

under the effect of iron, but in that case the transition is non-congruent. Indeed, the
boundary plane initially planar in pure magnesia locally dissociates in presence of
iron into sinuous parts surrounded by sub-boundaries indicating local changes not
only of the boundary plane but also of the misorientation [48]. The planar boundary
regions are described by a grid of screw dislocations parallel to the
110 directions
of the f.c.c. magnesia, while the regions that have rotated by 0.5◦ at most display a
distorted hexagonal network of dislocations. This network is composed of a family
of screw dislocations parallel to

110 and two families of mixed dislocations parallel
to
120. The oxidation degree of iron in magnesia seems to play a non-elucidated
role in the transition that is favoured in the presence of Fe+++ .
The elastic interaction between a low-angle twist grain boundary and a solute
atom has been simulated using the Monte Carlo method with an EAM (Embedded
Atom Method) potential and on the basis of the isotropic elasticity theory [49]. This
interaction comes from two effects: the difference between the sizes of the solvent
and the solute atoms and the difference between their elastic moduli, giving rise to
two energetic terms, E mis (mis for misfit) and E mod . (mod for modulus). A positive
value of each of these terms indicates a repulsive interaction. More precisely, a
solute atom larger than the solvent atom is repelled by a hydrostatic compression
stress (positive) and vice-versa. Similarly, a hard elastic inhomogeneity gives rise to
a repulsive interaction and vice-versa. The misorientation angle is limited such that
elastically deformed regions exist between the intrinsic dislocations forming a grid.
We do not recall here the linear elasticity formulae; we just give simplified forms
that enable to distinguish two important parameters:
E mis = K mis εα Dmis (6.30a)

E mod = K mod εμ Dmod (6.30b)
K mis and K mod are constants that depend on the solvent parameters (μ, ν, r : the
elastic moduli and the atom radius, respectively). εα is equal to a −1 (da/dc) with a
the solvent lattice parameter and (da/dc) its change with the solute concentration,
εμ = μ−1 (dμ/dc). Finally, Dmis and Dmod are spatial quantities function of the
distance d between dislocations; d depends on the misorientation angle according to
(2.7) and on the position (x, y, z) of the solute atom in a square of the grid. The value
of the interaction energy E int = E mis + E mod gives the local estimates of the solute
concentration at the point (x, y, z) and for a given distance d.
The spatial functions Dmis and Dmod have high values along the dislocation cores
and rapidly decrease in the interior of the grid meshes (Fig. 6.21). The factor Dmis
takes its maximum value at the intersections between the screw dislocations while
Dmod = 0; the latter reaches its maximum value midway between the dislocations.
At the centres of the grid squares, the modulus effect is strictly null, while the size
effect is minima but preserves a finite value. The two energetic terms E mis and E mod
rapidly increase when the misorientation angle θ decreases and reach characteristic
values of individual screw dislocations for a value of d equal to about 10b.
Note that these calculations are strictly valid only for a very dilute solution and for
areas beyond the dislocation cores. However, the interaction between solute atoms
and dislocation cores may be preponderant; it is constant along a dislocation line
and independent of the atom position in the grid [50, 51]. It is only after dislocation
core saturation than the segregation in the neighbouring elastic regions may start.
There, a question must be raised: is it possible that a repulsive elastic interaction
between a given grain boundary and a given solute exists while the core interaction
is attractive?
Fig. 6.21 Evolutions of the spatial functions Dmis (d, x, y, z) and Dmod (d, x, y, z) in function of
the solute position in the (yOz) plane parallel to the grain boundary plane and for two values of x:
a Dmis for x = 0.5b and b Dmis for x = 1.5b; c Dmod for x = 0.5b; d Dmis for x = 1.5b. The
distance between the screw dislocations is 10b; these dislocations intersect at the centre of each
square, this point is chosen as mathematical origin of the two functions [49]
Segregation has a very weak effect on the level of the shear stresses associated to
screw dislocations in high-angle twist grain boundaries. We recall that these disloca-
tions are centred on the filler units and they intersect on the minority units where the
stress maxima are localised (Figs. 3.23, 3.35) [52]. The study by Monte Carlo simu-
lation of ten [001] twist grain boundaries in a dilute Ni–Cu alloy shows that copper
segregation is well maximum on the minority units. Hydrostatic tensile stresses are
reduced, the copper atom being larger than the nickel one; but the shear stresses,
linked to the symmetry of the atomic sites in the boundary, are quasi unchanged by
copper segregation [53].
More generally, segregation appears more linked to the core structure of the
intrinsic dislocations, i.e. the grain boundary core structure. This relationship is
discussed in the next section.
6.5.3 Grain Boundary Segregation and Grain Boundary

Atomic Structure
First simulations of the grain boundary segregation in relation with the atomic
boundary structure have been made at 0 K. They enable to determine the segre-
gation energy associated to each boundary site, to highlight the large anisotropy of
segregation from one site to the other, and to predict the formation of an intergranular
ordered compound and the occurrence of structural transformations induced by the
segregation. All these phenomena have been predicted by the classical and statistical
thermodynamic approaches previously described.
6.5.3.1 Segregation Anisotropy from a Grain Boundary

Site to Another
The first known examples of the site-to-site anisotropy concern symmetrical [001]
tilt grain boundaries in the substitutional alloys Cu–Bi, Cu–Ag and Au–Ag [54, 55],
with a particular attention to the = 5, 36.87◦ that has the advantage to have been
widely experimentally studied. The = 5 (210) boundary possesses two metastable
structures described by the units B and B , the ratio of their energies being 1.04 in
favour of the B unit [56]. The segregation energy calculations, site-by-site, are made
on a first solute atom by exploring all the possible substitutional sites. Figure 6.22
shows the different sites in the B structure of the = 5 (210) boundary with the
corresponding segregation energies E So [55]. The sites Z, J, K, although no strictly
symmetrical to the C, E, G sites with respect to the boundary plane, display the same
energy value. The energy is the most negative (the most favourable to segregation) for
the sites associated to a high hydrostatic tension; this suggests that the main driving
force for segregation comes from the size effects, the bismuth atom being larger
than the copper atom. The first bismuth atom being placed in a site with negative
segregation energy, we introduce the second atom on one of the possible segregation
sites until to find the new most favourable site. The segregation energies of the second
atom are reported in Table 6.2.
Table 6.2 Segregation energies for different sites of the = 5 (210) boundary in copper in the
case where two bismuth atoms enter the boundary elemental unit
Site b1 bz c1 e e1 ez1
E S1 (eV/at) −3.00 −1.15 0.39 0.11 −0.83 0.42
The first atom is placed on a site B. The sites bz and ez1 are displaced along z compared to b and
e1 [55]
This procedure is repeated for several bismuth atoms, so increasing the solute
concentration of the grain boundary. The segregation energy remains negative until
the bismuth addition does not exceed 13 atoms per unit cell; this corresponds to a
concentration of 1.25 monolayer in agreement with the bismuth level experimentally
Fig. 6.22 Structure B of a symmetrical = 5 (210) tilt grain boundary with the different segre-
gation sites of a first solute atom indicated by letters. To differentiate the structure B of the site B, a
small capital letter indicates the latter. On the right of the scheme, a list of the segregation energies
associated to the different sites is given [55]
detected on a fractured boundary [57]. The resulting structure is ordered with each
bismuth atom surrounded by four copper atoms and vice-versa. The formation of such
a structure, stable although different from the B and B predicted for the = 5 (210)
boundary in pure copper, is closely linked to the site segregation anisotropy. The
segregation occurring in the second possible boundary structure B starts on a site
with a segregation energy E So = −1.6 eV/at; but, as soon as a bismuth atom is
located on that site, the B structure begins to change in the B structure. The structural
transition is complete when two types of sites are filled; this corresponds to a bismuth
concentration of 0.19 monolayer. Thus, segregation may reduce the multiplicity of
the possible grain boundary structures. The predominant influence of the steric factor
on the grain boundary segregation in the Cu–Bi system may be generalized to any
metallic system, but the variations from one site to another may be weaker as it is
the case for the Au-Ag alloys [54].
In principle, the segregation anisotropy existing in the favoured grain boundaries
can affect the intermediary boundaries [55]. In particular, the segregation may con-
tinuously vary with the misorientation angle for symmetrical
110 and
100 grain
boundaries with the same median plane that obey the structural unit model. However,
this argument ignores three facts, at least, that make it obsolete in many cases:
• The existence of boundaries with short periods that break the continuity of the
descriptions from one reference structure to the other; these boundaries are often
chosen as delimiting boundaries.
• The deformation of the structural units in intermediary grain boundaries: the stress
state differs for a given site in a given unit according to the location of this unit,
in the favoured boundary or not.
• The formation, in low stacking fault energy materials, of 3D grain boundaries (See,
Sect. 3.3.1).
The following examples show how the segregation is a complex phenomenon,
unpredictable on the basis of any geometrical criterion; each case requires simulation
studies, as much as possible linked to experimental investigations.
The segregation width is limited to about one interatomic distance on each side
of the boundary plane in singular grain boundaries, but may be slightly extended
in general boundaries and more clearly in dissociated boundaries in low stacking
fault energy materials. The Monte Carlo method had been used to simulate the
segregation of palladium in three grain boundaries in nickel in which the solute atoms
are concentrated with very different extensions. Nickel is considered as a low stacking
fault energy metal [40] despite the relatively high SFE value experimentally found,
assumption already used in the calculations of several
110 tilt boundaries in nickel
(see Sect. 3.3.1). The palladium atom is much larger than the nickel atom (10 %),
but has the same number of valence electrons; the elastic effect is thus dominant.
Figures 6.23, 6.24 and 6.25 show the palladium distribution in the nickel boundaries
site-by-site (a) and in function of the distance from the boundary plane (b) for =
11 {113}, = 33 {441} and = 33 {225}, respectively. Two types of sites may be
distinguished:
• The sites in the grain boundary core that generally contain the larger palladium
concentration.
• The sites in the elastically deformed region near the boundary; these sites signifi-
cantly contribute to the total level of segregation in the = 33 {225} boundary.
The solute distribution, homogeneous and limited to 0.3 nm on each side of the
singular = 11 grain boundary, becomes heterogeneous and extended around the
two general boundaries. It is particularly true for the {225} boundary that undergoes
dissociation with emission of a Shockley dislocation on one side of the boundary. The
resulting stress field implies that the sites above the dislocation are in compression
and thus depleted in palladium (white sites) while the sites below the dislocation are
in tension and thus enriched in solute (black sites); the two regions are separated by
the (111) glide plane of the dislocation. A stacking fault is clearly visible between
the boundary and the partial dislocation; the segregation in this region is due to an
elastic attraction and not to a Suzuki effect.
In the previous example, several degrees of freedom simultaneously change. If the
misorientation is now fixed and only the plane orientation varies, a strong anisotropy
from one plane to another appears as it was observed, at the microscopic level, for a
= 3 boundary in nickel [28] and a general boundary in alumina [36]. The grain
boundary plane plays a major role in the segregation–structure relationship, as it
imposes the boundary atomic structure. A study using the technique of quenched
molecular dynamics simulations enables to detail the differences in the behaviours
to segregation of two = 11 grain boundaries (same misorientation) with either a
{113} plane or a {332} plane for two binary systems Ni (Ag) and Ag (Ni) [58]. The
retained parameters to characterize a grain boundary site i are:
Fig. 6.23 Palladium distribution in a = 11 {113} grain boundary in nickel: a site-by-site

(a darker site corresponds to a higher palladium concentration level); b in function of the distance
from the boundary plane [40]


• The Voronoï volume νi that mathematically defines the local atomic volume; a
positive value of the difference νi − ν0 (ν0 being the calculated Voronoï volume
in the perfect crystal) indicates a dilatation at the site i.
• The local hydrostatic pressure pi equal to −1/3 of the trace of the stress tensor
associated to the atom i; a positive pressure indicates a site in compression.
The different terms of the segregation energy (elastic and cohesion) are calculated
site-by-site in order to specify the driving force that locally controls segregation.
The elastic contribution was calculated in two ways depending on whether the site is
locally in hydrostatic stress condition or not, the latter case being attached to a highly
anisotropic extended defect. Such an analysis, very rare, provides an interesting light
on the segregation phenomenon. We do not reproduce here the tables giving the
values, site-by-site, of the geometrical and physical characteristics of two boundaries
in the systems so-called direct for Ni (Ag) and inverse for Ag (Ni) (see [58]). But, as
it is the detailed consideration of the signs and the values of the elastic and cohesion
terms that enables to discuss the origin of the driving forces, we extract the following
few examples of this analysis.
Figures 6.26 and 6.27 show the atomic structures of the = 11 {113} and =
11 {332} with various possible sites of segregation; in parallel, the variations of
the segregation energies are given in function of the distance of the site from the
boundary plane, for each of the binary systems Ni (Ag) and Ag (Ni). As previously
for the system Ni (Pd) (Fig. 6.23), the variation of the segregation energy (or of
the solute concentration) is limited to about 0.5 nm from the boundary plane; it is
regular over this distance for the singular = 11 {113} (Fig 6.26). On the contrary,
the segregation energy value fluctuates on each side of the boundary plane for the
= 11 {332}; the fluctuations are especially important for the Ni (Ag) system
(Fig. 6.27). In the = 11 {113}, the site a is under high tensile stress, the sites b1
and b2 under low tensile stress. There are no sites under high compression in this
structure. The segregation of the silver atom, larger than the nickel atom, is thus
favoured by a steric effect. Conversely, nickel does not segregate to this boundary in
silver.
To explain more in details the segregation behaviour of each site, we must refer
to the values of its different characteristics in order to identify and explain a possible
role of the cohesion factor in the segregation. We give an example of this analysis
for the site b1 of the = 11 {113} grain boundary that weakly attracts nickel atoms
in the Ag (Ni) system. This effect is due to its cohesion energy, of same value but
opposite signs to the energy associated to the size effect. In the Ni (Ag) system, the
cohesion factor also contributes somewhat to the total driving force, but it has the
same sign than that of the size factor. The noticeable value of the cohesion energy
at the b1 site comes from its particular environment, very different from that of other
sites (Fig. 6.26a) that leads to a stress state far from the hydrostatic character. Similar
reasoning has been made to explain the singular behaviour of the sites b1 , b4 and d
in the = 11 {332} grain boundary. In particular, the high compression at site D
has no equivalence in the {113} boundary. The values of the cohesion energies to
this site in the two systems are opposite to the values of the elastic energies and no
negligible. For the other sites b1 and b4 , the cohesion factor is involved in the total
segregation energy, although likely of the same sign that the size factor; it is the case
for site b1 in Ni (Ag). More generally, the sites in the general {332} boundary are
Fig. 6.26 a Calculated atomic structure of the = 11 {113} grain boundary. The sitea belongs to
the boundary plane, the sites b1 , b2 and c are in the first, the second, then the third plane parallel to
the boundary plane, respectively; b variation of the segregation energy with the distance from the
boundary plane (plane 0). The positions of the different sites are reported in the ±1, ±2 and ±3
planes [58]
Fig. 6.27 a Calculated atomic structure of the = 11 {332} grain boundary. All the sites are at the
exterior of the boundary plane represented by a vertical dashed line; b variation of the segregation
energy with the distance from the boundary plane (plane 0) given in angstroms. Only the variation
on one side of the boundary plane is reported for a given system: on the right for the Ni (Ag) system,
on the left for the Ag (Ni) system [58]
more distorted than in the singular {113} boundary and the hydrostatic condition is
rarely satisfied.
The main conclusions, resulting in particular from the previous study, are the
followings:
• The distinction between the size effect and the cohesion effect in the total segre-
gation energy calculation is valid for all the grain boundary sites.
• The consideration of the steric factor alone is most often sufficient to account for
the segregation level per site; the cohesion effects really occur when it is relatively
large and displays an opposite sign to the size effect.
• The cohesion effect is significant for sites where the linear relationship between the
local volume and the local pressure is not respected. This effect is explained by
the character locally non-hydrostatic of the stress or by a high local compression.
However, these conclusions must be relativized; they are only valid for two metals
of same structure modelled using the same type of potential, but cannot be generalized
to more complex systems. It must be noted that numerous simulations are performed
on model systems, not necessarily the most interesting from a practical point-of-
view. In particular, the segregation of elements like sulphur, phosphorus, antimony
and boron at grain boundaries of several metals (iron, nickel copper …) in which they
have a noticeable influence on the properties, remains few studied at the atomic scale
even if these systems have been well experimentally investigated at higher scales.
6.5.3.2 Change of the Grain Boundary Structure Induced

by Segregation
Congruent phase transformations induced by segregation have been previously

evoked, either experimentally observed as in the Fe–Au system [47], or obtained by
simulation as in the case of bismuth in the structure B of a = 5 boundary in cop-
per [55]. They depend on the site-by-site segregation anisotropy. This phenomenon
deserves to be detailed at the atomic scale, what we propose to do through some
examples.
A phase transformation of the congruent type is obtained by simulation of the
palladium segregation to a = 9 {221} grain boundary in nickel [40]. The structural
unit E of this boundary is stable from 0 to 800 K in pure metal. Another structure,
with a higher energy, composed of units E in unstable at high temperature. The
situation is inversed in presence of palladium; the unit E transforms, under the √
effect
of segregation, in the unit E , deformed and displaced by a vector τ = a0 2/4
along the tilt axis (Fig. 6.28).
Fig. 6.28 a Stable structure E of a = 9 {221} grain boundary in nickel; b the structure E is
transformed in the structure E under the effect of a palladium segregation. The white and black
symbols indicate atoms in the alternated {220} planes along the
110 tilt axis [40]
The following example confirms that no geometrical parameter may predict the
propensity to segregation of a grain boundary; moreover, for a given boundary, the
segregation depends on the thermal treatment underwent by the material. Bismuth

does not segregate to coherent twins when they are present in copper before the segre-
gation treatment; but, if these twins are created by transformation under annealing of
other boundaries containing bismuth, then a monoatomic layer of bismuth is present
in the = 3 {111} facets. Each bismuth atom occupies an area equal to three times
that of the copper atom in the {111} plane. This configuration corresponds to the
densest packing of the bismuth atoms, avoiding their superposition. The spacing
of the bismuth atoms in the 2D arrangement is similar to that it adopts in the bis-
muth bulk of hexagonal structure [59]. Even a coherent twin is not immune to the
introduction of a solute in its structure, the presence of impurities depending on the
thermo-mechanical history of the grain boundary.
A congruent phase transition also explains the appearance of an intergranular
film observed when phosphorus segregates to the symmetrical = 5 (310) and
= 9 (114) tilt grain boundaries in b.c.c. α-iron [60]. The phosphorus atoms,
segregated on the form of a monolayer, provoke a complete restructuration of the
boundary. These atoms, initially placed in substitution in the structural units, then
occupy the centres of the trigonal prisms of iron atoms newly formed in the two
types of boundary; this leads to a decrease of the interfacial energy. The obtained
configuration is very close that of phosphorus in the 3D compound Fe3 P (Fig. 6.29),
as predicted by the thermodynamic models [4].
Fig. 6.29 Polyhedral units of iron atoms in [1-10] tilt grain boundaries: a = 5 (310) and b
= 9 (114); the different symbols indicate the different positions along the [1-10] axis. When
the phosphorus atoms enter these structural units, on the basis of one atom to one unit, they induce
a structural change (deformation and displacement of the structural units); in the two cases c
= 5 (310) and d = 9 (114), each atom occupies an interstitial site (*) surrounded by three
iron atoms to form Fe3 P [60]
The previous intergranular order appears in contradiction with the assumption

of Lejcek and Hofmann about a disordered structure of a grain boundary contain-
ing phosphorus [42]. This dichotomy is probably due to a difference between the
boundaries involved in the two studies and supports the necessity to compare more
closely the studies performed at different scales.
When the segregation is similar to the occurrence of a 2D phase in the grain
boundary, may we still speak of segregation? Or is it a beginning of precipitation?
This question remains totally open.
6.5.3.3 Experimental Evidence of the Grain Boundary Segregation

at the Atomic Scale
One of the best recent technique used to reveal segregation at the atomic level is
the high-angle annular dark field (HAADF) transmission electron microscopy [61]
which supplies a map sensitive to the atomic number of the atoms in a thin foil
(Z-contrast map); it enables to detect the atom distribution on and in the vicinity of
the boundary defects with a spatial resolution of the order of the interplanar distance.
HAADF images of a grain boundary near the rhombohedral twin in alumina
clearly reveal the presence of yttrium on the disconnections and its absence between
these defects in the boundary region displaying a twin relationship (Fig. 6.30a, b).
A detailed image of each disconnection shows that the distribution of yttrium (white
columns on the image) somewhat differs with the type of defect (Fig. 6.30c, d)
[62]. But in any case, the segregation extends from 3 to 5 nm along the boundary;
it remains localized onto two rhombohedral planes in the direction perpendicular
to the boundary plane and follows the boundary steps (see horizontal brackets on
Fig. 6.30d). Moreover, the yttrium distribution adopts a two-dimensional structure
as proved by the alternation of very bright spots with less bright spots, the brightest
spots being translated from one crystal to the other. This Y-rich structure corresponds
to the YAlO3 compound in the [10-1] projection [62].
Despite its power, the HAADF technique gives only a 2D image of the solute
distribution; in particular differences of content along an atomic column may not be
revealed.
The unique technique that enables to map, in 3D and at the atomic scale, the distri-
bution of the chemical species in the vicinity of a grain boundary and to quantitatively
measure the interfacial Gibbs energy excess is the atom probe tomography [63]. The
analysis does not require any calibration; the concentrations are directly proportional
to the number of collected ions. Figure 6.31 shows a 3D image of a general grain
boundary between two γ crystals in an N18 superalloy; it clearly reveals the segre-
gation of chromium, molybdenum and boron and the depletion in aluminium in the
boundary core. The segregated atoms form a continuous film of about 1nm thickness
along the boundary [64]. The large size γ particles, near the boundary, are enriched
in aluminium. The probe resolution allows to highlight the {001} plane sequence in
the grain at the left of the ordered γ phase.
Fig. 6.30 HAADF images of a rhombohedral twin displaying three disconnections in alumina: a in
a non-doped alumina; b yttrium-doped alumina: the presence of yttrium only on the disconnections
(white atomic columns) is clearly revealed; images c and d illustrate the yttrium distribution at the
disconnection cores (see text) [62]
Fig. 6.31 3D images obtained by atom probe tomography of the segregation of chromium, molyb-
denum and boron to a grain boundary in a super alloy N18 (each point corresponds to one atom).
The regions of the γ phase close the grain boundary are enriched in aluminium. The volume is
orientated such that the grain boundary plane and {011} planes in the left crystal are perpendicular
to the plane of the figure [64]
To conclude, the relationship between segregation and atomic structure of grain

boundaries has still been few studied from an experimental point-of-view. Apart from
the results obtained by HAADF or by atom probe tomography, the required tools
are in development. Until now, the link between segregation and structure has been
mainly established by the atomistic simulation methods: energy minimization at 0 K
and, for finite temperature, molecular dynamic and Monte Carlo simulations. Only
the latter enables to establish a heterogeneous spatial distribution of the equilibrium
species, as it is the case in a bicrystal. How realistic are the calculations? This
depends on the use of a potential that well describes the interactions between the
atomic species. For more information on the different simulation approaches of the
intergranular segregation, the reader may refer to the review article by Pontikis [65].
We must note that, whatever the method and the potential refinement, all the results
converge on the predominant role of the size effect in the grain boundary segregation.
Furthermore, segregation on the atomic sites in a structural unit is sensitive to the
site distortion, when the unit is included in an intermediary grain boundary; as a
result, concentration variation appears in a same structural unit in function of the
misorientation θ . However, the structural unit model is still relevant to describe
impure boundaries.
Generally, the segregation phenomenon is better approached for metallic sys-
tems of neighbouring species Cu–Ni, Ag–Au …than for alloys where a non-metallic
element segregates, such as sulphur, carbon, phosphorus in metals. But, this is actu-
ally those systems implying high enrichment factors that have the most important
consequences on the grain boundary behaviours (embrittlement, hardening…) and
thus on the material properties. The cohesive effect that occurs in some grain bound-
ary sites in metals, even secondary with respect the size effect, may also have con-
sequences on the boundary behaviours and take a growing importance in ionic and
covalent materials. The segregation approach on the basis of the grain boundary
electronic structure is thus necessary.
6.5.4 Grain Boundary Segregation and Grain Boundary

Electronic Structure
The segregation effect on the grain boundary electronic structure covers two phenom-
ena: a change of the electronic bonds that may occur in any material and a change of
the band structure in semiconductors. In several metallic alloys or intermetallic com-
pounds, segregation induces a weakening or a strengthening of the grain boundary
cohesion that has important practical consequences. To explain these effects the
knowledge of the atomic arrangement after segregation is not sufficient, we must
consider the change of the electronic bonds resulting from the presence of solute
atoms in the boundary. In semiconductors, the modifications of the electronic band
structures by foreign atoms may change the grain boundary electronic properties;
this is a crucial problem for the use of polycrystalline silicon in the photovoltaic cells.
6.5.4.1 Change of Electronic Bonds on the Effect of Segregation

in Metals
From the earliest simulations of segregation in metals, the electronic structure cal-
culations have been coupled to the atomic structure calculations. Despite a fairly
large number of works, the theoretical explanation of the effects of a non-metallic

solute (sulphur, phosphorus, boron…) on the atomic bonds in the grain boundaries
of metals and of intermetallic compounds remains controversial. Three models have
been proposed [66]. The first one directly implies the force of the metal/solute chem-
ical bond; the second suggests that the nature of the chemical bond, rather than its
force, is modified: from metallic the bond becomes covalent; the third one takes into
account the change of the metallic bonds in the close vicinity of the solute atoms
[60, 67]. The latter seems to be preferred to explain the cohesion changes in grain
boundaries of metals.
The effects of boron and phosphorus on the = 5 and = 9 in iron totally
differ, strengthening for the first one, embrittling for the second. Whereas phosphorus
induces a structural change (Fig. 6.29), boron only lightly modifies the structural
unit of iron; the maps of the corresponding hydrostatic stresses confirm these effects
[60]. Calculations of the relaxed atomic structures are followed by calculations of the
density of states that clearly indicate the effects of each solute. In pure iron, the atoms
at grain boundaries as well as in matrix are weakly linked with their neighbours; thus,
there is no influence on the grain boundary cohesion. In presence of phosphorus, the
Fe–Fe bonds surrounding the Fe3 P clusters are weakened while boron forms strong
bonding with iron along the boundary without weakening the nearest Fe–Fe bonds
(Fig. 6.32). This difference may explain the embrittling role of phosphorus and the
strengthening role of boron in the grain boundaries of pure iron and iron alloys.
Similarly in the Cu–Bi system, the differences in charge density and in the density
of states between the boundary doped with bismuth and the pure boundary show that
the presence of bismuth in a twin leads to an electronic redistribution resulting in a
weakening of the Cu–Cu bonds [68].
The following example, that we especially detailed, simultaneously involves the
metal–metal and the metal–solute bonds in order to explain the influence of a solute
on the cohesion of a grain boundary in nickel. The density functional theory (DFT)
in the local density approximation (LDA) coupled to the model of polyhedral atomic
cluster allows the differences in the electronic charge and in the density of states
for a = 11 {113} in pure nickel and in boron or phosphorus doped nickel to be
established [69].
The charge density in the boundary containing boron remains similar to that of
the pure boundary, but decreases in the interstitial sites of the boundary containing
phosphorus (Fig. 6.33). The densities of states are not significantly modified in the
presence of boron compared to the pure grain boundary; however, with phosphorus
they are displaced towards higher energies and the bandwidth becomes narrower
(Fig. 6.34). In both cases, the solute forms a strong bond with the solvent; but boron
does not affect the bonds between two neighbouring nickel whereas phosphorus
decreases the bonding tendency of the metallic atoms and the boundary stability.
Moreover, the orbital hybridization, and therefore the covalent nature of the bond
between the solute and the solvent, is lower in the case of phosphorus than in the case
of boron. The tendency to bonding between the solute atom and the solvent atom
also differs depending on whether the nickel atom is in the grain boundary plane
(Ni1) or out of that plane (Ni2), showing a directional bonding character.
Fig. 6.32 Schematic repre-

sentation of the effect of boron
or phosphorus segregation on
the iron atomic bonding state.
The double, simple and dotted
lines indicate strong, nor-
mal and weak atomic bonds,
respectively [60]
The interaction energies for different pairs of nickel atoms (Fig. 6.33) in the pure
boundary and in the boundary doped with boron or phosphorus are reported in
Table 6.3. The atomic bonds are weakened in both cases but relatively little under the
effect of boron compared to the strong attenuation of the nickel–nickel interaction
in presence of phosphorus, and this especially as the pair of nickel atoms is close to
the solute atom.
The effects of the solutes on the = 11 {113} grain boundary in nickel are well
associated to modifications of the metal–metal bonds, but also to the nature of the
bonding between solute and metal atoms. Furthermore, site competition between
boron and other solutes (or impurities) explains that this element drives out nitrogen,
oxygen, phosphorus and sulphur from grain boundaries in nickel [69]. The finding
of a change in the local electronic structure of a grain boundary does not allow,
however, to infer its behaviour with respect to the intergranular fracture; the latter
must also strongly depends on the interaction between the matrix dislocations and
grain boundaries.
(a) (b)
Fig. 6.33 Difference of charge densities between the = 11 {113} grain boundary in pure nickel
and the same boundary in nickel doped: a with boron; b with phosphorus. The spaces between the
contours are 0.002 e/(u.a.)3 . The full and the dotted lines represent a gain and a loss of charge,
respectively [69]
Table 6.3 Interaction energies (in eV) between nickel atoms in a = 11 {113} pure grain boundary
and in the same boundary doped with boron or phosphorus [69]
Pairs of nickel atoms Pure boundary B-doped boundary P-doped boundary
1–4 −1.3 −1.55 −1.28
1–6 −1.20 −1.14 −1.01
1–13 −1.50 −1.47 −1.01
4–5 −2.13 −2.10 −1.73
6–9 −1.82 −1.63 −1.30
7–9 −1.57 −1.30 −1.06
6.5.4.2 Change of Electronic Bonds on the Effect of Segregation in Oxides
The segregation of different cations X (X = Sc, Y, La) in substitutional position in

grain boundaries of α-alumina (corundum) is analysed for three twin boundaries:
prismatic = 3 (10-10), rhombohedral = 7 (10-12) and pyramidal = 13
(10-14). The modifications of the atomic structure, of the interfacial energy and
of the chemical bonding X–O with the presence and the quantity of cation atoms
are investigated by combining ab initio calculations of the electronic structure with
empirical ionic model [37]. Cations induce a reduction in the interfacial energy in
the three boundaries, more intense than the ionic of the solute is large. Like in matrix,
relaxation is mainly due to the increase of the distances X–O between the cation and
the neighbouring oxygen ions compared to these distances Al–O in pure alumina. For
example, the distances between nearest neighbours equal to 0.184 and 0.195 in pure
alumina increase to 0.209 and 0.226, respectively, when an yttrium atom replaces
an aluminium. The boundary is more open and more deformable than the matrix,
allowing the point defects and the neighbouring atoms to better relax. The results are
similar with scandium and lanthanum solutes. A linear relationship exists between,
in one hand, the difference in ionic radii of the solute and of aluminium and, on
Fig. 6.34 Densities of states

of nickel atoms, Ni1 and Ni2
(Fig. 6.33) in a pure boundary,
a boundary doped with boron
and a boundary doped with
phosphorus, compared to the
densities of states in the crystal
[69]
the other hand, the X–O bond length or the interfacial energy or the intergranular
segregation energy (Fig. 6.35); this relationship extends to Ti and Ca solutes.
Moreover, the propensity to segregation is lower when the boundary is more
ordered. The change in the local electronic structure in = 3 and = 13 is
represented by the projection of the density of states on the different sites of the
solute X, for O atoms localized at increasing distances from the boundary and, for
comparison, for O atom in the crystal. Such a representation is given for = 3
(Fig. 6.36); the relative positions and heights of the peaks, as well as the hybridization
between the atomic states of the solute and of the oxygen atom are very similar in the
three grain boundaries. The electronic structures for the three cations X and for the
three environments ( = 3, = 13 and matrix) are qualitatively and quantitatively
equivalent. A certain hybridization appears between the p states of the solutes and
Fig. 6.35 A linear relation-

ship links the misfit between
the ionic radii (rX − rAl ) [Å]
to different quantities: a the
average length of the X–O
bonds [in Å] for X = Sc,
Y and La in = 13; b the
interfacial energy (in Jm−2 )
for = 3, = 7 and
= 13 boundaries; c the
absolute value of the segrega-
tion energy (in Jm−2 ) for the
same boundaries [37]
the 2s and 2p states of the neighbouring oxygen layers; it extends to the first two
layers for scandium and yttrium and until the third layer for lanthanum. However,
despite the covalent character brought by this hybridization, the X–O bond preserves
a high degree of ionicity.
The new electron distribution resulting from segregation may be visualized
by plotting the difference of electron densities in the Al–O and the X–O bonds
Fig. 6.36 Projections of the densities of states of Sc, Y and La solutes segregated to a = 3
(10-10) twin, for the first four neighbouring layers of oxygen and, by comparison, for an oxygen
atom in pure alumina crystal [37]
(Fig. 6.37). Given the similarity between the electronic structures in both boundaries,
only the redistribution caused by the cation Y segregated at the prismatic = 3 twin
boundary is illustrated; it is obtained by cutting the density of states of the bicrystal
by a basal (0001) plane. Like in matrix, the covalent character taken by the Y–O
bond is obvious: the density of the valence electrons between yttrium and oxygen
atoms increases; it decreases in the plane perpendicular to the bond, but this change
concerns only the closest oxygen atoms.
Fig. 6.37 Section (0001) of

the difference of densities
of states of the Al–O and
Y–O bonds showing the
redistribution of the electrons
due to the substitution of an
Al cation by a Y cation in a
= 3 twin in alumina. The
electronic densities (e/Å3 ) are
given in grey scale [37]
In conclusion, the structural rearrangement induced by the segregation of different

cations in different grain boundaries of alumina is governed by the size effect. Indeed,
the segregation energy depends linearly on the misfit between the ionic radii of the
solute and of aluminium. The concept of ionic radius must be used with caution
as very simplistic. The partial degree of covalence acquired by the cation–oxygen
bonding at grain boundaries is similar to that it takes for a single cation in the α
alumina bulk.
6.5.4.3 Band Structure Changes in Semiconductors on the Effect

of Segregation
A nickel atom in interstitial position in the = 13 and = 25 tilt grain boundaries

slightly modifies the band structure and the semiconductor character of silicon. Con-
versely, nickel in substitution leads to the appearance of deep electronic levels occu-
pied in such a manner that the boundary locally acquires a semi-metallic character
( = 13) or metallic ( = 25) (Fig. 6.38) (M. Torrent, Thèse d’Université Paris VI
(1996). Olivier Hardouin Duparc (communication personnelle)).
Ab-initio calculations show that segregation of arsenic in a = 5 {310}
001
of silicon may occur without defect by a cooperative incorporation of atoms whose
Fig. 6.38 Diagrams showing

the density of electronic states
(full lines) of a = 25 grain
boundary in silicon containing
a nickel atom. The dotted lines
correspond to the density of
states of a perfect crystal: a Ni
in insertion in the boundary;
b Ni in substitution in the
boundary (M. Torrent, Thèse
d’Université Paris VI (1996).
Olivier Hardouin Duparc
(communication personnelle))
Fig. 6.39 a Schema showing how the atoms of an arsenic dimer in substitution on two neighbouring
sites of silicon, move away from each other in a tri-coordinated configuration; b Projection along
the tilt axis of the relaxed atomic structure of the = 5 {310} in silicon with the A unit (hatched);
c section of the map of the electronic charge density in a plane containing arsenic on two atomic
sites (b, c) of the structural unit A. Each arsenic atom is surrounded by density contours whose
intensity decreases away from the atom [70]
the tri-coordination (Fig. 6.39a) explains the loss of electric activity [70]. Dimers or
ordered chains of arsenic atoms (or of dimers) are formed along the intrinsic edge
dislocation cores. The relaxed structure of this grain boundary is constituted by A
units (Fig. 6.39b). Figure 6.39c shows a section of the map of the electronic charge
density in a boundary, going through a plane containing arsenic atoms at the b and
c sites of the A unit. No significant change occurs between the nearest neighbour
atoms in the dimers indicating that each arsenic atom is tri-coordinated.
Finally, the relationship between segregation and electronic structure is still less
studied and remains qualitative. In metallic systems, the presence of a solute in a
boundary induces a change of the chemical bonds, in terms of occupied bonding
states or anti-bonding states, of covalent or ionic type. But we must be careful not to
quickly infer an effect on the intergranular embrittlement or consolidation. The ionic
oxides globally preserve their ionicity, even if some bonds take a covalent character.
The change of the diagram of the density of states in the boundary, in terms of states
created in the band gap, may locally transform a semiconductor in a semimetal.
6.6 Pre-wetting Transition Upon Segregation at Grain

Boundaries
We already evoked a change in the geometrical boundary parameters, particularly

the boundary plane, in presence of a solute, and the formation of an ordered 2D com-
pound. But, another important effect of the segregation deserves a special interest:
the solid-state pre-wetting that leads to the grain boundary wetting [71].
Pre-wetting is the adsorption of solutes on the form of a finite thickness layer
at an interface. This layer is not a phase in the sense that it has no autonomy; it
does not exist without the neighbouring crystals. The terminology pre-wetting is
used because the formation of a new phase that completely wets the interface occurs
when the chemical potentials of the elements contained in that layer increase. Wetting
6.6 Pre-wetting Transition Upon Segregation at Grain Boundaries 201
is a generic term that does not necessarily implies the formation of a liquid, but works
for entirely solid systems.
Investigation of the pre-wetting behaviour has provided the means for construc-
tion of the 2D phase diagram of a grain boundary [71]; we must take care of this
terminology because the so-called 2D phase is not strictly a phase in the usual thermo-
dynamic sense. However, such grain boundary phase diagram has been obtained for
the Cu–Bi system [72]. Diffusion experiments were performed on Cu–Bi polycrystals
with different bismuth content; an abrupt increase of the grain boundary diffusivities
of both copper and bismuth appears for certain bismuth contents which are unequiv-
ocally in the single-phase region of the bulk phase diagram. Figure 6.40 shows the
concentration dependence of the bismuth grain boundary diffusivity P Bi = sδ DGB
in Cu–Bi alloys at two temperatures with s the solute segregation factor, δ the grain
boundary width and DGB the grain boundary diffusion coefficient; the same curve is
established for copper diffusivity (Fig. 6.41); a large enhancement of grain boundary
Fig. 6.40 Concentration

dependence of the Bi
grain boundary diffusivity
P Bi = sδ DGB in Cu–Bi
alloys at T = 1116 K (a)
and 1093 K (b). The solid and
dashed lines represent the bulk
and grain boundary solidus
concentrations, respectively
(after [73]), The dotted lines
represent the bismuth diffu-
sivity in pure Cu and in the
two-phase (solid + liquid)
Cu–Bi alloys [72]
Fig. 6.41 Concentration

dependence of the Cu grain
boundary diffusivity P Cu =
δ DGB in Cu–Bi alloys at
T = 1116 K. The dotted
lines represent the copper
diffusivity in pure Cu (self-
diffusion) and in the two-phase
(solid + liquid) Cu–Bi alloys
[72]
diffusion is observed at a certain temperature dependent Bi content which is clearly

below the bulk solubility limit; after exceeding this limit, no significant change in
the diffusivity is observed.
These results together with those obtained by Auger Electron spectroscopy [73]
enable to plot a grain boundary solidus line that we superimpose to the phase diagram
(Fig. 6.42) despite the fact that it is not a bulk-equilibrium phase as it does not exists
without grain boundaries.
Fig. 6.42 Cu-rich side of the

Cu–Bi phase diagram (after
[73]). The solid curve is the
(retrograde) bulk solidus line.
The dashed line is the grain
boundary solidus line obtained
for Cu–Bi polycrystals. The
annealing temperatures (1116
and 1093 K) and the Bi con-
centrations (see squares) used
in the diffusion experiments
are indicated. L = liquid;
(Cu) = Cu-rich solid solution
[72]
Such observations allow convincing conclusions on the existence of a pre-wetting

grain boundary phase transition and therefore, of a disordered (liquid-like) grain
6.6 Pre-wetting Transition Upon Segregation at Grain Boundaries 203
boundary layer between the bulk and the grain boundary solidus lines. This layer
results from an advanced state of segregation.
Pre-wetting phenomenon within the boundary induces a large modification of
the boundary structure; various states of this structure lead several authors to intro-
duce a new concept of complexions to characterize different boundary structures and
properties.
6.7 Concept of Complexions
We have seen that segregated solutes may exist in grain boundaries on different
forms, starting from few atoms distributed on a monolayer until a continuous
quasi-liquid film. The different grain boundary configurations could be considered as
separate phases with associated thermodynamic quantities: excess volume, entropy,
enthalpy… However, arguing the fact that the thermodynamic properties of intergran-
ular phases cannot be separated from the neighbouring bulk phases. Tang et al. have
proposed to name these interface equilibrium features: complexions [74]. The reason
why they are called complexions and not phase transitions is that they do not satisfy
the Gibbs definition of phase. Different boundaries having the same complexion
may have different atomic structures but have the same thermodynamic characteris-
tics. A complexion may transform into other complexions by chemistry and/or heat
treatments. Segregation and melting are examples of a complexion transition. Gibbs
recognised the existence of such transitions in films, and Cahn’s critical wetting the-
ory [75] is an excellent example of a first-order transition in the value of interface
absorption.
Different properties are associated to the different complexions; in particular,
the transport of atoms may differ by many orders of magnitude according to the
type of complexions. The grain boundary mobility, that is strongly dependent on the
transport property, has been used to distinguish the different complexions on the basis
of their migration kinetics [76]. Alumina was chosen as a model system for this study.
The evolution of the average grain boundary mobility as a function of temperature
in alumina polycrystals with different microstructures has been investigated. The
different kinetic behaviours have been correlated to the different types of complexions
that constitute, in average, the polycrystal grain boundary network. Six different
types of complexions have been distinguished; they have been observed by high-
angle annular dark-field (HAADF) transmission electron microscopy (see Sect. 6.5.3)
for three of them and by high-resolution transmission electron microscopy for the
three others (Fig. 6.43); they are schematically represented on Fig. 6.44. The different
complexions result from the segregation of different solutes, but a given solute may
have different effect according to the doping level and to the grain growth mode,
normal or abnormal; magnesium induces complexion I, neodymium complexions I
and III, silicon and calcium complexions III, IV, V and VI.
The grain boundary geometrical parameters have an influence on the stability of
a particular complexion; for example, whatever the temperature, the basal plane in
Fig. 6.43 High-angle annular dark-field (HAADF) transmission electron microscopy of a com-
plexion I, b complexion II, c complexion III, and high-resolution transmission electron microscopy
of d complexion IV, e complexion V and f complexion VI. Bright spot in the HAADF-STEM images
indicate the presence of neodymium. Complexion I shows sub-monolayer segregation; complexion
II shows no segregation; complexion III shows bilayer segregation; complexion IV show a dis-
ordered layer solute; complexion IV contains an intergranular film and complexion VI a thicker
wetting intergranular film with arbitrary thickness [76]
calcium-doped alumina more likely adopts a complexion III than the other planes
that contains complexion IV–VI. Finally, the formation and the stability of a given
complexion depend on several factors: chemistry, thermal treatment, misorientation
and grain boundary plane orientation. This multiplicity of complexions and thus
of possible grain boundary behaviours has been rationalized through a complexion
diagrams that is superimposed to the bulk-equilibrium phase diagram for the binary
system Al2 O3 –SiO2 (see Fig. 6 in Ref. [76]). Investigations of complexions in other
material systems appear as future challenges.
Complexion may serve as a new concept for kinetic engineering in material science
[76]. This kinetic consideration is part of the grain boundary engineering proposed
in 1984 by Watanabe [77]. On the contrary of most grain boundary distributions
that rest on geometrical criteria (see Chap. 12) the kinetic engineering may use a GB
classification in terms of complexions based on GB transport properties, thus close
the material performance; Note that a distinction between grain boundaries based on
their diffusivity has already been proposed and the role of segregation in the different
grain boundary diffusion behaviours has been discussed [78, 79]. Other distributions
supported by different grain boundary behaviours are also mentioned in part III. In
any case, the management of the grain boundary behaviours can be used to obtain
and control a given application requirement of a given material. The fact to relate
6.7 Concept of Complexions 205
Fig. 6.44 Schematic configurations of the six different grain boundary complexions: I, submono-
layer segregation; II, clean grain boundary; III bilayer segregation; IV, multi-layer segregation; V,
equilibrium thickness amorphous intergranular film; VI, wetting film. This representation is based
on the images in Fig. 6.43 [76]
this kinetic engineering to a new concept of complexion gives a great importance to

the segregation phenomenon and explains the part it takes in that book.
6.8 Role of Extrinsic Dislocations in Grain Boundary

Equilibrium Segregation
The approach of the segregation of point defects on extrinsic dislocations is simi-

lar to that of their segregation on crystal dislocations. The elastic strain and stress
fields extend to large distance in both cases, their expressions in function of the
distance from the dislocation line take the same forms. Note here that we deal with
equilibrium segregation although extrinsic dislocations are non-equilibrium defects
in gain boundaries.
The solute atom is considered as an elastic sphere with a radius r* that differs from
the radius r0 of the grain boundary site in which it takes place; the host site may be
that of a solvent atom for a substitutional solute or an interstitial site in the boundary.
The relation between r* and r0 involves the elastic deformation ε associated to the
point defect: r ∗ = r0 (1 + ε).
In the vicinity of the extrinsic dislocation, an elastic interaction occurs between
the stress fields of the two defects, point and linear, the energy of which is inversely
proportional to the distance R between the defects. As a result, the solute concen-
tration near a dislocation line differs from the average volume concentration C0
and from that CGB of the boundary in which the dislocation is located. The general
equation giving the solute concentration in a site of energy E GB is:
C = C0 exp −E GB /kT (6.31)
The interaction energy between a solute and a dislocation being generally nega-
tive, a force attracts the solute towards the dislocation line: C > C0 . But a limit value
of C exists, it cannot be superior to one atom per available site. When the distance
R between the defects decreases, C increases until a limit value linked to a criti-
cal radius R0 , almost equal to the dislocation core radius. The solute is condensed
on the dislocation line forming Cottrell cloud or a Cottrell atmosphere. The inter-
action energy has reached a minimal value yielding a stabilization of the extrinsic
dislocation. We say that the dislocation is trapped by the Cottrell atmosphere.
Although these reminders are not specific to the segregation of an extrinsic dis-
location line, we find it necessary to address two issues later in this section:
• Intergranular precipitation on grain boundary defects.
• Accommodation processes of the intergranular stresses associated to extrinsic
dislocations in which the segregation/precipitation phenomena and stabilization–
destabilization of dislocations play a predominant role.
Figure 6.45 shows an example of yttrium overconcentration in the close vicinity
of two extrinsic dislocations that have been stabilized in a general grain boundary
in alumina. A variation in the yttrium content along a curved general boundary has
also been detected: in the region displaying several extrinsic dislocations, the yttrium
content is clearly higher than in the neighbouring regions without dislocations [35].
The relationship between segregation and grain boundary defects is also recipro-
cal: the defect attracts the solute and the solute trapped on the dislocation line changes
the dislocation core structure. An example of this reciprocal effect has been revealed
by experiments of the sulphur segregation to nickel grain boundaries. A same near
= 3 grain boundary contains extrinsic dislocations, dissociated or not according
to its sulphur content. Dissociated dislocations with Burgers vector a/6
112 have
been only observed in the pure boundary. In all the other boundaries, close or not
= 3, the extrinsic dislocations preserve the Burgers vector of the perfect lattice
dislocations [34].
Note that, in the case of an extrinsic dislocation, the solute enrichment has two
origins and two kinetics: coming from the matrix by a volume diffusion mechanism
thus relatively slow and coming from the grain boundary by an intergranular dif-
fusion mechanism, thus generally more rapid. This significant difference between
the segregation on a dislocation depending on whether it is located in the matrix or
in the grain boundary takes a great importance in the stress relaxation phenomena
(see Chap. 9) and, more generally, in all properties involving migration and /or grain
boundary sliding.
6.9 Non-equilibrium Segregation at Grain Boundaries 207
Fig. 6.45 X-ray spectra

obtained by STEM on a grain
boundary in yttrium-doped
alumina: a the yttrium peaks
are associated to the presence
of two extrinsic dislocations
in the grain boundary; b the
dislocations are visualised by
their intense strain field (dark
region) [35]
6.9 Non-equilibrium Segregation at Grain Boundaries
Non-equilibrium segregation (also named kinetic segregation) generally results from

an interaction between the atoms of one component of a solid solution with the
thermal vacancies in excess in this solution due to quenching or high-rate deformation
and/or with the interstitial defects created by irradiation. It also accompanies grain
boundary migration that occurs during recovery and recrystallization of a material
or under the effect of an electric field. Then the solutes are dragged by the moving
boundaries. These complex phenomena are described here without enter in details
in the theories; the reader may refer to [6, 80, 81].
The occurrence of a non-equilibrium segregation phenomenon is first proposed
by Aust et al. [82] to explain the local increase or decrease of the microhardness near
a boundary depending on whether it is enriched or depleted in solute after quenching.
Soon after, Anthony [83] directly reveals and explains this phenomenon by consider-
ing two types of processes depending on the existence or not of a coupling between
vacancy and solute. In a material maintained at high temperature then quenched,
vacancies in overconcentration in the matrix migrate to the sinks constituted by
the grain boundaries during quenching (if not too rapid) and/or during subsequent
annealing. If a strong positive interaction exists between vacancies and solute atoms,
the two species migrate together towards a grain boundary on the form of complex or
solute-vacancy pair. As a result a solute accumulation extends over a region of several
micrometers on both sides of the grain boundary. If the interaction is weak, the solute
and vacancies fluxes may be in opposite directions, leading to a solute depletion near
the grain boundary. Similar processes occur during irradiation but, in that case, solute-
interstitial complexes on the shape of dumbbells are formed. Another possible phe-
nomenon occurs when the fluxes of solute and solvent atoms are both opposed to the
vacancy flux, the intergranular region is then depleted or enriched in solute depending
on whether the volume diffusion coefficient of the solute is higher or lower than that
of solvent [80]. Due to its similarity with the Kirkendall effect in which the uneven
fluxes of solutes create vacancy concentration gradients, the previous phenomenon
is called Inverse Kirkendall effect. Analogous processes may occur with interstitials.
As for equilibrium segregation, the driving force for non-equilibrium segregation
is the energy decrease associated to the atom transfer from the bulk to the grain
boundary. Although the energy of a system at equilibrium differs from that in a non-
equilibrium state, this difference is negligible compared to the segregation energy of
a solute. The two types of segregations essentially differ in their kinetics that depends
more on the rate of the microstructure relaxation (non-equilibrium => equilibrium)
than on the solute diffusion in the matrix [6]. This yields a drastic rise in the appar-
ent diffusivity of the solute suggesting that the phenomenon will stop as soon as
the vacancy excess in the volume is eliminated. The non-equilibrium segregation is
thus a transitory phenomenon. The concentration profiles of vacancies or vacancy–
solute complexes in the intergranular region have been deduced from studies of the
segregation kinetics by solving the second Fick equation with appropriate condi-
tions for grain boundaries. By considering, in the quasi-steady state, the diffusion of
vacancy–solute pairs, the equation takes the form [84]:
Cx − CGB x
= erf (6.32)
Cg − CGB 2(Dc t)1/2
where Cx , CGB and Cg are the vacancy–solute complex concentrations at a distance

x from the boundary, in the grain boundary and in the grains, respectively, at the
temperature after quench during the time t. Dc is the volume diffusion coefficient
of the complex. This approach is often limited by the ignorance of the magnitudes
of the quantities attached to the energy and the diffusion of the complexes. However,
it enables to predict an inverse motion of the solute, from the boundary to the grain,
after a critical time tc corresponding to the depletion of complexes in the grain
centre; this may occur during quenching or during subsequent ageing, if the quench
is too rapid. The non-equilibrium segregation is then reduced and the conditions for
equilibrium segregation to grain boundaries are restored. A more rigorous approach
to non-equilibrium segregation is the simultaneous numerical resolution, with the

computer, of the equations giving the variations in concentration of all species i
involved in the process [85]:
∂Ci ∂ 2 Ci
= Di 2 (6.33)
∂t ∂x
with i = v (for vacancy), s (for solute) and c (for complex) … However, this method
neglects the effects of microstructure and the relation between segregation and grain
boundary diffusion.
The vacancy–solute pairs migrate more rapidly than the isolated solute atoms;
the concentration of the latter is thus not affected by the non-equilibrium segrega-
tion. The solute concentration profiles in the vicinity of the grain boundary differ
according to the segregation type: equilibrium, non-equilibrium after quenching and
non-equilibrium after irradiation (Fig. 6.46) [81].
Fig. 6.46 Solute segregation profiles on one side of a grain boundary according to the segregation
type: a equilibrium segregation; b non-equilibrium segregation after quenching; c non-equilibrium
segregation after irradiation (o/s and u/s mean that the solute atom is larger and smaller that the
solute atom, respectively) [81]
The parameters that influence the mechanisms of segregation after quenching are:
• The temperature of the initial annealing that determines the vacancy percentage
after quenching.
• The cooling rate that allows a more or less large number of atoms to reach the
intergranular region.
• The interaction energy between a vacancy and a solute forming a complex (a pair)
and the activation energy for the diffusion of this complex.
The results on the non-equilibrium segregation of boron in an austenitic steel
quenched from temperatures beyond 800 ◦ C well illustrate the influences of the first
two parameters (Fig. 6.47) [85]. The higher the initial temperature, the higher the con-
centration of solute in the boundary due to the large number of vacancies formed and
quenched. This result strongly distinguishes the non-equilibrium segregation from
the equilibrium segregation to grain boundaries, the latter being more elevated at low
temperature. Furthermore, a segregation maximum always occurs for a quenching
rate that depends on the initial temperature. Synergic effects between boron and
molybdenum have been found, similar to those discovered between phosphorus and
molybdenum or between phosphorus and boron in equilibrium segregation.
Fig. 6.47 Evolutions of the

enrichment factor β of boron at
grain boundaries of austenitic
steel (grain size 150 μm)
in function of the cooling v
and for different annealing
temperatures T0 [85]
In case where segregation occurs during annealing subsequent to quenching, the

controlling factors are reduced to the activation energy for migration of complexes
and the annealing temperature. The acceleration of the kinetics of vacancy elimina-
tion during an isothermal treatment of a Fe-40%Al alloy doped with boron compared
to this kinetics for a non-doped alloy may be then explained by the formation of
boron–vacancy complexes that migrate more rapidly towards the boundaries than
the isolated vacancies (Fig. 6.48) [86].
The non-equilibrium segregation is also characterized by a possibility of
desegregation with time, resulting from the inverse diffusion of the segregated species
to their stable positions in the matrix. This is observed in the case of boron segregation
in the Fe–Al intermetallic compound (Fig. 6.49); after a high content, mainly due to
Fig. 6.48 Kinetics of vacancy

elimination during an isother-
mal annealing at 380 ◦ C for
three grades of Fe-40Al: non-
doped alloy (◦), doped with
40 ppm () and 100 ppm of
boron () [86]
Fig. 6.49 Influence of the

holding time at 400 ◦ C on the
average boron concentration
at grain boundaries of a Fe-40
Al alloy doped with 200 at.
ppm of boron [86]
a non-equilibrium segregation mechanism, the grain boundary boron concentration

is maintained to a fixed low value corresponding to equilibrium [86].
The approaches previously described may always be used; the amplitude of the
segregation depends on the relative diffusion rates in the matrix of the complexes
and of the free solute atoms. Under irradiation, the acceleration of the diffusion
and the strong interstitial–solute bonding must be taken into account. Generally,
the vacancy–solute complexes are formed when no difference exists between the
atoms of the two elements. On the contrary, a strong bonding between the interstitial
and the solute is established when the solute atom is smaller than the solvent atom
leading to a high segregation level induced by radiation. When the solute atom is
larger than the solvent atom, a solute depletion occurs at the boundaries under the
effect of irradiation, as it is the case for chromium in austenitic steel (Fig. 6.50) [87].
Before irradiation, chromium is strongly segregated to grain boundaries of the steel
that underwent quenching, vacancy–solute complexes being created.
For a given species and a given energy of the incident particles (particularly neu-
trons), the stationary segregation level depends on the particle flux (in displacement
per atom or dpa/ second) and the irradiation temperature. At low temperature and high
flux, the defect overconcentration is elevated (high production and low mobility):
the vacancy–interstitial recombinations reduce the flux of defects arriving to the
boundary and the induced segregation is low. At elevated temperature and low flux,
the overconcentration of defects and their flux to the grain boundaries are again low.
Fig. 6.50 Redistribution in the vicinity of a grain boundary of the austenitic steel components,
induced by neutron irradiation in a light water reactor: a concentration profiles of the components
for a given radiation dose; b Chromium distribution according to the radiation dose: initial state (),
after a high radiation dose (5 dpa) () and after an intermediary radiation dose (1.5 dpa) () [87]
Only for intermediate temperatures and fluxes, a high segregation level is achieved;
thus, there exists a range of irradiation conditions giving rise to this phenomenon.
Finally, for a given flux and a given temperature of neutron irradiation, induced
segregation is even less important than the defects are produced in dense cascades:
many recombinations reduce the effective level of free defects, able to migrate over
long distances and only capable of inducing non-equilibrium segregation [88, 89].
Modelling of grain boundary segregation induced by irradiation has been first
performed in the case of dilute solutions by neglecting the direct interaction between
solutes. In concentrated solid solutions, the assumption of isolated solute atoms is
not realistic. A model of chemical kinetics that takes into account the effects of
the alloy chemistry on the point defects properties, on intergranular diffusion and
segregation, is described in two joined papers [88, 89]; it implies a large number
of parameters (pair potential interaction parameters, attempt frequencies and saddle
point energies) fitted on the equilibrium properties and on experimental diffusion
data. The model well accounts for the effect of interstitial solutes on the segregation
induced by irradiation in the Fe–Cr–Ni alloys: these solutes enhance the chromium
depletion and the nickel enrichment in grain boundaries, in good agreement with
experiments.
Non-equilibrium segregation may also results from a high-rate deformation when
the vacancy excess is generated by the non-conservative motion of jogs along the
screw dislocations at low temperature or on thermal jogs at high temperature. Non-
equilibrium segregation practically occurs in all systems.
6.10 Concluding Remarks
The distinction between the quantity of solute resulting from an equilibrium

mechanism and that coming from a non-equilibrium mechanism is not obvious.
But whatever the mechanism, segregation induces a grain boundary embrittlement
or strengthening, depending on the geometrical boundary parameters. Controlling
grain boundary resistance to fracture requires to control grain boundary segregation.
We may play on the competition between the segregated elements to favour segrega-
tion of a strengthening element at the expense of that of embrittling elements. Then,
we enter the domain of Grain boundary Engineering.
As a transition to the next chapter, grain boundary segregation, equilibrated or not,
often constitutes a preliminary stage for precipitation. The transition between the two
phenomena is not easy to identify; in particular, when a 2D film with the same com-
position and structure than the 3D compound appears at the boundary. Precipitation
strongly affects the grain boundary behaviours, either with a complementary effect
to the segregation or in an opposite way as noted for the dislocation stabilization. An
approach of the interfacial precipitation is thus a prerequisite to the understanding
of the grain boundary properties and constitutes the subject of the following chapter.
References
1. J.O. Arnold, J. Iron Steel Inst. 45, 107 (1894)

2. A.P. Sutton, R.W. Balluffi, Interfaces in Crystalline Material (Oxford Science Publications,
Oxford, 1995)
3. E.D. Hondros, J. de Physique 36, C4–117 (1975)
4. M. Guttmann, D. McLean, in Interfacial Segregation, ed. by W.C. Johnson, J.M. Blakely (ASM
Metals Park, Ohio, 1979)
5. P. Lejcek, S. Hofmann, Crit. Rev. Solid State Mater. Sci. 20, 1–85 (1995)
6. G. Saindrenan, R. Le Gall, F. Christien, Endommagement interfacial des métaux - Ségrégation
interfaciale et conséquences (Ellipses, Paris, 2002)
7. W. Hume-Rothery, R. Smallman, C.W. Haworth, The Structure of Metals and Alloys (Institute
of metals, London, 1969)
8. P. Guyot, J.-P. Simon, J. de Physique 36, C4–141 (1975)
9. I.S. Sokolnikoff, Mathematical Theory of Elasticity (Mc Graw Hill, New York, 1956)
10. D. Wolf, J.F. Lutsko, J. Mater. Res. 6, 1427 (1989)
11. J.B. Adams, W.G. Wolfer, S.M. Foiles, Phys. Rev. B40, 9479 (1989)
12. P. Wynblatt, D. Chatain, Metal. Mater. Trans. 37A, 2595 (2006)
13. D. McLean, Grain Boundaries in Metals (Clarendon Press, Oxford, 1957)
14. M.P. Seah, E.D. Hondros, Proc. R. Soc. Lond. A 335, 191 (1973)
15. P. Lejcek, O. Schneeweiss, Surf. Sci. 487, 210 (2001)
16. J.O. Vasseur, P.A. Deymier, B. Djafari-Rouhani, L. Dobrzynski, Interface Sci. 1, 49 (1993)
17. R.H. Fowler, E.A. Guggenheim, Statistical Thermodynamics (Cambridge University Press,
Cambridge, 1939)
18. M. Guttmann, Surf. Sci. 53, 213 (1975)
19. H. Erhart, H.J. Grabke, Metal Sci. 15, 401 (1981)
20. C.L. Briant, Philos. Mag. Lett. 73, 345 (1996)
21. C. Pichard, M. Guttmann, J. Rieu, C. Goux, J. de Physique 10, C4–151 (1975)
22. S. Hofmann, J. de Chimie Physique 84, 141 (1987)

23. C.D. Terwilliger, Y.M. Chiang, Acta Metall. Mater. 43, 31 (1995)
24. J.W. Cahn, J. de Physique 43, C6–199 (1982)
25. C.L. Briant, Acta Metall. 31, 257 (1983)
26. T. Ogura, C.J. Mc Mahon, H.C. Feng, V. Vitek, Acta Metall. 26, 1317 (1978)
27. S. Susuki, K. Akibo, K. Mikura, Scr. Metall. 15, 1139 (1981)
28. D. Bouchet, L. Priester, Scr. Metall. 20, 961 (1986)
30. D. Wolf, J. Phys. 46, C4–197 (1985)
32. A.P. Sutton, Philos. Mag. A 63, 793 (1991)
33. M. Juhas, L. Priester, M. Biscondi, Mater. Sci. Eng. 185, 71 (1994)
34. D. Bouchet, B. Auffray, L. Priester, J. Phys. 49, C5–417 (1988)
35. D. Bouchet, F. Dupau, S. Lartigue-Korinek, Microsc. Microanal. Microstruct. 4, 561 (1993)
36. W. Swiatnicki, S. lartigue Korinek, J.-Y. Laval, Acta metall. mater. 43, 795 (1995)
37. S. Fabris, C. Elsässer, Acta Mater. 51, 71 (2003)
38. P. Lejcek, S. Hofmann, Interface Sci. 1, 163 (1993)
39. L.S. Shvindlerman, B.B. Straumal, Acta Metall. 33, 1735 (1985)
40. J.D. Rittner, D.N. Seidman, Acta Mater. 45, 3191 (1997)
41. P. Lejcek, Surf. Interface Anal. 26, 800 (1998)
42. P. Lejcek, S. Hofmann, Interface Sci. 9, 221 (2001)
43. J. Stolarz, J. Le Coze, J. Phys. 51, C1–146 (1990)
44. T.G. Ference, R.W. Balluffi, Scripta Metall. 22, 1929 (1988)
45. C.W. Park, D.Y. Yoon, J.E. Blendell, C.A. Handwerker, J. Am. Ceram. Soc. 86, 603 (2003)
46. J. Friedel, Adv. Phys. 3, 446 (1954)
47. K. Sickafus, S.L. Sass, Scripta Metall. 18, 165 (1984)
48. J.A. Eastman, S.L. Sass, J. Am. Ceram. Soc. 69, 753 (1986)
49. D. Udler, D.N. Seidman, Scripta Metall. Mater. 26, 449 (1992)
50. R.E. Peierls, Proc. Phys. Soc. 52, 23 (1940)
51. F.R.N. Nabarro, Proc. Phys. Soc. 57, 256 (1947)
52. D. Schwartz, V. Vitek, A.P. Sutton, Philos. Mag. A 51, 499 (1985)
53. H.Y. Wang, R. Najafabadi, D.J. Srolovitz, R. Lesar, Acta Metall. Mater. 41, 2533 (1993)
54. A.P. Sutton, V. Vitek, Acta Met. 30, 2011 (1982)
55. V. Vitek, G.J. Wang, Surf. Sci. 144, 110 (1984)
56. G.J. Wang, A.P. Sutton, V. Vitek, Acta Metall. 32, 1093 (1984)
57. E.D. Hondros, D. McLean, Philos. Mag. 29, 771 (1974)
58. B. Lezzar, O. Khalfallah, A. Larere, V. Paidar, O. Hardouin Duparc, Acta Mater. 52, 2809
(2004)
59. M. Yan, M. Sob, D.E. Luzzi, V. Vitek, G.J. Ackland, M. Methfessel, C.O. Rodriguez, Phys.
Rev. B 47, 5571 (1993)
60. M. Hashimoto, Y. Ishida, R. Yamamoto, M. Doyama, Acta Metall. 32, 1 (1984)
61. P.M. Voyles, J.L. Grazul, D.A. Muller, Ultramicroscopy 96, 251 (2003)
62. S. Lartigue-Korinek, D. Bouchet, A. Bleloch, C. Colliex, Acta Mater. 59, 3519 (2011)
63. D. Blavette, F. Vurpillot, B. Deconihout, A. Menand, Atom probe tomography: 3D imagning
at the atomic level, Chapter 9 of Fabrication and characterization in the micro-nano range, in
New Trends for Two and Three Dimensional Structures, vol. 10, eds. by F.A. Lasagni, A.F.
Lasagni (Springer, New York, 2011), pp. 201–222
64. E. Cadel, D. Lemarchand, S. Chamberland, D. Blavette, Acta Mater. 50, 957 (2002)
65. V. Pontikis, J. Phys. IV France 9, Pr4-29 (1999)
66. O. Hardouin Duparc, Communication au 41ème colloque de métallurgie “Ségrégation Interfa-
ciale dans les solides”, CEA Saclay/INSTN, 1998
67. C.L. Briant, R.P. Messmer, Acta Metall. 30, 1811 (1982)
68. V.J. Keast, J. Bruley, P. Rez, J.M. Maclaren, B.D. Williams, Acta Mater. 46, 481 (1998)
69. L.G. Wang, C.Y. Wang, Mat. Sci. Forum, Trans. Tech. Pub. Switz. 294–296, 489 (1999)
References 215
70. A. Maiti, M.F. Chisholm, S.J. Pennycook, S.T. Pantelides, Phys. Rev. Lett. 77, 1306 (1996)
71. P. Wymblatt, D. Chatain, Mater. Sci. Eng. A 495(1–2), 119 (2006)
72. S. Divinski, M. Lohmann, C. Herzig, B.B. Straumal, B. Baretzky, W. Gust, Phys. Rev. B71,
104104–1 (2005)
73. L.S. Chang, E. Rabkin, B.B. Straumal, B. Baretzky, W. Gust, Acta Mater. 47, 4041 (1999)
74. M. Tang, C. Carter, R.M. Cannon, Phys. Rev. Lett. 97, 075502 (2006)
75. J.W. Cahn, Critical point wetting. J. Chem. Phys. 66, 3667 (1977)
76. S.J. Dillon, M. Tang, W.C. Carter, M.P. Harmer, Acta Mater. 55(18), 6208 (2007)
77. T. Watanabe, Res. Mechanica 11, 284 (1984)
78. W.A. Swiatnicki, M.W. Grabski, Acta Metall. 34, 817 (1986)
79. W.A. Swiatnicki, M.W. Grabski, Mater. Sci. Eng. 100, 85 (1988)
80. J. Philibert, Diffusion et transport de matière dans les solides (Les éditions de Physique, Paris,
1985)
81. R.G. Faulkner, Int. Mater. Rev. 41, 198 (1996)
82. K.T. Aust, R.E. Hanneman, P. Niessen, J.H. Westbrook, Acta Metall. 16, 291 (1968)
83. T.R. Anthony, Acta Metall. 17, 603 (1969)
84. P. Doigt, P.E.J. Flewitt, Acta Metall. 19, 1831 (1981)
85. L. Karlsson, Acta Metall. 36, 25 (1988)
86. A. Frackiewicz A.S. Gay, M. Biscondi, Mater. Sci. Eng. A 258, 108 (1998)
87. S. Bruemmer, E.P. Simonen, P.M. Scott, P.L. Andersen, G.S. Was, J.L. Nelson, J. Nucl. Mater.
274(3), 299 (1999)
88. G. Martin, M. Nastar, J. Phys. IV France 9, 59 (1999)
89. M. Nastar, P. Bellon, G. Martin, J. Phys. IV France 9, 69 (1999)
Chapter 7
Precipitation at Grain Boundaries
Several reasons explain why precipitation at grain boundaries occurs preferentially

compared to precipitation within the crystals:
• First, a thermodynamic reason is the decrease of the energy barrier of the heteroge-
neous nucleation with respect to its value for homogeneous nucleation. Nucleation
at grain boundaries is especially favoured when the chemical driving force is low
and, simultaneously, the ratio between the grain boundary energy and that of the
“nucleus/bulk” interface is high.
• Then, the self-diffusion coefficient at any grain boundary is generally higher than
in the volume; this allows the solute atoms to rapidly migrate towards any new
phase nucleus within the grain boundary.
• Finally, a strong segregation yields solute atom saturation in the grain bound-
ary, then formation of second-phase nuclei. For example, garnet (YAG) precipi-
tates appear just after yttrium saturation of grain boundaries in yttria-doped alu-
mina polycrystals (Fig. 7.1) [1]. Grain growth leads to redistribution of yttrium
atoms in grain boundaries; the average boundary yttrium concentration increases
until saturation followed by formation of intergranular YAG particles. The critical
size for which the “segregation/precipitation” transition occurs decreases with the
global yttrium content in alumina.
Apart from the previous reasons that concern the whole grain boundary, local pre-
cipitates may appear on intergranular defects, dislocations and steps. For an extrinsic
dislocation, the driving force is derived from its elastic field and the grain boundary
precipitation phenomenon is akin to that occurring on a lattice dislocation.
The main goal of this chapter is not to deal exhaustively with the grain boundary
precipitation, but to give some insights into a phenomenon that locally affects the
grain boundary structure and chemistry, and appreciably modifies the polycrystal
responses to various stimuli. Recrystallization and plastic deformation are particu-
larly affected because they imply grain boundary migration, generally impeded by
intergranular precipitates. Corrosion is also affected as the second-phase particles or
the neighbouring regions depleted in solutes constitute local electrochemical cells.

218 7 Precipitation at Grain Boundaries

average yttrium concentra-
tion at grain boundaries in
alumina polycrystals as the
grain size increases e.g. the
volume occupied by the inter-
granular regions decreases.
Yttrium saturation just cor-
responds to the appearance
of garnet precipitates at grain
boundaries, all the more rapid
as the yttrium doping is higher
[1]
In other respects, the analysis of the precipitate shape and distribution at a grain
boundary may bring information on the plane, on the energy and on the defects of
that grain boundary.
7.1 Energetic Aspect
The formation of a second-phase nucleus is usually heterogeneous and is favoured

at the grain boundaries of the parent phase. The total free enthalpy variation G that
accompanies the heterogeneous nucleation of an embryo is given by:
G T = VG V + Aα/β γα/β −Aα/α γα/α (7.1)
G V is the free enthalpy variation per unit volume (chemical term plus, eventually,
a stress contribution, due to coherent stresses for example). Aα/β is the area occupied
by each interface α/β and γα/β the energy per unit area of each interface. Aα/α γα/α =
G D is the energy released by the destruction of the defect constituted by the grain
boundary region where nucleation occurs. The first and the third terms are negative,
thus favourable to nucleation; the second term is positive as it corresponds to a surface
creation, and thus is opposed to phase transformation. Nucleation occurs when the
total G T term is reduced by an increase of the nucleus size.
Consider first a β nucleus, the interfacial energy of which being relatively high and
independent of interface orientation. Such a nucleus is incoherent with the α matrix,
its equilibrium shape, so-called “allotriomorph”, is constituted by two spherical caps
symmetrical to each other with respect to the grain boundary plane (Fig. 7.2). The
terms expressing the nucleus shape in Eq. (7.1), volume V, interface area Aα/β and
7.1 Energetic Aspect 219
Fig. 7.2 Totally incoherent

nucleus, so-called
“allotriomorph”, at a grain
boundary
grain boundary area Aα/α now occupied by the nucleus, depend on the radius r and
on the ϕ angle between the nucleus and the grain boundary surfaces.
It is shown that the free enthalpy variation G het necessary for heterogeneous
nucleation at a grain boundary is a fraction of the homogeneous nucleation enthalpy
G hom , which only depends on ϕ:
G het = S(ϕ)G hom (7.2)

with S(ϕ) = 1/2(2 + cos ϕ)(1 − cos ϕ)2 (7.3)
and G hom = 4/3πr 3 G V + 4πr 2 γα/β (7.4)
Below a critical radius of curvature, the nucleus is not stable; it is only a solute atom
cluster that forms and dissolves in the matrix upon the effect of thermal vibrations.
Indeed, in expression (7.4) and for low radius sizes, the positive G surface term
dominates; it increases until the radius size reaches a critical value r ∗ , then decreases
(Fig. 7.3). By differentiating this equation with respect to r, we get the critical r ∗
value and the corresponding energy barrier G ∗ hom to overcome in order that a
stable nucleus appears:
2γα/β
r∗ = − (7.5)
G v
The critical r ∗ value for heterogeneous nucleation is obviously the same than that
for homogeneous nucleation, but the energy barrier G ∗het differs according to
the ϕ angle value, more or less lower than G ∗hom (Fig. 7.3): for ϕ = 45◦ ,
G ∗het ∼= 0.1G ∗hom and for ϕ = 10◦ , G ∗het ∼= 3 · 10−4 G ∗hom . When ϕ = 90◦ ,
the sum of the two parts of the nucleus, in both crystals, forms an entire sphere and
there is no energy profit.
Until now, we have considered an incoherent nucleus with respect to the matrix.
In fact, the shape of a β nucleus appearing at a αI /αII grain boundary of the α phase
depends on several parameters:
• The energy of the grain boundary, function of its atomic structure;
• The energy of each α/β interface depending on the existence (or not) of preferential
habit plane relationships between the nucleus and the matrix, for example dense
parallel planes or directions leading to lattice coherency;
• The orientation of the grain boundary plane on one side and on the other side of
the α parent phase: this plane may be parallel (or quasi-parallel) to a dense plane
of the β phase or may have a random orientation;
Fig. 7.3 Schematic variation

of the total free enthalpy
in function of the radius of
curvature of the sphere (or of
the spherical cap) showing
that the energy barrier for
heterogeneous nucleation
is well lower than that for
homogeneous nucleation, for
a same critical radius value
• The presence of extrinsic dislocations with long-range stress fields that strongly
attract the solute atoms.
As a result, we generally observe a non-homogeneous precipitate distribution from
a grain boundary to the other and along a same boundary.
Among the above-mentioned parameters, we have previously treated in part I of
the grain boundary energy (see Chap. 4). We must now analyze the other parameters
before to deal with the various morphologies of the second-phase nuclei at grain
boundaries. First, we review the classification of interfaces that may appear between
two crystalline phases, classification based on the notion of coherency. The role of
extrinsic dislocations will be approached later on.
7.2 Different Types of Interfaces and Precipitates
Three types of hetero-phase interfaces are generally considered: coherent or com-

mensurate, semi-coherent (or “dis-commensurate”) and incoherent or incommensu-
rate [2].
7.2.1 Coherent Interface
A coherent interface arises when the two crystals match perfectly at the interface
plane so that the two lattices are continuous across the interface (Fig. 7.4). Whatever
7.2 Different Types of Interfaces and Precipitates 221
the chemical species, the interface plane has the same atomic configuration in the
two phases, requiring the existence of a preferential orientation relationship between
the two crystals.
Most often, the interface plane is a dense plane in each crystal and the dense direc-
tions are parallel. The atomic distances along the dense directions in the interface,
dα and dβ , are equal or almost equal. If they are strictly equal, the interfacial energy
only depends on the difference between the chemical species from one interface side
and from the other one, then γcoherent = γchemical . The energy of a perfectly coherent
interface is very low, of the order of few mJ·m−2 . If the atomic spacing in the two
phases lightly differs, the coherency may be maintained provided the presence of
so-called “coherency strains” that account for the distortions at the interface vicinity
(Fig. 7.5). These strains introduce a volume term G c in the energetic balance given
by Eq. (7.6):
G v = V (G chem. + G c ) (7.6)
Then, the energy of a coherent interface may increase until values of the order of
200 mJ·m−2 .
With respect to the grain boundary classification, an analogy may be proposed
between a perfectly coherent interface and a perfectly two-dimensional coincident
grain boundary as far as they do not contain misfit dislocations in one case and no
intrinsic secondary dislocations in the other case. However, this analogy is restricted
by the existence of high-energy coincident grain boundaries. The preferential rela-
tionships between two phases are akin to the coincidence relationships between two
crystals that may lead to favoured grain boundaries, but in any case, the interface
plane determines the energy.
Fig. 7.4 Perfectly coherent interfaces: a each crystal has a different chemical composition but the
same crystal structure; b the two phases have different lattices
Fig. 7.5 Coherent interface with a slight mismatch leading to coherency strains in the crystals at
vicinity of the interface
7.2.2 Semicoherent Interface
When the misfit between the atomic distances from each side of the interface
increases, periodic dislocations so-called “misfit dislocations” appear (Fig. 7.6).
These dislocations are analogous to intrinsic dislocations in grain boundaries. The
misfit δ is defined by:
2(dβ − dα )
δ= (7.7)
(dβ + dα )
where di is the atom spacing in the phase i along the “coincident” directions within
the interface (not to be confused with the interplanar distance), α is the parent phase
and β is the new phase. If δ is very small, the distance D between the dislocations is:
|b|
D∼
= (7.8)
|δ|
with |b| the Burgers vector of the interfacial dislocations.

The interface displays a good fit everywhere apart from the regions near the
dislocation core; its energy is given by:
γsemi-coherent = γchem. + γstress (7.9)
γstress is the elastic energy of the dislocations per unit area of the interface. For low
δ values, the elastic energy is approximately proportional to the dislocation density,
i.e. inversely proportional to their distance and thus proportional to the misfit. Semi-
coherent interfaces are akin to vicinal grain boundaries; their energies vary between
200 and 500 mJ·m−2 .
7.2 Different Types of Interfaces and Precipitates 223
Fig. 7.6 Semi-coherent interface: the misfit parallel to the interface is accommodated by a periodic
arrangement of edge dislocations
As in the case of grain boundaries, the network accommodation at the interface

may proceed by formation of pure steps or of so-called “structural” steps displaying
a dislocation character, also named “disconnections”.
When the misfit increases, the distance between the dislocations decreases.
Beyond a certain δ value, generally ±0.25, the dislocation cores overlap; it is then
difficult to distinguish the good fit regions from the bad fit regions and the interface
becomes incoherent.
7.2.3 Incoherent Interface
An incoherent interface generally results from a random orientation relationship

between the nucleus and the matrix (Fig. 7.7). But, it may also appear between crystals
displaying a preferential misorientation if the interface plane is random (analogy with
a grain boundary where, for a given coincidence, the plane may be symmetrical or
not, constituted of dense planes or random planes of the crystals).
The energies of the incoherent interfaces, between 500 and 1000 mJ·m−2 , are
relatively insensitive to variations of the interface plane orientation. In an incoherent
interface, the misfit dislocation spacing is comparable to the dislocation core width.
From the two previous comments, incoherent interfaces and general grain boundaries
may be considered as analogous.
7.2.4 Different Types of Precipitates
According to the interfaces that encounter a precipitate, we distinguish:

Fig. 7.7 Incoherent interface
• Fully coherent precipitates appear when all their interfaces are coherent. They
generally exist only in the first precipitation stages, when structural changes do
not occur. If they result from homogeneous nucleation, they take a spherical shape.
• Partially coherent precipitates possess some coherent or semi-coherent interfaces,
the other ones being incoherent. Most often, only one family of planes in the β
crystal may have the same atomic configuration than another family in the α
crystal; only one interface orientation may then display a relative low energy. This
planar interface gives to the precipitate a faceting shape. The prediction of the
facet number and orientations will be discussed later on.
• Incoherent precipitates are surrounded only by incoherent interfaces. The interfa-
cial energies of all the interfaces being elevated and quasi-identical, the precipitate
ought to have a spherical shape in the matrix and an “allotriomorph” shape at grain
boundaries. However, energy differences independent of the coherency degree
appear, for example interfacial energy decreases in presence of solute segregation;
then the incoherent precipitates often take a polyhedral shape.
7.3 Generalized Wulff construction for Nucleus Equilibrium at

Grain Boundary: Various Shapes of Nuclei
7.3.1 Equilibrium Condition for a Nucleus at a Grain Boundary
Starting with the nucleus allotriomorph shape (Fig. 7.2), the interfacial free energy or
surface tension (see Sect. 4.2) equilibrium at the triple point A requires the following
condition:
7.3 Generalized Wulff construction for Nucleus Equilibrium at Grain Boundary 225
γα/α
cos ϕ = (7.10)
2γα/β
The tendency for a β nucleus to lie on the support constituted by a α/α grain boundary
is stronger as the ϕ angle tends to 0, e.g. cos ϕ, also named wetting coefficient m, tends
to 1. The grain boundary efficiency to be a heterogeneous nucleation site depends
on its relative energy compared to those of the interfaces that may appear between
the second phase and the parent phase.
In order to fulfill the equilibrium (7.10) at a high-energy grain boundary, nucle-
ation of an incoherent allotriomorph is favoured. This shape is adopted whatever the
grain boundary energy in case of amorphous precipitates; the relative dimensions of
the amorphous nuclei depend on the wetting angle, their measurement allows to go
back to the gain boundary energy, according to:
a−b γα / α
= (7.11)
a+b γα / β
a and b are the half axes of the ellipse corresponding to the projected image of the
precipitate (Fig. 7.8).
This relation has been used for determining the relative energies of different grain
boundaries in copper on which amorphous SiO2 particles precipitate (Fig. 7.8) [3].
The energies obtained using this method are in good agreement with those deduced
from other experiments.
According to the equilibrium Eq. (7.10), coherent or partially coherent nuclei must
preferentially appear at low-energy grain boundaries. The coherent or semi-coherent
interfaces being planar, the precipitate is faceted. Several shapes are possible and
observed according to the grain boundary plane. In the most favourable case when
the boundary plane is a dense habit plane of the β phase in the two α crystals, a
plate-like nucleus appears (Fig. 7.9). This situation occurs for an f.c.c. nucleus at
Fig. 7.8 Observation in transmission electron microscopy of SiO2 precipitates at two grain
boundaries in copper: a = 11 {311} and b = 9 {122} (In the centre, the precipitate shape) [3]
Fig. 7.9 β plate-like precipitate formed at a symmetrical α/α grain boundary when the boundary
plane is a habit plane of the β phase in the α matrix
a symmetrical {110} grain boundary of a b.c.c. parent phase and, reciprocally, for
a b.c.c. nucleus at a symmetrical {111} boundary of an f.c.c. parent phase; in both
cases, the atomic distances in the two phases must be very close. If the grain boundary
is asymmetrical with a {110}CC or {111}CFC plane in only one crystal, its energy is
higher and a thin layer of β phase with a coherent interface lies parallel to the dense
plane on only one side of the grain boundary.
Apart from these two extreme situations, intergranular precipitate morphologies
may greatly vary according to the relative energies of the different α/β interfaces
compared to the α/α grain boundary energy; they also depend on the low-energy
facet orientation with respect to the grain boundary plane. These shapes are derived
graphically through a generalization of the Wulff construction.
7.3.2 Principle of the Generalized Wulff Construction
In the same way that we may derive the equilibrium shape of a particle in a homoge-
neous matrix by using the Wulff construction (See Sect. 4.3.2), we may predict the
equilibrium shape of a nucleus at a grain boundary by minimizing the total inter-
facial energy with respect to a pertinent parameter, such as the contact angle. The
calculations, performed at two or three dimensions, rest on the same hypothesis: the
surface and/or the volume strain energy resulting from the misfit between the matrix
and the nucleus may be neglected. This assumption is verified when the nucleus takes
an ellipsoidal shape and when the elastic constants of the two phases are identical;
in that case, the strain energy is independent of the morphology. Even in case of
anisotropy, provided that the misfit is small, the previous hypothesis still holds good.
We do not detail here the mathematical developments that yield the nucleus shape.
We only attempt to show how, using the generalized Wulff construction, it is possible
to graphically predict the shapes of various particles at grain boundaries.
This generalized Wulff construction has been proposed almost at around about
the same time by two independent author teams [4, 5]. The first formulation by Cahn
and Hoffmann introduces the capillarity vector ξ such as ξ n = γ (n is the unit
vector on the normal to the plane of energy γ ) with ξ·dn = dγ and n·dξ = 0. The
component of ξ normal to a surface is equal to its interfacial tension (or energy). This
approach based on the capillarity concept, although very useful, is not necessary to
understand the principle of the generalized Wulff construction, which we introduce
now following the Lee and Aaronson approach [5, 6].
The principle consists in plotting the Wulff constructions to predict the β nucleus
shapes in both α crystals from one side and from the other of the α/α grain boundary.
The γ -plots of centres OI et OII (for crystals I et and II, respectively), are situated
at a distance OI OII proportional to the αI /αII grain boundary energy in a direction
perpendicular to the grain boundary plane. For simplification, we denote the grain
boundary by α/α. The region where the two constructions superimpose gives the
equilibrium shape of the nucleus. The so-called “auxiliary γ -plot” of centre OII is
added to the normal Wulff construction in order to find the shape of the nucleus in a
crystal I coupled to a crystal II via a grain boundary. But, we may also consider that
the curve of centre OI is the auxiliary γ -plot to derive the nucleus shape in crystal II.
It is the reason why, in the following, we simply refer to curves of centres OI and OII .
7.3.3 Equilibrium Shape of Two-Dimensional Nuclei at Grain

Boundaries
7.3.3.1 Nucleus Without Facets
This is the simplest case illustrated in Fig. 7.10. The nucleus interfaces in both crystals
are incoherent. The Wulff construction in each crystal is thus reduced to a circle
whose the radius is proportional to the incoherent γα/β energy (to avoid complexity
in the notations, we write γα/β for γ α/β inc. ). The overlap of the two Wulff circles
(normal and auxiliary) well defines an allotriomorph. The existence of such a nucleus
requires the intersection of the two circles e.g. γα/α < 2γα/β , then cos ϕ < 1 and
the equilibrium condition (7.10) may be fulfilled. Whenever γα/α ≥ 2γα/β , complete
wetting is achieved.
The grain boundary plane is located midway between OI and OII as the two
spherical caps are symmetrical to each other with respect to that plane. If that is not
the case, when a faceted nucleus may form, a major problem of the generalized Wulff
construction is to locate the horizontal line corresponding to the grain boundary plane
position.
7.3.3.2 Nucleus with One Facet in One Crystal
We consider that the facet may only appear in the nucleus part located in crystal I of the
parent phase, because only there, the orientation relationship between the precipitate
and the matrix may be realized. The shape in crystal II is a semi-allotriomorph without
any preferential orientation with this crystal. In that case, the nucleus is generally a
single crystal.
Let us denote φ the angle between the facet and the grain boundary plane (not to be
confused with ϕ, the angle between the spherical cap and the grain boundary plane).
Fig. 7.10 Wulff construction for an allotriomorph at a planar grain boundary. The shape of the
nucleus is delimited by a hachured line (see text)
Two cases are distinguished according to the energy value γ C α/β of the coherent or
semi-coherent interface, upper or lower than 1/2 γα/α .
• When γ C α/β > 1/2γα/α
A range of φ orientations exist such as the α/β interface does not intersect the grain
boundary plane (the intersection lies inside the γ -plot of centre OII ). Beyond a critical
angle φC , the facet cuts the grain boundary plane. Note that when the γ C α/β value
strongly increases, higher than 1/2γα/α , another critical angle φext appears above
which the facet is totally located outside the γ -plot OII ; then, the nucleus takes the
shape of a spherical cap in each crystal.
When φ < φ C , the construction is very simple as the facet does not touch the grain
boundary (Fig. 7.11). All the contact angles of the interfaces with the grain boundary
plane are equal to ϕ and the equilibrium conditions at the triple points A and C are
independent of φ. The grain boundary plane is still located midway between the
Wulff circle centres. One only has to trace the vector corresponding to γ C α/β with
its orientation and its magnitude in order to locate a sharp cusp in the γ -plot OI . A
second sharp cusp, diametrically opposed to the first one, exists in the γ-plot OI , but
it does not account for the nucleus geometry.
When φC ≤ φ ≤ φext , the facet, easily reported in the Wulff construction (OH =
γ C α/β and the angle OII OI H = φ), cuts the grain boundary plane. Then, the problem
Fig. 7.11 Wulff construction for a nucleus with one facet when φ < φC and γ C α/β > 1/2γα/α ; φ
is the angle between BC and AA . The OH segment length is proportional to the energy of the BC
facet
is to locate this plane in the Wulff construction, knowing that (Fig. 7.12):
γα/α = OI OII = OI D + DOII = γα/β cos ϕI + γα/β cos ϕII (7.12)
An additional relationship, established by Lee and Aaronson, indicates that, for

continuity reasons, the distance AC between the junctions nucleus/grain boundary
in crystal I must be equal to the distance A C of the same junctions in crystal II [5].
Such a condition requires that CC = C P (= AA by construction). The only manner to
determine the location of the C point is to build a curve equidistant from the BQ facet
and from the BPQ circle arc (dotted line on the Fig. 7.12). The C point corresponds
to the intersection of this curve with the circle of centre OII . The equilibrium shape
of the nucleus is given by ABC in crystal I and by A TC in crystal II.
• When γ C α/β ≤ 1/2γα/α
The previous analysis may apply for any value of φ > 0. Even in the extreme case
where φ tends to 0, the facet being superposed to the grain boundary plane, one may
prove that ϕI = γ C α/β /γα/β ; this value being retained, Eq. (7.12) may be rewritten
as:
γα/α − γ C α/β = γα/β cos ϕII (7.13)
Fig. 7.12 Wulff construction

for a nucleus with one facet
when φ > φC and γ c α/β >
1/2γα/α (see text)
This relation is equivalent to the equilibrium condition of a liquid nucleus on a

substrate: γα/s −γβ/s = γα/β cos ϕ; in that case, the grain boundary plane is equivalent
to the substrate.
7.3.3.3 Nucleus with Two Facets in One Crystal
When φ increases, the opposite facet B Q in the Wulff construction also cuts the
grain boundary provided the ratio γα/βc /γ
α/β is sufficiently small. The equilibrium
shape is given on Fig. 7.13. The dihedral angle ϕII for the nucleus part in crystal II is
fixed by the equilibrium conditions, it is then possible to locate the grain boundary
plane AC. As for the BQ and B Q facets, they have the same length γ c α/β and are
equally distant from the centre OI ; they make an angle φ with the horizontal line
passing through OI (parallel to the grain boundary plane).
7.3.3.4 Nucleus with Facets in the Two Crystals
The shape of the nucleus in each crystal is deduced as previously. In addition, the
orientation relationship between the two phases must be respected from one side
and from the other of the initial grain boundary; this condition generally yields
Fig. 7.13 Wulff construction for an intergranular β nucleus displaying two facets in the same α
crystal (see text)
Fig. 7.14 a Lead inclusions at twin boundaries in silicon, each morphology corresponds to the
intersection by the grain boundary plane of octahedral particles. b Some shapes adopted by the
inclusions by assuming that the octahedra have the same size in both crystals. The schematic drawing
shows that each bicrystal inclusion preserves the twin relationship between its two parts [7]
the formation of bicrystalline nuclei. The study of lead inclusions at twin grain
boundaries of silicon well illustrates the formation of bicrystalline particles having
a morphology composed of truncated octahedra (Fig. 7.14) [7].
7.3.4 Influence on the Grain Boundary Plane on the Nucleus

Shape
The orientation of the grain boundary plane is implicitly considered in the Wulff
construction, it fixes the φ angle between this plane and the usual habit plane between
the two phases. In the extreme case when φ is null in the two crystals, the nucleus
takes the shape of a thin layer on each side of the grain boundary (Fig. 7.9). In the case
when φ = 0 in one crystal only, equilibrium respects Eq. (7.13) and the precipitate
lies on only one side the grain boundary. The nucleus shape in the other crystal
depends on the grain boundary plane orientation on that side with respect to the habit
plane. According to the φ value, the nucleus takes a semi-allotriomorph shape with
or without facets.
The following example concerns an austenite precipitate at a ferrite grain boundary
(Fig. 7.15) [8]. The growth of the precipitate yields an evolution of the initial nucleus
shape; however, the section observed by transmission electron microscopy illustrates
pretty well the morphology predicted by the Wulff construction. The grain boundary
plane is oriented about 5◦ from a dense {110} plane of the ferrite in crystal C, it
is close to {210} in crystal A. The precipitate develops in crystal A where its EF
facet does not cut the boundary plane. The φ angle value being 16–19◦ , the facet
plane is such as {110}ferrite // {211}austenite , in agreement with the Kurdjumov-Sachs
relationship.
Fig. 7.15 Austenite precipi-

tate at a ferrite grain bound-
ary. The semiallotriomorph
shape with a facet inclined
from about 18/19◦ on the
grain boundary plane may be
explained by the position of
the boundary plane, close to
{110} in crystal C and not
far from {210} in crystal A
(see text) [8]
7.3.5 Equilibrium Shape of Three-Dimensional Nuclei at Grain

Boundaries
The calculations and graphical constructions used to determine the two-dimensional

nuclei are at the basis of the research to predict faceted three-dimensional nuclei.
The grain boundary may remain planar or must pucker such that the junction angle
with the nucleus is closer its equilibrium value. Figure 7.16 gives an example of
construction for a three-dimensional nucleus: the nucleus shape in crystal I is a
spherical cap with a facet determined by using the Wulff construction, the nucleus
shape in crystal II, without facet, is obtained by calculation [6].
Calculation of the activation free energy G ∗ for nucleation shows that the
mechanism of planar grain boundary is favoured when the φ angle between the facet
and the boundary is higher than 18◦ + φC ; instead, the pucker mechanism occurs for
the lower φ values. The G ∗ energy rapidly increases with φ, in particular when the
relative energy of the facet is low. Thus, the formation of a facet on a general grain
boundary must preferentially occurs on a plane parallel to one among the equivalent
habit planes, only.
Fig. 7.16 Equilibrium shapes

for the two parts of a three-
dimensional nucleus (γα/α =
1.07 γα/β et γ C α/β = 0.3γα/β )
with one facet in contact with
the grain boundary plane in
crystal I (φ = 15◦ ): a nucleus
shape crystal I deduced from
the Wulff construction; b
calculated shape in crystal II;
the intersection of the facet
with the grain boundary is
indicated by an arrow [6]
7.3.6 Grain Boundary Puckering Phenomenon
The Young equilibrium equation applied at triple junction does not take into account
the components of the interfacial tensions in the direction perpendicular to the grain
boundary plane. This shortcoming has no consequences on the allotriomorph or on
the nucleus with one facet that does not intersect the grain boundary because these
components annihilate. On the contrary, when a nucleus displays one or two facets
that cut the grain boundary plane, the equilibrium at the junctions between the inter-
faces and the grain boundary requires re-examining the previous Wulff constructions.
In the case of one facet (Fig. 7.12), balanced components of the interfacial tension
exist only in the direction parallel to the grain boundary plane at the point A (A )
ϕI = ϕII , whereas unbalanced components, either parallel or perpendicular to the
boundary plane occur to the point C (C ). To achieve complete force balance, the
grain boundary must pucker such that one side moves up or down relative to the other
side. Figure 7.17 shows an example of such a puckering: the grain boundary side near
the C junction moves down with respect to the side near the A junction. An addi-
tional condition must be fulfilled in a order such that the grain boundary preserves
a average null curvature, e.g. the chemical potential remains constant all along the
grain boundary; the puckered grain boundary must preserve its initial orientation at
the junctions with the nucleus.
Fig. 7.17 Wulff Construction analogous to that of Fig. 7.12, modified to account for the grain
boundary puckering phenomenon (see text) [6]
But the reality is more complex; for a three-dimensional grain boundary, the
triple junctions are lines and thus it is a plane that must preserve its curvature.
Generally, no exact solution is found to fulfil the condition of an average null
curvature. The best approximation is to describe the puckered grain boundary by
a catenoid, a section of which being represented on Fig. 7.18a [6]. The primary
position of the grain boundary is given by dotted line, whereas a 90◦ section of
that plane is drawn in full line. This puckering phenomenon of a grain bound-
ary at its intersection with a precipitate may be observed by transmission elec-
tron microscopy when the grain boundary plane is parallel to the electron beam
(Fig. 7.18b) [8].
Fig. 7.18 a Section perpendicular to the grain boundary showing its puckering at the junctions with
a nucleus and the catenoid shape it locally adopts: φ = 10◦ , γα/α = 1.07 γα/β and γ C α/β = 0.3γα/β ;
b Austenite precipitates (denoted 2 and 3) at a ferrite grain boundary. Straight above precipitate 3,
the grain boundary seems to present (point A) the puckering phenomenon [8]
7.4 Grain Boundary Precipitate Growth
A precipitate, when observed by electron microscopy, rarely presents the primary

shape of its nucleus. On the one hand, growth yields coherency loss of some inter-
faces and, on the other hand, interactions with obstacles (dislocations, solutes, other
precipitates) modify the particle shape. Only high-resolution electron microscopy
allows us to visualize nucleus shape. However, intergranular precipitates, observed
by conventional electron microscopy after limited growth, display morphologies
which may be explained on the basis of the generalized Wulff construction. This is
due to the fact that the growth mechanisms strongly depend on the types of interfaces
that surround the precipitate. Without entering in the detailed growth mechanisms
and in the various approaches of the growth kinetics, we will shortly explain why
an intergranular precipitate may or not inherit its shapes from that of the primary
nucleus after growth.
7.4.1 Migration of a Curved Incoherent Interface
Two processes must occur in order that an incoherent interface migrates: atom
transport across the interface and solute redistribution from the matrix towards the
precipitate or vice versa. The first process controls the very first stages of growth
whereas the precipitate extension is governed by bulk diffusion. The growth rate is
proportional to (D/t)1/2 with D the diffusion coefficient of the solute in the parent
phase and t the growth time. The (D/t)1/2 law is no longer respected when the dif-
fusion fields of several precipitates overlap. The growth rates take different orders
of magnitude according to the initial oversaturation state of the solid solution and to
the temperature. When an allotriomorph is located at a grain boundary, a complex
mechanism named collector plate mechanism operates in addition to the lattice dif-
fusion towards the nucleus [9]. This mechanism takes place in three stages: solute
segregation towards the grain boundary that reduces its free energy according to the
Gibbs equation, then rapid solute diffusion along the grain boundary (DGB DV )
especially at moderate temperature and for a substitutional solute), finally solute
diffusion (coefficient Di ) along the interface between the nucleus and the matrix. As
a result, the precipitate growth longer than thicker (Fig. 7.19) and the angles at triple
junctions no longer represent the nucleus equilibrium.
Fig. 7.19 Growth of an incoherent precipitate at a grain boundary by the “collector plate” mecha-
nism (see text)
7.4.2 Migration of a Planar Interface (Coherent or Semicoherent)
Coherent interfaces being dense crystalline planes, they may not accept additional
atoms and the growth mechanism by diffusion across the interface is unlikely. Solute
atoms may be only adsorbed on steps that are generally present along an average
planar interface. Thus, growth occurs by lateral displacement of the steps (Fig. 7.20).
The step displacement rate is proportional to the volume diffusion coefficient and
to the solute oversaturation; it is inversely proportional to the step height h and
may be relatively high in case on nanometric steps. However, a rapid lateral step
displacement only induces a low movement of the whole interface perpendicular to
itself. The global rate does not depend on the step height, but is inversely proportional
to their length λ. An interface without any step is totally immobile; this is the case for
7.4 Grain Boundary Precipitate Growth 237
Fig. 7.20 The migration of a coherent or semi-coherent planar interface occurs by displacement of
nanometric steps. The average grain boundary plane (dotted line) moves slowly in the v direction
a fully coherent interface. Moreover, the number of steps has a tendency to decrease
as the growth progresses; then, the step motion will stop if there is no continuous
step formation.
This mechanism applied to a partially coherent precipitate, displaying a thin disc
or a thin plate-like shape with facets parallel to the grain boundary plane, explains
that the precipitate maintain for a long time the shape of its nucleus. However, the
ratio between the area occupied by the coherent (or semi-coherent) interfaces and
that occupied by the small incoherent interfaces increases, excluding any estimation
of the relative interfacial energies.
We may conclude that the shape of an intergranular precipitate often accounts
for the characteristics of the grain boundary on which it appears, at least from a
qualitative point-of-view; in particular, it may inform us about the orientation of
the grain boundary plane. Moreover, a planar interface most likely preserves its
initial orientation. But, apart from particular cases, the greatest care must be taken in
deducing the relative energies of interfaces and grain boundaries from the precipitate
dimensions.
7.5 Localization of Grain Boundary Precipitates on Extrinsic

Dislocations
Extrinsic dislocations, associated or not with steps, and pure steps are preferential
nucleation sites. Nucleation on an extrinsic dislocation is favoured compared to that
on a lattice dislocation as a stronger solute concentration exists in the grain boundary
and the solute diffusion is more rapid. It is then the consequence of a preferential
segregation on an extrinsic dislocation line. In that case, nucleation results from a
double heterogeneous effect; with respect to the crystal and with respect to the grain
boundary. We deal with the nucleation on an extrinsic dislocation by considering the
same processes that those occurring on a lattice dislocation. The nucleation kinetics
depends on the energies of the interfaces between the two phases and on the elastic
constants of the material. Two nucleation kinetics exist that may be distinguished
by calculating the total free enthalpy variation that accompanies the formation of a

total free enthalpy of a cylin-
drical embryo surrounding
a dislocation line with the
embryo radius R. Curve
a is characteristic of a nucle-
ation and growth process with
R0 the radius of the Cottrell
atmosphere and RC the criti-
cal radius for a stable nucleus;
Curve b corresponds to the
instantaneous nucleation on
the dislocation line
cylindrical embryo of radius R surrounding the dislocation line. This variation (per
unit dislocation length) is:
G T = πR 2 G V + 2πRγ − A.lnR (7.14)
The term A.lnR corresponds to the elastic energy of the dislocation that must be
subtracted from the total energy as the defect is released. A is an elastic constant
depending on the material. In order that the nucleation occurs, the total enthalpy
must decrease when R increases. Figure 7.21 shows the two types of possible G T
evolution according to the relative values of G V , A and γ . If the energy decreases
continuously, nucleation on the dislocation line occurs instantaneously. If a local
minimum followed by a maximum exists, the first augmentation of the radius up
to R0 corresponds to the formation of a Cottrell atmosphere with stabilization of
the dislocation. In order that R increases again beyond R0 , necessary condition to
form a second-phase nucleus, an energy barrier equal to GRC − GR0 must be
overcome. The nucleus thus appears when a critical radius RC is reached, yielding a
destabilisation of the dislocation. The condition for which nucleation occurs instan-
taneously or via an energy barrier may be easily obtained by differentiating equation
(7.14) with respect to R and by cancelling its first member; we have then to find the
solutions of the following quadratic equation:
2πR 2 G V + 2πRγ − A = 0 (7.15)
If |2A G V | < πγ 2 , there are two roots to Eq. (7.15): the values of R0 and RC . The
nucleation requiring activation seems to be favoured for elevated interfacial energies,
e.g. for incoherent nuclei.
If, on the contrary, |2A G V | > πγ 2 , Eq. (7.15) has no roots, the energy decreases
continuously when R increases. For certain values of the two material characteristics
A and GV , instantaneous nucleation occurs for low γ values e.g. it preferentially
implies coherent or semi-coherent precipitates.
7.5 Localization of Grain Boundary Precipitates on Extrinsic Dislocations 239
Numerous observations by transmission electron microscopy show precipitates

that have been formed on extrinsic dislocation lines. Several grain boundaries in a
Fe-9% Cr-0.01% C display Cr23 C6 microcarbides elongated parallel to the extrinsic
dislocations lines, located at the intersection of the slip plane of one crystal with
the grain boundary plane (Fig. 7.22) [10]. In the same way, the association of nio-
bium carbides with extrinsic dislocations has been revealed in stainless steel grain
boundaries (Fig. 7.23) [11].
Copper micro-precipitates on intergranular dislocations have also been observed
in = 25 grain boundaries in silicon leading to an electric activity [12].
To sum up, intergranular precipitates have an appreciable influence on the
grain boundary properties; in particular, they impede grain boundary migration and
constitute preferential sites for corrosion. Localized on extrinsic dislocations, they
Fig. 7.22 Dark field electronic images showing Cr23 C6 microprecipitates aligned parallel to the
extrinsic dislocations in grain boundaries of a Fe-9% Cr-0.01% alloy: a vicinal grain boundary
deviated from 2◦ of = 5; b general grain boundary [10]
Fig. 7.23 a NbC intergranular precipitates observed on a replica of a stainless steel sample. All
the particles are parallel and elongated along a preferential direction; b dark field electronic image
showing NbC particles associated to extrinsic dislocations displaying a weak contrast (along AB,
CD and EF for example) [11]
pin them; the resulting stabilization has important consequences on all the processes
requiring dislocation motion either within the grain boundary or from the boundary
to the crystals. These phenomena are the subjects of the two last chapters of part II
of this book.
References
1. P. Gruffel, P. Carry, J. Euro, Ceram. Soc. 11, 189 (1993)

2. A.P. Sutton, R.W. Balluffi, Interfaces in Crystalline Material (Oxford Science Publications,
New York, 1995)
3. A. Otsuki, Ph.D. thesis, Kyoto University (1990)
4. J.W. Cahn, D.W. Hoffman, Acta Metall. 22, 1205 (1974)
5. J.K. Lee, H.I. Aaronson, Acta Metall. 23, 799 (1975)
6. J.K. Lee, H.I. Aaronson, Acta Metall. 23, 809 (1975)
7. S. Hagège, Interface Sci. 7, 85 (1999)
8. N. Benfetima, L. Priester, Revue Phys. Appl. 21, 109 (1986)
9. H.I. Aaronson, J.K. Lee, in Lectures on the Theory of Phase Transformations, ed. by H.I.
Aaronson (TMS, New York, 1975), p. 103
10. S. Lartigue, L. Priester, Acta Metall. 31, 1809 (1983)
11. A.R. Jones, P.R. Howell, B. Ralph, J. Mater. Sci. 11, 1593 (1976)
12. A. Broniatowski, C. Haut, Phil. Mag. Lett. 62, 407 (1990)
Chapter 8
Interactions Between Dislocations and Grain
Boundaries
Two types of stresses act on a lattice dislocation at vicinity of a grain boundary

(Fig. 8.1)
Fig. 8.1 The forces acting on a dislocation at vicinity of a grain boundary: a the distance between
the two defects, linear and planar, is superior to the grain boundary periodicity; b the dislocation is
extremely close the grain boundary (see text)
• The long-range elastic stresses: applied stress and internal stresses caused by other
dislocations present within the grains (Fd ) and/or in the boundary. In case of an
elastically anisotropic material, (or in case of an interface between two different
media), an additional stress must be considered: the interface causes the self-stress
exerted by a straight dislocation on itself to be non-zero. The effect of this stress
is named image force Fi .
• The short-range stresses: friction stress (Ff ) and stress linked to the grain boundary
core (Fc ), the effect of which is only felt by a dislocation located very near the
boundary, at a distance less than the grain boundary period.
Among these stresses, three characterize the situation of a dislocation at vicinity
of an interface. First, the image force may help or impede the motion of the dislo-
cation towards the grain boundary, all the stronger as the distance between the two
defects decreases. The contribution of the image force to the interaction between
dislocations and grains boundaries justifies that we devote a section to this question,
rarely approached on other books dealing with interfaces. Then, stress concentration
close or in the grain boundary yields formation of dislocation configurations in the

242 8 Interactions Between Dislocations and Grain Boundaries
neighbouring crystalline regions that are specific of the deformation of a bicrystal or

a polycrystal at low or moderate temperature. This action of the grain boundary on
its close environment is briefly presented. Finally, the interactions between the cores
of the two defects, linear and planar, are detailed; they give to the grain boundary a
fundamental role in the plastic deformation at high temperature.
8.1 Long-Range Elastic Interaction: Image Force
A dislocation near a free surface is always attracted to the surface by a force whose
theoretical approach requires to introduce (as in electrostatic) the notion of “image
dislocation” symmetrical from the real dislocation, but located outside the solid. The
image force is then the interaction force between the two dislocations, real and image.
Similarly, a dislocation near an interface in an anisotropic medium is submitted to a
force, attractive or repulsive according to the elastic constants of the two half-media
from each side of the interface and to the defect geometry. In that case, reference to an
“image dislocation” has no sense, the lattice dislocation is considered as interacting
with a continuous distribution of infinitesimal dislocations in the interface. For grain
boundaries, the difference between the elastic constant of the crystals comes from
their misorientation and only acts in case of an elastically anisotropic material.
From a theoretical point of view, the image force on a dislocation has been solved
whatever the dislocation orientation with respect to the grain boundary plane [1, 2].
But, numerical results have only been obtained in the case of a dislocation parallel to
the grain boundary plane (Fig. 8.2); they rest on the image force theorem of Barnett
and Lothe [1]:
E I/II − E I E
Fi = − =− (8.1)
d d
d is the distance between the dislocation and the grain boundary, E I is the pre-
logarithmic energy factor for the dislocation energy in an infinite crystal I and E I/II
is the pre-logarithmic factor for the dislocation at the I/II interface. For a given
Fig. 8.2 Geometrical config-

uration used to calculate the
image force for a dislocation
near a grain boundary
8.1 Long-Range Elastic Interaction: Image Force 243
misorientation between crystals I and II, the factor E I/II depends only on the direction
t of the dislocation in the interface (and not on the orientation of the interface plane);
therefore the image force is the same for any interface belonging to the same zone t.
To calculate the E I/II factor, it is necessary to obtain the pre-logarithmic factor for
the dislocation in an infinite crystal II. The direction and the Burgers vector of the
dislocation in crystal I are t and b, respectively. The rotation between crystal I and
crystal II being described by the matrix R, t becomes Rt and b becomes Rb referred
to the cubic axes of grain II.
E I , E II and E I/II are quadratic forms in the Burgers vector b as follows:
E I , = b BI b E II = b BII b E I/II = b BI/II b (8.2)
where B is the southwest block of a certain 6 × 6 matrix N which plays a central

role in the anisotropic elastic theory of line defects. Various methods of computing
it have been discussed by Condat and Kirchner [3]. The matrix NI/II is equal to 2
[NI + NII ]−1 . For a fixed Burgers vector, the problem has five degrees of freedom:
two for the direction t of the dislocation, two for the direction r around which the
grains are turned relative to each other, and one for θ , the rotation angle. The axis r is
the eigenvector of the rotation matrix obtained when (R − I)r = 0 and the rotation
angle is given by Eq. (1.6): θ = cos−1 [Tr R − 1]/2.
In order to get a very approximate quantitative estimate of E it suffices to
recall that, with μ being the shear modulus of an isotropic material, of the order of
magnitude of elastic constants, E I = μb2 /4π. The difference between E I and E I/II
will be a function of elastic anisotropy, but if one takes E I/II = 0, 9E I , it is found
that Fi ∼= 10−2 μb/d. Even at a distance of d = 100b, this becomes Fi ∼ = 10−4 μb,
which is comparable to the force caused by the flow stress of f.c.c. metals at any
temperature, or b.c.c. metals at high temperature.
The E I/II factor may have a smaller or higher value than E I and E II or a value
between them. Care must be taken to ensure that this relative order is correctly
interpreted. To overcome a boundary, a dislocation is necessarily submitted to crys-
tallographic effects in the intergranular region and in its vicinity. The use of the image
force must be restricted to the possibility for a lattice dislocation to move towards or
away from a grain boundary. Any conclusion about crossing is not straightforward.
Thus, we only consider the difference between the energy of the dislocation in the
initial crystal and that in the grain boundary. The dislocation is attracted to the grain
boundary is E I > E I/II ; it is repelled from the grain boundary if E I < E I/II and there
is no image effect if E I = E I/II .
An exhaustive study of several geometric configurations implying dislocations
with different Burgers vectors and different line directions and for different
misorientations (axes and angles) of the grain boundaries has been performed for
cubic materials [4–6]. The main results are as follows:
• The image force is null when the dislocation line t is parallel to a high symmetry
rotation axis r: fourfold 100 or threefold 111. This is true whatever the Burgers
vector, but no longer verified when t is parallel to a twofold 110 symmetry axis.

interaction energy E
between a perfect disloca-
tion b = a/2 [111] and a grain
boundary in b.c.c. iron in
function of the misorienta-
tion angle θ: r = [100], t =
[011] (cur ve 1), r = [110],
t = [110] (cur ve 2), r =
[111], t = [010] (cur ve 3) [5]
• In f.c.c. materials, FI = 0 for a screw dislocation when b and t are parallel to a

twofold 110 rotation axis [6].
• Coincident grain boundaries do not display special behaviours of the image force
that continuously varies with the misorientation angle θ around a given axis and for
a fixed dislocation (Fig. 8.3). Each curve giving the evolution of E with θ presents
a symmetry in agreement with the corresponding misorientation axis: fourfold for
100, threefold for 111 and twofold for 110, reducing the investigation fields
to θ ≤ 45, θ ≤ 60 ◦ and θ ≤ 90 ◦ , respectively [5].
• The influence of the bicrystallography on the image force appears for symmetrical
tilt grain boundaries in f.c.c. materials: the image force is null for all the dislocations
parallel to the twin plane [6] .
• For a fixed grain boundary θ [uvw] and a fixed Burgers vector b, the E evolutions
with the angle ϕ between b and t are similar for different metals with the same
f.c.c. or b.c.c. structure. The maximal and minimal E values always correspond
to the same dislocation characters Fig. (8.4) [5].
• For cubic materials, the magnitude of the image force for a fixed “dislocation/grain
boundary” configuration increases with the anisotropy factor H = C44 − (C11 −
C12 ) and not with the anisotropy ratio A = C44 /(C11 −C12 ). This force is repulsive
for most configurations when H is positive (all the f.c.c. metals and numerous
b.c.c. metals) and attractive when H is negative (some b.c.c. metals as chromium,
molybdenum…) (Fig. 8.4 and Table. 8.1).
• When H > 0, (in most cases), it is important to mention that screw dislocations are
repelled from almost all the grain boundaries. Insofar as any other force acting on
the dislocations may overcome this repulsion, screw dislocations are confined to
the interior of each grain. The image force appears as an obstacle to the polycrystal
deformation.
8.1 Long-Range Elastic Interaction: Image Force 245
Fig. 8.4 Evolution, for b.c.c. ΔE (pJ/m)

metals, of the interaction
energy E between a θ 45 ◦ 50
[100] grain boundary and a
b = a/2 [111] dislocation
in function of the angle ϕ 30
between the fixed Burgers
vector and the varying dislo- 10
cation line. The E values are
ordered by increasing values
of the anisotropy factor H (see -10
Table 8.1) [5]
-30
-50
211 100 111 011 111 211
Table 8.1 Values of the

Metal A H (1010 Pa)
anisotropy ratio A and of the
anisotropy factor H for Cr 0.69 − 10
different b.c.c. metals [5] Nb 0.51 − 5.4
Mo 0.91 − 2.4
W 1 0
K 6.35 0.44
Na 8.15 1.03
Ta 1.56 5.9
Fe 2.36 12.9
• Finally, in most cases, the magnitude of the image force is comparable to those of
the other forces acting on the dislocation [6]; this force must be taken into account
when analysing the interaction between a dislocation and a grain boundary in
anisotropic materials (silver, copper, nickel, chromium, silicon…).
More punctually, calculations have shown that the image force may help or impede
the movement of a dislocation towards the grain boundary [7, 8]. As an example,
in a Fe-4%Si alloy, the image force magnitudes for perfect dislocations located at
about 10 nm of = 3, = 9 and = 15 grain boundaries vary from −4 to 100 ×
10−4 N·m−1 , in agreement with the estimation. They are of the same order than the
forces between dislocations. These effects are enhanced for interfaces between two
different phases and, more particularly, for thin layers and multilayers.
8.2 Dislocation Configurations in the Vicinity of a Grain

Boundary
At relatively low temperature (T < 0, 4 Tm ), the grain boundary most often plays
the role of an obstacle to the deformation. Dislocation pile-ups are created along the
boundaries of any material in the very first plasticity stages. These pile-ups rapidly
evolve to form various configurations, usually observed in the crystals at advanced
deformation stages. The studies, by X-ray topography and by transmission electron
microscopy, of silicon bicrystals deformed just above the elastic limit will show
different arrangements attributed to the presence of = 9 and = 25 110 tilt
grain boundaries [9, 10]. They take the form of more or less long dislocation pile-
ups (Fig. 8.5), of networks with a scale shape linked to the activation of a second
slip system and to lock formation (Fig. 8.6) and of cross slip traces whose repetition
may activate prismatic dislocation loops (Fig. 8.7). Finally, the resolved shear stress
may be reversed during the deformation, leading a dislocation moving towards the
grain boundary to go backwards. The curvature of a lattice dislocation at vicinity of
a grain boundary cannot inform us about its origin: crossing of a dislocation coming
from the other crystal, emission by an intergranular source or reflection by the grain
boundary.
All the observed configurations seem to appear not randomly, but try to preserve
the bicrystal symmetry and to minimize the long-range stresses. In any case, the
regions near the grain boundary are in an advanced deformation stage. This conclu-
sion may be extended to other grain boundaries and other materials. In principle,
the dense structure of a singular grain boundary may constitute a stronger obsta-
cle compared to a general grain boundary, however in the reality it seems that the
Fig. 8.5 Silicon = 9 bicrystal deformed until the upper elastic limit (ε = 8.10−3 , σcr =
33 MPa): a two long dislocation pile-ups in the primary slip planes of crystals I and II are stabilized
at points A and B (crystal I) by secondary dislocations with, simultaneously, shorter pile-ups in the
primary slip planes of each side of the grain boundary; b Schematic representation of the observed
configuration [9]
8.2 Dislocation Configurations in the Vicinity of a Grain Boundary 247
Fig. 8.6 Silicon = 9 bicrystal that underwent the same deformation than previously: a scale
shape networks have been formed at vicinity of the grain boundary in one crystal; Schematic
representation of the configuration showing that it is constituted by primary dislocations (full line),
by secondary dislocations (curved dashed lines) and by Lomer Cottrell locks (straight dashed
lines) [9]
Fig. 8.7 Prismatic disloca-

tions created by repeated cross
slips at vicinity of a = 25
grain boundary in silicon.
The primary slip plane (trace
parallel to the boundary plane)
and the secondary slip plane
are both almost perpendicular
to the image plane in transmis-
sion electron microscopy [10]
structural differences weakly influence the mechanical behaviour at low temperature.

The differences between the regions near the grain boundaries and the crystal centres
may vary according to the number of slip systems of the material and to the value of
the stacking fault energy, but in any case grain boundaries strongly contribute to the
polycrystal hardening.
8.3 Short-Range Interactions Between Linear

and Planar Defects
When lattice dislocations impinge at low temperature on grain boundaries, the latter
play the role of strong obstacles for slip transfer. When the applied stress and or
the temperature increases, some lattice dislocations may enter the grain boundaries
leading to the formation of extrinsic grain boundary dislocations (see Sect. 5.2). The
boundary is now in a non-equilibrium state. The entrance of a dislocation in a grain
boundary may generally occur provided a decrease of the total energy, implying the
elastic energy and the core energy of each defect. Furthermore, the lattice dislocation
incorporation requires the relaxation of the stresses associated to the previously
formed extrinsic dislocations. The relaxation mechanisms proposed until now are
thermally activated and differ according to the characteristics of the grain boundary
and of the incoming dislocation. In the following, although they are very similar, the
entrance mechanisms and the very first reactions of a dislocation in a grain boundary
will be presented separately from the stress accommodation mechanisms leading
the grain boundary to a new equilibrium (see Chap. 9). This is a formal distinction
insofar as it just corresponds to different relaxation stages, but it allows us to better
describe the total process, often incomplete in the reality.
The entrance of a dislocation in a grain boundary may be explained by differences
between the crystal and the grain boundary elastic constants and/or by changes in
its core configuration. Some authors consider that isolated dislocations with lat-
tice Burgers vector bl may exist only as “trapped” dislocations and may be easily
extracted from the grain boundary. This probably refers to the difficulty to distin-
guish, by transmission electron microscopy, a dislocation within from a dislocation
near a grain boundary. However, experiments have revealed dislocations that have
reacted with the intrinsic network, while preserving a bl lattice Burgers vector. Thus,
to avoid confusion, we consider here that any isolated dislocation (not part of a
periodic network) is “extrinsic” whatever its Burgers vector. Extrinsic dislocations
that preserve the lattice Burgers vector bl are analogous to “fresh” dislocations in
crystals.
Under the effect of a thermal and/or a mechanical stimulus, three types of
interactions between a dislocation and a grain boundary may occur yielding a
“dislocation/grain boundary” relaxation: Combination, decomposition and transmis-
sion (Fig. 8.8) [11].
First, each type of process is analyzed and supported by results. Then, more com-
plex entrance processes of a dissociated dislocation in a grain boundary are described,
implying a temporary energy increase. Finally, simulations of the interaction between
a grain boundary and a perfect or a dissociated dislocation are presented.
8.3 Short-Range Interactions Between Linear and Planar Defects 249
Fig. 8.8 The three types of interactions between a lattice dislocation (bl ) and the intrinsic grain
boundary structure to form an extrinsic grain boundary dislocation EGBD (be ): a before reaction;
b combination with an intrinsic dislocation with Burgers vector bi ; c decomposition into two
dislocations, glissile bg et sessile bc ; d transmission of the dislocation from crystal I to crystal II
with blI = blII + be
8.3.1 Combination Processes
Combination between a lattice dislocation and a grain boundary may easily occur
in a small angle grain boundary, the intrinsic dislocations Burgers vector being that
of the lattice dislocations. In f.c.c. crystals, the combination reaction is analogous to
that leading to the formation of a Lomer lock; it is very favoured from an energetic
point of view, the resulting vector being one of the crystal. It has been observed in a
2◦ [001] twist grain boundary in a deformed gold bicrystal (Fig. 8.9) [12].
Combination is rarely mentioned in high angle grain boundary. However, it may
occur without difficulty in a general grain boundary where the secondary disloca-
tion Burgers vectors are very small. In that case, the small elastic energy increase is
balanced by the diminution of the dislocation core energy in the boundary. Com-
bination reactions occurring in singular and vicinal grain boundaries under thermal
activation will be detailed later (see Chap. 9). Only one example is given here, it
describes a reaction between a b1 lattice dislocation and a bi intrinsic dislocation in
a near = 3 (11-1) tilt grain boundary in nickel [13]:
bl + bi = be (8.3)
a/2[01 − 1] a/3[−1 − 11] = a/6[−21 − 1]
This reaction is accompanied by a strong decrease of the elastic energy. The intrinsic
dislocation belongs to a sessile dislocation network that accounts for the deviation
between the real grain boundary and the exact =3 coincidence grain boundary. The
resulting extrinsic dislocation is glissile and may displace under internal stresses. In
Fig. 8.9 a Extrinsic grain boundary dislocation AB along 110 in a twist [001] grain boundary
(θ = 2◦ ) in a deformed gold bicrystal; b Schematized configuration of the AB dislocation [12]
small angle and in vicinal near =3 grain boundaries, when the lattice dislocations
are dissociated in Shockley partials, combination reactions may yield dislocation
annihilation.
8.3.2 Decomposition Processes
The stress accommodation occurs within the grain boundary by decomposition of the
lattice dislocation in two (or more) products, the Burgers vectors of which being those
of the DSC lattice. We prefer the term “decomposition” to dissociation for avoiding
confusion with the formation of partial dislocations in crystals. The application of
the Frank criterion is less restrictive insofar as the total “dislocation/grain boundary”
system energy must be considered. Figure 8.10 shows two examples of decomposi-
tion observed by transmission electron microscopy in coincident = 3 {112} and
= 29 {113} grain boundaries [14, 15]:
a/2[1 − 1 0] → a/6 [2 − 1 − 1] + a/6 [1 − 21] (8.4)
a/2[110] → 3 × a/58[370] + 2 × a/58 [1 0 4 0] (8.5)
Fig. 8.10 Decomposition of two lattice dislocations in a grain boundary: a two dislocations A and
B in a = 3 grain boundary in aluminium (at the top) decompose into two products (at the bottom)
after one month at room temperature [14]; b a lattice dislocation decomposes into five products
(starting from point X) in a = 29 grain boundary in stainless steel [15]
The decomposition depending on the intrinsic structure, the products are likely
smaller when the grain boundary displays a higher coincidence index associated
to smaller DSC elementary vectors. Due to the latter assumption, decomposition in
a great number of dislocations with very small Burgers vectors occurring in general
grain boundaries has been considered as a “spreading” process for several years.
This was supported by the widening contrast of extrinsic dislocations observed by
conventional electron microscopy. However, fine observations by high-resolution
microscopy reveal that the decomposition products are never negligible; they rather
result from a linear combination of the elemental DSC vectors [16]. The observations
have been made on a series of symmetrical 110 tilt grain boundaries in silicon, the
high degree of purity excluding segregation effects on the interactions. They allow us
to better analyse the entrance steps and the following intergranular reactions by con-
sidering the link between the structural units and the grain boundary dislocations.
The atomic structures of coincidence boundaries in silicon have been previously
described in terms of structural units (see Sect. 3.3.4); the reader may also refer to
the review paper [17]. The entrance, at moderate temperature (T ∼ = 0.6Tm ), of a
dissociated dislocation in a = 9 {122} grain boundary clearly shows the decom-
position process of the leading partial into two products: bg = a/18 411 glissile
dislocation and bc = a/9 122 sessile dislocation. The rapid motion of the glissile
component far from the impact point and the slight displacement by climb of the
sessile component, under the internal stress effect, allow the b30 = a/6 211 trailing
partial to integrate the grain boundary structure, in its turn (Fig. 8.11) [18].
Fig. 8.11 Decomposition of a dislocation in a 9 symmetrical tilt grain boundary in silicon: a

dissociated dislocation before the entrance in the grain boundary; b decomposition of the b90 leading
partial in a bc sessile dislocation and a bg glissile dislocation; c the second b30 partial dislocation
has also entered the grain boundary owing to small displacement of the bc sessile dislocation and
the large motion of the glissile dislocation (out from the image zone) [18]
It is likely that the same relaxation processes occur in case of non-dissociated

dislocations, when a dislocation impinges on a grain boundary that already contains
an extrinsic dislocation.
Decomposition in a = 51 grain boundary requires two steps and its products
are not elementary vectors of the DSC lattice (Fig. 8.12) [19].
Fig. 8.12 Decomposition in two steps of a perfect lattice dislocation in a = 51 tilt grain boundary
in silicon: a decomposition into two products: a D1 = 3bc + 2bg + bd sessile dislocation and a D2 =
2bg glissile one; b decomposition of D1 and motion by glide of D2 . Note that the Burgers vectors
of the dislocations resulting from successive decompositions are not elementary vectors of the
DSC [19]
The decomposition is controlled by the intergranular structure. Owing to the defi-

nition of an extrinsic dislocation in the “SU/GBD” model (see Sect. 5.2), its Burgers
vector must be compatible with the unit (or the units) added to or suppressed from
the equilibrium grain boundary structure. As a result, the addition (or suppression) of
a same unit in different grain boundaries is associated to an extrinsic dislocation, the
Burgers vector of which varying. An example of such a variation, given in Table. 8.2
[16], concerns the effect of adding a T unit of the favoured = 3 grain boundary
(Fig. 3.24) in different symmetrical tilt grain boundaries between = 9 (38.9◦ ) and
= 3 (70.55◦ ). In any case, the extrinsic dislocation Burgers vector is perpendicu-
lar to the grain boundary plane; then, its magnitude may be compared to that of the
be DSC vector, also normal to this plane. The br vector of the extrinsic dislocation
associated to the T unit decreases from its value in the = 9 grain boundary (b/a
= 0.333) to zero in the = 3 grain boundary where, obviously, adding a = 3
proper unit cannot create an extrinsic dislocation. Addition of one M unit of = 9
in the same grain boundaries leads to an increase of the b/a ratio from its value in
= 9 (0.333) until its value in = 3 (0.7). The latter corresponds to a/2 110
Burgers vector that may always be considered as a DSC vectors sum. This approach
also takes into account the step heights associated to the extrinsic dislocations.
Table 8.2 Magnitudes of the br Burgers vectors linked to the addition in different symmetrical
110 tilt grain boundaries in silicon of one structural unit T of = 3 compared to the magnitudes
of the elemental be DSC vectors of these grain boundaries (a is the lattice parameter of silicon) [16]
θ◦ br /a br /be
9 38.9 0.333 1
187 42.88 0.292 4
337 44.8 0.272 5
59 46 0.260 2
11 50.5 0.213 1
123 53.59 0.180 2
41 55.88 0.156 1
3 70.55 0 0
The importance of this description is that an in-coming dislocation may decom-

pose into stable structural units in the grain boundary which do not necessarily cor-
respond to the smallest DSC vectors: moreover, the residual structural units created
during the process although topologically equivalent may correspond to different
Burgers vectors depending on the misorientation angle. Then, the Burgers vector
magnitudes in high coincidence index grain boundaries may be non-negligible. This
is in contradiction with the geometrical model conclusions (see Sect. 2.3.2) and with
the interpretation of the widening of extrinsic dislocations in terms of spreading.
This point is very important for the understanding of the dislocation accommodation
processes in grain boundaries.
However, the results in Table 8.2 must be carefully considered, insofar as only
periodic grain boundaries with low index misorientation axes have been observed.
What happens in mixed tilt/twist, in quasi periodic and finally in general grain bound-
aries possessing five macroscopic degrees of freedom? It is likely that the latter con-
stitute a large majority of the real grain boundaries in polycrystal, those where the
extrinsic dislocation contrast widening has been revealed. Incorporation of lattice
dislocations in general grain boundaries is still misunderstood.
8.3.3 Transmission Processes
The accommodation of the stresses associated to an extrinsic dislocation occurs

by transmission of the dislocation or to one of its decomposition product in the
neighbouring crystal. The interfaces play their role of sources for lattice dislocations.
Sources may also be initiated in the same crystals giving rise to reflection. The
transmission phenomenon strongly depends on the geometric configuration formed
by the dislocation and the grain boundary. Two types of transmission are generally
considered (Fig. 8.13):
• The so-called direct transmission that imposes that the traces of the two slip planes,
from each side of the grain boundary, are common (or almost common) in the grain
boundary plane. If the Burgers vectors of the incoming and the emitted dislocations
Fig. 8.13 The two types of dislocation transmissions that may occur across a grain boundary: a
“direct” transmission with or without product left in the interface; b “indirect” transmission or,
occasionally, reflection
are parallel, no dislocation is left in the grain boundary and the energy for crossing
is simply that of cross slip. Otherwise, a product is left in the interface and the
energy for crossing is the energy of this residual dislocation. The preferentially
slip systems activated in a bicrystal for a direct transmission obeys a geometric
criterion or “transmission factor” defined by [20]:
(II · III )(bI · bII )

M= (8.6)
bI bII
where I are the directions of the slip plane traces in the grain boundary plane and
b are the Burgers vectors of the dislocations in the two crystals. Higher is M (close
1) easier is the transmission. This criterion has been clarified by Lee et al. [21],
the three following conditions must be fulfilled in order that a direct transfer may
occur:
• The angle between the slip plane traces in the grain boundary plane must be as
small as possible (ideally null)
• The residual product must have a Burgers vector as small as possible (ideally null)
• Finally, the resolved shear stress on the emitted dislocation must be very high.
• “Indirect” transmission may occur even if no common traces exist in the grain
boundary plane. Dislocations in one crystal pile up on the grain boundary leading to
a high stress concentration in the boundary and in the opposite grain. A dislocation
source is then activated, either in the grain boundary itself or at its close vicinity
but, in any case, it is due to the grain boundary barrier. It is not necessarily located
in front of the pile-up. The “indirect” transmission mode, that does not require any
geometric condition, occurs more currently in polycrystals. It likely influences the
Hall-Petch law [22].
The most favourable situation for a direct transmission is realized in a = 3
grain boundary in f.c.c. or b.c.c. materials when an incoming screw dislocation is
parallel to the misorientation axis. In that case, the dislocation could cross the inter-
face without leaving any residual product. Nevertheless, transmission has never been
found to be easy. Slip bands are stopped on a = 3 grain boundary in a Fe-Si alloy
although the deformation condition fulfills the ideal transmission factor (M = 1) for
the primary slip planes in the two grains [8]. The slip impediment may be due to the
stresses associated to the dislocations at the head of the pile-up; these stresses affect
the glissile dislocation distribution in the deformed grain and the formation of slip
bands in the opposite grain. In case of anisotropic material, it may also be attributed,
at least partially, to an image force effect [8]. In other respects, a direct transmission
is only observed when no climb motion is authorized; if not, the trapped dislocation
tends to decompose and the products move away from each other, as a result the
screw dislocation cannot be transmitted. In a = 9 grain boundary in silicon, screw
dislocations crossing the boundary have been observed by in-situ high voltage elec-
tron microscopy at relatively low temperature (about 950 ◦ C). No residue is detected
on the high-resolution electron microscopy image; on the contrary, decomposition
products are observed if the deformation is performed at elevated temperature (about
1200 ◦ C). Thus, the decomposition of a lattice dislocation in a grain boundary appears
contradictory with its transfer in the neighbouring crystal [23].
For no screw dislocations, direct transmission leaves residual dislocations in the
grain boundary. This has been observed in a near = 3 grain boundary in a
nickel bicrystal [7]. Numerous grain boundary dislocations, in favour of a strong
deformed state, are well revealed by using the electron microscopy weak beam tech-
nique (Fig. 8.14a). The similarity of the two configurations, on one hand of the
F1 /F2 dislocations, on the other hand of the E /B dislocations (Fig. 8.14b) as well as
the equality of the Burgers vectors (bF1 = bE and bF2 = bB ) support the occurrence
of a lattice dislocation decomposition in the grain boundary followed by emission of
products in one crystal.
Fig. 8.14 a Electronic micrograph, obtained by using the weak beam technique, of a near =3
grain boundary in a nickel bicrystal containing a high dislocation density (g = [1-3-1]I = [113]II ).
The F dislocation is decomposed into two products, the B dislocation seems to be emitted in crystal
I. b Schematic drawing of the dislocation configurations [7]
A detailed analysis of the forces acting on the configuration (applied force on

the B dislocation, interaction force between E and B, image force on B and friction
force) allows us to better describe the reaction. A dislocation coming from crystal II,
analogous to the C or D dislocation, has entered the grain boundary, has decomposed
in two products, then the B product has been emitted in crystal I according to:
F → F1 + F2
a/2[10 − 1]II a/6[−1 − 1 − 2]II a/6[41 − 1]II
⇓ (8.7)
a/2[1 − 10]I
This type of reaction is similar to that previously mentioned by Smith [24]; the
experiment gives the detailed processes that take place within the grain boundary.
It must be noted that decomposition alone is not energetically favoured, the grain
boundary is temporarily and locally submitted to a high stress that is partially relieved
by dislocation emission in the crystal. The resolved shear stress acting on the emitted
dislocation has been estimated to be two times the elastic limit of a nickel polycrystal
and thus largely superior to the elastic limit of a bicrystal.
When an angle exists between the slip plane traces, transfer may require prismatic
glide in one crystal at vicinity of the grain boundary. Such a phenomenon has been
found in stainless steels, allowing the transfer of a lattice dislocation across a general
grain boundary misorientated around an axis very close 110; this is only possible
if a very small Burgers vector is left in the interface and requires that the angular
deviation between the slip traces is less than 8◦ [25].
Finally, in the grain boundaries favourable to direct transmission, the latter may
only operate if there is no intergranular dislocation climb; if not, accommodation
occurs by decomposition and motion of the products in the grain boundary. For the
twin = 3, owing to its low diffusivity, a range of intermediate temperatures exists
for which decomposition and transmission are two competitive phenomena. More
generally, extrinsic dislocation accommodation processes take place by decomposi-
tion at high temperature and by indirect transmission at low temperature.
8.3.4 Entrance of a Dissociated Dislocation in a Grain Boundary
In f.c.c materials with low stacking fault energy, a perfect dislocation is dissociated
into two Shockley partial dislocations. On the contrary to the Burgers vector of a
perfect dislocation that may always be considered as the sum of DSC vectors, thus
compatible with the grain boundary structure, a Shockley partial may or not possess a
Burgers vector belonging to the DSC lattice. In favourable cases, partial dislocations
may enter separately in the grain boundary and decompose, similarly to perfect
dislocations; such process has been reported for silicon [18]. When an incoming
partial dislocation has a Burgers vector not part of the DSC lattice, its entrance in
the grain boundary requires complex processes that have been modeled, simulated
and observed by transmission electron microscopy to a slight extend.
8.3.4.1 Models for the Interaction Between a Dissociated Dislocation

and a Grain Boundary
The entrance of a lattice dislocation dissociated into two partials incompatible with
the grain boundary translation lattice may occur according to two types of mecha-
nisms (Fig. 8.15) [26]:
• A partial recombination starts by a constriction in the grain boundary plane fol-
lowed or not by the decomposition of the recombined dislocation into DSC dis-
locations. This mechanism is comparable to a cross slip with, more probably, a
lower energy barrier for two reasons: the grain boundary is often considered as a
narrow region of soft material with respect to the volume and the dislocation core
undergoes widening in the boundary.
• Each partial separately enters the grain boundary where it decomposes into a
DSC dislocation and a non-DSC dislocation with formation of a stacking fault,
the energy of which being probably smaller than that in crystal. Later, the non-
DSC dislocations may combine to form a DSC dislocation with suppression of
the stacking fault. This mechanism requiring less energy than the recombination
is likely favoured.
Fig. 8.15 Entrance processes in a grain boundary of a lattice dislocation dissociated into two
Shockley partials possessing non-DSC Burgers vectors: a two partials in a crystal; b recombination
process by constriction (i) followed by decomposition into DSC dislocations in the grain boundary
(ii); c Separated entrance on each partial with formation of non-DSC products and of a stacking
fault (i and ii). Schematic representation extracted from [26]
Finally, whatever the process, there is only a transient energy increase; the total
energy of the grain boundary/crystals system always decreases by absorption of a
lattice dislocation in a grain boundary. On the contrary, the transmission that most
often left a residual dislocation in the grain boundary occurs with a total energy
increase [26].
Despite the favoured energetic considerations for the separated decomposition of
each partial, the simulated and observed entrance of a dissociated dislocation in a
low stacking fault energy metal, such as copper, may proceed by recombination. But,
the transmission across a coherent twin predicted by the simulations has never been
observed [27].
8.3.4.2 Observations of Interactions Between Dissociated Dislocations

and Grain Boundaries
In the case where the partial Burgers vectors are compatible with the grain boundary
structure, the operating mechanisms are the same as those for a perfect dislocation.
When this condition is not fulfilled, the experimental results must be compared to
the models [26] and to the simulations [27]. The interaction mechanisms have been
elucidated for a near = 3 {111} grain boundary in copper by coupling conventional
and high-resolution transmission electron microscopy techniques. In a twin, the DSC
vectors never are the Burgers vectors of the incoming dislocations; indeed, the latter
must intersect the boundary and thus, must be situated in a {111}slip plane inclined
with respect to the grain boundary plane. Then, the interaction mechanisms differ
according to the fact that the grain boundary plane is or not a possible slip plane for
the perfect dislocation [28].
• If the grain boundary plane is a possible glide plane for the perfect dislocation,
the two partial dislocations recombine to enter the grain boundary and further
decompose into DSC dislocations; this phenomenon, analogous to a cross slip,
requires a transient increase of the total dislocation/grain boundary energy.
• If the grain boundary plane is not a possible glide plane for the perfect dislocation,
the entrance of the partials, each one in its turn, seems to be favoured. This is
suggested by the analysis of an extended defect in the (1-1-1) slip plane of crystal
II near the (11-1) grain boundary plane (Fig. 8.16). The total Burgers vector of the
defect, normal to the grain boundary plane, is b = a/3 [11-1]II ; its formation may
be described as follows:
– Dissociation in the (1−1−1)II plane of the perfect dislocation:
a/2 [01-1]II → a/6 [-11-2]II + a/6 [12-1]II (8.8)
– Entrance in the grain boundary then decomposition of the leading partial to

form a glissile DSC dislocation and a non-DSC dislocation with a stair-rod
Burgers vector:
a/2 [-11-2]II → a/6 [-21-1]II + a/6 [10-1]II (8.9)
– The glissile dislocation moves away the impact point; the second partial at
the close vicinity of the grain boundary and the non-DSC dislocation form
a configuration with a total Burgers vector a/3 [11−1]II well measured on
the micrograph:
a/6 [12-1]II + a/6 [10-1]II → a/3 [11-1]II (8.10)
The second partial is impeded to approach closer the grain boundary because it is
submitted to a repulsive force coming from the stair-rod dislocation.
Fig. 8.16 Image in

high-resolution electron
microscopy of a partial
Shockley dislocation in a
(11−1) plane of crystal CII
at vicinity of a = 3{111}
grain boundary in copper.
This dislocation is repelled
from the grain boundary by
a non-DSC dislocation in (or
extremely close) the grain
boundary. The total configu-
ration (Shockley + non-DSC)
displays a b = a/3 [11−1]II
Burgers vector normal to the
grain boundary plane (sessile
configuration) [28]
Alternatively, an analogous configuration may arise from the relaxation of an

extrinsic dislocation a/3 111 associated to a step with formation in a crystal of a
partial Shockley dislocation and maintenance of a stair-rod dislocation at the grain
boundary; this reaction has been observed in a twin grain boundary in aluminium [29].
But, what could be the origin of an extrinsic dislocation with a/3 111 Burgers vector?
It probably results from the absorption of a perfect lattice dislocation, followed by
its combination with a glissile DSC dislocation or by its decomposition into two a/6
{112} and a/3 {111} DSC dislocations. The latter reaction cannot occur in a crystal,
but it could be possible in a grain boundary. In case of copper, the entrance of a
perfect dislocation requires first a recombination of the partials. Whatever the retained
process, the presence of a dissociated Frank dislocation near the grain boundary may
in fine result from the incorporation of a perfect dislocation.
The two entrance mechanisms of a dissociated dislocation have been revealed in
the case of a coherent twin in copper; their occurrence according to the character,
glide plane or not, of the grain boundary plane may be supported by two following
remarks. In the first case, the mechanism is akin to the cross slip process largely
observed in copper. In the second case, the configuration is similar to that observed
for the entrance of a dissociated dislocation in a = 3 grain boundary in silicon,
material which also displays a low stacking fault energy [30].
To conclude, whatever the operating mechanism that yields the existence of extrin-
sic dislocations within a grain boundary, transmission, decomposition or combina-
tion, the result is the formation of a non-equilibrated grain boundary. Return to
equilibrium requires the intergranular stresses to be relieved. Then, new dislocations
may enter the grain boundary and the deformation may go on. Relaxation is a major
grain boundary process under various mechanical and thermal stimuli.
8.3.5 Simulation of the Interaction of a Lattice Dislocation

with a Grain Boundary
One among the first molecular dynamic simulations of the interaction of a perfect
dislocation with a grain boundary has been performed in the case of a = 9 grain
boundary in iron [31]. The absorption of the perfect dislocation a/2 [111] is accom-
panied by core relaxation. The dislocation decomposes into two DSC dislocations,
glissile b1 and sessile b2 . The glissile dislocation moves away the entrance region
(Fig. 8.17). In the same grain boundary containing phosphorus, the lattice dislocation
is repelled from the grain boundary; initially located in the grain boundary, it does
not decompose and its core remains very localized. This “locking” segregation effect
on extrinsic dislocations is confirmed by experiments. It has a particular importance
in the relaxation phenomena and strongly influences the grain boundary behaviour
in polycrystal deformation at high temperature.
Other molecular dynamic simulations, using a potential for aluminium derived
from the embedded atom method, show that a perfect screw dislocation in a =
11 {113} grain boundary decomposes into two glissile dislocations with Burgers
vectors a/22 417. The elastic energy reduction is about 45 % according to the
Frank criterion; furthermore, the grain boundary energy is very weakly affected by
the presence of the a/22 417 dislocations [32].
The core interaction has also been simulated for a screw dislocation parallel to the
misorientation axis for a =9 {221} grain boundary in aluminium [33]. The excess
Fig. 8.17 Interaction of a perfect dislocation b = a/2 [111] with a = 9 grain boundary in iron:
a the dislocation is absorbed within the grain boundary; b it decomposes into two dislocations, b1
glissile and b2 sessile; c the b1 dislocation glides in the grain boundary plane [31]
energy evolution with the distance between the dislocation and the grain boundary
is reported on Fig. 8.18 for different positions of the glide plane impacts in the grain
boundary period. Below a distance less than the period in the direction normal to the
grain boundary plane, the dislocation is always attracted to the boundary; beyond
this distance, there is no core effect. The core interaction is thus very localized. An
estimated value of the necessary stress for extracting the dislocation from the grain
boundary is very elevated, about 5.10−2 μ.

excess energy of a system
composed by a screw dis-
location parallel to the
misorientation axis of a
= 9{221} symmetrical tilt
grain boundary in aluminium
with the “dislocation/grain
boundary” distance. The three
curves correspond to different
localizations of the slip plane
trace in the grain boundary
period [33]
The interaction between a dissociated dislocation and a grain boundary has

been simulated, using a N-body potential adapted for pure copper, then for f.c.c.
intermetallic compounds [27]. The simulations are restricted to symmetrical tilt
grain boundaries around [110] and to screw or 60◦ dislocations parallel to the
misorientation axis. In any case, the interaction is attractive and very elevated stresses
are required for the dislocation reacts with the grain boundary core. The interaction
mechanisms are function of the fine grain boundary structure. They are detailed below
in the cases of = 3 {111}, = 11 {311} and = 27 {511} grain boundaries in
copper. Dissociated screw lattice dislocations move on the effect of a resolved shear
stress of 500 MPa.
In the case of Σ = 3 {111} grain boundary, the leading partial is stopped by the
coherent twin. Under the effect of stress, the two partials come closer; they enter the
grain boundary provided the resolved stress reaches a minimum value of 1500 MPa.
Then, transmission of the screw dislocation in the symmetrical glide plane of the
neighbouring grain occurs. After transmission, the two partials move away from
each other. In the Σ = 11 {311} grain boundary, the recombined screw disloca-
tion decomposes into two a/22 147 DSC dislocations that are glissile in the grain
boundary plane. These two dislocations are associated to steps in opposite directions,
the height of which being equal to one interplanar distance. In the Σ = 27 {511} grain
boundary, the boundary response is more complex, depending on the place where
the lattice dislocation glide plane intersects the = 27 structural unit. If the dislo-
cation impinges on the grain boundary at the middle of a structural unit, transmission
occurs under a resolved shear stress of 1300 MPa; the partial dislocation contraction
in the course of transmission is not so pronounced than in the case of = 3 and
strong relaxations are observed in the neighbouring structural units (Fig. 8.19). If the
glide plane cuts the structural unit at its extremity, the lattice dislocation undergoes
a cross slip at the grain boundary vicinity, then the leading partial is absorbed. When
the stress increases, the trailing partial also enters the grain boundary while the first
partial core extends in the other crystal. No transmission is observed even for a final
stress level of 3000 MPa.
Fig. 8.19 a Different dislocation configurations found in = 27 grain boundary in copper; b

transmission of the screw dislocation in the middle of the structural unit (configuration 1) with
indication of the strong relaxations occurring in the neighbouring units. The different symbols
indicate atoms on successive {110} planes along the [110] axis [27]
Different responses to the entrance of lattice dislocations in grain boundaries

depending on their structure have also been obtained using multiscale modelling
[34]. The simulations concern the symmetrical = 3 {111}, = 11 {113} and
= 9 {114} tilt grain boundaries in aluminium. Whatever the number of screw dis-
locations in a pile up, they are absorbed then decomposed in the = 3 and = 11
boundaries and transmission is never observed. On the contrary dislocations are trans-
mitted in the = 9 at the cost of high stresses; the transmission occurs over planes
where the Schmid factor is not maximal. The antagonism between decomposition and
transmission already discovered by high-resolution electron microscopy studies [23]
is well confirmed by the simulations [34]. Widening of the dislocation core has been
revealed in an asymmetrical = 11 {252} // {414} tilt boundary in aluminium [35].
In case of metals, the behaviour of a screw lattice dislocation entering a coherent
twin strongly depends on the metal stacking fault energy [36]. If the dislocation is
spontaneously decomposed in aluminium, metal of high stacking fault energy [34], it
recombines by constriction in the boundary and then is transmitted into the adjacent
grain in copper and nickel, metals with lower stacking fault energies.
Fig. 8.20 a Intersection region of a pile up of five edge dislocations with a = 11 {113} boundary
in aluminium. 26 dislocations have been emitted into the gain boundary, leading to its migration and
to he formation of micro-cavities; b atomic positions corresponding to (a) and providing evidence
of damage in the interaction region [37]
The answer of a given boundary also differs according to screw, edge or mixed
character of the incident lattice dislocation. Edge dislocations entering a = 11
{113} in aluminium provoke the nucleation of intergranular dislocations [37] that
modify the grain boundary stress state and yield damages on the form of micro-
cavities (Fig. 8.20).
On the basis of their results, the authors propose three criteria for nucleation of
grain boundary dislocations [37] that could be added to those already proposed by
Lee et al. [21]:
• The normal stress on the grain boundary must be low
• The step associated with the residual defect must be small
Fig. 8.21 Interaction mechanisms between a dissociated mixed 60◦ dislocation and a = 3 {111}
coherent twin in copper deduced from simulations: a the leading partial is γA, the perfect dislocation
is recombined within the boundary and partials are emitted in the right grain; b the leading partial
is γD, after recombination in the boundary, a partial dislocation associated to a stacking fault is
emitted in the right grain and forms a lock with the sessile residue retained in the boundary [36]
• If the Shockley partial remains close the intersection region in the boundary and
is not absorbed, the shear stress acting at the pile-up head must be high.
Nucleation of grain boundary dislocations have also been obtained by simulations
in 100 and 110 tilt boundaries in copper [38] and in asymmetrical tilt boundaries
in copper and aluminium [39]
For mixed dislocation, the interaction processes are more complex. They differ
according to the sign of the Burgers vector of the leading lattice dislocation. Multiple
mechanisms may occur, they are illustrated on Fig. 8.21 for a mixed 60◦ dislocation
entering a = 3 {111} coherent twin in copper [36]. The incident dislocation DA is
dissociated into two partials Dγ and γA. If γA is the leading partial, the two partials
recombine into the boundary; the resulting dislocation is emitted in the opposite
grain leaving an intergranular residue Cγ (Fig. 8.21a). If γD is the leading partial,
after recombination within the boundary, a partial dislocation associated with an

intrinsic stacking fault is emitted in the grain and a sessile product is maintained in
the boundary (Fig. 8.21b). The configuration attached to the boundary forms a lock,
also observed in aluminium and nickel. The stresses required for such processes are
very large, of the order of a gigapascal [36]
Finally, in any case, dislocations dissociated or not, the calculated stresses for
transmission or for decomposition are very elevated. These two phenomena are
rarely observed at low temperature, they require thermal activation. The necessary
relaxation phenomena are the subject of the next chapter.
References
1. D.M. Barnett, J. Lothe, J. Phys. F4, 1618 (1974)

2. A.Y. Belov, V.A. Chamrov, V.L. Indenbom, J. Lothe, Phys. Status Solidi B. 119, 565 (1983)
3. M. Condat, H.O.K. Kirchner, Phys. Status Solidi B 144, 137 (1987)
4. O. Khalfallah, M. Condat, L. Priester, H.O.K. Kirchner, Phil. Mag. A61, 291 (1990)
5. M. Khalfallah, M. Condat, L. Priester, Phil. Mag. A67, 231 (1993)
6. L. Priester, O. Khalfallah, Phil. Mag. A69, 471 (1994)
7. S. Poulat, B. Décamps, L. Priester, Phil. Mag. A77, 1381 (1998)
8. M. Polcarova, J. Gemperlova, J. Bradler, A. Jacques, A. George, L. Priester, Phil. Mag. A78,
105 (1998)
9. M. Martinez-Hernandez, H.O.K. Kirchner, A. Korner, A. George, J-P. Michel, Phil. Mag. A56,
641 (1987)
10. A. George, Rev. Phys. Appl. 23, 479 (1988)
11. L. Priester, dans “Les joints de grains dans les matériaux”, Les Editions de Physique (1985)
p. 231
12. L. Priester, R.W. Balluffi, J. Microsc. Spectrosc. Electron. 4, 615 (1979)
14. R.C. Pond, D.A. Smith, Phil. Mag. 36, 353 (1977)
15. W. Bollmann, B. Michaut, G. Sainfort, Phys. Status Solidi A. 13, 637 (1972)
16. J. Thibault, J.-L. Putaux, A. Jacques, A. George, H.M. Michaud, X. Baillin, Mat. Sci. Eng.
A164, 93 (1993)
17. L. Priester, J. Thibault, V. Pontikis, Solid State Phenomena, vol. 59–60, (Scitec Publications,
Switzerland, 1998), pp. 1–50
18. M. Elkajbaji, J. Thibault-Desseaux, Phil. Mag. A58, 325 (1988)
19. H.M. Michaud, X. Baillin, J. Pelissier, J.-L. Putaux J. Thibault. Microsc. Microanal.
Microstruct. 4, 221 (1993)
20. Z. Shen, R.H. Wagoner, W.A.T. Clark, Sripta Metall. 20, 921 (1986)
21. T.C. Lee, I.M. Robertson, H.K. Birnbaum, Phil. Mag. A62, 13 (1990)
22. J.C.M. Li, Y.T. Chou, Metall. Trans. 1, 1145 (1970)
23. J. Thibault-Desseaux, J.-L. Putaux, A. Bourret, H.O.K. Kirchner, J. Phys. 50, 2525 (1989)
24. D.A. Smith, J. Phys. 43, C6–225 (1982)
25. C.T. Forwood, L.M. Clarebrough, Phil. Mag. A44, 31 (1981)
26. A.H. King, F.R. Chen, Mat. Sci. Eng. 66, 227 (1984)
27. B.J. Pestman, J.Th. De Hosson, V. Vitek, F.W. Schapink, Phil. Mag. A64, 951 (1991)
28. L. Priester, J-P. Couzinié, B. Décamps, J. Thibault, in Proceedings of the 25th Risö International
Symposium on Material Science “Evolution of Déformation Microstructures in 3D” (2004),
p. 79
29. D.L. Medlin, C.B. Carter, J.E. Angelo, M.J. Mills, Phil. Mag. A75, 733 (1997)
References 267
30. J-L. Puteaux, Thèse Grenoble (1991)

31. Y. Ishida, M. Mori, J. Phys, 46, C4–46.5 (1985)
32. R.J. Kurtz, R.G. Hoagland, J.P. Hirth, Phil. Mag. A79 (1999) 683
33. A. Aslanides, thèse de Doctorat, Université Paris VI (1998)
34. M.P. Dewald, W.A. Curtin, Phil Mag. 87, 4615 (2007)
35. M. De Kooning, R.W. Kurtz, V.V. Bulatov, C.H. Henager, R.G. Hoagland, W. Cai, M. Nomura,
Workshop on modelling and experimental validation (les Diablererts, Switzerland, 2002),
J. Nucl. Mater. 323(2–3), 281 (2003)
36. Z.H. Jin, P. Gumbsch, K. Albe, K. Lu, H. Gleiter, H. Hahn, Acta Mater. 56, 1126 (2008)
37. M.P. Dewald, W.A. Curtin, IUTAM symposium on plasticity at the micron scale, Tech. Univ.
Denmark Lyngby, Denmark, Modell. Simul. Mater. Sci. Eng. 15(1), S193 (2007)
38. D.E. Spearot, M.A. Tschopp, A. Mark, K.L. Jacob, D.L. McDowell, Acta Mater. 55(2), 705
(2007)
39. M.A. Tschopp, D.L. McDowell, Int. J. Plast 24(2), 191 (2008)
Chapter 9
Intergranular Stress Relaxation
The relaxation processes at grain boundaries control the polycrystal mechanical

behaviour at high and moderate temperature and the recrystallization phenomena.
They necessarily imply cooperative thermally activated mechanisms (Fig. 9.1).
Simultaneously to the absorption of the extrinsic dislocations within the grain
boundary structure, rotation of the adjacent grains and migration of the grain
boundary occur. Dislocation pile-ups created along the boundary on steps or triple
junctions are relieved by emission of lattice dislocations. In extreme cases, the relax-
ation proceeds by cavitation, especially at triple junctions. All the mechanisms are
necessarily coupled, but they may be formally distinguished between those implying
reactions with the neighbouring grains and those occurring within the grain boundary.
The latter are now described by analyzing the grain boundary structure and energy
evolutions in the course of the extrinsic dislocation accommodation, i.e. the return of
the disturbed grain boundary (non-equilibrated) towards an equilibrium state [1, 2].
Initially, the grain boundary equilibrium is given by the Eq. (2.4): B = (I−R−1 )X.
If n extrinsic dislocations are formed in the boundary, the Burgers vector density
becomes B = B + nbe (5.2) with be = bDSC or be = bDSC (in particular
bDSC = bm ). The grain boundary structure is then locally disturbed, but the global
misorientation is preserved. A new equilibrium is reached when the Frank–Bilby
equation is again satisfied:
−1
B = (I − R )X with R = R (9.1)
The extrinsic dislocation accommodation in the grain boundary is accompanied

by a change in the orientation relationship between the crystals. The extrinsic dis-
locations incorporate the intrinsic structure and lose their long-range elastic strain
fields. In the previous equations, the Burgers vector density B or B is not neces-
sarily discretized. Discrete products formation depends on the relaxation processes
within the grain boundary; indeed, a correlation must exist between the local stress
distribution and the relaxed structure adopted by the grain boundary.

270 9 Intergranular Stress Relaxation
Fig. 9.1 The different stress accommodation mechanisms at grain boundaries in a polycrystal
9.1 Extrinsic Dislocation Accommodation Models
The stress relaxation may occur within the interface by two alternative processes:
core delocalization or incorporation within the interface structure, the interfaces
playing their role of sinks. Whatever the relaxation mode, it strongly depends on the
interface structure. The accommodation models either take or do not into account the
intrinsic structure and thus are more or less relevant according to the grain bound-
ary type. The distinction between singular, vicinal and general grain boundaries
based on their energy, is not sufficient to understand the different relaxation modes.
A distinction based on the periodicity appears more appropriated: periodic symmet-
rical grain boundaries (most of the grain boundaries until now studied in bicrys-
tals), quasi-periodic or incommensurate grain boundaries and even amorphous grain
boundaries as proposed by Rosenhain [3] and recently re-discovered [4, 5]. The latter
may constitute a non-negligible proportion in real materials.
9.1.1 Extrinsic Dislocation Core Delocalization
The model rests on the basic principle that any dislocation has a tendency to delocalize
its Burgers vector density distribution in order to reduce its elastic energy. Concerning
an extrinsic dislocation, its energy reduction is accompanied by a grain boundary
energy increase. The delocalization goes on until the two energetic variations balance;
this happens for a certain value s of the dislocation core width (Fig. 9.2) [6].
If only one dislocation is present in a grain boundary, its accommodation is
achieved when the delocalization extends on the overall grain boundary length.
Delocalization is equivalent to a continuous distribution of infinitely small Burg-
ers vector dislocations in a grain boundary region or in the total grain boundary
plane area.
The core width in a general grain boundary, without intrinsic dislocation network,
has been deduced from the Peierls equation as it was calculated in a crystal [6, 7].
The grain boundary structure is considered as disordered with respect to a perfect
9.1 Extrinsic Dislocation Accommodation Models 271
Fig. 9.2 Evolution of the extrinsic dislocation elastic energy in the course of its core delocalisation
in a grain boundary
crystal, its disorder varying periodically. Beyond a certain temperature, the dislo-
cation core may widen with time; indeed, the stress field induced by the Burgers
vector distribution yields the diffusion of matter from the grain boundary regions
in compression towards those in tension. In other words, the infinitely small dislo-
cations repelled each other and spread by climb in the grain boundary plane. The
delocalization is a thermally activated phenomenon; its kinetics is given by:
kTs s 3
ts = A (9.2)
μΩ DGB δ
with ts the time at the temperature Ts for the dislocation core reaches a width s, μ is
the shear modulus of the material, k the Boltzman constant, Ω the atomic volume, δ
the grain boundary core width (often considered equal to 0.5 nm) and DGB the grain
boundary self-diffusion coefficient at the temperature Ts . A is a constant that seems
to depend on the operating mechanism [7].
For a long time, the delocalization phenomenon has been considered as respon-
sible for the extrinsic dislocation contrast widening when the grain boundary is
submitted to a thermal treatment. The dislocation image width for which the contrast
disappears has been compared to the core width corresponding to the total dislocation
relaxation. The image width just before the contrast disappearance is about two times
the extinction distance ξg , linked to the imaging condition in electron microscopy.
But, this comparison is now questioned: recent weak beam electron images have
revealed discrete products in a grain boundary region where a contrast widening was
observed under bright field conditions [8]. More generally, the contrast evolution
is not directly linked to the dislocation core evolution; widening and disappearance
of dislocation contrast only show how the processes look from “outside” the grain
boundary and the “spreading” phenomenon is still misunderstood. Nevertheless, the
electron microscopy results on the spreading kinetics have been used to estimate
the grain boundary diffusion coefficient and activation energy [9, 10]. The obtained
values are of the same order than those determined by other methods.
The fluid like region model of a deformed grain boundary [11] may be considered
as an extension of the delocalization model. The authors start with the hypothesis
that a grain boundary without long-range order may exist. When dislocations enter
such a grain boundary at elevated temperature, they are submitted to a continuous
core delocalization yielding an excited state of certain regions named “quasi-fluid”
or “quasi-liquid” regions. At the extremities of these regions, stresses concentrate;
their relaxation proceeds by dislocation emission in the crystals. As the deformation
goes further, the number of extrinsic dislocation increases, the “fluid-like” regions
overlap; most of the grain boundaries are in an excited state and a viscous Newtonian
flux takes place (Fig. 9.3).
Fig. 9.3 The “fluid-like” or

“quasi-viscous” accommoda-
tion model: a formation of
“quasi-liquid” small islands in
course of the extrinsic dislo-
cation core delocalization, the
stress variation along the grain
boundary is also reported;
b dislocation emission in
the crystals; c under a more
important dislocation flux
(arrows), the excited regions
partially overlap to form a
“quasi-viscous” layer along
the grain boundary [11]
Although there is no experimental observation to valid the existence of quasi-

viscous regions in a deformed grain boundary, the model allows its authors to
quantitatively account for the role played by the extrinsic dislocation formation
and relaxation in the superplastic deformation of materials [11]. It may be reason-
ably taken into account to explain the accommodation phenomenon in general grain
boundaries, beyond a certain advanced deformation state.
9.1.2 Extrinsic Dislocation Decomposition and Product

Reorganization
The Extrinsic dislocation decomposition model has been first developed for tilt grain
boundaries and edge dislocations [12]. A bm extrinsic dislocation decomposes into
n products with Burgers vectors bDSC , regularly spaced along a certain distance H
in the grain boundary (Fig. 9.4).
Fig. 9.4 a Initially isolated

dislocation; b decomposition
in several products located
at a distance h from each
other along a grain boundary
length H (unstable state); c
equilibrium is reached again
when a regular distribution of
the products occurs all along
the grain boundary (H = L)
[12]
The grain boundary energy evolution, from its initial state to its deformed state then its
relaxed state, is followed by considering the dislocation wall as an edge disclination
dipole of width H. Generally, the energy is composed of three terms:
E = E θ0 + E el + E GB (9.3)
with E θ0 the initial energy of the grain boundary without defect, E el the elas-
tic energy increase due to extrinsic dislocations and E GB the interfacial energy
change. The elastic energy evolution E el is a logarithmic function of L/H with L
the grain boundary total length and H the dipole width or the dissociation distance.
The E GB variation is inversely proportional to the product number n (H = nh)
which, according to the authors, increases with the coincidence index . The evolu-
tion of the difference between the initial energy and the energy all along the relaxation
process is reported on Fig. 9.5a. Each curve corresponds to a different grain bound-
ary: n is taken equal to 3 for a = 3 grain boundary and n = 100 for a general grain
boundary, considered as displaying a high coincidence index. Equilibrium is reached
when the decomposition products are evenly distributed all along the grain boundary
length and become intrinsic. For a vicinal or general grain boundary, the final energy
may be superior or inferior to the initial energy; indeed the angular deviation resulting
from the extrinsic dislocation surplus may increase or decrease the initial θ misorien-
tation. Furthermore, the energy difference is the function of the grain boundary type:
for a singular grain boundary, the energy after relaxation may be extremely different
than the initial one; on the contrary it weakly differs for a general grain boundary .
These results are in agreement with the curve giving the energy evolution in function
of the misorientation for symmetrical tilt grain boundaries around 110 (Fig. 9.5b).
In that model, the relaxation kinetics is controlled by the sessile dislocation climb
in the grain boundary and thus, as in the previous model, by grain boundary diffusion.
For n 1, the grain boundary is general (equivalent hypothesis to that of the
delocalization model), the kinetics law takes the form of Eq. (9.2), the relaxation
Fig. 9.5 a Evolution of the difference between the grain boundary energy E in course of the
relaxation process and its initial energy E θ0 (coordinate without dimension) in function of the
ratio between the extrinsic dislocation decomposition distance and the total grain boundary length
L = 103 b (see text) [12]; b schematic evolution of the calculated energy of symmetrical tilt grain
boundaries in function of their misorientation showing that the energy differences (E 0 − E) or
(E 0 − E ∗ ) associated with small misorientation changes are small in general grain boundaries;
they are large and always positive for singular grain boundaries in agreement with the predicted
variations in (a)
time is proportional to s 3 , only the numerical coefficient A differs from the previous
one [12].
The model only considers sessile extrinsic dislocations; moreover, it is limited for
two additional reasons:
• In the case of singular or vicinal grain boundaries, it ignores the presence of
primary and/or secondary intrinsic dislocation networks,
• In the case of general grain boundaries, it rests on the hypothesis that the number
of dislocations n is high and their Burgers vectors extremely small. However,
high-resolution electron microscopy has revealed non-negligible Burgers vectors
in high index coincidence grain boundaries (see Sect. 2.3.3).
9.1.3 Extrinsic Dislocation Incorporation Within the Intrinsic

Structure
The model starts by a decomposition process of the initial extrinsic dislocation, but
it has the great advantage over the previous model to take into account the grain
boundary structure in terms of intrinsic dislocations and in terms of structural units.
Even if it constitutes an improvement of the previous model, it is also restricted to
pure tilt grain boundaries and to extrinsic dislocations parallel to the intrinsic ones
[13]. The authors consider three components of the non-equilibrium grain boundary
structure (Fig. 9.6):
Fig. 9.6 Schematic representation of the formation of a non-equilibrated grain boundary: a lattice
dislocations enter a grain boundary; b the three components of the non-equilibrated grain boundary
structure: glissile dislocations, sessile dislocations and disclinations at triple junctions; c disordered
arrangement of sessile dislocations [13]
• The glissile extrinsic components with Burgers vectors bg , coming from the lattice
dislocation decomposition, are very mobile (this was observed by high-resolution
electron microscopy) and form pile-ups to the triple junctions,
• The edge components with Burgers vectors bc may easily annihilate by reac-
tion with the intrinsic dislocation displaying opposite Burgers vectors, the climb
distance being inferior to the h0 distance that characterizes the initial state. This
process particularly operates when h0 is small, thus in general grain boundaries.
As a result, the bc component density decreases. The balance between the dis-
locations + bc et − bc being rarely realized, a dislocation surplus appears in the
boundary that leads to disclination formation at triple junctions,
• Finally, the sessile extrinsic dislocations superimpose the intrinsic network;
together, intrinsic and extrinsic, they form disordered arrangement of parallel
intergranular dislocations. This configuration is taken as the initial model of a
non-equilibrated grain boundary.
The stresses associated with the three components must be relieved in order that
the grain boundary finds again an equilibrium state. Only the reorganization of the
sessile component disordered arrangement contributes to the grain boundary struc-
ture change. It occurs by dislocation climb over distances of the same order of
the dislocation spacing and thus depends on the grain boundary diffusive proper-
ties. An equation similar to (9.2) gives the necessary time for the disappearance of
the dislocation contrasts, the A numerical coefficient being smaller than in the two
previous models [14].
The disordered arrangement of dislocations may be described on the basis of the
“intrinsic dislocations/structural units” model by the following sequence:
[m1 A] B [m2 A] B . . . . . . [mi A] B. (9.4)
The non-equilibrium degree is given by the variance D:
Dm = m2 − m20 (9.5)
where m0 is the number of majority units A in the equilibrium structure of the grain
boundary with a period given by [m0 A] B.
The non-equilibrium degree may also be represented by the dislocation spacing
variance:
Dh = h2 − h20 = Dm d2A (9.6)
with h = md A + d B (d A and d B are the lengths of the A and B structural units,

respectively).
A return towards equilibrium must yield a structure described by a structural unit
periodic arrangement with a period [m f A]B that differs from the initial one [m0 A]B.
However, the reorganization of the sessile dislocations does not lead to a total
relaxation of the long-range stresses within the grain boundary; the two non-
equilibrium components formed at triple junctions are still present: glissile prod-
ucts and disclination dipoles. The latter, that control the last relaxation stage, have
a particular importance in the polycrystal properties. The motion of the glissile
dislocations induces grain boundary sliding; it results in plastic incompatibilities at
triple junctions whose relaxation implies reactions with the crystals. The relaxation
mechanisms and their kinetics have been proposed on the basis of the continuous
theory of plasticity or on the basis of the discrete dislocation approach [15]. What-
ever the theory and for the two types of defect arrangements, the mechanisms are
also controlled by grain boundary diffusion. Their kinetics also obeys an equation
similar to (9.2) with a very low value of the constant A (≈0.007) that weakly varies
with the approach model and with the dislocation type.
To sum up, it is remarkable that all the grain boundary stress relaxation models
(excluded the triple junction processes) yield the same kinetics (9.2). The differences
between the numerical coefficient A, of two orders of magnitude according to the
model, delocalization or decomposition with dislocation rearrangement or incorpo-
ration in the intrinsic network, may indicate the more or less long-range character
of the diffusion processes. But, the DJ δ products appearing in the Eq. (9.2) are not
well known and it must be better to consider the P = k DJ δ products in which
the k coefficient accounts for a segregation effect on the diffusion. As a result of
these uncertainties, the calculated relaxation times at a given temperature differ from
the determined experimental times, even if care is taken to not confuse the process
manifestation (dislocation contrast variation, for example) from the process itself. In
a recent review of the kinetics of the different models, the unified value A = 0.036
has been proposed whatever the relaxation approach, continuous or discrete; it seems
to pretty well account for a large number of experimental results [16].
Finally, the relaxation phenomenon understanding requires first the knowledge
of the long-range stress distribution in the grain boundary and how it evolves with
time when the gain boundary is submitted to thermal treatments. The two following
sections develop the models of the extrinsic dislocation stress field evolution with
time and distance from the grain boundary. These mechanical approaches allow us to
partially answer the question: how happens the change of the dislocation character,
from extrinsic to intrinsic? But, the physical processes have still to be specified. This
question is fundamental to approach the grain boundary role in plastic deformation.
9.2 Evolution of Extrinsic Dislocation Stress Fields

with the Distance from the Grain Boundary
Two approaches have been developed: the first one leads to a diminution of the shear
stress in function of x−1/2 [13, 17], the other one in function of x−3/2 [18], accord-
ing to the non-equilibrium degree considered. In the first case, important and really
random disorders are authorized. In the second one, the dislocations may displace
a little bit from their equilibrium positions; such an arrangement has been named
quasi-equidistant grain boundary. Other difference comes from the dislocation con-
tent origin: in one case, lattice dislocations are introduced in the initial grain bound-
ary, in the other case, only fluctuations of the intergranular dislocation s from their
equilibrium positions are considered.
9.2.1 Random Disordered Dislocation Wall Model
The disordered arrangement of dislocations is modelled as a correlated quasi-uniform

wall of dislocations. The intergranular dislocations are considered as disclination
dipoles and the pseudo-random distribution of mi , given by the relation (9.4), is
generated using Monte Carlo technique. The infinite grain boundary is divided in
periods of length H; each of them contains a great number N of disclination dipoles
(see Fig. 3.36).
The diminution law of the long-range stress fields is given by the dependence of
the field variance Dσ(x,y) with the distance x from the wall [13]:
Dσ(X,Y ) = σ(X,Y
2
) − σ(X,Y )
2
(9.7)
σ(x,y) is the average value of the shear stress. This average value is obtained by
realizing a set of possible random stress fields linked to the random distributed dislo-
cations. By considering that the average value is null, that all the dipole arrangements
have the same probability to occur along the boundary (direction y) and that the elas-
tic field variance is uniform in a plane located at a distance x of the grain boundary,
the previous formula may be simplified:
Dσ(x,0) = σ(x,0)
2
(9.8)
Calculations are made for distances s such as h 0 < x ≤ H (for h0 see Fig. 9.6a) in
order to avoid the short-range and the periodic effects.
The dependence of Dσ(x,0) with the distance x from the grain boundary is given
by par [17]:
(Dσ(x,0) )1/2 ≈ Abn [(ρGBD /x)]1/2 (9.9)
where bn is the dislocation Burgers vector component normal to the grain boundary,
ρGBD is the dislocation linear density and A = μ[2π(1 − v)]. This formula is only
valid for long-range stresses (x ≥ ρGBD −1 ).
As an example, in an f.c.c. material, the [001] √ = 5(θ = 36.9◦ ) tilt
√ grain
boundary is composed of A and B units with d A = a 2/2 and d B = a 5/2 (a
is the lattice parameter). By taken N = 1200 and the number or realizations of the
stress fields equal to 1000, the expression of the stress variance with the distance
x ≥ h 0 = 4d A + d B ∼= 4a is given by [13]:
Dσ(x,0) = CDm (a/ h 0 )3 (a/x) (9.10)
C is a constant easily determined from the experimental curves and Dm represents

the grain boundary non-equilibrium degree according to expression (9.5). The stress
field variance in the function of the distance from the grain boundary is shown on
Fig. 9.7 for different non-equilibrium degrees when m0 = 4.
The expression of the variance is also given by considering the disordered
configuration of dislocations with Burgers vector b and an average distance from
each other equal to h. Moreover, by assuming that two dislocations (+b and −b)
are absorbed for n extrinsic dislocations, the extrinsic dislocation density is deduced
(ρ = 2/nh 0 ) and a simple equation for the variance is:
(Dσ(x,0) )1/2 ≈ 0.14[μb/(1 − v)](3ρ/4x)1/2 (9.11)
The effects of the disorder resulting from the extrinsic dislocations are analyzed when
the density of the dislocations entering a grain boundary after a 2–3 % deformation
at room temperature is equal to 107 m−1 . By using Eq. (9.11) for a grain boundary
in aluminium, elevated internal stresses (>10−4 μ) are found up to relatively large
distances (x ≈ 60b), thus they may likely influence the dislocation glide in a relatively
thick region or mantle along the grain boundary. The authors also consider the excess
grain boundary energy, equal to the difference between the non-equilibrated and the
equilibrated grain boundary energies, the misorientation being preserved. With the
9.2 Evolution of Extrinsic Dislocation Stress Fields 279
Fig. 9.7 Stress field variance

in function of the distance
from the grain boundary for
different non-equilibrium
degrees of the intergranular
structure with m0 = 4; the
different curves 1, 2, 3, 4,
5 et 6 correspond to Dm =
0.1 − 0.3 − 0.5 − 0.7 − 1 − 2
[13]
same data that those previously used for the stresses, a simple expression may be
obtained:
3μb2 ρ
γex = (9.12)
8 (1 − v)
For aluminium deformed of 2 or 3 %, the excess energy is 1.2 × 10−2 J · m−2 ; it only
corresponds to about 2 % of the maximal energy of the tilt [001] grain boundaries
(γ ≈ 0.6 J · m−2 ). On the contrary, for more elevated deformation, the energy of the
non-equilibrated grain boundary may reach two times the value of the energy of the
equilibrated ones; in that case, the grain boundary seems to be in a quasi-amorphous
state; it is likely to occur in submicron grain-sized materials.
9.2.2 Quasi-Equidistant Grain Boundary Model
9.2.2.1 Infinite Wall of Dislocations
The term quasi-equidistant introduced for defining this model is due to a first
approach of the lightly deformed tilt grain boundaries using anisotropic elasticity
[19]. A periodic distribution of identical dislocations (screw, edge or mixed) parallel
to each other in an infinite wall constitutes the starting point. Then, the dislocations
are authorized to move over a small distance h0 δi (with h0 the equilibrium distance
Fig. 9.8 Description of a

disordered dislocation wall:
a ordered wall; b disordered
dipole wall; c disordered wall
obtained by superposing (a)
and (b)
between dislocations or period and δi 1), without changing the average α inter-
face dislocation density tensor [20]. The disordered configuration is obtained by
superimposing a disclination dipole network to the equilibrated network (Fig. 9.8);
its distortion field is thus the sum of the fields of the perfect and of the wall of discli-
nation dipoles randomly distributed [18]. The β d average value of the distortion
field, obtained for all the possible realizations of the disordered disclination wall and
for large distances x > h0 , is null (h0 is the distance between the dislocations in
the ordered wall). The δ 2 average deviation is calculated by considering that the
disorder is small and non-correlated, i.e. δ (m) δ (n) = 0; m = n allows to locate
the dipole and δ 2 has a finite value. Then, the standard deviation of the stress field
(which is the square root of the variance) is equal to:
[Dσ(ij) ]1/2 = [σij2 − σij 2 ]1/2 = Fij [μ2 π(1 − v)](δ 2 1/2 )(b/ h 0 ) (h 0 /x)3/2
(9.13)
Fij is a coefficient, which has been calculated for particular configurations.
Expression (9.13) is valid for all the non-null components of the stress field of
quasi-equidistant grain boundaries whatever the dislocation Burgers vector
orientation. In any case, the average elastic field remains equal to that of the strictly
periodic wall, but the fluctuations decrease proportionally to x −3/2 . Each realization
of a disordered wall leads to a stress field which weakly decreases.
The fluctuation values have been calculated on the basis of formula (9.13) and for
some particular values of Fij corresponding to physically meaningful situations. For
example, if 10−12 < b/ h 0 < 10−1 , any component of the fluctuating stress field is
superior to 2.5 × 10−4 μ at distances x of the order of 6h0 . This situation may exist
at vicinity of low angle grain boundaries.
Unified approaches, proposed by the two author groups [21, 22], have resolved the
apparent dichotomy between the elastic field evolution laws in x −1/2 or x −3/2 for a
9.2 Evolution of Extrinsic Dislocation Stress Fields 281
non-correlated disorder: the first one corresponds to a strong disorder, the second one
to a small disorder. From a physical point of view, it seems that extrinsic dislocations
formed during deformation or recrystallization arrange themselves randomly, the
disorder tends to infinite and the expression of the stress evolution varies as x −1/2 .
However, if the deformation takes place at high temperature or if the relaxation is
in a sufficiently advanced state, the disorder diminishes and an extrinsic dislocation
pseudo-periodic network exists in the grain boundary; then, the dislocation stress
field components vary as x −3/2 .
Such pseudo-periodic networks have been observed in metals [23–25] and in
ceramics [26]. In the intermediate situations that accompany the evolution of grain
boundary from a very disturbed state to equilibrium, the stress fiends likely obey
other laws.
9.2.2.2 Finite Wall of Dislocations
The authors of the previous approach [18] have also been interested by finite arrange-
ments of dislocations that better correspond to real situations. A tilt grain boundary
contains, on a part H of its length (parallel to Ox), a finite wall of n identical edge
dislocations (b perpendicular to the wall) all parallel to Oz and equidistant from h0
(Fig. 9.9). The stress field of such a configuration is that of a disclination dipole. The
remaining part of the grain boundary is described as a distribution of infinitesimal
dislocations.
Fig. 9.9 Finite tilt wall in the yOz plane formed by edge dislocations parallel to Oz with Burgers
vector b along Ox. The grain boundary is completed by a distribution of infinitesimal dislocations
The contribution of the different grain boundary parts to the σx y shear stress has been
analytically calculated. Figure 9.10 presents the evolutions of the calculated stress
for y = 0 with the distance x from the grain boundary in three cases: for an infinite
periodic wall, for a wall containing only 11-edge dislocations forming a dipole and
for one dislocation with Burgers vector 11 b. The perfect wall contribution decreases
rapidly with the distance as soon as x > 2π h 0 .
The wall behaves as a single dislocation with Burgers vector nb for distances
x > 2H . In between, the stress variation is controlled by that of the dipole; a
minimum appears for x ≈ h 0 , which results from the combination of the rapid
decrease due to the perfect wall and the rapid increase due to the dipole; the stress
also presents a maximum value for x ≈ π 2 H .

σxy stress component of a wall
containing equidistant edge
dislocations whose Burgers
vector is normal to the wall.
The stress is expressed in unit
μb/2 π h0 and the distance
is expressed in unit 2 π h0
(h0 : the distance between
dislocations); a infinite wall;
b finite wall of 11 dislocations;
c single dislocation with
Burgers vector 11b [18]
9.3 Evolution of Extrinsic Dislocation Stress Fields with Time
The question takes on a particular importance for the understanding of the progres-
sive disappearance of the extrinsic dislocation contrast on the electron microscopy
image upon the effect of thermal activation. To analyze this phenomenon which
accompanies the incorporation of extrinsic dislocations in the intrinsic network, the
evolution with time of the grain boundary long-range stresses is considered akin to
that of a finite dislocation wall, the length of which progressively increasing [14].
A tilt grain boundary contains edge intrinsic secondary dislocations equidistant
from h; periodic extrinsic edge dislocations equidistant from H = 2Nh are super-
imposed to the intrinsic network (Fig. 9.11). A high value of N corresponds to the
existence of only one extrinsic dislocation in the grain boundary.
Under the effect of the stresses due to the extrinsic dislocations and under
annealing, all the dislocations move by climbing to create a new periodic network.
Figure 9.12 shows the evolutions with the distance from the grain boundary of the
shear stress σx y for y = 0 (position of an extrinsic dislocation located midway
between two intrinsic dislocations) as the relaxation processes are going forward.
The curves 1 to 9 correspond to different dimensionless times τ of relaxation, thus
to different reorganisations of all the dislocations, from the initial disturbed state to
a periodic state. They are valid for values of N less than 100, i.e. equivalent to a low
density of extrinsic dislocations. Remarkably, for intermediate times, the curves take
the same shape than that obtained for a finite wall of dislocations (Fig. 9.10). For
small x values, the stress exponentially decreases with the distance. Then the curves
pass by a minimum and a maximum and become very close the curve for of single
dislocation as x increases. When the relaxation time increases, the maximum stress
value is displaced towards increasing x values, while its magnitude simultaneously
decreases.
This stress evolution with the distance from the grain boundary at a given tem-
perature may be compared to the evolution of the extrinsic dislocation contrast seen
as a spreading process. The evolution kinetics has been calculated and takes the
9.3 Evolution of Extrinsic Dislocation Stress Fields with Time 283
Fig. 9.11 Initial structure of a tilt grain boundary containing intrinsic dislocations equidistant from
h and extrinsic dislocations equidistant from H [14]
Fig. 9.12 Evolution with time of the curves giving the shear stress (for y = 0) in function of
the distance x from the grain boundary; this evolution is due to the reorganization of the extrinsic
and intrinsic dislocations within the grain boundary yielding stress relaxation. The curves 1 to 9
correspond to increasing relaxation times [14]
same form than that determined, by electron microscopy for the disappearance of
the extrinsic dislocation contrast. The necessary time trel for a complete relaxation
of the intergranular structure is given by a similar expression to (9.2), with H instead
of s. More precisely, H is the distance between extrinsic dislocations and thus is a
physical parameter; on the contrary, s is the image width of the dislocation just before
its contrast disappearance (s ≈ 2ξg ) and thus depends on the observation conditions.
Although, the expressions giving ts and trel are similar, the time for dislocation image
extinction does not indicate the relaxation process achievement and, more probably,
under estimates the real time. The expression of the relaxation kinetics deduced from
the incorporation model avoids using the s parameter. It leads to relaxation times of
the same order than those experimentally determined.
9.4 Experimental Studies of Extrinsic Dislocation

Accommodation
Extrinsic dislocation accommodation has been extensively studied, but most often
no correlation has been made between the results obtained at different observation
scales. The processes have been elucidated for well-characterized grain boundaries
in bicrystals by electron microscopy studies, using conventional and high-resolution
techniques, while their kinetics have been determined on polycrystals, by In situ
annealing in the microscope.
To approach the relaxation processes in real grain boundaries, we must
characterized extrinsic dislocations even when the SU/GBD model cannot describe
the grain boundary structure. How the evolution of a disturbed grain boundary
towards an equilibrium structure may really proceed by comparison to the mod-
els? What may be its real kinetics? In the following, we try to answer these questions
by describing the observed phenomena that accompany the return to equilibrium of
different grain boundaries in metals and semiconductors.
9.4.1 Accommodation in Symmetrical Tilt Grain Boundaries

in Semiconductors
High-resolution transmission electron microscopy observations enable to emphasize

the incorporation processes in the intrinsic structures of several symmetrical tilt grain
boundaries in semiconductors [27–30], but they cannot give information about these
processes in mixed tilt/twist grain boundaries and/or in those displaying a random
plane.
Expression (9.1) indicates that a deformed grain boundary returns to equilibrium
by changing its misorientation θ. If the incoming lattice dislocations integrate the
intrinsic structure by an incorporation process, the misorientation change may be
9.4 Experimental Studies of Extrinsic Dislocation Accommodation 285
related to the number of dislocations that enter the grain boundary. This has been
proved for symmetrical 110 tilt grain boundaries in bicrystals submitted to plastic
deformation, in both tension and compression, the strain axis being situated in the
grain boundary plane. The extrinsic dislocations come from the primary slip planes
of the two crystals symmetrically located with respect to the grain boundary plane.
A tension strain yields a decrease of the misorientation angle, the inverse happens
for a compression test. By referring to the Read and Shockley expression that gives
the distance between the primary dislocations in function of the misorientation angle
for a low angle tilt grain boundary (d = b/θ ), it has been shown that dN dislocations
distributed along the t unit vector in the grain boundary modify its misorientation
of dθ = b dN. The result may be extended to high angle symmetrical tilt grain
boundaries. The increase of the number of dislocations coming from both crystals is
2dN by unit length along the grain boundary. Only the edge components promote a
misorientation change, which, in case of tension test, may be written:
dθ = −2 (b · n) dN (9.14)
With n the unit vector normal to the grain boundary plane.

The dε deformation, associated to these dislocations is:
dε = (b · t) dN (9.15)
The total angular variation is:

2 (b · n)
θ = dθ = − dε (9.16)
(b · t)
Such an evolution has been confirmed for a germanium bicrystal deformed by tension
at 490 ◦ C, the initial misorientation (38.9◦ ) of the = 9 110 tilt grain boundary
decreases until 20◦ [29].
The incorporation processes in the same = 9{122} grain boundary in sili-
con have been studied in function of temperature [30]. The structure of the dis-
turbed grain boundary after 1.7 % deformation at 1120 K results in a non homo-
geneous distribution of extrinsic dislocations (Fig. 9.13), the Burgers vectors of
which being those determined at the beginning of the lattice dislocation incorpo-
ration (Fig. 8.11); the opposite glissile products are annihilated. At this temperature,
the relaxation is incomplete. On the contrary, after a 1.5 % deformation at 1470 K, a
periodic arrangement of secondary sessile dislocations are superimposed to the initial
structure leading to a misorientation change θ , analogous to a sub-grain bound-
ary in a boundary. The total relaxation may be explained by the decomposition, at
high temperature of the b30 dislocations, the glissile product annihilation and the
reorganization by climb of the sessile products.
This evolution well fits with the Nazarov et al. model [13] because the conditions
of the model are realized in the experiments: pure tilt grain boundary and sessile
Fig. 9.13 After compression of a silicon bicrystal, a more or less perfect sub-boundary superim-
poses itself to the initial = 9 grain boundary structure: a deformation at 1120 K, heterogeneous
distribution of extrinsic dislocations with DSC Burgers vectors; b deformation at 1470 K, periodic
configuration of edge dislocations that became intrinsic [30]
extrinsic dislocations parallel to the intrinsic ones. But, these conditions are rarely
fulfilled in most practical cases.
9.4.2 Accommodation in Singular, Vicinal and General Grain

Boundaries in Metals
Although less detailed and precise than the previous experiments, investigations
of metals at different scales allow us to compare the accommodation phenomena
occurring, on one hand, in different grain boundary types for a given metal, on
the other hand, in different metals for a given grain boundary. They show that the
accommodation processes differ according to the grain boundary microscopic degrees
of freedom and question the necessary occurrence of a diffusion process.
The first studies of the extrinsic dislocations relaxation performed on random

grain boundaries in metals support the delocalization model [31, 32]. Recently, the
dislocation behaviour in nickel and copper grain boundaries (singular, vicinal or
general) have been investigated by coupling several electron microscopy techniques:
conventional (bright field and weak beam), In situ (with heating and/or deformation
holder) and high-resolution, in order to get a multi-scale view of the phenomena
[8, 33–36]. Copper and nickel have been chosen owing to the difference between
their stacking fault energy that has been previously shown to strongly influence the
entrance of a lattice dislocation in a grain boundary (see Sect. 8.3.4).
9.4.2.1 Accommodation in Near = 3 {111} Grain Boundaries
A very near = 3 {111} grain boundary in a nickel bicrystal contains an intrinsic

dislocation network that accounts for the very small tilt deviation (θ = 0.09◦ ) from
the exact coincidence relationship. Two extrinsic dislocations with a lattice Burgers
vector cut the intrinsic network (Fig. 9.14a). It must be noted that this situation is
not taken into account by the existing relaxation models. Under the effect of a short
annealing at moderate temperature, the B extrinsic dislocation reacts with an intrinsic
dislocation giving rise to a short segment MN, the Burgers vector of which being the
sum of the Burgers vectors of the two dislocations (Fig. 9.14b). The evolution of the
configuration, submitted to a long thermal treatment, is followed by In situ electron
microscopy and schematically represented on Fig. 9.15. The dislocations move by
climb in the grain boundary. First, the MN segment combines with the B2 part of
the extrinsic dislocation to form a dislocation with a Burgers vector equal to that of
the A intrinsic dislocation. Then, a lattice dislocation L enters the grain boundary
and reacts with the A dislocation giving rise to a glissile dislocation that rapidly
moves towards a grain boundary extremity. Finally, after several combination and
Fig. 9.14 a Bright field electron microscopy micrograph of a very near = 3 grain boundary
in nickel containing a periodic network of intrinsic dislocations A, A , A (fine lines) cut by two
extrinsic dislocation B and B (coarse lines); the B extrinsic dislocation reacts with the A intrin-
sic dislocation to form the MN segment; b schematic representation of the configuration after
interaction [8]
Fig. 9.15 Evolution under the effect of a thermal treatment of the configuration coming from two
secant dislocations, an intrinsic and an extrinsic, in a = 3 grain boundary in nickel. Observations
under heating have been performed by In situ electron microscopy (see text) [8]
annihilation reactions, the configuration is reduced to only one extrinsic dislocation

B1 parallel to the intrinsic network. This situation corresponds to the initial one in
the incorporation model [13] that may now operate. Indeed, the residual extrinsic
dislocation decomposes into a sessile product and a glissile product that moves apart
from the sessile component by glide [8].
Thus, when an extrinsic dislocation is not parallel to the intrinsic network, the
Nazarov model [13] may in fine operate, provided first the occurrence of a series of
reactions is not predicted by the model. However, the evolution of the disturbed = 3
grain boundary after a long treatment at 0.7 Tm does not reveal complete dislocation
reorganisation in a periodic network. The stress relaxation remains incomplete, the
= 3 {111} grain boundary cannot easily return to a new equilibrium state.
A similar study concerns a very near = 3 {111} grain boundary in a copper
polycrystal [34–37]. On the contrary to the previous case, all the intergranular dislo-
cations are glissile and, in principle, may easily move in the grain boundary. How-
ever, in most real cases, their glide motion is impeded by configurations resulting
from the intersection with the grain boundary of several lattice dislocations coming
from different slip planes in both crystals. Figure 9.16 shows sequences of evolution
under in situ annealing of numerous dislocations and their corresponding schematic
representations. The isolated F et D dislocations begin to glide in the grain boundary
at 420◦ (≈0.5 Tm ) and run into an obstacle constituted by a complex dislocation con-
Fig. 9.16 Evolution of a dislocation configuration in a near = 3 {111} grain boundary in

copper: a after annealing at 0.5 Tm for 1 h, two glissile dislocations F and D come close the
complex configuration formed by intergranular glissile dislocations and lattice dislocations that
have partially entered the grain boundary; b the obstacle is overcome by the first dislocation D ; c
interaction of the mobile dislocation F with the configuration; d after the obstacle has been crossed,
the grain boundary appears less disturbed than in (a) [35]
figuration. They overcome this obstacle via node motions accompanied by numerous
reactions whose the details are not given here. After this crossing process, the grain
boundary appears less disturbed in favour of an energy decrease (compare Fig. 9.16a
and d).
Further evolution, after annealing at 0.74 Tm during two hours, does not lead
to a totally equilibrated grain boundary; some glissile dislocation arrangements are
visible, probably moving towards a grain boundary extremity (Fig. 9.17) [37].
Fig. 9.17 Arrangements of

glissile dislocations obtained
after about 2 h at 0.74 Tm in a
deformed near = 3 {111}
grain boundary in copper [37]
In both metals, copper and nickel, the twin boundaries containing extrinsic dis-
locations relax by complex reactions with the formation of discrete products that
slowly evolve to form more simple arrangements of sessile or glissile dislocations.
However, this evolution does not yield a perfect equilibrium structure, even after
maintaining the grain boundary for a long time at high temperature, up to 0.7 Tm .
The difficulty for the dislocations to reorganize themselves may be attributed to a
low grain boundary self-diffusion coefficient, close to the bulk diffusion coefficient.
Thus, it appears probable that extrinsic dislocation accommodation in = 3{111}
grain boundaries occurs by dislocation emission in the crystals, but this process
requires a high level of local stresses. In any case, twin boundaries constitute very
efficient barriers to the deformation, despite very favourable geometric factors for
transmission (see Sect. 8.3.3).
9.4.2.2 Accommodation in Near Coincidence Grain Boundaries Displaying

Intrinsic Dislocation Networks
Reactions akin to the previous ones also occur in a vicinal symmetrical Σ = 11{311}
in nickel (Fig. 9.18). As the deviation from the exact coincidence misorientation is
relatively large, the intrinsic dislocation network is very dense. The lattice dislo-
cation trapped in the grain boundary reacts with a great number of intrinsic dis-
locations yielding the formation of small discrete segments, well revealed by the
electron microscopy weak beam technique. Observed in bright field conditions, the
configuration displays a contrast widening that could be interpreted as the manifes-
tation of a spreading process. This example is significant of the possible errors due to
the observation and reinforces the necessity of multiple investigations. The discrete
configuration may be explained by the decomposition of the extrinsic dislocation
into two products that react with the intrinsic network according to the schematic
representation of Fig. 9.19. In agreement with the proposed processes, a honeycomb
network is well observed after further annealing of the thin foil (Fig. 9.18d).
Fig. 9.18 Electron bright field image showing: a an extrinsic dislocation that cuts the dense intrinsic
network in a vicinal 11 {311} grain boundary in nickel; b the evolution of the configuration
under annealing; c electronic weak beam micrograph revealing the discrete products coming from
the interaction between the extrinsic dislocation and the intrinsic ones after 40 min at 250 ◦ C;
d honeycomb aspect of the configuration after further annealing [8]
Fig. 9.19 Interpretation of the image observed after annealing (Fig. 9.18c and d): a decomposition of
the D extrinsic dislocation into two products; b and c configuration after reaction of the two products
with the intrinsic network (the NM segments are those visible by the weak beam technique)
The grain boundary maintained during 1 h at 0.4 Tm displays a quasi-periodic

arrangement of extrinsic dislocations (Fig. 9.20) prefiguring the strictly periodic net-
work predicted by Nazarov [13]. On the contrary of a near singular = 3 boundary,
this vicinal = 11 grain boundary may reach a state very close to equilibrium after
moderate annealing temperature and time.
A general near Σ = 9 {221} grain boundary in a copper bicrystal is observed
after annealing during 30 min at 300 ◦ C. It presents a network of sessile dislocations
A, accounting for the tilt deviation from the coincidence and an extrinsic dislocation
Fig. 9.20 Electronic weak

beam image showing, after
1 h annealing at 0.4 Tm , a
quasi-periodic network of
intergranular dislocations,
in favour of an advanced
relaxation state (very near
equilibrium) of a vicinal
11 {311} grain boundary in
copper [8]
B that cuts the network (Fig. 9.21) [37–39]. The near = 9 {221} grain boundary
is considered as general by reference to its high energy (Fig. 1.4). The observed
asymmetrical grain boundary contains few extrinsic dislocations. The B dislocation
displays a strong continuous contrast as observed in bright field but, it does react
with the intrinsic network when observed in weak beam conditions. Again, the
interpretation of spreading is questioned, depending on the investigation technique
accuracy. The observed configuration is explained by the decomposition of the B
dislocation into two DSC dislocations that react with each intrinsic dislocation, as
previously illustrated on Fig. 9.19. The evolution of this configuration towards a
quasi-periodic network occurs after few days at room temperature (∼0.2 Tm for
copper) (Fig. 9.22). Extrinsic dislocation accommodation in non-identified grain
boundaries in copper has already been mentioned for temperature about 240 K
(0.18 Tm ) [40].
The relaxation processes appear identical for the near = 11 {311} grain
boundary in nickel and for the near = 9 {221} grain boundary in copper, but
their kinetics differ. Moreover, return to equilibrium is better realized in the general
= 9 boundary in copper [39] than in the vicinal = 11 in nickel [8]. This kinetics
aspect will be discussed later on.
9.4.2.3 Accommodation in General Asymmetrical Grain Boundaries
Finally, the investigations of the relaxation phenomena concern general grain

boundaries (the general character is defined with respect to several degrees of
freedom) that do not present intrinsic dislocations in electron microscopy.
In nickel, an asymmetrical near Σ = 11{332} grain boundary contains two
families of extrinsic dislocations whose evolution under annealing yields a contrast
widening of their image following by their disappearance (Fig. 9.23). No discrete
product has been revealed, even in weak beam conditions, all along the relaxation
process [8].
A general grain boundary (43◦ 111) in copper was investigated in a polycrystal
under In situ tensile test at room temperature. Dislocations belonging to long pile-ups
Fig. 9.21 Observation

in transmission electron
microscopy of a B extrinsic
dislocation cutting a network
of A intrinsic dislocations in
a near = 9 grain boundary
in copper: a bright field image
showing a continuous contrast
of B; b weak beam image
revealing the products of the
reaction between the B and
the A dislocations [39]
enter the grain boundary where their contrast disappears instantaneously without any
alteration of the grain boundary fringe contrast (Fig. 9.24) [38, 39]. These reactions
suggest that spreading is the operating mechanism. But, on the contrary of nickel, it
is difficult to attribute this phenomenon to a diffusion process.
9.4.2.4 Interpretation of the Observed Relaxation Phenomena

by Referring to the Models
Before to try to understand the processes involved in the accommodation phenomena,

it appears necessary to compare the results on the reactions and contrast evolution
of the extrinsic dislocations, obtained at the microscopic scale, with the characteris-
tics of the residual defects determined at the atomic scale. The comparison is made
possible because the same grain boundaries in high purity metal are concerned; purity
Fig. 9.22 Evolution of the configuration resulting from the interaction between the B and the A
dislocations: a initial situation; b after 2 days at room temperature; a periodic network of dislocations
is being restored in the grain boundary [39]
condition is very important, as solute segregation would modify the intergranular

reactions. High-resolution transmission electron microscopy brings new and neces-
sary information to interpret the conventional electronic observations.
At the atomic scale, all the symmetrical = 3 {111}, = 9 {221}, = 11 {311}
and = 11 {332} near coincidence grain boundaries in nickel and copper display
extrinsic defects clearly identified by the presence or absence of one structural unit in
the perfect structure. These defects are very well localized, as illustrates in the near
= 11 {332} in nickel (Fig. 5.3c). On the contrary, an asymmetrical = 11 grain
boundary in nickel, similar to that observed in conventional electron microscopy
(Fig. 9.23), appears composed of asymmetrical {111} // {331} incommensurate
facets that alternate with half-periods of the {332} symmetrical grain boundary
(Fig. 9.25). Each asymmetrical facet may be considered as an extended defect with
respect to the symmetrical {332} grain boundary the procedure proposed by King
Fig. 9.23 a Weak beam electronic image showing two families of extrinsic dislocations: A (white
contrast) and B (black contrast) in an asymmetrical near = 11 {332} grain boundary in nickel; b
after annealing at 0.2 Tm during 15 min, the B dislocation are no longer visible and the A dislocation
contrast is strongly widened [8]
Fig. 9.24 a Pile-up of lattice dislocations on a general grain boundary in a copper polycrystal; b
the two leading dislocations successively enter the grain boundary without changing its contrast;
c the magnification of the entrance region image just when one dislocation incorporates the grain
boundary clearly shows the instantaneous disappearance of the dislocation contrast [38, 39]
et Smith [41] allows us to describe this facet as an extrinsic dislocation with a total
Burgers vector equal to 2 b3 (b3 = a/22 332 is a DSC vector of the = 11 grain
boundary) associated to a step, the height of which being equal to a DSC period [33].
The comparison of the results obtained at the nanoscopic scale with those deduced
from microscopic studies allows us to associate the discrete defects, observed in any
symmetrical grain boundary whatever its energy, to very localized dislocation cores.
Inversely, the contrast widening of dislocations in incommensurate grain
boundaries may be linked to the presence of defects, the cores of which being not
localized. But, whatever their core structure, all the defects possess DCS Burgers
vectors, elemental or not. The results clearly indicate that the extrinsic dislocation
relaxation does not depend on the grain boundary energy, but seems to be correlated
to its degree of periodicity. This proposal may be well explained starting with the
notion of cell of non-identical displacements (c.n.i.d.) and referring to Fig. 5.10. In
periodic grain boundaries, any delocalization of a dislocation core is impeded by a
force analogous to the friction force in crystals, the grain boundary periodicity being
given by the c.n.i.d.. The energy barriers to overcome for displacing a dislocation
in the grain boundary plane are plotted on the form of “γ surfaces” (see Fig. 4.20).
When a c.n.i.d. exists, the relaxation processes may be described by the Nazarov
model [13], although a little bit more complex in the reality. On the contrary, in a
Fig. 9.25 a High-resolution

electron microscopy image
of the structure of a general
= 11 grain boundary
composed of {111}//{331}
asymmetrical facets alternated
with symmetrical ones, the
length of which being equal
to half a period E+ DE− D of
the {332} symmetrical grain
boundary (see Fig. 5.3c); b
determination of the charac-
teristics of a defect constituted
by an asymmetrical facet
along [121]: bDSC = 2b3 and
h = 1 period of the CSL
lattice (5 cells of this lattice
are drawn in each crystals)
[33]
quasi-periodic (or even amorphous) grain boundary Grain boundary, the dislocation
core extension is easy, as it does not require any energy increase. The relaxation
is well described by the delocalization model, including the “quasi-viscous” model
[6, 11].
It is obvious that the stacking fault energy influences the entrance of a lattice
dislocation in a grain boundary, then its emission in a crystal (see Sect. 8.3.4); but,
so far, it is impossible to know its role in the relaxation phenomena. It most probably
plays a role in the decomposition and the delocalization phenomena as it favours the
dislocation core widening; we have already mentioned that an extrinsic dislocation
in a grain boundary is really equivalent to a dislocation in a crystal.
At this stage, we can raise the question about the possibility to predict the
relaxation mode for a grain boundary, knowing only its macroscopic parameters.
The two near = 11{332} grain boundaries previously studied, one symmetri-
cal and periodic, the other asymmetrical and non-periodic, possess high energies.
The first one is general with respect to its energy and singular with respect to its
plane; the second one is general with respect to all its macroscopic degrees of free-
dom. These two grain boundaries that are general according to the usual classi-
fication (see Sect. 4.6.1) possess different relaxation modes. Here, we consider as
really general the mixed tilt/twist grain boundaries, with irrational misorientation
and/or irrational plane that are submitted to relaxation by delocalization. These grain
boundaries are most probably in high proportion in materials. Until now, their fine
structure has not been observed; only, some general grain boundaries have been
modelled in terms of quasi-periodicity.
Simulations of the structures of twist grain boundaries in bicrystal and other grain
boundaries in nanocrystalline silicon [4, 5] suggest the formation of an equilibrated
amorphous phase in high-energy grain boundaries. These results, that re-honour the
Rosenhain ideas from the beginning of the twentieth century [3], could be extended
to a large majority of random grain boundaries in polycrystals.
It is relatively easy to conceive that a discrete or continuous relaxation process
operates, depending on the translation symmetry in the grain boundary plane, but
it remains to discuss the necessary (or not) occurrence of diffusion mechanisms.
Until now all the accommodation models imply grain boundary diffusion; but this is
questioned by the instantaneous disappearance of the extrinsic dislocation contrast in
a general grain boundary in copper at room temperature. This extreme rapidity can-
not be explained by taking into account the grain boundary diffusion parameters (see
Sect. 9.4.3). It might result from the quasi-periodic structure, without translational
symmetry in the plane (c.n.i.d. volume equal to zero) of the general grain boundary
(here general means with high-energy independent from any macroscopic parame-
ter). In that grain boundary , an extrinsic dislocation may be described in terms of
deformation density and the stress relaxation in terms of elastic wave propagation
[8, 42].
To conclude, whatever the mechanism responsible for the instantaneous stress
relaxation, implying diffusion or not, the phenomenon gives rise to the general grain
boundaries the role of perfect sinks for the lattice dislocations. The latter remark takes
on importance with the perspective to bring closer the observations at the microscopic
scale and the behaviours at the macroscopic scale (first deformation stage, creep
behaviour…); it may also help for studying the properties of nano-grained materials
in which the proportion of general grain boundaries is likely elevated.
9.4.3 Accommodation Kinetics
We have already seen that the kinetics of all the extrinsic dislocation accommodation
mechanisms within the grain boundaries takes the same form (9.2). By another way,
the experiments previously described show that the degree of relaxation, after a long
treatment at a given temperature, strongly differs according to the grain boundary
type, singular, vicinal or general.
Numerous studies by In situ electron microscopy allow us to emphasize the dislo-
cation contrast evolution versus time and according to the grain boundary character.
In fact, the detailed processes are not accessible; only the disappearance of the dislo-
cation contrast is observed, whatever the operating process, core widening or decom-
position in very small products. Generally, the kinetics will vary with the inverse of
the diffusion coefficient [43]. For example, the accommodation of extrinsic disloca-
tions in near = 3 {111} grain boundaries, always incomplete even after annealing
at high temperature, may be understood by the low diffusivity of the coherent twin,
almost equal to the bulk diffusivity. Reorganization of dislocations, even not totally
achieved, occurs more rapidly when the grain boundary departs from singularity.
The necessary times for the disappearance of the extrinsic dislocation contrast have
been used to evaluate the self-diffusion coefficients and the activation energy for
diffusion in different grain boundaries [43]. The results are in good agreement with
those obtained using more usual techniques.
But the instantaneity of the relaxation process observed in a general grain
boundary in copper at 293 K (0.2 TF ) is still questioned. Until now, all the inves-
tigations of the dislocations contrast widening in copper, even at low temperature
as 0.18 Tm , reveal that the phenomenon occurs progressively [40]. To estimate the
necessary relaxation time ts , we use the relation 9.2 by expressing the grain boundary
diffusion coefficient at the temperature Ts and by using the unified value 0.036 for
the A constant [16], we obtain:
kTs s 3 Q
ts = 0.036 exp (9.17)
μD0J δ RTs
with k = 1.38 × 10−23 J · K−1 , R = 8.314 J · mol−1 · K−1 , μCu = 5 × 1010 Pa, =
3 = 0.253 10−27 m3 . We take s = 100 nm, reasonable value if we take into account
bCu
experimental results already used for the same type of calculations. The characteristic
values, pre-exponential factor and activation energy, for the “average” grain boundary
diffusion in pure 99.9998 copper are: D0J δ = 1.16 × 10−15 m3 s−1 and Q = 84.75
kJ · mol.−1 , respectively [44]. With these data, the relaxation time in a random grain
boundary in copper at room temperature is about 6 h. By using diffusion data obtained
on bicrystals [45], the relaxation time in a = 9 {221} grain boundary is found of the
same order. Even by using a coefficient A lower than the previous one by two orders
of magnitude, the relaxation time remains superior to 1 min. We could also consider
that the diffusive properties vary from one region to the other along a same grain
boundary, for example, a local increase of the diffusion coefficient might occur in
the “quasi-fluid” islands [11]. Whatever the hypothesis, the instantaneous relaxation
in the general grain boundary of copper is opposed to the occurrence of diffusion.
Alternately, we may consider that the local reorganization in a grain boundary is not
necessarily described by macroscopic values as D J δ, related to long-range diffusion
measurements.
In the cases where diffusion seems to be the controlling mechanism for stress
relaxation kinetics, the latter strongly depends on the intergranular segregation
level, itself linked to the misorientation [46] and to the grain boundary plane [47].
Surprisingly, for several near = 11 grain boundaries in nickel, the time for extrin-
sic dislocation accommodation increases when the effective interplanar spacing deff
decreases (Table 9.1), in contradiction with the hypothesis that the atomic density
increases with deff (see part 1). In fact, the relaxation time augmentation is attributed
to the diminution of the grain boundary diffusion coefficient under the influence of
segregation, the latter being stronger as the atomic density decreases (deff small).
Moreover, for a given grain boundary , the kinetics varies with the dislocation line,
as revealed by the contrast disappearance times for two extrinsic dislocation families
in = 11 in nickel (Fig. 9.26, Table 9.1). This remark supports the existence of an
intrinsic anisotropy of the grain boundary diffusion, more rapid along the 110 tilt
axis than along any other directions [48]. It may also reveal a segregation anisotropy
linked to differences in the elastic energies of interaction between the solute and
different dislocations.
Fig. 9.26 Contrast

disappearance times td (s)
for extrinsic dislocations
in function of the effective
interplanar spacing, during a
thermal treatment at 250 ◦ C of
different grain boundaries in
nickel [47]
The extrinsic dislocation accommodation kinetics has been evaluated for grain
boundaries in metals and alloys, but differences have also been found between grain
boundaries in alumina, yielding different behaviours under high temperature defor-
mation [26]. Extrinsic dislocations are only observed after creep in certain near
Table 9.1 Contrast disappearance times td (s) for two families of extrinsic dislocations in nickel:
X parallel or quasi-parallel to 112 and Y parallel to 011 in grain boundaries near = 11
(deviation θ ◦ ) [47]
θ ◦ Grain boundary plane deff /a Segregation level Dislocation line td (s) at 250 ◦ C
0.50 {113} // {113} 0.300 Weak 112// 112 40
011// 011 320
2.20 {112} // {117} 0.170 Medium 123// 134 120
2.20 {116} // {112} 0.145 High 257// 123 160
011// 011 1260
2.10 {225} // {114} 0.100 High 347// 235 630
011// 011 5000
coincidence grain boundaries in a magnesia-doped alumina; these grain boundaries

seem to keep the memory of processes occurring in all grain boundaries [49], allowing
us to precise the creep/grain boundary dislocations model [50]. Addition of yttrium
in alumina leads to a decrease of the grain boundary self-diffusion coefficient, that
modifies the stress relaxation processes and, finally, affects creep and sintering of
alumina polycrystals [49]. Like in metals, a predominant influence of the chemistry
on the grain boundary stress relaxation appears in ceramics.
9.5 Conclusion on the Extrinsic Dislocation Relaxation

Phenomena
Even without considering the quantitative results on extrinsic dislocation accom-

modation, the importance of the phenomenon must be pointed out, whatever the
material. Indeed, the large majority of the grain boundaries in real materials are in a
non-equilibrium state, although only a limited number of them display extrinsic dis-
locations. The non-visualization of intergranular defects does not mean the absence
of residual stresses within a grain boundary. The non-equilibrium degree of a grain
boundary may appreciably affect its properties (diffusion, segregation, migration,
grain boundary sliding, electric conductivity…). The defect relaxation mode plays
an important role in the grain boundary responses to various stimuli. It is thus nec-
essary to improve our knowledge on the relaxation of general grain boundaries. The
mechanisms that allow the intergranular defects to play on material properties are to
be established.
In polycrystals, the complete grain boundary relaxation involves phenomena at
triple junctions. However, until now, we have only considered a grain boundary free at
its extremities; we have just evoked the existence of stress concentration at junctions.
A better knowledge of the role of grain boundaries in the material properties implies
to approach the structure and the defects of grain boundaries “constrained” at triple
junctions.
Finally, to attempt to relate the elementary mechanisms to the macroscopic laws,
we must study the grain boundaries at the mesoscopic and macroscopic scales; in
particular, we must establish their global and spatial distributions in polycrystals.
These preoccupations constitute a crucial step towards “Grain Boundary Engineer-
ing”. We approach these questions in part III of this book by considering grain bound-
ary networks, since the triple junction until the complex grain boundary ensemble in
real materials.
References
1. L. Priester, Interface Sci. 4, 205 (1997)

2. L. Priester, Mater. Sci. Eng. A 309–310, 430 (2001)
3. W. Rosenhain, D.J. Ewen, J. Inst. Metals 8, 149 (1912)
References 301
4. P. Keblinsky, D. Wolf, S.R. Philipot, H. Gleiter, Phys. Rev. Lett. 77, 2965 (1996)
5. P. Keblinsky, D. Wolf, S.R. Philipot, H. Gleiter, Phys. Lett. A 226, 205 (1997)
6. W. Lojkowski, H.O.K. Kirchner, M.W. Grabski, Scr. Metall. 11, 1127 (1977)
7. W. Lojkowski, M.W. Grabski, in Déformation of polycrystals: Mechanisms and microstructures
ed. by N. Hansen, A. Horswell, T. Leffers, H. Lilholt, Riso Nat. Lab., Roskilde, Denmark (1981),
p. 329
9. M.W. Grabski, J. de Physique 46, C4–567 (1985)
10. W. Swiatnicki, W. Lojkowski, M.W. Grabski, Acta Met. 34, 599 (1986)
11. V.N. Perevezentsev, V.V. Rybin, V.N. Chuvil’deed, Acta Metall. Mater. 40, 887 (1992)
12. R.Z. Valiev, V. Yu Gertsman, O.A. Kaibyshev, Phys. Stat. Sol. (a)78, 177 (1983)
13. A.A. Nazarov, A.E. Romanov, R.Z. Valiev, Acta Metall. Mater. 41, 1033 (1993)
14. A.A. Nazarov, A.E. Romanov, R.Z. Valiev, Scr. Metall. 24, 1929 (1990)
15. A.A. Nazarov, Interface Sci. 8, 315 (2000)
16. A.A. Nazarov, Interface Sci. 8, 71 (2000)
17. A.A. Nazarov, Phil. Mag. A 69, 327 (1994)
18. G. Saada, E. Bouchaud, Acta Metall. Mater. 41, 2173 (1993)
19. E.E. Zacimchuk, S.I. Selitser, Sov. Phys. Solid. St. 26, 695 (1984)
20. E. Kröner, Physics of Defects (North Holland, Amsterdam, 1981), p. 215
21. G. Saada, D. Sornette, Acta Metall. Mater. 43, 313 (1995)
22. A.A. Nazarov, A.E. Romanov, B. Baudelet, Phil. Mag. lett. 5, 303 (1993)
23. S. Lartigue, L. Priester, Acta Metall. 31, 1809 (1983)
24. R.J. Kurtz, R.G. Hoagland, J.P. Hirth, Phil. Mag. A 79, 683 (1999)
25. W.A. Swiatnicki, S. Poulat, L. Priester, B. Décamps, M.W. Grabski, Acta Mater. 46, 1711
(1998)
26. S. Lartigue-Korinek, F. Dupau, Acta Metall. Mater. 42, 293 (1994)
27. M. Elkajbaji, J. Thibault-Desseaux, Phil. Mag. A58, 325 (1988)
28. J. Thibault-Desseaux, J.L. Putaux, A. Bourret, H.O.K. Kirchner, J. Phys. 50, 2525 (1989)
29. J.J. Bacmann, M.O. Gay, R. de Tournemine, Scr. Metall. 16, 353 (1982)
30. J. Thibault, J.-L. Putaux, A. Jacques, A. George, M. Elkajbaji, Microsc. Microanal. Microstruct.
1, 395 (1990)
31. P.H. Pumphrey, H. Gleiter, Phil. Mag. 30, 593 (1974)
32. R.A. Varin, J.W. Wyrzykowski, W. Lojkowski, M.W. Grabski, Phys. Stat. Sol. (a) 45, 565
(1978)
33. S. Poulat, J. Thibault, L. Priester, Interface Sci. 8, 5 (2000)
34. L. Priester, J-P. Couzinié, B. Décamps, J. Thibault, in Proceedings of the 25th Risö International
Symposium on Material Science. "Evolution of Déformation Microstructures in 3D" (2004) p.
79
35. J.-P. Couzinie, B. Décamps, L. Boulanger, L. Priester, Mater. Sci. Eng. A 400–401, 264 (2005)
36. J.-P. Couzinié, B. Décamps, L. Priester, Int. J. Plast. 21, 759 (2005)
37. L. Priester, J.-P. Couzinié, B. Décamps, Adv. Eng. Mater. 12–10, 1037 (2010)
38. J.-P. Couzinié, B. Décamps, F. Pettinari-Strurmel, L. Priester, in Proceedings of the Interna-
tional Conference Copper’06 in "Better properties for Innovative Products" , France 2006, ed.
by J.M. Welter
39. L. Priester, J.-P. Couzinié, B. Décamps, S. Lartigue-Korinek, Int. J. Mater. Res. 101(10), 1202
(2010)
40. K.J. Kurzydlowski, W. Zielinski, J. Wyrzykowski, Mater. Sci. Technol. 2, 420 (1986)
42. C. Solenthaler, Phys. Stat. Sol. (a) 149, 21 (1995)
44. T. Surholt, Acta Mater. 45, 3817 (1997)
45. A. Suzuki, Y. Mishin, Interface Sci. 11, 131 (2003)
46. B. El M’Rabat, L. Priester, Mater. Sci. Eng. A 125, 31 (1990)
47. W.A. Swiatnicki, S. Poulat, L. Priester, B. Décamps, M.W. Grabski, Acta Mater. 46, 1711
(1998)
48. M. Biscondi, Physical Chemistry of the Solid State: Application to Metals and their Compounds,
ed. by P. Lacombe (Elsevier, Amsterdam, 1984), p. 225
49. S. Lartigue Korinek, C. Carry, L. Priester, J. Eur. Ceram. Soc. 22, 1525 (2002)
50. R.C. Pond, D.A. Smith, P.W.J. Southerdern, Phil. Mag. A 77, 27 (1978)
Part III
From the Free to the Constrained
Grain Boundary
La vraie force de l’esprit se mesure au degré d’incertitude
qu’il est capable de surmonter.
(F. Nietzsche)
One can appreciate the true strength of mind according to the

amount of uncertainty it is able to overcome.
(F. Nietzsche)
Until now we have described an ideal grain boundary, considered infinite in extent,
then a faulted grain boundary, localized in a bicrystal where the intergranular
stresses may be relieved at the boundary extremities abutting on free surfaces. But
the final goal in materials science is the understanding of the engineering
polycrystal properties, and in this context grain boundaries are not infinite but
terminated on triple junctions. Moreover, the stresses in each boundary do not
easily relax at the junction, as it was the case in bicrystals. Grain boundaries in a
polycrystal cannot be treated as isolated bicrystal pairs. The assemblage of grains
constitutes a connected system of grain boundaries.
Two questions of interface science, essential from a practical point of view,
must now be discussed:
- How a grain boundary may be connected to its neighbours? How is the triple
junction structure?
- How are the grain boundaries distributed in a polycrystalline network, accord-
ing to their geometrical characteristics or to one (or many) of their properties?
All non-exhaustive answers that we can bring to each of the previous questions
have not been reviewed in a book until now; they constitute the two chapters of
Part III, whose main purpose is the transition from bicrystal to polycrystal.
It is well known that the response to a given stimulus of a crystal included in a
polycrystal differs from that of a single crystal with the same crystallographic
orientation. Similarly, the behaviour of a grain boundary must differ depending on
whether it is free in a bicrystal or constrained at a triple junction (or multiple
junction) in a real material. Knowledge of the properties of grain boundaries
included in a polycrystal is only useful if it enables to go back to global properties.
But, the properties of a whole may never be easily inferred from those of its
elements.
We may argue that the crystalline texture plays an important role in numerous
polycrystal properties, such as deformation and recrystallization, as well as in
304 Part III: From the Free to the Constrained Grain Boundary
physical properties (magnetic properties of Fe-Si alloy sheets, for example). In the
same way, before analyzing the influence of a grain boundary network on any
property, we have to characterize its organization in a polycrystal i.e. we have to
determine the grain boundary texture. Knowledge of the junctions between grain
boundaries and their distribution are even more crucial that the fundamental role of
grain boundaries in the material properties is widely recognized. Presumably this
role increases as the grain size decreases, however, the extension of knowledge on
the grain boundary structures and behaviours in nano-crystalline materials gives
rise to much controversy.
Questions concerning grain boundary ensembles are discussed for many years,
since the idea proposed by T. Watanabe in 1984 to control the grain boundary
distribution in polycrystals in order to control the material properties, an idea
known as Grain Boundary Design or Grain Boundary Engineering. Behind the
idea of grain boundary engineering, pierces the old reductionist concept of
analyzing the behavior of a system based on those of its constitutive elements. The
method prevails in science, for many systems, but the discovery of the chaos
theory reveals the existence of some systems whose complex behaviour arises
from nonlinear interactions of a very small number of constituents. To what extent,
a property of a material is the classic view or not? What are the relevant degrees of
freedom of the system? a poorly understood phenomenon occurring in the material
can meet the deterministic chaos? The answers to these questions are still in the
infancy of materials science. Therefore, waiting to better tame the disorder of
crystalline materials out of equilibrium, we adopt here a traditional approach.
The responses to the current wide spread problem of the role of the grain
boundary ensemble in the overall material behavior rest on many approximations
and simplifications in terms of fundamental knowledge about the subject ‘‘grain
boundary’’. Therefore we limit ourselves in each of the two chapters of Part III, the
triple junction and the grain boundary network, to give information elements to
progress in the direction of engineering without drawing firm conclusions. Real
opportunities exist for the use of grain boundary knowledge for practical purposes,
substantial efforts can and should be made in this direction.
Chapter 10
The Triple Junction
The triple junction is a one-dimensional defect that appears as soon as we consider a

system of connected grains and grain boundaries forming the microstructure of most
crystalline materials used in the industry; in that sense, it may be seen as the elemental
configuration of a polycrystal. Although it is well known to influence most of the
material properties, it remains a microstructural defect that is little studied compared
to grain boundary. It may have opposite effects on the material properties: obstacle
to the deformation at low temperature, preferential path for corrosion, for wetting or
for cavitation under creep.
In a manner similar to that adopted in the approach of the grain boundary order
(see Chap. 1), the triple junction is discussed successively in three ways:
• Its geometry: degrees of freedom and crystallography of the boundary junction
• Its equilibrium structure in terms of stress order, then in terms of atomic arrange-
ments
• Its energy, determined by calculations or by experiments
Then, the main defects occurring at triple junction are described. Finally, the stage
that enables to go from tricrystal to polycrystal is briefly considered.
The triple junction properties are not discussed here; much more than the grain
boundary properties, they are controversial and are ultimately very little known
despite the fact that they should likely have an important contribution to the
properties (especially transport properties) of polycrystals. Their knowledge appears
as a prerequisite to go towards grain boundary engineering [1].
10.1 Triple Junction Geometry
10.1.1 Geometrical Parameters and Triple Junction Classification
A triple junction is defined by the confluence of three crystals or by the conflu-

ence of three grain boundaries. In general, 11 independent macroscopic geometrical

306 10 The Triple Junction
parameters are required to describe a triple junction: 6 for the grain-to-grain rota-
tions (3 × 2 as the third one is not independent), 5 for the position of the three planes
abutting at the junction on the form of a line. If the triple junction is equilibrated, the
product of the three rotation matrices, expressed by reference to a same coordinate
system, must be equal to the identity matrix. A simple combination rule of the three
coincidence indices then applies to the triple junction between three coincidence
grain boundaries:
1 ·2 = d 2 ·3 (10.1)
with 3 the coincidence index of the boundary 3 and d a common divisor to 1

and 2 .
If a junction displays two grain boundaries, = 3 and = 9, the third boundary
may have two indices: 27 or 3, depending on the d value d = 1 or 3 (3 × 9 = 12 × 27
or 3 × 9 = 32 × 3). In principle, the combination rule is always valid for cubic
materials, as a coincidence relationship always exists between crystals. In that case,
the DSC lattices of two boundaries are sub-lattices of the DSC lattice of the boundary
with the higher coincidence index. Practically, the consideration of a limit value
of ( < 25 or 41) restricts the application of this rule to three experimental
misorientations equal or very close to exact coincidence misorientations. Otherwise,
a rule on the allowable coincidence deviations must complete the rule (10.1) in order
to define the triple junction; this issue becomes important when classifying the triple
junction in polycrystals (see Sect. 11.7.1).
Each boundary involved in a triple junction possesses N equivalent descriptions
linked by the bicrystal symmetry (see Sect. 1.1.2); as a result a triple junction displays
N3 equivalent descriptions (N3 = (242 )3 = 13824 for the cubic systems. Similarly
to the notion of grain boundary disorientation, a set of three angles θ1 , θ2 and θ3
and one rotation axis uvw may represent the set of N3 triple junction descriptions.
This leads to distinguish triple junctions possessing a common crystallographic axis
from those where the triple line is random with different indices in the three grain
boundaries. Akin to the junctions between two crystals that most often are random
(irrational misorientations, asymmetrical planes) in real material, the triple junctions
are generally random in polycrystals. The existence of a common axis reduces to 9 the
number of macroscopic geometrical parameters required to define a triple junction.
Some triple junctions possess more than one possible common axis, for example
311, 210 and 211 for the junction = 3, = 5, = 15. Currently, most
studies concern only the triple junctions with a common crystallographic axis.
On the basis of the coincidence concept, triple junctions may be classified into
three types: general junction between three general boundaries, mixed junction
between general and coincidence (or near coincidence) boundaries and special junc-
tion between three special boundaries [2]. Note immediately that the junction between
three grain boundaries can be random if these boundaries have no common rotation
axis ( = 3, = 13a, = 39b, for example). The term special used in this con-
text has a very restricted meaning, as the grain boundary plane orientation is not
taken into account. This does not imply any special grain boundary property and,
10.1 Triple Junction Geometry 307
a fortiori, of the triple junction where it ends. However, we must mention that grain
boundary sliding is impeded at certain triple junctions between three coincidence
grain boundaries in fine-grained alumina submitted to superplastic deformation [3].
The limitations of the concept of speciality applied to grain boundaries are also found
in the classification of a triple junction that is truly “special” only if it has special
properties.
Another distinction between two types of triple junctions named I and U, proposed
by Bollmann [4], rests on the consideration of a preferential description so-called
Nearest Neighbour Relationship (NNR) among the equivalent descriptions of the
grain boundary structure (see Sect. 2.3.3). It raises a fundamental question on the
description of grain boundaries in terms of dislocations and will be discussed later
(see Sect. 10.2.2).
On the basis of the usual distinction between grain boundaries (see Sect. 4.6.1),
a new triple junction classification is proposed: singular junction between three
singular grain boundaries, vicinal junction between vicinal or singular and vicinal
boundaries, general junction as soon as one grain boundary implied in the junction
is general. Singular or vicinal junctions, such as = 3 − = 3 − = 9 and
= 3 − = 9 − = 27, have often been observed in f.c.c. metals with low
stacking fault energy; more generally, the coincidence indices of the three boundaries
abutting at the singular junction are on the form = 3n .
In this chapter, we essentially deal with the triple junctions having a common
crystallographic axis in materials of cubic symmetry. We focus on the high symmetry
junctions 100, 110 or 111 and, among them, the symmetrical junction between
three symmetrical boundaries; these junctions likely possess particular properties.
10.1.2 Tricrystallography
As triple junction is the meeting of three crystal lattices, the approach of the symme-
tries of the configuration is first needed. The tricrystallography predicts which are
the elements of the point group symmetry of a tricrystal. It is based on the theory
of the three-coloured symmetry groups [5] that is an extension of the dichromatic
groups, the usefulness of which has been reported for the description of the grain
boundary structure; but the triple junction crystallography is much less developed
than the bicrystallography. Its application is often restricted to the bi-dimensional
point symmetry allowing the approach of triple junctions in thin films. In the absence
of local relaxations, it is likely that the symmetry elements of the tricrystal tend to
be preserved during the formation of a triple junction. This hypothesis rests on the
Neumann principle of symmetry dictated extrema [6]. This type of arguments can
be used to locate the geometric variables describing the triple junctions, which by
symmetry should lead to extrema in the properties, such as energy that is minimum
at equilibrium [7]. Figure 10.1 presents à tri-chromatic pattern consisting of three
interpenetrated f.c.c. crystals, turned against each other by 30◦ rotation or equivalent
(60◦ , 120◦ or 150◦ ) around a 001 axis [8]. Two types of symmetry operations are
Fig. 10.1 Three-coloured pattern for an f.c.c. material from which triple junctions of 001 rotation
axis are generated. Only atoms of two layers are shown with different symbols according to the
crystal to which they belong: • for crystal I, ◦ for crystal II and for crystal III. Full and dashed
lines correspond to classical and coloured mirror planes, respectively (see text) [8]
clearly visible in the figure: the classical operations that transform a crystal in itself
(mirror planes whose traces are drawn in full lines) and the coloured operations for
which a crystal of a given colour is transformed into a crystal of another colour
(dashed lines).
To generate a tricrystal with an expected point symmetry, we superimpose to
the three-coloured pattern an operator Y formed by three branches separated by
120◦ angles, each branch representing a grain boundary. Then we rotate the operator
around the rotation axis in order to align its branches on the symmetry elements that
are desired for the three-coloured object; finally we put all the points of same colour,
and uniquely, inside a region located between two branches. Two triple junctions of
different symmetries then obtained are schematized in Fig. 10.2: one is a junction of
point symmetry 3m with a three-fold symmetry axis, each plane is a mirror for the
whole configuration; the other displays a point symmetry m, with only one mirror
plane [8].
Observations by high-resolution transmission electron microscopy of triple
junctions in aluminium thin layers, deposited on a silicon single crystal substrate
with (111) orientation, illustrate the symmetry role [9]. The epitaxial relationship
10.1 Triple Junction Geometry 309
Fig. 10.2 001 triple junctions, of symmetry 3m (a) and m (b), are obtained by superimposing
an operator Y to the three-coloured patterns and placing atoms in their respective crystal. For the
symmetry descriptions, see text. The mirror plane in case (b) is the II/III plane. Only two atom
layers are represented (symbols of different sizes). Compared to Fig. 10.1, atoms of crystal III are
represented here by empty circles surrounded by bold lines [8]
between the substrate and the aluminium film leads to the formation of three {100}
orientation variants of the aluminium grains. The 111 three-fold axis is preserved
as symmetry axis for the whole tricrystal. The facets adjacent to the junctions are
100 tilt grain boundaries, symmetrical or not, whose planes tend to be parallel to
the {100} or {110} planes of the third crystal. When the boundaries are symmetrical,
two types of junctions can be observed (Fig. 10.3):
• The symmetrical triple junction possesses 120◦ dihedral angles. It is formed by
three identical grain boundaries, the plane of each of them being parallel to {100} or
{110} of the opposite crystal. These planes are mirror planes for the tricrystal. This
is the case of the junction TS , equivalent to the 3m point symmetry configuration
described in Fig. 10.2a.
• The asymmetrical triple junction, such as TA , has two 105◦ and one 150◦ dihedral
angles. Only one plane is parallel to the {001} plane of the opposite crystal, the two
others are parallel to a {110} of an adjacent crystal. Only the first plane is a mirror
plane for the whole configuration. Generally, if a boundary between two grains is
a mirror plane for the third grain of the junction, this boundary is asymmetrical.
Due to the epitaxy between the thin film and the substrate, all the grain
boundaries have the same misorientation of 120◦ or 30◦ (equivalent rotations by
symmetry around [100]). Practically, the misorientation of the grain boundary paral-
lel to a {110} mirror plane of the tricrystal slightly differs from 30◦ , in order to adopt
the periodic structure of the nearest coincidence boundary, for example, that of the
= 53 (θ = 31.9◦ ) {072} boundary. Its period is then alternately composed of one
unit of = 5 (θ = 36.9◦ ) and one unit of = 17 (θ = 28◦ ), in agreement with the
structural unit mode [10]. Other parts of the same type of boundary have a misori-
entation of 30.5◦ 100 corresponding to a coincidence = 65 and a symmetrical
{074} plane.
Fig. 10.3 Image in high-resolution transmission electron microscopy of two adjacent triple junc-
tions between aluminium grains deposited on a (111) single crystal silicon substrate and rotated
with respect to each other from 30◦ around 100. The junction TS presents a 3m symmetry with
three equivalent grain boundaries rotated from 120◦ , each boundary plane being parallel to a {110}
plane of the opposite crystal. The junction TA presents a dihedral angle of 150◦ and two of 105◦
separated by a grain boundary parallel to a {100} plane of the opposite crystal; the latter plane is
the only mirror plane of the triple configuration [9]
Note that twist deviations may occur during the aluminium layer growth, grain
boundaries being no longer perpendicular to the silicon surface. It is remarkable
that the symmetry ruptures observed in the aluminium thin layer only occur when
the three-fold symmetry axis is absent; this indicates a energy minimum for the 3m
symmetry junctions, type TS , deeper than that of the junctions type TA with only one
mirror plane m. Thus, crystallographic arguments seem to explain the configuration
of the triple junction between grain boundaries in thin layers in epitaxy with a single
crystal substrate. Their extension to three-dimensional materials raises numerous
questions relating to the triple junction equilibrium, since the consideration of the
dihedral angles at the mesoscopic scale to the atomic description of the junction,
going through the connection, at the microscopic scale, of the intrinsic dislocations
of the three boundaries forming the junction.
10.2 Triple Junction Equilibrium
A large dichotomy exists between the mesoscopic and microscopic approaches of the
triple junction equilibrium, taking into account the average grain boundary parame-
ters, and the nanoscopic description that considers the atomic structures of the three
grain boundaries in the vicinity of the triple junction. These different approaches are
10.2 Triple Junction Equilibrium 311
discussed below; the relevance of each of them to describe the equilibrium at the
junction is faced to experiments as much as possible.
10.2.1 Thermodynamic Approach: Limits
10.2.1.1 Herring’s Equilibrium Rule
The equilibrium condition between grain boundaries at a triple junction has been
derived by Herring [11]:
3 3
σi ti + (∂σi /∂αi ) ti∧n = 0 (10.2)
i=1 i=1
where σi is the interfacial tension of the grain boundary i, ti is a unit vector normal
to the junction in the boundary i, n is a unit vector along the junction and αi is an
inclination angle of the boundary plane i with respect to a given reference. Note
that in part I of the book, the inclination angle (named ϕ) was measured from the
initial position of the grain boundary with respect to the free surface. The first term in
expression (10.2) is equivalent to the tangential force FT that acts on each boundary
when the interfacial tension is independent of the boundary plane inclination; in that
case, we have seen (see Sect. 4.2) that the tension parallel to the boundary plane is
equivalent to the boundary energy γ . The balance of the interfacial energies γi at an
equilibrated triple junction (Fig. 10.4a) is thus given by:
γ1 γ2 γ3
= = (10.3)
sin α1 sin α2 sin α3
If the energy of a grain boundary is function of its misorientation, a force FN

normal to the boundary plane acts on its extremity resulting in its rotation towards
a lower energy position. This is the origin of the second term in expression (10.2).
If the position of one of the grain boundaries abutting at a triple junction is fixed,
this force provokes a motion of the junction from the position O to P (Fig. 10.4b);
formulae (10.2) then applies by considering the angles β and δ measured by referring
to the fixed boundary (β + δ = α1 ). No torque term acts on a singular grain boundary
when its energy is at a minimum on the curve γ = fn(α). However, a force larger
than FN acting on the boundary induces its displacement.
Note that normal forces must exist at triple junctions between symmetrical tilt
grain boundaries such as those observed in an aluminium thin film [9]. The stability
of the junction may be only established by comparing the geometry derived from the
symmetry to that predicted by the boundary energy balance. Such a junction is stable
only if the energy of each boundary strongly depends on its plane inclination, favour-
ing symmetrical planes. It is then submitted to a twist-torque in order to be stabilized
and possesses an elastic strain field. Other configurations (or classes) of symmetrical
Fig. 10.4 a Balance of the

interfacial tensions at a triple
junction in a metastable equi-
librium state. b Rotation of
two grain boundaries of a
junction under the effect of
the normal forces, the third
boundary (between grains II
and III) being considered as
fixed
junctions around 001 have been predicted on strictly geometric bases and the forces
acting on some of these junctions in aluminium have been calculated [12].
Formula (10.3), similar to that which governs the junction of a grain boundary with
a free surface, enables to determine the relative energies of three grain boundaries
at a junction, but only in cases where the energy of each boundary weakly depends
on its plane orientation. However, this formula is often used by neglecting the exis-
tence of a torque term. In particular, it cannot be applied to the junctions between
coherent and incoherent twins, often present in f.c.c. metals. Such negligence may
lead to serious contradictions between the ratios of the interfacial energies deduced
from the dihedral angles observed at the junction and those obtained by using the
energies calculated from the geometric grain boundary parameters; furthermore, the
studies of polycrystals only refer to the coincident (or not) character of the grain
boundaries. But we know that no reciprocal relationship exists between coincidence
and energy. Therefore, we must use with caution any conclusion, even qualitative, on
grain boundary energies deduced from triple junction configurations in polycrystals

and only based on a coincidence criterion (see Sect. 11.7).
Note that quadruple junctions (not to be confused with quadruple node) may exist,
as a transitory state, during grain growth; they are generally unstable and dissociate
into two triple junctions [13].
10.2.1.2 Observation of a Triple Junction at Different Scales: What

Equilibrium?
Consider four triple junctions between singular and/or vicinal = 3n grain

boundaries, classified on the basis of their misorientation (Fig. 10.5) [14]. These
junctions have been observed in a thin plate of a high purity nickel polycrystal, after
annealling such that the grain size is larger than the plate thickness and the grain
boundaries are perpendicular to the two free surfaces. They are especially interest-
ing to analyze because the twin boundaries are well known for their violation of the
Herring rule and residual stresses can be concentrated at junctions between multi-
ple twins. In the case (a), the rule (10.3) seems to be verified semi-quantitatively:
it leads to an extremely low energy for one of the = 3 boundaries, likely with a
{111} plane; the energy of the other = 3 (probably an incoherent twin) is larger
than previously but lower than that of the = 9 boundary, which indeed is a high
energy boundary in metals. Its application to example(c) gives a lower energy for the
= 27 grain boundary than for the = 9, in agreement with the relative energies
of the 110 tilt boundaries in f.c.c. metals. However, applying the rule (10.3) leads
to equal energy for the = 3 and = 27 boundaries; this is impossible for two
boundaries close to singularity. The dihedral angles in cases (b) and (d) cannot be
explained by considering only the boundary misorientations. The no-relevance of
the rule (10.3) cannot be attributed to a modification of the interfacial energies under
the effect of segregation, nickel used in this study being very pure. It indicates that
torque terms must be included in the energetic balance.
It is expected that the knowledge of the average boundary planes enable to
better approach the triple junction equilibrium. In this perspective, all the geometrical
parameters of the two = 3 and the = 9 boundaries have been determined by
transmission electron microscopy (Fig. 10.6) [14]. The order of the energies seems
to be respected semi-quantitatively. The values of the dihedral angles, compared
to those measured at the mesoscopic scale, are closer the values imposed by equi-
librium: value largely superior (150◦ ) for the angle opposite to the boundary 1/2
( = 3 {111}) and similar values for the two boundaries 1/3 and 2/3 whose energies
are expected to be comparable. We suggest that the boundaries in a thin foil rotate
more easily than in a thick sample under a torque effect, so they tend towards a more
stable position. However, if we apply Eq. (10.3), the energy of the coherent twin
1/2 is only about half those of asymmetrical boundaries with high index planes, in
contradiction with the usual data on the relative energies.
Note that the grain boundary plane observed at the microscopic scale is often
faceted and differs from the mean macroscopic plane only relevant for the application
Fig. 10.5 Four junctions of = 3n grain boundaries in a high purity nickel polycrystal. Rule
(10.3) is semi-quantitatively respected in cases (a) and (c) but not in the two other cases [14]
Fig. 10.6 a Electronic image

of a triple junction = 3 −
= 3 − = 9 observed in a
high purity nickel polycrystal;
b schematic representation
that specifies the mean grain
boundary planes [14]
of a formula giving a thermodynamic equilibrium. Furthermore, a constrained grain

boundary likely adopts a structure that differs from that of the free boundary in a
bicrystal; in particular, the triple junction may impede the rigid body translation
that leads to the energy minimization. Finally, the presence of defects in one of the
boundaries, at least, may yield changes of the dihedral angles.
Fig. 10.7 Image in high-resolution transmission electron microscopy of a triple junction =

3 − = 3 − = 9. The {111} planes of the three crystals and the two = 3 grain boundary
planes are perpendicular to the image surface; the = 9 grain boundary plane is slightly inclined
with respect to this surface. The {111} boundary planes seem to be favoured in the close vicinity of
the junction implying dihedral angles of 109.5◦ on each side of a strictly coherent twin (vertical on
the figure) [14]
To go further in comparison between triple junction configuration and grain

boundary characteristics, a = 3 − = 3 − = 9 junction is observed at
the nanoscopic scale (Fig. 10.7). We note that the grain boundary planes reorient to
be parallel to {111} in each crystal (except for = 9 where the plane is parallel to
{111} in only one crystal); this requires two dihedral angles of 109.5◦ and one of
141◦ in front of a strictly coherent twin. The dihedral angles measured at this scale
cannot obviously provide information on the relative grain boundary energies. The
plane of the second twin displays steps in the vicinity of the junction. A pretty good
matching of the {111} planes is observed at the junction, despite a slight inclination
of the = 9 boundary with respect to the incident electron beam.
A similar configuration has also been observed by transmission electron micro-
scopy at the triple junction of garnet crystals (Y3 Al15 O12 ) welded such that the
dihedral angles of 90◦ and 180◦ give the initial tricrystal a T-shape (Fig. 10.8). The
initial boundary planes {112}I/II , {110}I //{112}III and {111}II //{112}III have turned
during the diffusion annealing to form two dihedral angles close to 105◦ and 115◦ ,
the third one being about 135◦ –140◦ . The grain boundary curvatures to adopt the
boundary configuration at the triple junction extend on about 10 μm. Beyond, the
boundaries are straight, they follow the T- profile initially defined and their structures
Fig. 10.8 a Formation of a garnet tricrystal starting from a bicrystal and a single crystal; b image
in transmission electron microscopy of the triple junction after annealing under vacuum at 1800 ◦ C
and under a compression stress of 1 MPa; the boundaries GB1 and GB2 as well as GB1 and GB3
form dihedral angles of about 110◦ , the angle between GB2 and GB3 is about 140◦ ; c image in
high-resolution transmission electron microscopy of the configuration [15]
are those of the corresponding boundaries in bicrystals. No trace of glassy phase or

porosity has been detected at the triple junction by high-resolution transmission
electron microscopy [15].
These triple junction observations at different scales raise a fundamental question:
what governs the configuration of the connection of three grain boundaries? Is it
the consideration of the total energy of the system or the local formation of low
energy planes (large interplanar distance) or even the realization of a symmetrical
configuration?
The fact that the dihedral angle rule is not sufficient to predict equilibrium has been
raised by the crystallographic approach and revealed by microscopic observations
of the triple junction. This is only when the variation of interfacial energy with the
dihedral angle for each grain boundary is small than that the Herring equation in
its simplified form (3.3) can be used. Strictly speaking, this rule only applies to the
junction of three boundaries defined as general with respect to the misorientation
and the boundary plane orientation; the balance is then realized with three dihedral
angles equal to or very close to 120◦ .
10.2.2 Equilibrium in Terms of Intrinsic Dislocations
A triple junction is also the locus where intrinsic grain boundary dislocation networks
meet together. By being cautious to describe the dislocation content of each grain
in a self-consistent manner, we easily show that the dislocation networks obey the
node Frank rule and the dislocation content at the triple junction is null. We apply the
Frank-Bilby equation (2.3) to each grain boundary by turning around their junction
line in a given sense. Si is the transformation matrices that relate the different crystals
to the reference coordinate system (see Fig. 2.1), p is a vector along the junction T
and Bi/j represents the dislocation content of each boundary i/j:
(S−1 −1
I − SII )p = BI/II
(SII −1 − SIII −1 )p = BII/III
(SIII −1 − SI −1 )p = BIII/I
0 = Bi/j (10.4)
If the Bollmann approach is used (Eq. 2.4), we must also define the rotations (or
more generally the transformations) in a self-consistent manner:
RI/II = SII SI −1 RII/III = SIII SII −1 RIII/I = SI SIII −1 (10.5)
And, for the triple junction:
RI/II RII/III RIII/I = I (10.6)

Fig. 10.9 Electronic images (a and b) showing a = 3 − = 9 − = 27b triple junction

in a copper alloy. The intrinsic dislocation networks schematized in (c) are well connected at the
junction; b examples of connections: X between a dislocation c of = 9 and two dislocations, d
and e, of = 27b, Y between b and c of = 9 and d and e of = 27b, Z between b of = 9, s of
= 3 and f of = 27b [16]
I is the identity matrix, then:
(I − RI/II )p = BI/II
(I − RII/III )p = BII/III
(10.7)
(I − RIII/I )p = BIII
0 = Bi/j
The perfect connection of intrinsic secondary dislocation networks, from a grain

boundary to another, is well illustrated in the case of a triple junction between near
coincidence = 3 − = 9 − = 27b grain boundaries in a copper-silicon alloy
(Fig. 10.9) [16].
The use of a non self-consistent description of grain boundary misorientations,
i.e. the absence of a unique reference system or the consideration of any one among
the multiple descriptions of a grain boundary, leads to an apparent rupture of the
dislocation equilibrium at the node [17]. Indeed, an equivalent rotation of a grain
boundary is obtained by multiplying Ri/j by a unimodular matrix U (Det. U = 1)
that represents one of the symmetry operations of the crystal.

If RI/II = RI/II U (10.8)

Then RI/II RII/III RIII/I =U (10.9)
And Bi/j = 0 (10.10)
Relation (10.10) cannot be obeyed if there is compatibility between the crystals.

In particular, the use of the description corresponding to the minimum content

in dislocations (NNR) for each grain boundary (see Sect. 2.3.3) leads to define two
types of triple junctions depending on the rotation product: equal to the identity
matrix (junction I) or to a unimodular matrix (junction U) [4]. In the latter case,
the triple junction is supposed to take a disclination character. But, we have seen
that all the descriptions of a same grain boundary are equivalent, they relate to
the same physical object. They all yield a same long-range elastic stress field (if it
exists) in a continuous elastic medium. The short-range displacement field depends
on the manner the dislocations are arranged in the boundary and not on their global
content determined by the Frank-Bilby (2.3) or Bollmann (2.4) equations. These
topological equations giving the defect contents are not simply related to the inter-
facial energy [17].
Numerous studies rely on the classification of triple junctions in U and I lines,
whereas this distinction is still the subject of serious controversy. Bollmann draws
attention to the fact that the notion of NNR, at the basis for the distinction between
U and I line, is purely geometrical; atomic forces can determine a relaxation mode
other than that suggested by the geometry alone. Certain differences in the junction
properties, attributed to their different character (I or U), may result from the presence
of defects at the triple line (see Sect. 10.4.2), a non-homogeneous deformation of the
crystals or an impurity concentration in the region common to the three crystals.
In conclusion, to geometrically describe a grain boundary, we may indifferently
use one or the other among the equivalent descriptions. On the contrary, the approach
of the interaction between three intrinsic dislocation networks imposes to define the
grain boundary dislocation contents in a self-consistent manner; this is the condition
for which no dislocation ends in the interior of the material and thus compatibility
is maintained between grains.
10.2.3 Equilibrium in Terms of Structural Units
There are few studies by atomistic simulations of triple junctions, they concern
singular = 3n junctions formed by three symmetrical 110 tilt grain boundaries
in materials with diamond cubic structure [18]. The junction of two coherent twins
= 3 {111} with a second-order coherent twin = 9 {221} is often observed in
diamond thin films, the two = 3 twins forming an obtuse angle of 109.47◦ [19].
The grain boundaries in diamond possess identical structure than those described
for silicon (see Sect. 3.3.4) consisting in two T units (6-atom ring) for one period of
= 3 and two M + M − units, symmetrically orientated with respect to the boundary
plane, in one period of = 9; each of the M unit is formed by a five-atom ring
connected to a seven-atom ring. The simulations of two triple junctions well show a
good geometric fit of the structural units in the junction cores (Fig. 10.10). However,
the stress fields are more extended in the vicinity of the junctions than around the
= 9 boundary [18]. An excess energy appears at each node in agreement with the
results of calculations performed for a = 3 − = 3 − = 9 triple junction in
Fig. 10.10 Atomic structures and hydrostatic stress distributions near two
= 3 − = 3 − = 9 triple junctions, at the extremities of a = 9 grain boundary in
diamond. The 110 axis common to the three crystals is perpendicular to the figure plane. The
stress level (in eV·Å−3 ) is: 0.15 (•) >, from −0.15 to −0.05 (•), from −0.05 to 0.05 (◦), from
−0.05 to 0.15 (∗), >0.15 ( ), respectively [18]
silicon [20]. Zones of high compression stresses are detected near the five-atom ring
and zones in tension are observed near the seven-atom ring at the extremities of the
= 9 boundary near the triple junctions.
The good fit between the structural units at the triple junction does not exclude the
presence of dangling bonds, some atoms having only three closest neighbours. Two
models of the atomic structure of the triple junction (on the left of Fig. 10.10), with
reconstructed core, are illustrated on Fig. 10.11. Atomic motifs C are introduced along
the rotation axis to satisfy the atom tetra-coordination. The reconstruction, giving rise
to the complete atom coordination, requires a double periodicity along the tilt axis,
similarly to what was observed for high-angle grain boundaries in semiconductors
[21]. The atomic motif reconstructed along the 110 axis consists of one eight-atom
ring that alternates with a five-atom ring. The energy of the model (b) is higher (by
0.67 eV·Å−1 ) than that of the model (a). The connection of the structural units of
= 3 to the seven-atom ring of =9, with only one reconstructed bonding, appears
favoured, in agreement with the consideration of all the possible core models for this
junction [19].
Fig. 10.11 Structure and energy distribution per atomic site for each of the two models of the triple
junction = 3 − = 3 − = 9 (on the left of Fig. 10.10), with reconstructed core. The atomic
motif C is introduced along the tilt axis in order to satisfy the atom tetra-coordination [18]
The results obtained for a = 3 − = 9 − = 27 junction are similar

but the core structure is more complex. In particular, three atomic rows, instead of
only one previously, present dangling bonds and the junction core is more extended
(Fig. 10.12) [18].
Fig. 10.12 Atomic struc-

ture of a = 3 {211} −
= 9 {122} − = 27 {255}
triple junction. Full circles
indicate atoms in the plane of
projection; empty circles indi-
cate atoms along the junction
axis at a distance a/4 110
from this plane. The atomic
rows, marked A, B and C,
contain dangling bonds [18]
Several theoretical studies indicate that ideal grain boundaries, of minimum

energy, cannot join together without forming a notorious atomic mismatch at the
triple junction. Triple junctions with empty cores are possible when the core energy
is large, then we must consider the balance between this energy and the surface
energy [22].
Experimental determinations of the core structures of triple junctions are very rare,
they concern junctions of crystals obtained by deposition of a polycrystalline film on
a substrate. Observations by high-resolution transmission electron microscopy reveal
the detailed crystallography of the configuration [9] but, rarely, the atomic structures
of the adjacent grain boundaries [9, 19]. Models are then proposed to describe the
connection between the structural units. In case of aluminium, we have seen that the
structure of a symmetrical {720} grain boundary, abutting at a symmetrical junction
such as TS (Fig. 10.3) and parallel to a tricrystal mirror plane, is well described on the
basis of the structural unit model [10] as a near = 53 grain boundary, intermediary
between = 5 and = 17. Otherwise, the asymmetrical (001)I //(035)II grain
boundary, of misorientation quasi-identical to the previous one and near the same
coincidence, appears rough at the atomic scale [9]. The core of type TS triple junction
has not been modelled, but we will see later that its energy can be estimated. From
observations of = 3− = 3− = 9 triple junctions in a diamond thin film, core
models have been proposed with the assumption that structural units must connect
with a minimum distortion and with the lowest number of dangling bonds [19], in
agreement with the simulation results.
Images in high-resolution electron microscopy of the triple junctions on bulk

materials (Fig. 10.7) show a pretty good fit between the dense planes from one crystal
to the other, but do not enable to visualize the structural units near the junction. The
difficulty for achieving resolution at the atomic scale more likely results from the fact
that the triple junction is not linear throughout the thickness of the sample. Variations
of the dihedral angles between the grain boundaries have been observed along a triple
junction by field ion microscopy.
10.3 Triple Junction Energy
A triple junction possesses a core with more or less large local stresses. It may also
contain defects that result in long-range elastic strains. Thus, we may define, as for
any linear defect, a line energy or line tension. This energy must be larger than that
of the reference state, the crystal, but a priori there is no restriction on its sign with
respect to those of the constituting grain boundaries. It is not excluded that a triple
junction may have a structure close to the crystal structure while those of the adjacent
grain boundaries are different [19]. A junction with a negative energy compared to
those of the boundaries would spontaneously extend; but this would provoke grain
boundary extension, such a configuration may thus resist to grain growth.
The answers of triple junctions and those of grain boundaries to solute segre-
gation or to liquid wetting differ: this fact supports the energy differences of these
two defects. Very often solute enrichment or wetting in a triple junction are more
important than in the constituting grain boundaries; this indicate a triple junction
energy generally superior to the interfacial energies.
10.3.1 Calculation of the Triple Junction Energy
Calculated line energies may be positive or negative, depending on whether the

triple junction is considered as a line at the intersection of sharp interfaces or a
junction between diffuse interfaces, more similar to those encountered in polycrystal
microstructures. To the first category, we must attach the junctions between multiple
twins whose atomic structures are well known, in particular in semiconductors where
they are numerous and clean.
Calculations of the total excess energy of six triple junctions, with different sym-
metries, included in a box constituted by seven crystals and grain boundaries of
misorientations 30◦ and 60◦ , give negative energy values [8]. Such grain boundaries,
observed in aluminium thin layers (Fig. 10.3), are close coincidence boundaries with
high index = 53 or = 65 and are general from an energetic point of view.
The excess energy is defined as the difference between, on one hand, the calculated
energy for a polycrystal simulation box (Fig. 10.13) and, on the other hand, the sum
of the energy of a single crystal containing the same number of atoms and the excess
10.3 Triple Junction Energy 323
Fig. 10.13 Schematic representation of the periodic simulation box of six triple junctions. The
grain boundaries abutting at the junction 3m and the boundary C (m)−P are misorientated from
60◦ , the other boundaries from 30◦ . The dihedral angles of the six junctions are indicated on the
drawing on the right [8]
energy of the grain boundaries. Simulations show that some atoms in the junction
have positive energies with respect to their energy in the crystal but inferior to the
energies of some atoms in a grain boundary. Although the result is the sum of six
energies, it indicates that one of the triple junctions, at least, has a negative energy [8].
These calculations rest on the hypothesis that grain boundaries and triple junctions
possess a diffuse structure. Otherwise, if each grain boundary and each triple junc-
tion are sharp, mathematically considered as two-dimensional and one-dimensional,
respectively, a negative energy then implies an unstable microstructure.
Energy calculations of different configurations of a = 3− = 3− = 9 triple
junction in silicon have been performed using molecular dynamic simulations. The
values of the line energies and of the stresses are measured in a cylinder surrounding
the triple line with increasing radius rTJ ; the energy values converge for rTJ ∼ = 8a0 (a0
is the lattice parameter of silicon equal to 0.549 nm). The results slightly differ
according to the junction configuration, i.e. according to the position of the triple
line in one or the other of the structural units of the adjacent boundaries. The excess
energy is in average equal to 0.35 eV/Å [20]. This value is lower by two orders
of magnitude to that experimentally determined for a triple junction in copper (see
Sect. 10.3.2), but remarkably, it is similar to the line energy of dislocations in silicon
crystals. Generally, the excess volume is negative or weakly positive. A map, at the
Fig. 10.14 Map of the prin-

cipal stress (σzz ) distribu-
tion around a = 3– =
3– = 9 triple junction. The
σzz = 0 contours are repre-
sented by dotted curves. The
white lobe indicates a high
compression stress that tends
to reduce the angle between
the two = 3 boundaries;
the black lobe indicates a high
tension stress that tends to
break the = 9 boundary.
The grey level scale is on the
right in kbar [20]
atomic scale, of the distribution of the principal residual stress σzz around the junction
is reported on Fig. 10.14. The white lobe indicates a high compressive stress which
tends to reduce the angle between the two = 3 boundaries; the black lobe indicates
a high tensile stress which tends to break the = 9 boundary. The non-balance of
the elastic stresses suggests that the junction between multiple twins is a singular
fault by comparison to the equilibrated junctions between three general boundaries.
The line energies, the excess volumes and the stresses at the atomic scale shows
that the = 3 − = 3 − = 9 triple junction is a real linear defect and not
simply the geometric locus of intersection of three planes. This junction is sharp on
the contrary to previous results where the junction had a diffuse character. In this
approach, the energy of a junction of multiple twins represents the upper limit for
the energy of any junction in a polycrystal [20]. In contrast, atomic simulations of
a nano-crystalline structure with random grain boundaries and triple junctions show
that the high-energy boundaries and their junctions have a disordered structure, close
to the amorphous state [23]. In that case, the excess energy per atom of the triple
junction is equal to the excess energy per atom of the grain boundary, itself equal
to the energy difference between amorphous and crystalline silicon. By comparing
the previous results, we may conclude that the connection between three low-energy
grain boundary gives rise to well ordered junctions, i.e. linear defects with a positive
line energy. The junctions between three twins have large line energies and residual
stresses. The triple junction energies tend to zero when the energies of the constitu-
tive grain boundaries increase. While the high-energy junctions between low-energy
boundaries are narrow, the low-energy junctions between high-energy boundaries
are diffuse; but their thickness never reaches a value such that it leads to a negative
line tension.
10.3 Triple Junction Energy 325
10.3.2 Experimental Determination of the Triple Junction Energy
The triple junction energy has been estimated by comparison to the energies of the
three grain boundaries abutting at the junction from the measurements of the depth
of the thermal etching grooves, using scanning tunneling microscopy (Fig. 10.15a)
[24]. The analysis rests on three hypotheses:
• The torque term of the interfacial tension is null
• The excess energy at the triple junction is constant
• The angle and the depth of the thermal etching groove along a grain boundary are
constant.
However, the groove characteristics (angle and depth) may differ from one grain
boundary to another; thus, the attack facies at the junction takes the shape of an
irregular tetrahedron.
The ratio q is measured between the groove depth ZT at the junction and that of
the deepest intergranular groove adjacent to the junction Z GB (Fig. 3.15b):
ZT
q= (10.11)
Z GB
This ratio is function of the surface, grain boundaries and triple junction energies.
The change of energy
E Z linked to the formation of a groove of depth
Z is
given by:
6 3

E Z =
AiS γSi −
AGB k γGB k −
Z·σ (10.12)
i=1 k=1
AS i is the area change of the surface i (walls of the groove) and γS i , the energy of
this surface;
AGB k is the area change of the grain boundary k (area defined by the
tetrahedron sides and the triple line) and γGB , the grain boundary energy; σ is the
line tension of the triple junction (same units than its energy). By considering a small
increase of the groove depth at the junction such that dE Z /dZ = 0, a maximal value
qmax has been derived for a triple junction formed by grain boundaries identically
attacked; the etching shape is then a regular tetrahedron. The value of qmax depends
on the opening angle θ of the grain boundary grooves, the line tension σ of the triple
junction and the distance c between the junction and the boundary region where
the groove is no longer affected by the junction (Fig. 10.15b). It is obvious that the
experimental value of the ratio q, referring to the deepest intergranular groove, must
always be less or equal to qmax . For a line tension σ = 0, qmax takes a maximal value
q0 = 1.33. If the experimental ratio q has a value less than q0 , the triple junction
has no own energy; only the connection between the three boundaries provokes
the depth difference. Otherwise, the triple junction possesses its own energy. For
example, for a ratio q = 1.46, this value is estimated to 5.10−4 mJ·m−1 , two orders
of magnitude larger than that of the line energy associated with a lattice dislocation
(≈5.10−6 mJ·m−1 ).
Fig. 10.15 a Topographic profile of a triple junction, that has been submitted to thermal etching,
obtained by scanning tunneling microscopy; b Schematic representation of the geometrical model
of a triple junction
Beyond 5 μm from the triple junction, the intergranular grooves are almost
constant (= θ∞ ). On the contrary, significant variations of θ are detected in the
vicinity of the junction along the three boundaries, indicating energy variations.
The latter are probably related to readjustment of the grain boundary planes when
approaching the junction, as observed by transmission electron microscopy. It is
remarkable that, for distances very close the triple junctions (<100 nm) such that
q ≤ q0 , the values of the three intergranular groove angles converge towards θ∞ .
They diverge in case of triple junctions such that q > q0 , suggesting the existence
of uncompensated stresses, characteristic of a disclination.
10.4 Triple Junction Defects
A defect at a triple junction may result, on one hand, from differences in the plastic
deformation of the crystals around the junction and, on the other hand, from grain
boundary misorientation mismatch at the junction. The latter is equivalent to an
incompatibility in the grain boundary connection. If only this type of mismatch
occurs, and by comparison to the connection of two crystals on each grain boundary
side, the defects at the triple junction may be considered as intrinsic in the sense that
they come from inside, without referring to the extension of the stress fields. If plastic
incompatibilities exist between the crystals, real extrinsic defects (dislocations or
disclinations) may be localized at the triple junction. These defects play an important
role in the first stages of plastic deformation, but their role becomes negligible when
the strain level increases.
10.4 Triple Junction Defects 327
10.4.1 Intrinsic Defects of a Triple Junction: Geometrical

Approach
The balance between the grain boundary intrinsic dislocation networks is always
realized at the junction, as it was the case for the junction of dislocations at a node and
according to the Frank rule. However, even if each boundary is equilibrated, stresses
may develop at the connection of three boundaries giving to the triple junction a
disclination or a dislocation character [17].
• A junction with a disclination character is obtained at the connection between
five twins in nanoparticles of f.c.c. or diamond cubic structure [18]. The crystals
are bordered by {111} planes and their common direction is parallel to [110]
(Fig. 10.16). After to introduce a Volterra cut along one of the boundaries, we
draw a circuit mapping around the junction according to the procedure proposed
by Pond [25]. The disclination character of the junction yields an increase of
the magnitude of the closure failure with the distance to the multiple junction
(SF or S F for example) [17]. This is the case of a wedge of 7.35◦ around a
common line [110]. This defect may be compensated by dislocation climb along
the interfaces in such a manner to form an edge dislocation wall that ends at the
√ The Burgers vector of these dislocations is a/3 111 and their spacing
junction.
d = a 3/2 sin 7.35◦ /2. The defect content is given by the Frank-Bilby equation
(2.3). The dislocation wall is not necessarily confined at a boundary, but may form
a low angle grain boundary within a crystal.
The atomic structure of the multiple junction core, studied by simulation of a
diamond nanoparticle, is in agreement with the disclination character of the junction
[18]. The core is well described in terms of structural units and stress distributions;
compression at the centre of the particle and tension at a certain distance from the
junction (Fig. 10.17) well correspond to what is predicted at the mesoscopic scale by
the disclination theory [26].
• A junction with a dislocation character results from the connection of boundaries
presenting rigid body translations τ between contiguous crystals. In the sum of
the transformations that accompany the circuit around a triple junction, we must
then enter a sum of rigid body translations that may be null or not (Fig. 10.18). If
this sum τij is different from zero, then the junction has a dislocation character
with a Burgers vector b = −τij . We must insist on the fact that the dislocation
character of the junction results from a translation defect that has its origin in
the equilibrium grain boundary structures. It exists even if the grain boundary
misorientations compensate themselves; it does not require faulted boundaries.
Junctions with dislocation character most often occur if they are composed of grain
boundaries with low indices. Indeed, if these boundaries preserve their infinite
structures, it is unlikely that rigid body translations are compensated. Relaxation
by emission of dislocation has been evoked to interpret an observation, by high-
resolution transmission electron microscopy, of an intersection between two coherent
Fig. 10.16 Schematic

drawing of a multiple junc-
tion where five twins of
f.c.c. material are connected.
Due to the positive wedge
disclination character of the
junction, an angular closure
failure of 7.5◦ appears after
applying a Volterra cut. The
wedge disclination character
is clearly revealed by the fact
that the closure failure of a
circuit around the junction
increases with its distance
from the junction. The intro-
duction of material would
eliminate the defect [17]
twins in a diamond thin film. The presence of an edge dislocation with a Burgers
vector b = a/9 221, vector of the DSC lattice of = 9 perpendicular to the
boundary plane, enables to explain the high contrast that develops in crystal I, in the
vicinity of the superior junction (Fig. 10.19) [19]. It must be noted that this junction
is not equilibrated, a balance of the forces may only occur if the three interfaces
have a positive tension; this junction results most probably from a growth accident
and is a metastable configuration. Another relaxation mode may occur to avoid
Fig. 10.17 Illustration of the structure of a diamond pentagonal particle formed by five tetrahedra
separated by twin boundaries. The distribution of the hydrostatic stresses per atom shows that these
stresses increase with the distance from the junction: from compression at the junction core, they
become tension at about 0.8 nm from the core [18]
or minimise the translation mismatch; one (or several) of the adjacent boundaries
adopts a structure with a rigid body translation different from that attached to the
infinite structure of minimum energy. This mode may be activated if the resulting
interfacial energy increase is less than the energy of the dislocation (s) introduced at
the junction. This mode seems to occur at the inferior junction of Fig. 10.19 where
a change of the II/IV twin boundary structure is accompanied by an expansion of
0.06 nm normal to the boundary plane, the interplanar spacing being preserved for
the other twin. Finally, grain boundary migration may constitute a third relaxation
mode.
According to King et al. [19], the relaxation mode of the translation mismatch
depends on the characters of the boundaries forming the junction (coincident or
not, values of the index…) but also on the grain size. In particular, for large
grain sizes, the boundaries tend to preserve their equilibrium structure (infinite
boundary) and the mismatch is taken off by one (or more) dislocation at the
junction.
The two types of previous defects, disclinations and dislocations, characterise a
triple (or multiple) junction; they belong to its structure as they are present in the
absence of deformation of each of the bicrystals that compose the junction. It is
why, in a certain manner, we can consider these defects as intrinsic. Other extrin-
sic defects may occur at the junction, resulting from thermo-mechanical treatments
underwent by the tricrystal (or polycrystal) during its elaboration and/or during its
deformation.
Fig. 10.18 A triple junc-

tion between two {111}
coherent twins and one {221}
incoherent twin: a the rigid
body translations are null, the
junction does not present any
defect; b if there is no balance
of the rigid body translations
existing in the different grain
boundaries, the junction then
takes a dislocation character
[17]
10.4.2 Extrinsic Defects of a Triple Junction: Mechanical

Approach
A triple junction does not present any extrinsic defect when the three grains are free of
stresses or homogeneously deformed. But, most often, during plastic deformation,
incompatibilities concentrate at interfaces. These incompatibilities may yield the
presence of extrinsic dislocations in the grain boundaries. Each boundary acquires a
non-accommodated additional misorientation and an angular mismatch, on the form
Fig. 10.19 a Image in high-resolution transmission electron microscopy of two coherent twins
(designated as grains II and III) in a diamond thin film. The intersection gives rise to a = 9 grain
boundary and to two triple junctions (to be compared with the simulated image on Fig. 10.10); b
schema explaining the crystallography of the two triple junctions—the meshing corresponds to the
projection of the DSC lattice of = 9 onto the (110) figure plane; the circles represent atoms
displaced 1/4 [110] with respect to this plane [19]
of a wegde disclination, appears at the triple junction. This defect has been studied
in detail in a mechanical approach [27]; we just give here a brief description.
The incompatibilities concentrated at a grain boundary have been described by a
dimensionless surface dislocation density tensor, according to the expression (2.1).
On a simplified form, this tensor may be written [27]:
BN = −N ∧ [β] (10.13)
N is normal to the grain boundary plane and β is the plastic distortion tensor.
The additional misorientation
θ that forms at the grain boundary may be sepa-
rated into two types associated with the decomposition of the plastic distortion tensor
β into two parts, symmetric and anti-symmetric, representing the plastic deformation
and the plastic rotation discontinuities, respectively:
[β] = [ε] + [ω] (10.14)

ε ω
θ = θ + θ (10.15)
with θ ω = −[ω], the vector representing the relative rotation between the crystals
and θ ε = −N ∧ [ε]·N.
Let us consider a linear junction, with unit vector η, between three crystals and let
us analyse the type of defect that may form at this junction by drawing a closed circuit
L K , in the clockwise sense, around the triple line (Fig. 10.20) [27]. The dislocation
content of the junction is given in function of the content in each grain boundary Bi
Fig. 10.20 Junction between

three grain boundaries and
definition of the parameters
taken into account in the
calculation of the elastic fields
of the defects [27]
and, by using expression (10.14), in function of the plastic distortion tensor [β]i :

3
3
Bη = − (Ni ∧ η)Bi = η [β]i (10.16)
i=1 i=1
When a complete circle is made around the junction, the sums of the discontinuities
are null:

3
3
3
[β]i = 0 [ω]i = 0 [ε]i = 0 (10.17)
i=1 i=1 i=1
and thus, Bη = 0.
Consequently, plastic deformations, homogeneous within each of the adjacent
crystals, do not generate any defect of the dislocation type at the triple junction.
Now we consider a triple junction, plastically incompatible, presenting an
additional misorientation θ between adjacent grain boundaries; we may search
for the defect that appears at the junction, by separating the effect of the rotation
components θ ω from that of the deformation components θ ε . The vector θ ω
is entirely governed by the difference of plastic rotations, [ω] is independent of the
normal N to the grain boundary plane. As a result:

3
3

ω = θω
i =− [ω]i = 0 (10.18)
i=1 i=1
The plastic rotations, homogeneous within each crystal, do not result in any defect
at the triple junction. On the contrary, the vector θε is governed by the orienta-
tion of the boundary plane and by the incompatibility of the plastic strains at grain
boundaries. As a result a misorientation defect is created at the junction:

3
3
= θεi = − Ni ∧ [ε]i ·Ni (10.19)
i=1 i=1
The linear defect of rotation is a wedge disclination having triple line for rotation
axis and a rotation vector (Frank vector) D = −.
Generally, the vector is not parallel to the triple line; the defect is then com-
posed of a wedge disclination with the coinciding rotation axis (Fig. 10.20) and a
twist disclination with a rotation vector perpendicular to the junction [27]. Same
approach may be applied to the junction between two facets along a grain boundary;
the triple junction is then considered as an assembly of double-junctions. It may also
be extended to a multiple junction of more than three boundaries.
The disclination is a real defect superimposed to the equilibrium triple junction
structure that plays an important role in the plastic deformation of polycrystals.
Junctions with a disclination character exist in a polycrystal before deformation as
the results of the thermo-mechanical history of the material. We could then explain
the different behaviours of certain junctions, previously attributed to an eventual I/U
character difference (see Sect. 10.2.2). We have seen that the deformation leads to the
formation of dipoles of disclinations in two adjacent triple junctions (see Sect. 9.1.3).
When the deformation increases, these junctions may generate partial disclinations
at the origin of new interfaces within the adjacent crystals.
In polycrystalline materials, a disclination in the sense of a rotational cut cannot
exist. The description of the triple junction in terms of disclination is formal,
equivalent descriptions in terms of dislocations may always be proposed [28].
However, it is proved useful to derive the energy in complex cases of defect con-
centration, to account for the deformation state of nano-crystalline materials and
to describe a dislocation arrangement in a grain boundary in relation to its atomic
structure (see Sect. 3.5).
10.5 From Tricrystal to Polycrystal
A polycrystal can be considered as an arrangement of tricrystals or as a network of

triple junctions that connect at a quadruple node. The configuration formed by four
grains, six grain boundaries and four triple junctions (Fig. 10.21) is the link in the
Fig. 10.21 Triple line

topology in a polycrystal.
The schema drawing shows a
node N formed at the junction
of four grains of six grain
boundaries (a, b, c, d, e, f) and
four triple junctions (A, B, C,
and D). The equilibrium angle
between two adjacent bound-
ary planes is 120◦ and that
between the triple junctions is
109.5◦ [29]
chain, superior to the tricrystal, for constructing a polycrystal. When the six bound-
aries sharing a common quadruple node have the same energy (general boundaries),
equilibrium at the node requires an angle of 120◦ between boundary planes and
109.5◦ between the triple junctions.
The Frank rule used for dislocation equilibrium at a node may be extended to a
polycrystal. Bollmann proves this equilibrium by considering triple junctions type
U [4, 29]. Even if you do not agree with the so-called U character of the defects (see
Sect. 10.2.2), the six grain boundaries of the configuration generally contain intrinsic
defects and balance requires that these defects annihilate at the node. The problem
is then how to apply the node rule at the quadruple junction [4]. First, the orientation
of each grain must be given with respect to a unique reference, grain 1 for example.
The sequence of the first three junctions, those adjacent to grain 1, is selected such
that it defines a right-hand screw when the triple junctions point out of the node.
The quadruple node may be considered at the centre of a tetrahedron, the four lines
joining this centre to the vertices are triple junctions that precise the situation of the
four crystals (Fig. 10.22a). It is schematized in two dimensions on a vertical section
10.5 From Tricrystal to Polycrystal 335
Fig. 10.22 a Schematic drawing of a quadruple node; b topologically equivalent representation

of the configuration at the node resulting from a vertical section of (a). The arrows represent the
sequential transitions through the grain boundaries
of the tetrahedron such that the intersections of the section plane with the grains, the
grain boundaries and the triple junctions are surfaces, lines and triple points, respec-
tively (Fig. 10.22b). The sequence of operations through the grain boundaries and
around the triple junction is given by arrows, starting from the origin O in crystal 1.
By considering the sequence of rotation matrices, the equilibrium condition is as
follows:
(R41 ) [R24 R32 R43 ] (R14 ) [R41 R34 R13 ] [R31 R23 R12 ] [R21 R42 R14 ] = I [III.20]
⇓ ⇓ ⇓ ⇓
around (d) around (c) around (b) around (a)
(10.20)
The condition (10.20) may be applied by replacing the rotation matrices R by trans-
formation matrices A.
If the triple junctions contain extrinsic defects resulting from a non-balance of the
grain boundary defects, the node is not equilibrated and can move.
Experimental studies generally concern polycrystals, usual form of the materials.
In the fundamental approaches, an intermediate state is sometimes considered: the
multicrystal. A multicrystal is conceived (for stress calculations for example) or
fabricated with a controlled organization of a limited number (10 to 20) of grains
(and grain boundaries). Controlled organization means that the orientation and shape
of each grain as well as the characteristics of each grain boundary are fixed. On
the contrary, a polycrystal displays a non-controlled distribution of its grains and
grain boundaries that depends on its thermo-mechanical history. The distribution
of the crystalline orientations or texture has been the subject of numerous studies
and we know how it influences the material properties. By analogy, we attempt to
understand how the grain boundary organization, more precisely, the distribution
of the geometrical characteristics or grain boundary texture influences the grain
boundary contribution to the properties of the whole material. But this approach is
much more complex than that of the crystalline texture.
References
1. T. Watanabe, Res. Mechanica 11, 284 (1984)

2. E.G. Doni, G.L. Bleris, Phys. Stat. Sol. (a) 110, 393 (1988)
3. S. Lartigue, L. Priester, J. Am. Cer. Soc. 71(6), 430 (1988)
4. W. Bollmann, Mat. Sci. Eng. A 113, 129 (1989)
5. A.V. Shubnikov, N.V. Belov, Colored Symmetry, part II (Pergamon Press/MacMillan, New
York, 1964)
6. G. Kalonji, J.W. Cahn, J. de Physique Colloque 43, C6–25 (1982)
7. J.W. Cahn, G. Kalonji, MRS Proceedings (1994)
8. S.G. Srinivasan, J.W. Cahn, H. Jonsonn, G. Kalonji, Acta Mater. 47, 2821 (1999)
9. N. Thangaraj, U. Dahmen, MRS Proc. 238, 171 (1992)
10. A.P. Sutton, V. Vitek, Phil. Trans. R. Soc. Lond. A 309, 1 (1983)
11. C. Herring, The Physics of Powder Metallurgy, ed. by W.E. Kingston (Mc Graw-Hill Pub.,
New York, 1951), p. 143
12. A.H. King, V. Singh, Mat. Sci. Forum. 207–209, 257 (1996)
13. M.A. Fortes, Interface Sci. 1, 147 (1993)
14. L. Priester, D.P. Yu, Mat. Sci. Eng. A 188, 113 (1994)
15. A.L. Vassiliev, E.A. Stepantsov, N.A. Kiselev, Phys. Stat. Sol. (a) 144, 383 (1994)
16. L.M. Clarebrough, C.T. Forwood, Phil. Mag. A 55, 217 (1987)
17. G.P. Dimitrakopulos, Th. Karakostas, R.C. Pond, Interface Sci. 4, 129 (1996)
18. O.A. Shenderova, D.W. Brenner, Phys. Rev. B 60(10), 7053 (1999)
19. A.H. King, F.R. Chen, L. Chang, J.J. Kai, Interface Sci. 5, 287 (1997)
20. S. Costantini, P. Alippi, L. Colombo, F. Cleri, Phys. Rev. B 63, 221 (2001)
21. A.-M. Papon, M. Petit, Scripta Metall. 19, 391 (1985)
22. F.C. Frank, Acta Crystall. 4, 497 (1951)
23. P. Keblinski, D. Wolf, S.R. Philipot, H. Gleiter, Phys. Rev. Lett. 77, 2965 (1996)
24. P. Fortier, G. Palumbo, G.D. Bruce, W.A. Miller, K.T. Aust, Scripta Met. Mater. 25, 177 (1991)
25. R.C. Pond, Dislocations in Solid, vol. 8, ed. by F. Nabarro (North Holland Pub., Amsterdam,
1989), pp. 1–66
26. A.E. Romanov, V.I. Vladimirov, Disclinations in crystalline solids, in Dislocations in Solids,
vol. 9, ed. by F.N.R. Nabarro (North Holland, Amsterdam, 1992), pp. 191–402
27. V.V. Rybin, A.A. Zisman, NYu. Zolotorevsky, Acta Metall. Mater. 41, 2211 (1993)
28. E. Kröner, Statistical Continuum Mechanics (Springer, Berlin, 1972)
29. W. Bollmann, Phil. Mag. A 57, 637 (1988)
Chapter 11
Grain Boundary Network: Grain
Boundary Texture
If we try to understand the role of grain boundary networks in polycrystal properties,

in order to promote special behaviours of the material, the first task is to establish then
to control the grain boundary organization in that network. This is the objective of
numerous studies dealing with the determination of the grain boundary characteristic
distribution in polycrystals. A grain boundary texture is established that results from
interconnected effects of several parameters: the crystalline texture, the energies
of the boundaries, their mobility, their possible rotation and/or sliding and their
interaction with solutes. This approach is based on the assumption that the properties
of a system can be deduced from those of its constituents; we approach the limits
mentioned earlier in the presentation of Chap. 10. Even if we consider that this
postulate applies to a grain boundary network, the understanding of the whole can
only be limited nowadays, due to the fact that the relationships between an individual
grain boundary and its properties remain to be clearly established. Furthermore, the
macroscopic grain boundary parameters being fixed, the intergranular structure and
properties may vary under the effects of temperature, in the presence of solutes and/or
extrinsic defects. Obviously, the different grain boundaries behave differently, but
no geometric criterion can satisfactorily predict the responses of a given boundary to
external stimuli. Only the 3 {111} grain boundary is really special from the point
of view of all its properties: no extrinsic defect, diffusivity almost equal to that of
the volume, no segregation, no intergranular corrosion …
Although we have, repeatedly warned the reader against any geometrical notion
of speciality of the grain boundaries, the numerous calculations and experiments
to determine the organization of various boundaries in polycrystals deserve to
be reported. We may reasonably hope that collecting a large number of data
characterizing the chemical and thermo-mechanical parameters that influence the
grain boundary distribution constitutes an empirical basis to control some material
properties. These studies are in the prospect to establish a link between the micro-
scopic and macroscopic scales, but they will always face the epistemological limit:
the whole is never the sum of its parts. An inherent property of a grain bound-
ary clearly identified in a bicrystal can be strongly modified when this boundary is

338 11 Grain Boundary Network: Grain Boundary Texture
connected to others. What are the exchanges of matter, the interactions between
defects and the stress transmission that occur in the interconnected network of grains,
grain boundaries, triple junctions, quadruple nodes? Questioning is vast.
In the domain of interfaces, as in any scientific domain, the approaches aiming at
knowledge are performed over time by a back and forth movement between those
that come under holism and those coming under reductionism. They constitute a
dialectical approach to reality, but a true overcoming of contradictions they contain
to achieve a synthesis does not seem feasible in the medium term. It is from these
heuristic considerations that we present results, among the most representative, on
grain boundary networks.
The studies on the Grain Boundary Character Distributions (GBCD) in
polycrystals first focus on the determination of the proportions of grain boundaries
with same misorientation so-called Grain Boundary Misorientation Distribution
(GBMD). This first distribution is followed by the grouping of boundaries into
two classes: coincident grain boundaries (exact and near coincidence relationship),
sometimes named special and general grain boundaries (random misorientation),
according to one of the criteria reviewed in Sect. 11.1. A distribution based on the
percentage of coincident boundaries is then generalized: Coincidence Grain Bound-
ary Distribution (CGBD). These two distribution modes GBMD and CGBD only
partially describe the grain boundary texture. Indeed, by comparison to each crystal
defined by three Euler angles that fixed its orientation, each grain boundary is defined
by five macroscopic geometrical parameters that account for its misorientation and its
plane. A true texture must consider the distribution, with a statistical aspect, of these
five parameters defining the grain boundaries in a polycrystalline space. However,
the statistic character of the texture implies a determination of these parameters at
the macroscopic or mesoscopic scale; this often reduces the texture to the misorien-
tation distribution. Few works on fine-grained polycrystals, by transmission electron
microscopy, lead to distributions of misorientations and grain boundary planes, but
often uncorrelated. Due to the small number of investigated grain boundaries, these
studies only give a tendency to the complete boundary network organization in the
polycrystal. More recently, statistical approaches of the grain boundary plane distri-
bution have been obtained by the technique of serial sectioning at different depth of a
polycrystal observed by scanning electron microscopy. The method allows acquiring
two-dimensional images of the microstructure as a basis for reconstructing three-
dimensional solids. Finally, if a distribution of the five macroscopic parameters has
been theoretically described in the mid 1980s, it was not until 2000 that simulta-
neous experimental distributions of the misorientations and the planes have been
established.
The misorientation distributions, that may be approached by calculations,
simulations and experiments, constitute the subject of Sects. 11.2 and 11.3. On the
contrary, the grain boundary plane distributions, presented in Sect. 11.4, can only be
experimentally studied, the grain boundary plane being not controlled by the geom-
etry of the neighbouring crystals. In the next section, we deal with the simultaneous
distributions of the misorientations and the boundary planes (Sect. 11.5). The final
goal in establishing these distributions being the improvement of the polycrystal
11 Grain Boundary Network: Grain Boundary Texture 339
properties through the grain boundary control, it is important to recall that no geo-
metrical criterion is able to serve as criterion of speciality. Starting with the assertion
that a boundary is special only if it displays special properties, some attempts were
developed to investigate the distributions of the grain boundary behaviours under a
given stimulus; they are briefly described in Sect. 11.6.
Whatever the retained criterion for classifying grain boundaries, we rapidly
conceive that the description of a boundary network is insufficient and that their
connections must be taken into account. Distributions of triple junctions (even
quadruple nodes), which are the subject of recent theoretical and experimental
approaches, are described in Sect. 11.7. They are sometimes associated with dis-
tributions of grain boundary energies.
All the previously mentioned approaches concern the overall grain boundary
texture. With the development of techniques, in particular the orientation imaging
microscopy (OIM), based on automatic indexing of electron backscatter diffrac-
tion patterns, we are more and more interested by the local texture, i.e. the spa-
tial arrangement of the grain boundary network in a polycrystal. A set of grains
bounded by grain boundaries of same misorientations and connected to each other
forms an ensemble or a colony or a cluster, the latter term being retained in the
book. A better geometric description of the polycrystal naturally results from the
knowledge of the local texture, although it is generally less statistical. It questions
about the pertinent dimension of a material: the average grain size or the average size
of the clusters? The existence of such clusters must have consequences on properties
implying propagation via grain boundaries (diffusion, intergranular fracture or cor-
rosion…). Section 11.8 presents some local arrangements of grain boundaries, often
limited to the mesoscopic scale. Without this microstructural vision, understanding
of a material property, taking into account the intergranular paths, can be addressed
by the mathematical theory of percolation; this is the subject of the last Sect. 11.9 of
this chapter.
11.1 Criteria for Grain Boundary Distribution
The use of geometric criteria to establish a grain boundary distribution in a poly-

crystal overcomes the inability to determine all the intergranular energies, i.e. the
structures at the atomic level of all the boundaries constituting the polycrystal; only
the latter determination enables to reach a distribution that would be significant,
but that is unrealistic. The relevant aspect of the energy is also not universal, low
intergranular energy does not necessarily imply a special behaviour of a grain bound-
ary. In particular, we have seen that the relaxation processes of the interfacial stresses
mainly differ with the degree of grain boundary symmetry; these processes are iden-
tical in periodic grain boundaries, whatever their energies (see Chap. 9). Moreover,
the energetic state of a given grain boundary in a bicrystal is not identical to that of
the same boundary constrained at a triple junction. Despite the previous fundamental
and experimental limits, the geometric criteria are at the basis of numerous studies
of the grain boundary texture; even if no of them is entirely satisfying. These criteria
have already been critically presented (See Sect. 4.5), they are reviewed here in order
to clarify the limits of their use to classify the grain boundaries. Using the five macro-
scopic geometrical degrees of freedom to predict energy constitutes a step forward
for the mesoscopic studies, but it does not allow us to describe the state of a real
grain boundary that also depends on chemical, mechanical, thermal factors. This is
why some classification criteria, based on the responses of various grain boundaries
in a polycrystal to external stimuli, are also discussed.
11.1.1 Misorientation Criterion
Generally, the misorientation distribution is followed by the selection of the

misorientations equal or near a coincidence position, according to a fixed criterion
on the angular deviation and by limiting the indices to a maximal value max .
This leads to a grain boundary distribution under the concept of coincidence, the
limits of which have been discussed in Chap. 1. We recall, in particular, that there
is no reciprocal and unique relationship between the intergranular energy and the
coincidence index. However, this classification rests on results of several studies
that empirically show an improvement of various properties when the proportion of
coincidence grain boundaries of low index increases in the material. They also
report the possibility to control the coincidence boundary frequency by playing on
different parameters, going from the purity of the material to its elaboration condi-
tions. One of the first studies of this type was performed in 1959 [1]; many studies
have subsequently developed from the 1980s and have continued to feed data in an
applied field of materials science.
The maximum value limiting the boundaries classified as coincident varies
from one study to the other, but is often fixed to 27 for cubic materials. To define
the near coincidence misorientations, the Brandon criterion [2] is most often used; it
considers as special any grain boundary whose the angular deviation from the exact
coincidence is less than an upper value θmax expressed below:
θmax = 15◦ −1/2 (11.1)
The value 15◦ is the limit of the low-angle misorientations allowing to select the
= 1 grain boundaries. Ishida and Mc. Lean consider a more restrictive criterion
with a variation in −1 , then reduce further the maximum deviation by taking 8◦
as limit for the low-angle boundaries [3]. Another criterion θmax = 15◦ −5/6 ,
proposed by Palumbo and Aust [4], is used in an increasing number of experiments.
The angular deviation from a coincidence misorientation is fully expressed by an
angle θd around an axis Rd that differs from the exact coincidence rotation axis Rc
and from that of the experimental coincidence axis Re . These two parameters may
be analytically deduced from the deviation matrix or geometrically determined from
the results obtained by transmission electron microscopy. Kokawa et al. propose a
11.1 Criteria for Grain Boundary Distribution 341
first expression of the total deviation θd [5]:
tg2 (θd /2) = tg2 (θx /2) + tg2 (θy /2) (11.2)
with
θx = |θe − θc | (11.3)

θy = θ0 the angle (R1 Rc ) = (R2 Rc )
A more precise value of θy = 2θ0 sin(θc / 2) is then reported; θy is the
angle between the rotation axes R1 and R2 , each of them differing from the exact
coincidence rotation axis. Furthermore, due to the low angles generally implied,
relation (11.2) may be simplified [6]:
θd = [θx2 + θy2 ]1/2 (11.4)
In most works that only attempt to select the near coincidence grain boundaries,
only the θx deviation is taken into account.
The grain boundaries described by the plane-matching model or, equivalently,
by a common axial direction (CAD) (see Sect. 1.2.4) are treated separately from
the previous coincident boundaries. We recall that this model corresponds to a good
matching of low index planes, most often {200}, {220} and {111} in the f.c.c materials
and only {110} in the b.c.c. materials. We define a maximal angular deviation between
a common axis yielding a good plane matching and a simple crystallographic axis:
θmax = θ0 (a/b)−1/2 (11.5)
with = h2 + k2 + l2 , a is the unit cell lattice parameter and b the intensity of the
lattice dislocation Burgers vector. θ0 takes the same values than those used for the
coincidence deviation and the variation of θmax may be function of −1 .
The misorientation criterion enables to establish a grain boundary distribution
that, without any doubt, is far to significantly represent the grain boundary ensemble
with a view to trace back to the polycrystal properties. Consideration of bound-
ary planes, which imposes the intergranular structure and energy, is necessary.
However, the misorientation distribution may indicate a first trend in the grain
boundary organization in a material; this is even better if the distribution is based
on a selection criterion of near coincidence boundaries more restrictive than that of
Brandon.
11.1.2 Grain Boundary Plane Criterion
The coincidence lattice has a physical reality only at the interface between the two
interpenetrated crystals, the position where the grain boundary plane is located in
these two lattices is thus fundamental; it determines the interfacial energy. However,
the selection of grain boundaries during the polycrystal formation is partly based on
the reduction of total interfacial energy, although other factors, in particular stresses
between crystals, are strongly involved [7]. The role of the grain boundary plane in
most of the boundary behaviours also appears fundamental. Note that the bound-
ary properties have been mostly studied on bicrystals displaying not only a coinci-
dence misorientation but also a particular low-energy plane; this is the condition of
boundary stability and thus for obtaining the expected bicrystal.
The tendency of a boundary to adopt low energy planes is well revealed by the
experiments of ball deposition on single crystal substrates and by the faceting phe-
nomenon. A relatively low energy is generally associated to a boundary plane pos-
sessing, in one crystal at least, a dense plane of the structure thus low {kkl} indices,
even if this is not a dense plane of the CSL lattice. If the boundary with a given
misorientation is symmetrical, its energy is often, but not necessarily, less than
that of the asymmetrical boundary. For copper and silver balls free to rotate on a
{111}, {110} or {100} single crystal substrate, the clearly favoured boundary planes
are {111}//{111}, {411}//{110} and {221}//{100}, respectively. Similarly, the faceted
boundaries in f.c.c. materials often adopt a {111}//{100} plane, despite it is incom-
mensurable [7].
This selection supports the efficiency of the high effective interplanar spacing
criterion (see Sect. 4.5.3) defined by Wolf [8] where deff is the arithmetic mean value
of the spacing between the planes parallel to the boundary plane in both crystals.
According to this criterion, a grain boundary may be considered as special if the
value of the deff /a ratio is higher a certain critical value (a is the parameter of the
crystalline unit cell). The critical value depends on the material, of the order of 0.150
for the f.c.c. structures [9]. Two geometrical grain boundary classifications based on
this criterion have been previously described [10, 11]. At the first and second levels of
each of these classifications, we find symmetrical and asymmetrical grain boundaries
with relatively elevated deff /a values (see Sect. 4.6.2). This dense plane selection not
only concerns metals and/or cubic materials; it is generalized to all materials; in
rhombohedral crystalline alumina, a large proportion of grain boundaries displays a
basal (0001) plane, in one crystal at least [12].
Closer is a boundary from the exact coincidence misorientation, better its plane
seems to adopt a symmetrical or asymmetrical position corresponding to low {hkl}
index values [7]. This experimental observation suggests that a classification based
on the misorientation, with a severe authorized angular deviation from the exact
coincidence, may indirectly give information on the boundary plane distribution.
The criteria low Σ index or large interplanar spacing are intrinsic to the
boundaries in that sense that they reveal their natural selection during elaboration
and/or grain growth of a polycrystal. But another criterion, that we classify here as
extrinsic, may also lead to a boundary texture, it consists in a preferential orientation
of the boundary planes with respect to a reference linked to the sample: free surface
for thin films or for sheets with thickness less than the grain size, solicitation axis for
polycrystals submitted to mechanical deformation or under the effect of a magnetic
field.
11.1 Criteria for Grain Boundary Distribution 343
11.1.3 Non Geometrical Criteria
One of the most often used criteria is the presence of intrinsic dislocations associated
to a particular grain boundary structure. It is at the basis of the definition of the upper
angular deviation from the coincidence misorientation (relations 11.1 and 11.5).
A second criterion rests on the occurrence of residual extrinsic dislocations after
deformation at high temperature, accounting for a particular mechanical behaviour
of some grain boundaries. The validity of this criterion is well illustrated in Fig. 11.1
that shows the grain boundary sliding quantity in function of the relative deviation
θ/θmax from the coincidence for different grain boundaries in aluminium [13].
Grain boundary sliding is accelerated only in boundaries where θ > θmax , and,
simultaneously, the presence of extrinsic dislocations is observed.
Fig. 11.1 Relation between

the grain boundary sliding
quantity and the magnitude
of the angular deviation from
the nearest coincidence given
by θ/θ max with θ =
θexp − θC S L and θmax , the
maximal deviation authorized
by the Brandon criterion. After
creep, no extrinsic dislocation
is observed in general grain
boundaries [13]
The time of relaxation of the stresses associated to extrinsic dislocations, inversely

proportional to the grain boundary diffusion coefficient DGB , may also be chosen as a
criterion for speciality [14]. The time tD for disappearance of the extrinsic dislocation
contrast at temperature TD , determined by In situ transmission electron microscopy,
is linked to the stress relaxation time (see Sect. 9.4.3) [15]; Its expression in function
of DGB and of the boundary width δ is given by an expression close Eq. (9.2):
tD = ATD /DGB δ (11.6)
From (11.6) the activation energy Q GB for grain boundary diffusion is deduced:
Q GB = RTD ln KtD /TD (11.7)

A grain boundary distribution based on the stress relaxation time is equivalent

to a distribution based on the boundary diffusivity; it thus has an evident physical
meaning compared to that established from geometrical criteria.
The three previous criteria do not enable to establish really statistical grain
boundary distribution because their use requires fine studies by transmission electron
microscopy, thus necessarily limited to a restricted number of boundaries.
In other few works, the grain boundary answers to an external stimulus are listed
and serve as basis for establishing a boundary distribution:
• Answers to an aggressive medium: wetting [16], corrosion [17] or mainly thermal

etching [18]. The grain boundary distribution is then established from the char-
acteristics of the etching groove (opening angle, depth . . .) and implicitly rests
on the assumption that a link exits between the answer of a grain boundary and
its energy. Despite all the limits evoked in Sect. 10.2.1, the Herring rule (10.3) is
often used to quantify the behaviours of different boundaries to wetting or thermal
etching.
• Answers to a mechanical solicitation: cavitation [19], rupture and grain boundary
sliding [13]. The evolution of these behaviours in function of the grain boundary
geometry is reported in the review paper [20].
In all cases, using these criteria lead to grain boundary distributions includ-
ing geometrical and chemical parameters and thus better reflects the real grain
boundary state in the polycrystal. However, their quantification is problematic and
any comparison between distributions remains limited to only one material.
Finally, grain boundaries may be classified on the basis of the complexion concept
(see Sect. 6.7 and Fig. 6.44 [21]).
11.2 Calculation of the Misorientation Distribution
Although the grain boundary misorientation distribution (GBMD or CGBD) only

partially results from the crystalline orientation distribution as it is also strongly
linked to the intergranular structure, a first approach of this distribution rests on
the existence of a coherency relationship between crystals in the polycrystal. It
is thus useful, as preliminary, to define some mathematical functions associated
to the crystalline structure, to specify the symbols used in the literature and the
existing relationships between the different functions that describe the material
microstructure. The second part of this section deals with the theoretical approaches
of the grain boundary misorientation distribution, calculated or simulated from a
random (or not) distribution of the crystalline orientations. Their results serve as
references to analyze the experimentally determined distributions.
11.2 Calculation of the Misorientation Distribution 345
11.2.1 Crystal Orientation and Grain Boundary

Misorientation Distribution Functions
The crystalline texture is quantitatively described by the Orientation Distribution

Function (ODF); similarly, the grain boundary misorientation distribution (GBMD)
may be quantitatively described by a distribution function named Misorientation
Distribution Function (MDF) that is not directly deduced from the previous one. The
MDF function gives the probability that grain boundaries have a given misorientation
between two adjacent grains in a polycrystal [22, 23]. The ODF function is supposed
to describe the anisotropy of a physical property that is insensitive to the interfacial
component of the microstructure. This is true, at first approximation, for magnetic
thermal and elastic properties, less true for the plastic properties and, a fortiori, for
intergranular fracture. Knowledge of the function MDF is thus complementary to
that of the crystalline texture in an attempt to understand the material behaviours.
The microstructure of any material is generally heterogeneous; this is the reason
why its characterization must include information on orientations and simultaneous
information on positions. An orientation correlation indicates that crystals with a
particular mutual orientation relationship are possibly adjacent.
The preference in polycrystals for grains to reside near other grains of a
particular orientation is quantitatively described by the Orientation Coherence Func-
tion or Orientation Correlation Function (OCF) [24]. The OCF function = c
(g, r, g ) constitutes the first approach for relating the grain boundary misorientation
distribution to the crystalline texture. It gives the probability density for the
simultaneous occurrence of crystallite orientation g at a point p and orientation g at a
point p (g and g represent each a set of Euler angles), the points p and p are located
independently from each other in a specific volume Vm and are separated by a vector
r. The dimension dm of the specific volume Vm is much larger that dc , the average
grain size. The OCF function also contains morphological information such as crys-
tallite size and shape as a function of the orientation. The measurements of the OCF
function must present a spatial stationary state; that means that the function must
not be changed by translation of the measurement volume. For a spherical volume
of radius rm , the variable r = |r| is restricted to the range 0 ≤ r ≤ 2 rm . Generally,
the OCF function is built for some r values that vary from a fraction until few multi-
ples of dc . From a practical point of view, a conditional OCF = c∗ (g, g |r), more
accessible by experiment, is defined as follows: c∗ (g, g |r)dg, dg is the probability
that point p lies in crystallite of orientation in the range (g, g+dg) and p lies in
crystallite of orientation in the range (g , g +dg ), p and p being separated by r. The
OCF and conditional OCF are linked by:
c(g, r, g ) = c∗ (g, g |r)R(r) (11.8)
with R (r) the probability density for vectors r occurring in the experiment [24].
When the microstructure is isotropic, the MDF function may be deduced from the
conditional OCF function. Let us consider g the misorientation between crystallites
of orientations g and g , in matrix notation:
[g ] = [g][g] (11.9)
The deviation g is represented in terms of Euler angles in the studies at the

mesoscopic level; it is equivalent to the θ angle previously used in microscopic
approaches and in simulations.
The misorientation distribution function MDF is given by f (g):
f (g) ∝ ∫∫∫ c∗ (g; g.g|r)dg (11.10)

with r
dc and the integration volume corresponds to a domain where all the
orientations are physically distinct in a cubic lattice [22]. The condition r
dc
required to calculate the MDF intuitively expresses that any coherency disappears
for very large distances compared to the average grain size. It is equivalent to say
that the processes that lead to coherency between crystals are supposed to operate
only on limited distances. If this assumption is verified, that may be confirmed only
by experiments, then, for r dc :
c(g, r, g ) = f (g)· f (g ) (11.11)
where f(g) is the ODF function.

More generally, for small values of |r|, the OCF function provides neces-
sary crystallographic details to establish a misorientation distribution. It may be
used to evaluate damages localized at grain boundaries and triple junctions or to
study the recrystallization mechanisms. For higher values of |r|, the OCF function
enables to quantitatively describe the local misorientation distribution function. The
dependence in r of the self-coherency function c(g, r, g) gives statistical information
on the grain shape and grain size in function of the crystallographic orientation g.
The correlation function OCF, for a given misorientation, may be obtained by
normalizing the experimental MDF with respect to the statistical MDF. The exper-
imental function f e (g) is determined by taking into account the length frac-
tion of the boundaries with misorientation g. The statistical function f s (g) is
calculated from crystalline orientation pairs randomly selected, without considering
their relative position in the microstructure. We obtain the OCF (g) function [22]:
f e (g)
OCF(g) = (11.12)
f s (g)
The OCF function is equal to 1 when non-correlation exists for the misorientation
(g) and satisfies OCF (g) > 1 (or OCF (g) < 1) if the misorientation g is
preferentially selected (or rejected) by comparison to what could be expected from
the relative abundance of orientations.
In the presence of a coherency orientation in a polycrystal, it seems appropriate to

identify a new microstructure scale named mesostructure that rests on the occurrence
of grain clusters (or aggregates). This first notion of mesostructure is incomplete; the
existence of grain clusters does not imply the existence of grain boundary cluster.
Indeed, even if each grain has several neighbours with which it is linked by a certain
type of grain boundaries, the latter are not necessarily connected each other. The
consideration of grain clusters may be important in view to understand properties
that depend on the grain boundary microstructure and on transport across boundaries
(electric resistivity for example). It may also enable to analyze properties sensitive
to propagation along grain boundaries (rupture, corrosion . . .).
11.2.2 Theoretical Misorientation Distributions
The theoretical proportions of grain boundaries so-called special including low

angle, coincident and CAD boundaries, are mainly estimated for materials of cubic
symmetry, by considering most often the Brandon speciality criterion. Very few
studies exist for non-metallic materials and/or with a non-cubic symmetry.
There are two manners to approach the theoretical misorientation distributions:
• The calculation of probabilities for a crystal, in a random polycrystalline material,

to be in an orientation with respect to a reference crystal such that it produces a
CSL or a CAD boundary.
• The random generation, on computer, of a series of rotation matrices simulat-
ing the formation of different grains of a random polycrystal, followed by the
determination of the resulting misorientations and their comparison with the ref-
erence matrices of the CSL and CAD boundaries.
11.2.2.1 Analytical Approach of CSL and CAD Grain Boundary

Probabilities in a Random Polycrystal
A grain boundary is defined by a vector V that describes the rotation (in orientation
and magnitude) required to bring a reference crystal into coincidence ( = 1) with a
second crystal. This vector with spherical coordinates, ν, θ and φ, lies within a sphere
of radius θ = 180◦ . ν is the maximal deviation authorized with respect to the exact
coincidence misorientation (v = v0 −1/n ). A distribution of rotations results in a
distribution of the density ρ of vector points in the sphere. A random distribution is
that which remains unchanged when any other crystal is chosen as reference crystal.
The proportion of coincidence grain boundaries with index in a random distribution
is obtained by multiplying the number of equivalent rotations (or equivalent regions
in the parametral sphere) by the probability ρ to find a random rotation within the
authorized deviation to the coincidence. The vectors V retained for the calculation
Fig. 11.2 Schematic illustra-

tion showing the parametral
sphere: a CSL zone is repre-
sented by a small sphere of
radius v at the extremity of
the rotation vector Vex = 0P;
a CAD zone is represented
by a cylinder centred on the
rotation axis [u,v,w]
of ρ have their extremities within a sphere of radius v, the centre of which is at

the extremity of the vector Vex corresponding to the exact coincidence (Fig. 11.2)
[25–27].
The probability ρ is given by:
ρ = (1/2π 2 )[(v − sin v)/2]0 [−cosθ ]π0 [φ]2π

0 (11.13)
with 0 ≤ ν ≤ π, 0 ≤ φ ≤ 2π and 0 ≤ θ ≤ π .
Table 11.1 gives, for the cubic materials, the percentages of boundaries near a
≤ 25 value, including low-angle grain boundaries ( = 1, θ ≤ 15◦ ), obtained
by using Eq. (11.13) and for two values of the authorized maximal deviation [27].
It is found that the fractions of these boundaries are small and decrease in a
non-monotonically way with the coincidence index , indicating an effect of the
lattice coincidence symmetry. The retained deviation criterion strongly influences
the obtained fractions. The proportion of twins is very low compared to that usually
encountered in materials. This result excludes a really random formation of polycrys-
tals that would be followed by a readjustment of the grain boundary angles during
growth or recovering in order to minimize the interfacial energy.
For material of hexagonal symmetry, the fraction of exact coincidence grain
boundaries (rotation axis parallel to the c axis) is identical, generally very low,
regardless of the material; but, the percentages of the different rational coincidence
boundaries √ vary with the c/a ratio (see Sect. 1.2.4). In the case of a compact structure
with c/a = 8/3, the total proportion of coincidence
√ boundaries with 7 ≤ ≤ 25
is equal to 0.9 %, it reaches 1.67 % for c/a = 5/2 [28]. In the case of rhombohedral
alumina, the theoretical proportion of rational coincidence grain boundaries is
0.22 %; it is 1.05 % for an ensemble of boundaries with 7 ≤ ≤ 31 [29]. Generally,
Table 11.1 Probabilities of occurrence of near coincidence grain boundaries in a random poly-
crystal for two maximal deviation criteria (Brandon’s criterion in −1/2 and −1 criterion)
Probabilities
−1/2 criterion (×10−2 ) −1 criterion (×10−2 )
1 1.98 1.98
3 1.53 0.29
5 1.07 0.10
7 0.86 0.05
9 0.88 0.03
11 0.68 0.02
13 0.59 0.01
15 0.82 0.01
17 0.51
19 0.48
21 0.66
23 0.43
25 0.48
Total 3 ≤ ≤ 25 9 0.5
the relative number of coincidence grain boundaries decreases with the diminution
of the lattice symmetry that leads to a smaller number of equivalent descriptions.
The vectors corresponding to near one-dimensional coincidence (CAD) grain
boundaries are located in a cylinder centred on a diameter parallel to the rotation axis
cutting the parametral sphere (Fig. 11.2). The consideration of symmetries for a cubic
crystal leads to numerous intersections of these cylinders and thus to an overlap of
the CAD zones. So the percentages of CAD boundaries obtained must be compared
to those derived from simulations. In the limit of Eq. (11.5) with θ0 = 15◦ , the
proportion of boundaries near a CAD description around 100 is 12 % for f.c.c.
structures and 9.3 % pour for b.c.c. structures. By considering rotations around 100,
110 and 111 axes, this percentage reaches 50 % in f.c.c. materials and it is very
sensitive to the slightest deviation from random distribution. This proportion is clearly
superior to that of the near CSL boundaries. A more reasonable percentage of 5 %
has been analytically found, by restricting to 2◦ the maximal deviation between exact
and real axes [29].
11.2.2.2 Simulation Approaches of CSL and CAD Grain Boundary

Probabilities in a Random Polycrystal
A series of rotation matrices are generated according to the procedure proposed by

Mackenzie and Thompson [30]; then, these matrices are converted in misorientation
matrices and compared to the theoretical matrices for the CSL and CAD boundaries.
The study by Warrington and Boon [27] deals with two sets of 400 rotations randomly
Table 11.2 Calculated average percentages of coincidence grain boundaries (using the Brandon
criterion) and of low-angle boundaries in a random polycrystal and in three polycrystals displaying
a fibre texture [31]
Grain boundary percentages
Random polycrystal 100 % 100 50 % 100 50 % 111 100 % 111
Grain boundaries 1 – 65 17.82 42.92 27.77 35.54
Grain boundaries 3 – 25 8.61 14.57 11.37 13.08
Grain boundaries 1 2.21 22.71 10.31 17.05
generated. The results, compared to those analytically established, show an overesti-

mation of the low-angle grain boundaries that the authors have corrected by dividing
by 3 the obtained percentage. For the other values, the percentages slightly differ
from one set to the other and from those predicted by the calculations, but the total
fraction of near coincidence boundaries with 3 ≤ ≤ 25 is 68/800 (8.5 %), thus
very near that found by calculation (9 %). This proportion and its possible variation
for non-random polycrystals seem difficult to measured by experiments apart from
the case of a very pronounced texture. The percentages of one-coincidence bound-
aries obtained by simulation are comparable to those previously calculated, reaching
50 % for boundaries around 100, 110 and 111 in a cubic material.
The same procedure of matrix generation than that previously used, allows
Garbacz and Grabski to establish a random distribution in a polycrystal constituted
by 10 layers of 20 × 20 grains, i.e. 4000 grains and 24290 grain boundaries [31].
The structure is cubic and the tri-dimensional space is filled with identical Kelvin
polyhedra. Note that these polyhedra are limited by eight hexagonal faces and six
square faces and that stacking requires an offset of a layer relative to the other
(Fig. 11.3). Consideration of Kelvin polyhedra is a reasonable approximation of a
real polycrystalline arrangement. The distributions of the axes and of the misorien-
tation angles are first determined in order to test the validity of the model. Both are
in good agreement with those analytically calculated [25–27]. More than 50 % of
the misorientation axes are in the zone V of the standard stereographic triangle and
about 25 % in the zone IV, i.e. far from simple crystallographic axes (Fig. 11.4). The
misorientation angles display a normal distribution with a maximum around 45◦
(curve a on Fig. 11.5); The effect of a fibre texture on the boundary misorientations is
simulated with the assumption of non-correlation between the orientations of neigh-
bouring crystals, it results a bimodal misorientation distribution (Fig. 11.5) [32].
Near coincidence grain boundaries with (1 ≤ ≤ 65), using the Brandon
criterion, are then selected. Their calculated and averaged proportions are gathered
in Table 11.2 for four types of polycrystals: random, with a 100 fibre texture, a
mixed 50 % 100/50 %111 texture and a 111 texture [31].
By comparison with the random distribution simulated by Warrington and Boon
[27], the percentage of low-angle boundaries is notoriously lower (comparable to
that analytically determined), the total percentage of coincident boundaries ( ≤ 25)
remaining of the order of Fig. 11.6. The overestimation of the simulated proportion of
Fig. 11.3 a Kelvin polyhedron; b tri-dimensional model of a polycrystalline aggregate built with
Kelvin polyhedra (N polyhedra per side); the polyhedra in grey correspond to the base layers and
to those laterally displaced
Fig. 11.4 Standard stereo-

graphic triangle cut in zones
in which the misorientation
axes are distributed; their
proportion in each zone is
indicated [31]
1 in [27] may be attributed to an insufficient number of rotation matrices generated

to model a random polycrystal.
The influence of the texture on the grain boundary character distribution is clearly
revealed (Table 11.2). Among the coincidence boundaries, the percentage of = 5
strongly varies from one fibre texture to the other: 8.6 % for 100, 0 % for the
111 instead of 1.07 for a random distribution. The proportions given in Table 11.2
diminish if we introduce a correlation between the crystalline orientations in the
simulation, but the main differences between textured and not textured materials are
preserved (higher percentage of low-angle boundaries whatever the texture and also
higher percentage of = 5 boundaries for the 100 fibre texture [32].
Fig. 11.5 Simulated grain

boundary misorientation
distributions: curve (a)
corresponds to a polycrys-
tal without texture; curves
(b), (c) and (d) correspond
to polycrystals displaying a
fibre texture 100, 110 and
111, respectively [32]
Fig. 11.6 Compared grain

boundary misorientation
(3 ≤ ≤ 25) distributions
in a random polycrystal [31]
In polycrystals of hexagonal structure, the misorientation angles, whose the

maximal value is 93◦ (instead of 62◦ for the cubic structures), display a random
normal distribution with a maximum near 60◦/70◦ . In this system, the random distri-
bution of grain boundaries based on their coincidence index in not unique, the rational
coincidences being specific to each material [28]. This is the same for rhombohedral
materials [29], and more generally for all materials of non-cubic symmetry.
Materials susceptible to multiple twinning constitute a particular class of materials
that gave rise to several models of grain boundary distributions. On one hand, these
materials have been largely studied and are widely used in industry; on the other
hand, they have specific grain boundary spectra. The principle to simulate these
distributions is briefly described hereafter [32]. The simulation always starts with
a single crystal; a new grain appears at each simulation step. Among the N grains
formed at the Nth step, a grain is randomly chosen and another grain in twin rela-
tionship is added to the previous, also randomly. This new grain is denoted N+1, all
the misorientations with the neighbouring grains are calculated again. Ten random
realizations are simulated at each step and the results are averaged. We remark that
beyond 50 simulation steps, the distribution spectrum of the = 3n grain bound-
aries is quasi-stationary; however, it slightly depends on the angular deviation to the
= 3 coincidence imposed in the simulation (Table 11.3).
Table 11.3 Simulated percentages of = 3n grain boundaries in materials susceptible of multiple

twinning in function of the number of simulation step (left) and in function of the angular deviation
from the exact 3 coincidence misorientation introduces in the calculation (right) [32]
Grain Exact = 3 misorientation Maximal angular deviation θmax
boundary Number of simulation steps from exact = 3 misorientation
100 200 500 1◦ 2◦ 3◦
3 43.7 ± 2.7 43.2 ± 2.2 43 ± 1.3 44.8 ± 3.9 45.4 ± 4.0 42.8 ± 2.9
9 26.8 ± 2.6 26.6 ± 1.6 26.3 ± 1.6 25.2 ± 2.3 18.8 ± 2.4 17.8 ± 2.2
27 14.4 ± 2.4 14.3 ± 1.6 14.1 ± 1.3 7.3 ± 1.7 7.3 ± 1.6 7.4 ± 1.2
81 7.9 ± 1.6 7.8 ± 1.2 7.7 ± 0.9 1.2 ± 0.6 3.1 ± 1.3 5 ± 1.3
The 3n grain boundaries cover a large fraction of the misorientation ensemble
in cubic materials with low or medium stacking fault energy. Indeed, if an interaction
between grain boundaries occurs, only = 3n boundaries may appear according
to Eq. (10.1) on the coincidence index combination at a triple junction. This rule
predicts a maximal proportion of twins equal to 2/3 in a polycrystals containing only
triple junctions (i.e. to the exclusion of any quadruple junction or higher order), the
resulting microstructure is controlled by twinning.
Practically, the experimental misorientation distributions may significantly
deviate from the calculated distributions (random or textured), but the latter provide
a benchmark for assessing the degree of balance of a polycrystal and understanding
its mesostructure.
11.3 Experimental Grain Boundary Misorientation

Distributions
The real misorientation distribution in a well-recrystallized polycrystal, in a

metastable equilibrium state, depends on the structure and the purity of the
material. Furthermore, it may be strongly affected by the elaboration proceedings
and the thermo-mechanical history of the material, leading to crystalline texture.

This distribution is often expressed by the relative number of grain boundaries
of a given misorientation fN (θ ) brought in % (θ ), but it can also be estimated as a
fraction of area occupied by one type of boundary fS (θ ), often obtained as a fraction
of the observed lengths fL (θ ). Indeed, the experimental distributions are generally
established on planar sections of samples (thin foils or polished free surfaces), but
according to simulations, it seems that these distributions are representative of those
existing in the volume. A comparison between the measured fraction of grain bound-
aries of a given misorientation and that which would result from the knowledge of the
texture sometimes completes the analysis of distributions and is estimated in terms
of correlation.
If the distribution is given in terms of coincidence, any comparison between the
various results requires checking that the same criterion for the maximal devia-
tion from the coincidence misorientation is used. This is usually the Brandon cri-
terion although it is very permissive, especially for = 3 grain boundaries. For
example, a boundary deviated by a total angle θd = 7.6◦ , less than θmax asso-
ciated to = 3 (θmax = 8.6◦ according to Eq. 11.1), may be described with
more accuracy if considered as deviated by 0.8◦ from = 81. The use of the
criterion θmax = 8◦ −1 [3] or θmax = 15◦ −5/6 [4] enables to avoid this ambi-
guity. But, whatever the retained criterion, the different data are difficult to directly
compare, due to the different approaches and to the experimental errors that impede
a true reliable distribution of all the boundary types. Any attempt to compare grain
boundary distributions between different materials must take into account by only the
results concerning polycrystals in comparable microstructural states. In particular,
near stable equilibrium states are considered: metals completely recrystallized (high
temperature close to the melting point Tm and/or prolonged annealing time), creep
leading to a minimal porosity for ceramics, slow solidification followed by annealing
for semiconductors. This is why we are firstly interested by grain boundary distri-
butions in relatively equilibrated polycrystals in materials with different structures,
purities and stacking fault energies. Then, the materials in industry being most often
in a non-equilibrium state, we consider the effects of various thermo-mechanical
treatments and of the resulting textures on the grain boundary ensemble in a given
material.
Each effect is illustrated by a limited number of examples. Many data from work
done in the years 1980–1990 (beginning of a considerable effort on this topic related
to technology development) are drawn from review articles [13, 32, 33], they are
supplemented by results of recent investigations.
11.3 Experimental Grain Boundary Misorientation Distributions 355
11.3.1 Different Types of Experimental Misorientation

Distributions
In this section, various results, sometimes contradictory, on the misorientation distri-

butions in different materials have been gathered by taking into account their common
features, according to a classification proposed by L. Fionova for the metals [33].
The results concerning metals are indeed by far the most numerous, in particular
for f.c.c. structures. We try to situate in this scheme some existing results on other
materials (semiconductors, oxides...).
Group I contains materials of f.c.c. structure, with a relatively low stacking fault
energy, susceptible to thermal twinning leading to lamellar twins. We add to metals,
first placed in this group (copper, nickel, some stainless steels…), the semiconductors
(silicon and germanium). Only one characteristic does not permit to place a material
in a group: for example, nickel has a medium stacking fault energy, but it displays
lamellar twins.
Group II is characterized by a distribution analogous to that observed in aluminium,
f.c.c. metal with a high stacking fault energy; twins only appear in the crystals corners.
Group III concerns b.c.c. metals that do not undergo thermal twinning and possess
a relatively high stacking fault energy.
An empirical relationship has been obtained for each type of distribution, on the
form [34]
F = ki −ni (11.14)
with F , the frequency of occurrence of coincidence boundaries. F is a sub-set

of the MDF function or f (g) according to Eq. (11.10).
For group I : kI ∼ = 100 and nI ∼=1
For group II : kII ∼= 20 and nII ∼= 1/2
For group III : kIII ∼
= 5 and nIII ∼
= 1/3
The grouping of metals according to their grain boundary distribution in
polycrystals appears to correspond to an electronic structure analogy. If we applied
to grain boundaries the formula of the interfacial energy E GB proposed for the free
surfaces [35], we obtain:
E GB ∼
= 0.25E c Z GB /Z v (11.15)
with E c , the cohesion energy, ZGB and Zv , the coordination number at grain
boundaries and in the bulk, respectively. For metals of electronic structure d with a
filling Nd of the layer d [35]:
Nd (10 − Nd )
E c = E cmax (11.16)
25
with E cmax the cohesion energy for Nd = 5.

Group I corresponds to metals having a relatively low cohesion energy with
Nd = 10 (Cu, Au, Ag) and Nd = 9 (Pt, Pd, Ni). On the contrary, the cohesion
energies for group III are stronger Nd = 4 (Nb), Nd = 5 (Mo) and at the limit
Nd = 7 (Fe). Aluminium, of electronic structure s.p., cannot be considered in this
analogy. As the macroscopic criteria retained to define the groups (structure, stacking
fault energy…), the electronic criterion suffers exceptions. As a result, overlap in the
classification occurs, affiliation of a material to a group requires a combination of
several criteria mentioned above.
11.3.2 Effects of the Structure and of the Stacking Fault

Energy of the Material
These two effects are gathered because they enable to deal with the misorien-
tation distributions according to the distinction previously defined. The results
analyzed hereafter mostly come from studies on well-recrystallized metals or metal-
lic alloys. Nevertheless, the comparisons are difficult since solute traces can sig-
nificantly change the boundary distribution in the case of almost pure metals and
the recrystallization conditions (or other elaboration conditions) affect this distri-
bution in all materials. Efforts are being made here, as much as possible, to ana-
lyze the results obtained on materials of controlled purity, having underwent similar
thermo-mechanical treatments.
11.3.2.1 Group I Materials
The results on materials of group I are the most numerous. They concern different
metals or alloys and different static recrystallization treatments for a same mater-
ial. The main common features to the different distributions with a grain boundary
number between 200 and 7000, depending on the investigation methods, are
summarized below:
• The distributions are essentially similar whatever the chemical composition and
the thermo-mechanical history of materials.
• The distributions are not monotonous, with a large predominance of = 3n grain
boundaries, in particular = 3.
• The proportions of grain boundaries other than = 3n are approximately those
predicted by the calculations of a random polycrystal.
To support the first common feature, we mention works on two types of stainless
steel polycrystals that underwent the same thermal treatments of static
recrystallization, but some of them have been obtained after deformation of a single
crystal, the others after deformation of a polycrystal. The resulting microstructures
noticeably differ by their grain size (6 and 24 μm, respectively) and by their distri-
bution of intergranular M23 C6 , precipitates, but the percentages either in number or
in = 3n boundary length are identical. The misorientation spectrum well appears
independent of the history of materials of group I [31].
Table 11.4 Percentages of different = 3n grain boundaries in some f.c.c. materials susceptible
of thermal twinning and having underwent the same static recrystallization treatment [31, 32]
Alloy Ni Ni-20% Cr 304 steel 316L steel Cr16Ni15Mo
Anneal 1573 K 993K + 1273 K 1366 K 1123 K 1423 K 1423 K 1373 K + 923 K
10 min 2 h 30 min 30 min 1 h 1h 30 min 10 min 2 h
% =3 41 34.5 34.6 42 35.5 33.1 34.6 34.8 37.5
% =9 9 7.5 7.4 15 5.2 4.7 6 3.6 3.6
% = 27 5.9 5 3.1 4 2.4 1.2 5.3 2.7 3.6
% = 81 1.8 3 – – 2.9 1.2 3.8 – –
The proportions of = 3n grain boundaries in different metals and alloys are

given in Table 11.4 [31, 32]. Compared to the values determined by
simulation (Table 11.3), they are clearly inferior for all the coincidence indices apart
from = 3.
We note a very small increase of the twin density with the annealing temperature.
This effect may be explained by an increase of the grain boundary migration rate that
yields a higher frequency of twinning events. The twin density reaches a maximal
value in the first stages of grain growth when the boundary migration rates are higher.
When growth goes further, the percentage of twins somewhat decreases and stabi-
lizes; this phenomenon has been attributed to twin annihilation by their interaction
with other grain boundaries [36].
The length fractions, related to the fractions of areas occupied by different
boundaries are reported for guidance in Table 11.5. It is only for = 3 that this
proportion is clearly superior to the grain boundary number proportion. It seems also
more sensitive than the number ratio to the nature of the material (stacking fault
energy) and to its thermo-mechanical history. The relative lengths for the other
= 3n grain boundaries are rather low.
Table 11.5 Relative lengths

Nichrome 304 steel 316L steel
(%) of = 3n grain
boundaries in various metallic 3 54.8 59.7 48.7
alloys 9 4.5 1.9 1.8
27 2.9 2 1
Generally, the relative length of = 3 grain boundaries exceeds 50 % and

increases with annealing time. This remark suggests that the boundary network tends
to a better equilibrium by multiplying the total length of = 3 boundaries, and not
their number, at the expense of higher energy boundaries. The = 3n boundaries
are not energetically favoured, their relatively high percentage is uniquely due to
geometrical reasons: they ensure connections between the = 3 grain boundaries
in a process of multiple twinning.
Fig. 11.7 Distribution of

the angular deviations from
exact coincidences in 316L
steel using the Brandon cri-
terion θmax = 15◦ −1/2 :
a deviations for the = 3n
boundaries; b deviations for
other boundaries [32]
In these studies, a tendency appears concerning the angular deviation from the
exact coincidence: it is generally lower for the = 3n than for the other boundaries
(Fig. 11.7) [31, 32]. This comment should be compared to another trend indicating
that the smaller is the deviation from the coincidence, the higher is the probability
for the grain boundary plane to be close a special plane, symmetrical or of low
indices [7].
Among the boundaries other than = 3n , that generally present percentages
close to the random distribution, the case of = 11 deserves discussion. Although
they constitute a large number of grain boundaries after sintering of small copper and
silver balls on a single crystal substrate [37], very few boundaries near = 11 are
observed in f.c.c. polycrystals [38]. The proposed explanation for such a difference
is that the = 11 boundary displays low energy in a bicrystal where it is free of
rotation stresses, this is the case in the sintering experiments; its rare appearance in a
polycrystal is attributed to stresses that impede grain rotation during annealing. Note
that this explanation also permits to justify the small percentage of any boundary other
than = 3 {111}. The extremely low energy of the coherent twin = 3 {111}
explains its selection in polycrystals, even in the presence of stresses; the formation of
the other = 3n boundaries results geometrically. Although the calculated energy
of the = 11 {311} grain boundary in f.c.c. metals corresponds to the second
minimum in the curve E = fn (θ ) (Fig. 1.4), it is significantly higher than that of the
twin. The non-selection of the = 11 boundaries seems to indicate that the effect
of the stresses developed during polycrystal elaboration overcomes the trend to the
selection of low energy boundaries. This is the evolution of the total energy of the
polycrystalline system that governs the final microstructure.
A comparison of the misorientation distributions in different metals and alloys
that have underwent the same static recrystallization treatment, clearly shows that
Fig. 11.8 Frequencies of = 3 grain boundaries in different materials susceptible to

multiple twinning. The averages are obtained for about 200 investigations on well-recrystallized
polycrystals [39]
the percentage of = 3 grain boundaries increases when the stacking fault energy
decreases (Fig. 11.8) [39].
For semiconductors attached to group I, massive polycrystalline silicon annealed
at high temperature (0.9 Tm ) presents a relatively high number of = 3 grain
boundaries, on the form of lamellar twins; this phenomenon is accompanied by
multiple twinning [33]. The length over thickness ratio of these twins is very elevated
(50–100) and it is the same for thin silicon films. The percentage of = 9 boundaries
does not exceed that predicted by the random distribution [28]. By comparison, the
misorientation distribution in germanium annealed at 0.9 Tm is more diversified and
the angular deviations from the exact coincidence are generally higher than in silicon.
To group I, we may add the intermetallic compound TiAl of L10 structure
that is susceptible to thermal twinning. Due to its rhombohedral structure, this
material displays coincidence grain boundaries with even indices. If the = 3n
and = 3n × 2 are gathered, the obtained percentages are comparable to those
indicated in Table 11.4: 33.8 % for the = 3 + = 6 boundaries, 12.4 % for the
= 9 + = 18 boundaries and 3.1 % for the = 27 + = 54 boundaries. Even
in the absence of a specific thermal treatment to develop crystalline texture, materials
of group I presents misorientation distributions very far from those theoretically pre-
dicted by a random distribution. The = 3 boundaries, possessing very particular
properties, are often set apart in the description of grain boundary networks. Their
significant presence in the microstructure likely indicates special properties of the
polycrystalline ensemble.
Table 11.6 Proportions of

Annealing conditions
coincidence grain boundaries
≈0.6 Tm , 500 h ≈0.9 Tm , 50 h
( ≤ 17) in high purity
aluminium that underwent 3 10.5 10
different recrystallization 5 5 –
treatments: ≈ 0.6 Tm , 500 h 7 – –
and ≈0.9 Tm , 50 h [33] 9 1.2 4.7
11 3.2 6.2
13a 2.7 –
15 – –
17a 2.2 4.1
17b 1.2 0.6
11.3.2.2 Group II Materials
For materials with high stacking fault energy, the misorientation distribution totally
differs; it is diversified with the appearance of non-negligible fractions of boundaries
other than = 3n , especially when the annealing is performed at low tempera-
ture. In well recrystallized (0.9 Tm ) pure aluminium, the total percentage of coin-
cidence grain boundaries is less than those encountered in the materials of group I
(25–35 % for ≤ 25), with however the presence of = 11 boundaries, quasi-
absence previously. The proportions of the different coincidence grain boundaries,
apart from = 3, are very sensitive to the temperature and duration of the
recrystallization treatment (Table 11.6) [33]. An augmentation of the annealing tem-
perature leads to an increase of the = 9 and = 11 boundaries and less dispersion
of the distribution. The distributions indicate that the coincidence boundaries other
than = 3n are more selected when the grain size decreases (low annealing T◦ )
[40]. This comment may also be deduced from a compilation of results showing that
most coincidence boundaries are no longer favoured beyond certain relative anneal-
ing temperature (T/Tm ≈ 0.8); on the contrary, the = 3n boundaries are selected
at any temperature [41]. This effect is explained by the reduction of the mobility
differences between grain boundaries when the annealing temperature is very close
to the melting temperature.
Other results also show that well-recrystallized aluminium presents misorienta-
tion distributions very sensitive to thermo-mechanical treatments. If we proceed to
recrystallization after a strong deformation of a single crystal plate of aluminium, the
misorientation distribution in the resulting polycrystal depends on the initial orien-
tation of the crystal before deformation (given by its surface orientation) and on the
preliminary deformation level (60, 70, or 80 % compression) (Fig. 11.9) [42]. Two
tendencies appear with increasing deformation level: a decrease of the low-angle
boundaries and a slight augmentation of the coincidence boundaries.
The increase of the percentage of coincidence grain boundaries with the defor-
mation level may be explained on the basis of the boundary migration processes
that depend on the absorption of lattice dislocations. We have seen that absorption
is favoured by the temperature, by the density of dislocations introduced in the
materials and depends on the boundary type, easier in general boundaries than in
coincident ones (see Sect. 9.4.2). So general boundaries absorbing lattice disloca-
tions may migrate, change their misorientation towards coincidence to become more
stable.
To sum up, the experimental misorientation distributions in materials of group II
differ from the random distributions and also from those obtained for materials of
group I. With respect to the latter, we mainly note a lower proportion of twins, a
larger dispersion of the coincidence boundaries and, above all, a high sensitivity to
the thermo-mechanical conditions of elaboration of polycrystals.
11.3.2.3 Group III Materials
For b.c.c. metals with a high stacking fault energy, the percentage of coincidence
grain boundaries is small, from about 2 % in niobium to about 10 % in high purity
iron and molybdenum annealed at high temperature [33]. The latter proportion is near
that theoretically predicted. Deviations θ from the coincidence misorientations are
generally elevated. Distributions are diversified, without preferential misorientations,
and are very sensitive to the purity of the material and to the thermal treatment it
undergoes. The results on b.c.c. metals and alloys (Fe-Si, molybdenum, tungsten …)
most often include texture effects; they are presented further (see Sect. 11.3.4).
Although non-mentioned in the classification by L. Fionova [33], we sug-
gest attaching to this group the hexagonal metals; the experimental misorientation
distributions of these metals, in the absence of texture, are similar to the random
distribution. This is the case of plates of high purity zinc (99.999 %) annealed at
350 ◦ C during 30 h for which the curve giving the misorientation frequency f (θ )
in function of θ does not present any peak; this curve presents a maximum situated
around 60◦ , as it was theoretically predicted [43].
Note that, whatever the group concerned, the previous statistics do not take into
account the CAD grain boundaries that mostly appear in the presence of a crystalline
texture (see Sect. 11.3.4).
Fig. 11.9 Frequencies of grain boundaries depending on the compression level (60, 70 and 80 %)
before annealing of aluminium single crystals with four different initial orientations 1, 2, 3 and 4 (see
standard stereographic triangle): a = 1 boundaries; b Coincidence boundaries 3≤ ≤ 29 [42]
Finally, the differences between the misorientation distributions in recrystallized

materials belonging to the different groups as well as the complex role of the solutes
in these distributions may be explained by three factors, energetic, kinetic and
geometric, that govern the final polycrystal microstructure [36]. The energetic and
geometric factors are more important for the pure metals of group I: the selection of
the = 3 grain boundary is primarily explained by its very low energy (the twin
energy is generally of the order of 5 % of the general grain boundary energy); the
presence of = 3n boundaries results from geometrical constraints at the connec-
tion with = 3. The kinetic factor also acts on the twin frequency by favouring, in
a certain limit, their presence when the temperature and the grain size increase. On
the contrary, and whatever the group to which a material belongs, the proportions of
the other coincidence grain boundaries increase when the grain size decreases. This
tendency has been detected in Fe-3% Si [13] and in pure aluminium [40].
Anneals at low temperature simultaneously lead to a fine-grained microstructure
and an elevated quantity of coincidence grain boundaries. A high temperature yields
a decrease of the activation energy for grain growth; the coincidence boundaries
are eliminated by the more mobile general boundaries; the resulting microstruc-
ture is then heterogeneous [40]. This selection mechanism may be modified in the
presence of solutes segregated to general boundaries that decrease their mobility (see
Fig. 11.11).
11.3.3 Effects of the Material Purity
The influence of a solute on the grain boundary distributions in a polycrystal of

a given material differs according to the chemical state and the repartition of the
solute: in solid solution in the matrix, segregated to grain boundaries, engaged in
intra-or intergranular precipitates. The energetic and kinetic effects that occur may
be opposed and develop more or less depending on the group at which the material
belongs. Many elements dissolved in matrix yield a decrease of the stacking fault
energy and favour the formation of = 3n grain boundaries; this effect must have
an important role in the materials of group I, even group II. In the presence of
segregation, another effect may oppose the previous: the energy differences between
grain boundaries vanish (analogous to a temperature increase effect) and one of the
reasons for selection of particular misorientations disappears. This has been observed
in b.c.c. metals ( group III): the percentage of coincidence boundaries, already low
for high purity metals, tends to zero when the solute concentration increases [33].
Segregation and precipitation generally impede grain boundary migration leading to
the finale microstructure after recrystallization. However, the influence of purity on
the kinetic factor is more complex. It depends on the solute quantity in the volume,
the segregation level that varies from one boundary to the other, the precipitate shapes
and the localizations. We describe here the influence of the material purity on the
grain boundary texture, in an attempt to highlight the major reason that governs a
change of this texture.
Table 11.7 Percentages of coincidence grain boundaries in nickel polycrystals containing different
sulphur quantities and submitted to a recrystallization treatment at 1450 K during 2 h
Ni − 0.3 ppm S Ni − 3 ppm S Ni − 10 ppm S Ni 99.97 % purity
1 1 4. 5 3.5 0
3 34 28.5 30.1 41
5 1.5 3 0.7 0.9
7 0.8 1.8 3.5 0
9 8.3 8.4 9.1 9
11 3 5.3 0.7 0.4
27 4.2 2.1 2.8 5.9
5≤ ≤ 25 9.4 19 11.9 2.3
(without 9)
The percentages determined on a less pure nickel recrystallized at 1573 K during 10 min are also
reported (numbers extracted from reference [36] and rounded to the first decimal)
From a historical point of view, one of the first investigations on the role of a solute
in the grain boundary distributions has been performed on molten lead (obtained by
zone melting) containing tin traces. The lack of solutes, as well as it excess, yields
an increase of the general boundary proportion; otherwise, an intermediary content
favours coincidence boundaries [1]. Similarly, the percentage of coincidence grain
boundaries ( ≤ 29) goes from 20 % in recrystallized pure aluminium to 50 % in this
metal containing 10 ppm of tin; adding 10 ppm of tin reduces the previous proportion
to 24 %.
A more precise investigation reveals the complex role of sulphur in the grain
boundary misorientation distribution in 99.999 % high pure nickel (Table 11.7) [36].
A very small addition of sulphur (3 ppm) yields a slight decrease of the = 3
boundaries and an increase of the total percentage of coincident boundaries (5 ≤
≤ 25) other than = 3 and = 9. The proportion of = 9 boundaries remains
almost constant whatever the sulphur content in nickel. In Table 11.7 the boundary
proportions determined in less pure nickel recrystallized at 1573 K during 10 minutes
are also reported: the twin percentage is then superior to that found after annealing
at 1450 K during 2 h. This difference is attributed to an effect of the temperature rise
on the boundary mobility. In nickel containing 3 and 10 ppm of sulphur, we note a
= 1 boundary frequency higher than that predicted by a random misorientation
distribution using the Brandon criterion.
The effect of a solute on the grain boundary distribution obviously differs, depend-
ing on whether this element may or not concentrate at grain boundaries during the
thermal treatment that leads to the observed microstructure. For example, when
the intergranular sulphur segregation in nickel occurs after or simultaneously to
the recrystallization treatment, the resulting boundary distribution strongly differs
(Fig. 11.10) [9, 20].
These results rest on transmission electron microscopy experiments, the number
of observed boundaries is then very inferior to that taken into account to establish
Table 11.7. However, we note similarities between the results: (a) proportions almost
similar of = 3 and coincident boundaries (including = 9) in bulk samples

and in thin foils of pure nickel (% S ≤ 3ppm). When the recrystallization treatment
does not allow sulphur to segregate to grain boundaries (b), the percentage of = 3
boundaries increases with the sulphur content in the matrix. On the contrary, if
sulphur segregation occurs during the boundary selection, the relative number of
twins greatly decreases, the number of general boundaries increases and low-angle
boundaries appear (c). This tendency is certainly explained by the fact that boundary
energy differences fade in the presence of solutes or impurities.
In recrystallized materials, the effect of impurities on the grain boundary mis-
orientation distributions, in particular for materials of group I, may be explained on
the basis of the three factors (energetic, geometric and kinetic) previously evoked
(Fig. 11.11).
The role of the energetic factor, that manifests itself by twin selection in the course
of multiple twinning, is mainly significant for low impurity contents. But, when the
impurity level increases, the contribution of this factor to the change of the grain
boundary texture diminishes. This evolution is observed not only for nickel but also
for the intermetallic compound Ni3 Al and for Ni-based super-alloys. The kinetic
factor has also a growing importance in the presence of impurities or solutes. A
maximal effect of this factor occurs for solute concentrations such that the mobility
of certain coincidence boundaries is significantly higher than that of general bound-
aries, the latter being preferentially affected by the segregation phenomenon. For
a very high purity or, inversely, when the segregation concerns all the boundaries
(excepted the coherent twin, the boundary mobility differences are low. The geomet-
ric contribution, which particularly results in the formation of = 3n boundaries,
can be considered independent of the purity of the material [36].
Fig. 11.10 Histograms showing the frequencies of = 3, CSL with ≤ 19, 1 (LA) and general
(G) grain boundaries in nickel polycrystals depending on their sulphur content and their thermal
treatment: a pure nickel recrystallized at 1050 ◦ C during 4 h; b Ni − 16 ppm S recrystallized as in
a then annealed at 625 ◦ C during 2 h to permit sulphur segregation; c Ni − 8 ppm S that underwent
a simultaneous treatment of recrystallization and segregation at 625 ◦ C during 12 days [20]
Fig. 11.11 Relative contribu-

tions of the three factors that
control the coincidence grain
boundary selection in the
misorientation distributions in
function of the solute content
of the polycrystals [36]
In the case of sintered materials (metals or oxides), the effect of a solute or doping
element on the grain boundary texture is mostly unknown. However, an interesting
result concerns rhombohedral alumina polycrystals (Fig. 11.12); as sintered pure
alumina presents two types of grain boundaries: general and one-dimensional (CAD)
boundaries with a proportion of about 13 %. After sintering of an alumina doped with
500 ppm of magnesia, the total fraction of coincidence grain boundaries becomes
important (∼=39 %), including a small percentage of low-angle (LA) boundaries,
20–22 % of exact or rational tri-dimensional coincidence (CSL) and about l5 % of
CAD boundaries [29]. The simultaneous doping with 500 ppm of magnesia and
500 ppm of yttria leads to a percentage of CSL boundaries (∼ =12 %) inferior to the
previous one, but the total fraction of coincidence boundaries (LA, CSL and CAD)
remains more elevated (∼ =27 %) than that obtained in pure alumina. These results are
correlated to the microstructures of the alumina polycrystals. Pure alumina displays
an heterogeneous microstructure with the local presence of grains of large size and
elongated shape, bordered by grain boundaries with straight planes (see Sect. 11.4);
doping with 500 ppm of magnesia and co-doping with 500 ppm of yttria results
in an equiaxed fine-grained microstructure with a high proportion of coincidence
boundaries; this effect is comparable to what has been observed in aluminium and
in a Fe-3% Si alloy when the grain size decreases.
Fig. 11.12 Histograms

showing the frequencies
of low-angle (L.A), tri-
dimensional coincidence
(C.S.L.) and one-dimensional
coincidence (C.A.D.) grain
boundaries in alumina
polycrystals after sintering:
a pure alumina; b 500 ppm
MgO doped-alumina [3]
The influences of solutes or impurities on the grain boundary texture, previously

evoked, are far to be exhaustive. It is often difficult to dissociate them from their
important effects on the crystalline texture. Some of the results on the thermo-
mechanical treatment effects, described below, indirectly include purity effects.
Furthermore, we do not report here on the role of solute engaged in intra- or
intergranular precipitates. The particles of second phase strongly affect the mobility
of boundaries, and thus their finale distribution in the material, but there are few
statistical studies on this subject.
To conclude, understanding the role of solutes (impurities) in the grain boundary
distribution requires an estimate of the combined effects of energy, geometry and
kinetics on the selection and the mobility of different boundaries. Knowing that
solutes also modify the crystalline orientation distribution, it is convenient to establish
a relationship between their effects on both textures (grains and grain boundaries).
Investigative efforts must be undertaken in this direction, given that the materials
used in industry are generally unclean.
11.3.4 Effects of Thermo-Mechanical Treatments: Relation with

Crystal Texture
From a rigorous point of view, the misorientation distribution cannot be determined

in a non-ambiguous manner from the knowledge of the crystalline texture. Indeed, the
mathematical orientation distribution function (ODF), that quantitatively describes
the texture, is a one-point distribution function, without any information about the
correlation of misorientations between adjacent grains. The MDF function, that
quantifies the grain boundary misorientation distribution, is a two-point distribution
function; it gives the fraction of pairs of adjacent grains with a given misorientation.
From a physical point of view, a two-point distribution function is reducible to a one-
point distribution function, only if the superposition principle is valid, in this case, if
all the orientations of adjacent grains are random [32]. From this principle, results a
predictable direct influence of the crystalline texture, determined by simulation, on
the boundary distributions in b.c.c. materials of group III, for which the assumption of
non-correlation of the adjacent grain orientations seems reasonable. This hypothesis
cannot be retained in the simulations of materials of group I because it is in contra-
diction with the formation of annealing twins; we must then expect a more complex
effect of the texture on the grain boundary distribution. Indeed, experimental results
show that f.c.c. materials, with low or medium stacking fault energy, may have sim-
ilar boundary distribution whereas they display different textures and, vice-versa,
different distributions may correspond to the same texture [38].
From a practical point of view, the effects of thermo-mechanical treatments on
the grain boundary distributions are diversified. Any synthesis requires a rigorous
effort to select the results in order to extract those common to several materials.
This is a necessary step to go towards grain boundary engineering. A relationship
between crystal and grain boundary textures is significant only in case of a strong
sharpness of the crystalline texture; a common feature to all the textured materials
then appears: this is a non-negligible number of low-angle grain boundaries ( = 1),
as predicted by the simulations (Table 11.2). In case of very strong textures, grain
boundaries are preferentially described by misorientations around an axis normal to
the average plane of the grains (for sheets) or parallel to the average grain direction
axis (fibre texture). This average orientation of the misorientation axis depends on
the dispersion around the principal direction of the texture.
This section presents results obtained from various materials gathered, as pre-
viously, in groups (I, II and III), knowing that orientation correlation effects may
strongly differ from one group to the other. Thermo-mechanical effects include those
resulting from manufacturing, deformation and recrystallization. For the latter, it is
convenient to well differentiate the effects of primary recrystallization from those of
normal growth and from those of abnormal growth or secondary recrystallization.
11.3.4.1 Group I Materials
The crystalline texture should have little influence on the misorientation distribution
of materials in group I, where a certain orientation correlation exists between grains.
However, a very strong texture may reduce the correlation effect, in particular in
the case where only one texture component develops; grains then display similar
orientations and the proportion of = 1 (low angle) grain boundaries significantly
increases.
A texture effect on the Ni3 Al compound non-enriched in boron has been observed
in correlation with the intermetallic ductility at low temperature (Fig. 11.13). The
percentages of = 1 and = 3 boundaries are clearly higher in the textured
materials obtained by unidirectional solidification than in the same materials without
texture resulting from deformation followed by recrystallization at 1300 K [44]. The
latter, with an elevated fraction of general boundaries (≈70 %) are more fragile at
room temperature.
The misorientation distributions are examined on high purity copper polycrystals
strongly deformed (95 %) by cold-rolling then annealed at 473 K (2 h), 773 K (2 h)
and 973 K (1 h), successively. The first treatment leads to a copper microstructure
typical of a primary recrystallization with a cube {001}100 texture and many island
twins, the average grain size being 6 μm. The percentage of twins is 56 %, mainly
due to the island contribution. Following treatments yield grain growth, the final
size after annealing at 973 K is 70 μm. Grains with cube orientation meet together
along common grain boundaries; the frequency of island twins diminishes and the
cube texture is reinforced with 65 % of grains orientated to less than 15◦ from the
ideal cube {001}100 orientation. For the two microstructures, the percentages of
low angle and coincidence boundaries resulting from additional annealing are given
in Table 11.8 [45]. With the strong reinforcement of the cube texture, the percentage
of low-angle boundaries clearly increases and, simultaneously, the twin percentage
decreases.
Fig. 11.13 Frequency of coincidence grain boundaries in Ni3 Al polycrystals non-doped in boron:
a after solidification; b after 25 % deformation and recrystallization at 1300 K during 20 min. The
number of investigated boundaries in each case is about 360. The crystalline texture of the sample
is given in the stereographic triangle associated to each histogram [44]
Table 11.8 Distribution of the coincidence misorientations in pure copper after grain growth result-
ing from annealing at 473, 773 and 973 K [45]
Coincidence index Annealing temperature after primary recrystallization
473 K ∼
= 0.34 Tm 773 K ∼ = 0.56 Tm (%) 973 K ∼
= 0.7 Tm (%)
1 – 23.3 53.9
3 56 % 32.5 23.0
5 − 27 – 5.8 4.6
The studies of Fe-50%Ni alloys show similar effects of the recrystallization texture
on the misorientation distribution as well as evolutions of the correlation function
OCF [46]. We recall that, for a given misorientation, this function (11.12) is equal
to 1 if no correlation exists, superior to 1 (inferior to 1) if the misorientation is
favoured (unfavoured) by comparison to what is predicted from the texture. Like
in the case of copper, the cube component dominates the crystalline orientation
distribution. After primary recrystallization, obtained by annealing at 0.5 Tm , 37 %
of grains are near the {001}100 orientation; generally high-angle grain boundaries
predominate (60 %) because, on the contrary of copper, the microstructure does not
display island twins. The percentage of cube-oriented grains widely increases with
the grain growth, reaching 59 % then 77 % under the effects of continuous anneal
of the deformed material until 0.62 Tm (800 ◦ C) and 0.68 Tm (900 ◦ C), respectively.
Simultaneously, the fraction of general boundaries decreases to reach 23 % and the
frequency of low-angle boundaries increases. Figure 11.14 presents, for each of the
previous microstructures, the misorientation distribution and the degree of correlation
for each misorientation. The proportions of = 1 and general grain boundaries
drastically change with growth, but the degrees of correlation remain quasi-constant.
On the contrary, a significant decrease of the correlation index accompanies the
Table 11.9 Degrees of correlation related to the misorientation distributions for general, low-
angle (θ ≤ 15◦ ) and = 3n grain boundaries in a Fe-50% Ni alloy after primary recrystallization
(φgrains = 5 μm) and after grain growth (φgrains = 12 et 15 μm) [46]
Sample grain Grain boundary type
size (μm) General θ ≤ 15◦ =3 =9 = 27
5 0.7 1.3 6.6 3.5 1.8
12 0.7 1.7 2.0 1.4 0.9
15 0.6 1.4 1.2 0.5 0.1
diminution of the proportions of = 3 (from 23 to 13 %) and = 9 (from 4 to

0.5 %) coincidence grain boundaries resulting from grain growth. The evolution of
the degree of correlation for general, low-angle and = 3n (n = 1 − 3) boundaries
is summarized in Table 11.9.
To try to explain the correlations observed in the different samples, the influence of
the crystalline orientations on the distribution of neighbouring grain pairs and on the
grain size distribution is analyzed. The crystalline orientation images show that low-
angle grain boundaries are mainly shared by cube-cube neighbouring orientations.
The formation of = 3 grain boundaries is associated with {001}100 cube -
{221}122 (cube twin) neighbouring orientations; this correlation matrix/twin is also
observed in weakly textured copper after recrystallization [45]. The grain boundaries
= 9 are mainly found between grains in twin–twin orientation or between two
random grains. Furthermore, the grain size of a given orientation is even larger
than this orientation is favoured, either after primary recrystallization or after grain
growth. This comment explains the differences between the experimental and statistic
misorientation distribution functions and the existence of a noticeable degree of
correlation for the = 3 boundaries between cube-oriented grains and grains of
{221}122 orientation.
In metals and metallic alloys displaying a low stacking fault energy, a certain
control of the thermo-mechanical treatments enables to increase the fractions of
= 3n grain boundaries, resulting in an improvement of the material behaviours
under various stimuli. First, the efforts to go towards grain boundary engineering
put the focus on the handling of the misorientations (change of the MDF function):
we report some results hereafter. But, to better understand the optimization of the
microstructure, an analysis of the grain boundary connectivity is necessary. This
aspect is presented later in terms of triple junction networks, of grain boundary
clusters and finally in terms of percolation (see Sects. 11.7 and 11.9).
Two sets of sequential thermo-mechanical proceedings are used to increase the
percentages of = 3 and = 3n grain boundaries [47]:
• A small deformation of the material (6–8 %) followed by an annealing series at
sufficiently low temperature, to avoid recrystallization, yields a grain boundary
re-orientation towards lower energy configurations. But these treatments are very
Fig. 11.14 Grain boundary misorientation distribution and degree of correlation for each
misorientation in a Fe-50%Ni alloy 90 % cold-rolled then recrystallized: a after primary recrys-
tallization at 600 ◦ C during 15 min; b after grain growth by continuous annealing until 800 ◦ C and
c until 900 ◦ C with a rate of 1000 ◦ C/h. A logarithmic scale is used to represent the boundary
frequencies. The white and black bars correspond to experimental and calculated distributions,
respectively. The evolution of the degree of correlation is represented by the curve (–•–) [46]
long (10–15 h) and not feasible in industry; moreover, they lead to an important
grain growth.
• Multi-cycle treatments alternating moderate deformations (ε < 30 %) and anneal-
ing at relatively high temperature during short times do not provoke (or few) grain
growth.
The results of each type of treatments on the boundary misorientation distributions
may be compared in the case of pure OFE copper with a high conductivity for
electronic industry (OFE for Oxygen Free Electronic):
• In the first type of treatments, the importance to proceed to sequential annealing
at low temperature is illustrated by the following example. Copper initially well
recrystallized then deformed 6–7 % in compression undergoes thermal treatments
in two steps and at more or less elevated temperatures (Table 11.10); the resulting
grain boundary distributions are given in Fig. 11.15 [48]. Before deformation, the
total percentage of coincidence grain boundaries ( ≤ 27) is about 70 %, among
them 55 % of = 3 boundaries, in agreement with this fraction in pure copper
(Table 11.8). After compression followed by thermal treatments, one of them being
performed at high temperature (0.63 Tm ), the grain size significantly increases in
the sample OFE-1, the percentage of coincidence boundaries decreases, and more-
over numerous low-angle boundaries appear. After two treatments at lower tem-
perature, the numbers of low angle boundaries drastically decrease in the samples
OFE-2 and OFE-3, while the fractions of = 3n boundaries become superior to
their initial values. Simultaneously, the angular deviations from the exact coinci-
dences diminish: for = 3, they are less than 1/10 of the authorized maximum
according to the Brandon criterion [2]. This remark is important given by the
finding that the grain boundary plane then tends to be very close to a low energy
plane [7].
• The second type of treatments consists in a series of 30 % deformation by com-
pression followed by annealing during 10 min at 673 K (∼ =0.5 Tm ). Between each
thermo-mechanical treatment (deformation and annealing), the copper samples are
quenched in water, characterized by optic microscopy, and then the grain bound-
ary distribution is automatically analyzed. Figure 11.16 shows the evolutions of
the percentages of = 3, = 3n and all the coincidence boundaries with
≤ 29 during three deformation/annealing cycles [47]. Similar evolutions have
been found for an Inconel 600 alloy annealed at much higher relative temperature
1273 K (0.8 Tm ).
An increase of the percentages of Σ = 3n boundaries may be obtained by a
judicious choice of the thermo-mechanical treatments imposed to the material. To the
extreme, a microstructure limited by multiple twinning is formed, entirely constituted
by = 3, = 9 and some low-angle grain boundaries. It is remarkable that the
fractions of = 3 grain boundaries in samples obtained by sequential annealing
at low temperature (Fig. 11.15) are very close of the limit two-third theoretically
predicted for this microstructure (see Sect. 11.2.2). Generally, the twin fraction is
less than two-third, a relation between this fraction and those of = 1 and = 9
Table 11.10 Thermo-mechanical treatments underwent by various samples of pure copper [48]
Sample
Proceeding OFE-1 OFE-2 OFE-3
Deformation % −7 −6 −6
Thermal treatments 0.45 Tm /8 h 0.40 Tm /14 h 0.37 Tm /14 h
0.63 Tm /14 h 0.48 Tm /7 h 0.44 Tm /6 h
Fig. 11.15 Histograms

showing the fractions of
= 1, = 3n and general
grain boundaries in samples
of OFE copper according
to the thermo-mechanical
treatments they underwent
(Table 11.10). The bars repre-
sent the boundary percentages
for recrystallized samples
OFE-1, OFE-2, and OFE-3,
successively [48]
Fig. 11.16 Coincidence mis-

orientation distributions in
OFE copper sequentially
deformed and recrystallized at
0.5 Tm [47]
boundaries has been proposed [49]:
f 1 + f 9 ≥ f 3 − 1/3 (11.17)
This relation has been obtained starting from an ideal microstructure uniquely
composed of triple junctions with only one = 3 boundary; in that case, the
percentage of = 3 grain boundaries is one-third. If this fraction is exceeded, the
interactions between twins lead to the formation of = 1 or = 3n , mainly = 9
boundaries. This relation is not verified for OFE copper after sequential annealing
(Fig. 11.16); the percentage of = 3 remains less than 1/3, whereas the fraction of
= 9 boundaries reaches an order of 12 %; it is the same for Inconel 600 alloy. This
suggests that the = 9 boundaries do not only result from interactions between
= 3 twins.
In conclusion, handling of = 3n grain boundaries in materials of group I
is possible via an appropriate choice of thermal treatments. We may then expect
an improvement of the responses to degradation (corrosion, fracture…) of these
materials due to a better resistance of their grain boundaries.
11.3.4.2 Group II Materials
In principle, there exists less correlation between adjacent grains in materials of

group II than in those of group I; we thus expect a stronger effect of the crystalline
texture on the grain boundary texture.
The changes in the boundary distribution in high purity aluminium (99.999
weight %) are investigated during grain growth after forging or cold rolling fol-
lowed by annealing at 473 K (0.5 Tm ), 673 K (0.73 Tm ) and 773 K (0.83 Tm ) [50].
After treatment at 0.5 Tm , there is a high percentage of general boundaries in the cold-
rolled sample that is not textured. Further annealing at 0.73 Tm yields the occurrence
of a pronounced texture and a significant increase of the relative number of = 1
boundaries (Fig. 11.17).
Same tendencies are found for forged samples but the changes are not so large
(Fig. 11.18).
Grain growth in recrystallized Al-Mg alloys, containing 0.3 and 2.7 weight
% magnesium, is studied by focussing on the relationship between the texture
development and the grain boundary distribution [51]. For the Al-0.3% Mg alloy, the
frequency of cube oriented (001) [100] grains, and simultaneously that of the low-
angle grain boundaries increase in the first growth stages then decrease (Fig. 11.19).
Conversely, for the alloy with 2.7 % Mg, the cube component of the texture continu-
ously decreases; the grain boundary distribution only changes a little bit (Fig. 11.20).
Clusters of cube-oriented grains containing = 1 boundaries are detected in
these alloys (Fig. 11.21); their spatial distribution plays an important role in the
microstructure change during grain growth [51].
The influence of magnesium on the microstructure is understood by its segregation
to aluminium grain boundaries leading to a reduction of the boundary energies and
mobilities. These effects are more pronounced for general boundaries; as a result,
their differences with = 1 boundaries vanish. The driving force for growth of clus-
ters of cube-oriented grains, containing mainly low-angle boundaries, decreases.
The observation of the Al-2.7% Mg alloy microstructure obtained after a thermal
Fig. 11.17 Changes of the crystalline orientation distribution and of the grain boundary coincidence
index distribution in an aluminium polycrystal, cold-rolled then annealed during 5 min at 473 K (a)
then submitted to further annealing at 673 K during 57 h (b) [50]

fraction of = 1 grain bound-
aries during grain growth
of aluminium, cold-rolled
(circles) or forged (triangles)
annealed at 673 K (◦, ) or
773 K (•, ) [50]
Fig. 11.19 {111} pole figure of samples Al-0.3%Mg annealed under different conditions: a 558 K—
1 h, d = 18 μm; b 773 K—200 s then 558 K—1 h, d = 160 μm ; c 773 K—16 min. then 558 K—
1 h, d = 480 μm ((001)[100], ♦(001)[310], (123)[634]); d Variation of the volume fraction
of cube-oriented grains and e change of the coincidence index distribution with the average grain
size d [51]

fractions of = 1, CSL
and general grain boundaries
in Al-2.7%Mg alloy with
the average grain size d
(in μm)—to be compared
Fig. 11.19e
treatment yielding a grain size of 540 μm well reveals the disappearance of the
cube-oriented grain clusters and the low fraction of = 1 boundaries (Fig. 11.22a);
this feature must be especially compared with the microstructure of Al-0.3% alloy
of almost same grain size (480 μm) that always presents regions formed by clus-
Fig. 11.21 Spatial grain and grain boundary distributions in Al-0.3%Mg alloy after the previous
(a), (b) and (c) thermal treatments. The cube-oriented {001}100 grains are shown in grey, those
orientated {001}310 on the form of hachured regions. The low-angle, general and coincident grain
boundaries ( ≤ 51) are drawn on the form of bold, fine and dotted lines, respectively [51]
ters and low-angle boundaries (Fig. 11.21c). The preferential migration of general
grain boundaries is also impeded by magnesium; this prevents a texture formation
(Fig. 11.22b).
To conclude, the grain boundary misorientation for materials of group II is
essentially managed by the augmentation of the low-angle grain boundaries, associ-
ated to a strong texture. For materials of group I, efforts to control their microstructure
especially consist to find adequate thermal treatments to favour the multiple twinning
process. A large difference thus appears between the two types of materials, even
with the same structure. It highlights the importance of the stacking fault energy that
not only influences numerous elemental mechanisms (see Sect. 8.3.4), but also the
mesoscopic organization of the polycrystal.
The effect of a deformation on the crystalline texture has implications on the grain
boundary texture. The boundary misorientation distributions, mixed in a random
way, have been calculated from the ODF functions of three aluminium samples:
Fig. 11.22 a Spatial distribution of grains (d = 540 μm) and grain boundaries in Al-2.7%Mg
alloy after a short thermal treatment at 773 K followed by maintain during 1 h at 573 K (same
representation than in Fig. 11.21); b pole figure associated to this sample (same symbols than in
Fig. 11.19) [51]
two of commercial purity (99.5 %) cold-rolled by 50–90 % and a pure-aluminium

(99.996 %) cold-rolled by 50 % (Fig. 11.23). For 50 % strain level, the samples dis-
play a weak texture whatever the aluminium purity; the misorientation distributions
present few differences with the random distribution with a maximum around 45◦ .
For 90 % deformation, the texture is more pronounced and the misorientation dis-
tributions displace towards the low angles. The experimental distributions strongly
differ from the calculated ones. They are clearly more concentrated after 50 % than
90 % cold rolling [52]. The high proportion of low-angle grain boundaries may be
explained by the formation of a cold-worked dislocation cell structure within grains.
Even if we exclude the cell interfaces from the evaluation, the percentage of bound-
aries with misorientation less than 20◦ remains elevated (higher than 70 % for pure
aluminium 50 % deformed). This large difference with the distribution predicted
from the ODF function suggests that various orientations generated within the initial
grains are not randomly distributed; correlations appear, sufficiently strong to affect
the misorientation distribution.
Studies concerning the relationship between crystalline texture and grain bound-
ary texture are still rare. Not only, the formation of cold-worked cells disturbs the
statistics on the = 1 boundaries, but the misorientation of a deformed grain
boundary may vary from one region to the other. More generally, consideration of
the geometrical parameters for deformed grain boundaries, strongly non equilibrated,
is questionable.
Fig. 11.23 Distributions, calculated from the ODF functions, of randomly mixed misorientations
(a, b and c) and experimentally determined distributions (d, e and f) for: a, d aluminium of com-
mercial purity (99.5%) cold-rolled by 50 %; b, e commercial aluminium 90 % cold-rolled; c, f pure
aluminium (99.996 %) cold-rolled by 50 % [52]
11.3.4.3 Group III Materials
Several experimental works clearly reveal that the effect of the texture obtained after
elaboration of materials of group III:
• Ribbons of Fe — 6.5weight % silicon alloys, rapidly solidified then annealed at
1363 K and at 1473 K during 3 h, present well defined {100} and {110} texture,
respectively. Table 11.11 shows the proportions of low-angle and coincidence grain
boundaries for textured ribbons and ribbons displaying a quasi-random distrib-
ution [53]. The observed distribution for the textured {100} alloy is close that
simulated (see Table 11.2), in particular for low angle boundaries; furthermore,
the differences between simulated and experimental percentages of coincident
grain boundaries may be reduced by limiting the values of the retained indices
to 25. The selected coincidence boundaries differ according to the type of tex-
ture: misorientations around 100: = 5, = 13 and = 25 or around
110: = 3, = 9, = 11, = 17 and = 19 (Fig. 11.24).
Table 11.11 Grain boundary misorientation distribution in ribbons of Fe-6.5% silicon alloys dis-
playing different textures [53]
Misorientation distribution
Annealing Grain size Texture % =1 %3 to 29 % general
conditions (μm) boundaries boundaries boundaries
1473 K—1 h 770 {110} 17 27.9 55.1
1363 K—1 h 600 {100} 24.8 20.4 54.8
1363 K—10 min 37 ∼
=random 3.4 10.3 86.3
1173 K—1 h 46 ∼
=random 5 7.6 87.4
Simulation Random 2.25 11.25 86.5
Fig. 11.24 Frequency of coincidence grain boundaries in function of in a ribbon of Fe-6.5% Si

alloy, rapidly solidified then annealed, displaying a fibrous texture {110} [53]
The effects of primary and secondary recrystallizations are compared for Fe-3%
Si alloys cold-rolled and annealed at 850 and 1200 ◦ C, respectively. After primary
recrystallization, the main texture component is {110}001 and the proportion of
= 1 boundaries in the sheets is of the order of 15 %. This texture clearly develops
and the percentage of low-angle boundaries reaches 50 % after secondary recrystal-
lization (Fig. 11.25) [54].
The microstructures of the samples (the low-angle grain boundaries being marked)
resulting from the two modes or recrystallization clearly reveal these changes
(Fig. 11.26) [55]. The augmentation of the low-angle boundaries during secondary
Fig. 11.25 Comparison of the frequencies of coincidence grain boundaries before (white bars) and
after (black bars) annealing that leads to a secondary recrystallization or b primary recrystalliza-
tion [54]
Fig. 11.26 Microstructure of Fe-3% Si alloy: a after primary recrystallization, grains in grey are
in Goss orientation (the symbol ∗ means that the two microstructures are linked at that point); b
after secondary recrystallization. In both cases thick lines denote = 1 boundaries [55]
recrystallization, intimately linked to the development of the Goss texture, seems to

occur even if the grains in {110}001 orientation are not the larger after primary
recrystallization.
• Two types of polycrystalline samples of high-purity molybdenum are developed,

one by sintering, rolling and annealing at 0.65 Tm (1873 K) during 4 h, the other
by zone melting, forging and annealing at 0.61 Tm during 4 h. The grain boundary
misorientation distribution for each type of polycrystal and the associated texture
(on the form of inverse pole figure) are reported in Fig. 11.27 [44].
A significant augmentation of = 1 boundaries accompanies the appearance of

a {114} crystalline texture in one type of samples. In the other, grains are randomly
oriented, the percentage of low-angle boundaries is similar to that predicted for a
randomly distribution. The coincidence boundary distribution is also more dispersed
for the non-textured sample. These differences result in different fracture behaviours.
• Studies performed on cylindrical bars of tungsten show that a fibre texture induces
a high percentage of grain boundaries described by the “plane matching” or “CAD”
model. The samples obtained by powder metallurgy, then rolled and annealed are
such that most grains have a 011 direction called primary axis almost parallel
to the 011 axis of the sample; it results in an important number of grain bound-
aries possessing a 011 coincidence axis [56]. The frequencies of 011 CAD
boundaries are compared to those of the boundaries displaying networks of intrin-
sic dislocations with Burgers vector almost parallel to a 011 direction in one or
more of the adjacent crystals. Such a vector is compatible with a b3 vector of the
DSC lattice, according to the one-dimensional coincidence model.
To sum up, the possibilities to play on the misorientation distribution in materials

of group III, akin to those evoked for materials of group II, mainly consist in the
development of a strong texture to which is generally associated a high proportion
of low-angle boundaries. The consequences of such a texture on material properties
have been less analyzed than those attached to multiple twinning for metals of group I.
However, the control of the = 1 boundary percentages seems to be a key factor
to improve the resistance of a material to intergranular fracture [44].
In conclusion, the grain boundary misorientation distributions, experimentally
established, complete the crystalline texture to give a first overview on the polycrys-
tal organization depending on the nature and purity of the material. They constitute a
pre-requisite to approach the overall grain boundary texture, including boundary and
triple junction distributions. But the establishment of the overall texture is still insuf-
ficient to trace the boundary contribution in the material properties, which requires
knowledge of the local grain boundary texture. This complementary information is
the subject to the following paragraphs.
Most statistics on the misorientations have been established by using the per-
missive criterion of Brandon [2] and must therefore be revised downward, but they
indicate real trends of neighbouring grain organization in polycrystals. They are more
significant for materials subjected to multiple twinning because the experimental
Fig. 11.27 Histograms showing the coincidence grain boundary ( ≤ 29) distributions on molyb-
denum polycrystals differently elaborated: a sintering, rolling and annealing at 1873 K, 4 h; b zone
melting, forging and annealing at 1773 K, 4 h. The inverse pole figure for each polycrystal is also
reported [44]
deviations from = 3n coincidences are generally low. Finally, the perspectives for
controlling misorientations mainly concern:
• Materials with low stacking fault energy by favouring the occurrence of = 3n
grain boundaries via a judicious choice of recrystallization treatments.
• All materials by developing a strong crystalline texture that imposes the formation
of a high number of low-angle grain boundaries.
11.4 Grain Boundary Plane Distributions 383
11.4 Grain Boundary Plane Distributions
Although the interfacial energy primarily depends on the grain boundary plane,
there are few investigations on the plane distributions by comparison to those on
misorientations; this is due to the experimental difficulties of such studies. They
accounted for only about 10 % of all grain boundary texture studies in the late twen-
tieth century.
The determination of the boundary plane rests on the observation of plane traces
on two surfaces of a sample; it is directly feasible on a bicrystal or a one-dimensional
multi-crystal (only one grain in the width and in the thickness—see Fig. 11.34), but it
is not easy in three-dimensional polycrystals. Two methods exist: either transmission
electron microscopy, but the number of observed grain boundaries remains too limited
to reach a statistical point of view, or the scanning electron microscopy (SEM) [57].
The goal is to measure the angles α and β on both surfaces of the sample (Fig. 11.28).
The angle α is directly measured on the viewing surface; to get the angle β it is
necessary to cut the sample vertically or to proceed to successive parallel sections
and to measure each time the displacement of the boundary trace on the surface
and the depth of the section The method requires that the plane is really planar at
the observation scale; the accuracy on its orientation is ±4◦ . With automation of
the techniques to proceed to parallel cuts and to obtain boundary trace orientations,
studies of grain boundary plane distributions develop, but the presentation of results
giving, simultaneously, misorientations and planes remains a challenge.
Fig. 11.28 Schema showing

the principle to measure by
SEM the grain boundary plane
orientation
Before to give some results on the average grain boundary plane orientations, let
us recall that boundaries are most often faceted at the microscopic scale, even at
the nanoscopic scale. These are the facets that adopt low-energy positions, the mean
plane observed at the mesoscopic scale being often random. The physical meaning
of the average grain boundary plane is thus questionable. However, in samples near
a metastable equilibrium, low-energy facets may preferentially develop such that the
average plane adopts an orientation close that of the facets. It is also remarkable that
the favoured boundary planes, resulting from sintering of balls on a single crystal
substrate [37] or from well-recrystallized polycrystals [7, 57], often correspond to
the facets that develop along boundaries of bicrystals observed by high-resolution

transmission electron microscopy.
The non-exhaustive review of the results presented here, voluntarily goes under
silence those dealing with the tilt or twist boundary character. Indeed, except some
particular cases of crystal growth on the shape of oriented columns or fibres strongly
elongated in a given crystallographic direction, we do not agree with the ability to
assign a unique character, tilt or twist, to a grain boundary in a polycrystal. In the case
of a more or less isotropic polycrystal growth, the final grain boundary plane orienta-
tions depend on complex effects of energetic and geometric constraints. Neither they
cannot be controlled with respect to the sample reference system, nor determined
in the neighbouring crystal systems. Indeed, the experimental determination of the
relative orientation of two crystals gives only one geometrical possibility among the
set of equivalent grain boundary descriptions. The experimental rotation axis has no
particular meaning; it is the same for the disorientation axis (see Sect. 1.1.2).
The physically meaningful description of a grain boundary must account of its
atomic structure that depends on local relaxations. Failing to reach the core structure,
the determination of the boundary in terms of intrinsic dislocations may inquire about
its tilt or twist character. However, such an analysis is impossible for the large number
of grain boundaries required in statistical studies. Then, how to decide that the well
defined normal to the boundary plane is parallel or not to one of the rotation axes,
randomly chosen to describe the grain boundary? This question comes under the
same ambiguity that arises in the distinction between U and I triple junctions; the
latter is supported by the choice of a preferred description of a grain boundary (NNR)
(see Sect. 10.2.2).
A compilation established by V. Randle reports the grain boundary plane
distributions in polycrystals of different metals [7]. The observations are averaged
over 150 planes, but they also include results from Orientation Imaging Microscopy
(O.I.M) concerning 104 –105 grain boundaries. The selection of particular boundary
planes in a polycrystal depends on two types of factors:
• Factors inherent to the material: its crystallographic structure and the nature of its
atomic bonds, its stacking fault energy leading or not to twinning.
• Factors depending on the sample: its micro-texture that results from thermo-
mechanical treatments (it particularly plays a role in the possibility for the bound-
aries to be close CSL misorientations), its geometry (surface effects, effects of a
stress axis…).
Numerous results concern f.c.c. materials of group I (low or medium stacking
fault energy), the boundary plane distributions are thus reported mainly for = 3,
= 3n and = 1 boundaries. In nickel, copper, gold and stainless steel, except
the coherent twin with the symmetrical {111} plane, most boundaries display asym-
metrical planes even irrational, and there are few incoherent {112} twins. On the
contrary, symmetrical grain boundaries predominate in Ni3 Al and silicon. The gen-
eral boundaries (in the sense non-coincident) most often have random planes [7].
The adoption of a symmetrical position also depends on the metal purity. A strong
selection of symmetrical = 3 {111} boundaries (≈65 %) occurs in pure cop-
Fig. 11.29 Histograms of the = 3 grain boundary plane orientations in pure copper annealed at
900 ◦ C during 1 h (white bars) then maintained at 540◦ during 97 h (black bars). The full line gives
the evolution of the calculated energy for 110 tilt grain boundaries [58]
per annealed for a long time to intermediary temperature (Fig. 11.29) [58]; this is
explained by the remarkable low energy of these boundaries, about one-third of that
of the coherent twin in nickel. In very high-purity copper annealed at elevated tem-
perature (≈0.9 Tm ), not only = 3 but also = 11 boundaries (not numerous)
possess a symmetrical plane, whereas the = 9 boundaries remain asymmetrical
with {111}//{115}. In nickel polycrystals annealed at 1000 ◦ C during 2 h, the = 3
boundaries appear with a frequency of 46 %; even if they present the straight shape
of the annealing twins, they generally are asymmetrical with planes in zone with
110, like {110}//{114}and{113}//{771} [57]. Calculations and experiments show
that 110 tilt grain boundaries are in an energy valley: the symmetrical {111} is
at the minimum (few tens of mJ·m−2 for metals of low stacking fault energy), the
other boundaries of the series have an energy that increases with the deviation from
the symmetrical {111} position until values of the order of 600 mJ·m−2 in copper
(Fig. 11.29). For comparison, the energy of a general grain boundary in nickel or
copper varies between 1 and 1.2 J · m−2 [59].
The length fractions of the different facets of the = 3n grain boundaries are
also reported for fine-grained polycrystals of commercial copper and stainless steel
(Fig. 11.30) [7].
The thermo-mechanical treatments that act on the plane re-orientation lead to
a diminution of the total grain boundary energy either by decreasing the interfacial
energy itself (selection of low-energy facets) or by decreasing the total grain boundary
area. The formation of low-energy facets is illustrated by the successive dissociations
underwent by = 9 {511} // {111} grain boundary to reach the symmetrical {211}
Fig. 11.30 Average lengths of the grain boundary facets generally observed in copper and stainless
steel for = 9 (a) and = 27 (b) [7]
Fig. 11.31 Diagram, drawn from transmission electron microscopy observations, showing the
successive dissociations of a = 9 grain boundary in silicon [60]
position (Fig. 11.31) [60]. The rotation of a grain under the annealing effect, leading to
a reduction of its total area, obviously depends on the situation of the boundary plane
compared to the free surfaces of the sample; it is efficient only for grain boundaries
abutting to one free surface, at least, having less adjacent crystals and then possessing
more freedom to re-orientate (Fig. 11.32) [7]. The latter are thus not representative
of the grain boundary network in usual three-dimensional polycrystals.
There are currently few statistical analyzes of the grain boundary planes for f.c.c.
materials of group II (aluminium and alloys), for b.c.c. metals of group III (iron,
manganese, tungsten . . .) and for materials of non-cubic structure. However, some
common features have been reported in the review paper [7]. It seems that grain
boundary planes in aluminium are generally irrational. In relatively pure iron poly-
crystals, containing only 40–100 ppm of phosphorus, the grain boundary planes are
never parallel to dense {110} planes of the b.c.c. structure [20]. The hypothesis that
the grain boundary planes adopt a position of minimum energy, often symmetrical, is
obviously not respected in metallic polycrystals. The deviation of the real plane with
respect to the simple index planes is generally higher for asymmetrical boundaries.
Fig. 11.32 Frequency of the grain boundary plane inclinations ϕ with respect to the normal to the
surface sample in annealed nickel: boundaries abutting at one free surface (white bars); boundaries
between grains in the sample centre (black bars). The inclination tendencies for the two types of
grain boundaries are given in dotted lines [7]
Fig. 11.33 Histograms showing in pure alumina and in 500-ppm MgO-doped alumina polycrystals
the frequencies of grain boundary planes parallel to the {0001} dense plane and to other dense planes
of the rhombohedral structure [3]
In ceramics, on the contrary of metals, grain boundaries with random misori-

entations often present low-index planes. The grain boundary plane distributions
in sintered alumina differ according to the oxide chemistry (Fig. 11.33) [29]. The
alumina purity favours the presence of dense (0001) planes while it does not
yield a selection of coincidence misorientations (see Fig. 11.12). The plane dis-
tribution differences have been related to the grain size differences. Pure alumina
presents a heterogeneous microstructure; in the regions where the grain size is large
(10–50 μm), a preferential grain growth on the form of disks occurs, the large disk
faces being parallel to the basal plane. The microstructure of sintered alumina doped
with magnesia is equiaxed with a small grain size (0.5 μm) and few dense planes.
A deformation by compression at 1500 ◦ C of the doped alumina leads to a local aug-
mentation of the grain size, and simultaneously the appearance of faceted boundaries
with a dense plane in one crystal at least. The selection of the {0001} dense planes
more likely results from growth kinetic factors rather than energetic ones.
Until now, we have considered grain boundary planes in three-dimensional bulk
materials. In one-dimensional (wires) or two-dimensional polycrystals (sheets or
ribbons) (Fig. 11.34), the grain boundary planes may re-orientate, under annealing,
more or less perpendicular to the free surfaces leading to particular properties of the
samples submitted to certain stimuli.
Fig. 11.34 organizations of a one-dimensional and b two-dimensional polycrystals
Other geometries may yield a selection of the grain boundary plane orientations
associated to a sharp crystalline texture. This is the case for the pancake microstruc-
ture obtained in aluminium leading to twist boundaries with {hkl} planes and for
samples with columnar grains (Inconel) or with a bamboo microstructure (iron and
gold) with formation of kkl tilt grain boundaries (Fig. 11.35) [7].
Fig. 11.35 Schematic representations of grain boundary geometries in relation with the sample
geometries that indicate the observed particular planes: a {hkl} twist boundaries for the pancake
microstructure of aluminium; b and c hkl tilt boundaries for the columnar microstructure of
Inconel (b) and for the bamboo microstructure of iron and gold [7]
In conclusion, a tendency for the grain boundary planes to adopt a high-density

position (low index planes) clearly occurs, whatever the material, when the purity
and/or the equilibrium degree increase. It appears associated to the coincidence for the
= 3n grain boundaries in materials of group I displaying a low or medium stacking
fault energy (f.c.c. metals, silicon, some intermetallic compounds. . .). In other mate-
rials, it depends on the conditions of recrystallization (metals and alloys) or sintering
(ceramics). A noticeable difference exists between the two types of materials: ran-
dom planes are quasi systematically associated to non-coincident grain boundaries
in metals, whereas coincidence and high planar density seem totally disconnected in
the case of ceramics. In massive polycrystals, only the boundary planes in contact
with the free surfaces tend to re-orientate under annealing to minimize their total
area; inside the sample, the energy reduction rather occurs by reactions between
grain boundaries or by dissociation; particular planes, with respect to the sample
macroscopic coordinates and with respect to references associated to neighbouring
crystals appear in one-dimensional or two-dimensional polycrystals.
11.5 Five Grain Boundary Macroscopic Parameters

Distributions
11.5.1 Theoretical Approach
The distribution of the grain boundary characteristics (GBCD) has been

theoretically described by the mathematical function so-called Intercrystalline
Structure Distribution Function (ISDF) [61]. This terminology raises an ambiguity:
indeed, the function only accounts for the five macroscopic parameter distribution
that does not reflect the grain boundary structure. The latter depends on local atomic
relaxations that certainly differ for a given grain boundary in a polycrystal, depend-
ing on its connection with other boundaries. The approach of the ISDF function is
performed by coupling together the stereological analysis of the distribution of the
normal orientations with respect to the sample reference system [62] and the MDF
function previously described [22, 23].
Let SV be the total area occupied by the grain boundaries per unit volume of a poly-
crystal. The ISDF function is described by SV (Ω, g) where Ω is the grain boundary
plane normal orientation and g the boundary misorientation. The description of g
requires three parameters: the Euler angles α, β and γ that put the referential of a
crystal in superposition with that of the other crystal; is given with respect to
a sample reference frame by two polar angles χ and η. The ratio SV (, g)/SV
represents the areal fraction of interfaces oriented at , within a range + d, and
misorientated by g, within the range g + dg.
dSV = SV (, g) d dg (11.18)
By integrating the dSV function over the entire range of misorientations and the
entire range of interface orientations, we obtain SV . The ISDF function is independent
of the choice of the crystal chosen as origin and independent of the sense of the
boundary plane normal, in one or the other grain, so that:

SV (, g) = SV −1 , g and SV (, g) = SV , g −1 (11.19)
For the series representation of the equation giving the ISDF and its derivation
starting from distribution analyzes on successive sections at different polycrystal
depths, the reader may refer to [61]. The principle of the method is briefly described
hereafter. The SV (g) function may be easily determined on a plane section of a
sample by X-ray or electron diffraction techniques. Then we established the function
j
L A (ω), giving the length per unit area in the jth section, of lines (grain boundary
plane traces) in a given orientation ω. Then, we derive the fundamental integral
j
equation linking SV (, g) to the LA (, g) function. Numerous coefficients are
involved in the series expansion of the ISDF function, but their number may be widely
reduced by taken into account the crystalline point symmetry and the symmetry of the
thermo-mechanical proceeding. Practically, the series representation of the function
is truncated at a certain order; this order and that of the symmetry group of the
proceeding P determine the number of plane sections in the analysis; it follows the
number of independent linear equations to simultaneously solve in order to get the
coefficients of the ISDF function. For example, if the truncation order Q is equal to
10 for a rolled polycrystal having an orthotropic symmetry, the minimum number
of sections is 20; this implies solving a system of 20 × 20 linear equations. The
number of measurements (number of boundary segments with different structures
to be analyzed), that must be realized to achieve an expected resolution level and
a statistical reliability, also depends on the crystal symmetry. For a point group
symmetry O (Q = 24) and for a material on the form of a rolled sheet (P = 8), the
number of measurements must be about 56,000 in order to get a resolution of 9◦ and
a reliability of 80 %. If we choose to truncate the series representation of the function
at the order Q = 18, then 60 plane sections must be analyzed; this is equivalent to
examine approximately 950 boundary segments and to determine the orientation of
about 300 crystals in each section. These numbers, given only as an indication, clearly
reveal the need for an automated procedure for collecting orientation measurements
and correlated topographies of lines on the different planar sections. Moreover, the
representation of the five-dimensional ISDF function on an understandable form
remains a serious challenge.
11.5.2 Experimental Approach
The challenge previously evoked probably explains the limited information brought
by the results of grain boundary distributions based on the experimental determination
of their five macroscopic parameters; but these studies are really promising as they
constitute a clear sign of progress to possibly link microscopic and mesoscopic scales.
The grain boundary distribution in function of the five macroscopic parameters
is analyzed on magnesia polycrystals obtained by hot-pressure then annealed, the
resulting grain size is 109 μm [63]. The parameters are measured on 4.106 grain
11.5 Five Grain Boundary Macroscopic Parameters Distributions 391
boundary segments making up 5.4 mm2 of boundary area. The method of serial
sectioning of the polycrystal at different depths, separated by 7 mm, is used; it enables
to only achieve an accuracy of 7.5◦ on the plane orientation. The grain geometries
and the grain orientations are controlled on each planar section. From the information
obtained on many sections, the distance between each section being small compared
to the grain size, it is possible to reconstruct the geometrical configuration of a
typical volume of the MgO oxide microstructure. The grain boundary distribution
function f (g, n) is the frequency of occurrence of a certain type of grain boundaries
characterized by g and n, given in multiple units (MRD) of this frequency in a
random distribution [63]. By comparison, the SV (, g) function (11.19) gives the
area fraction occupied by the boundaries of macroscopic parameters and g [61].
The presentation of the results of a five-dimensional function is not straightfor-
ward. The authors start to examine the misorientation distributions f (g) averaged
on all the grain boundary planes and the grain boundary plane distributions f (n)
averaged on all the misorientations, separately (Fig. 11.36). Each misorientation g
is represented by a Rodrigues vector (see Sect. 1.1.2); the normal denoted n corre-
sponds to a unit vector in the direction .
The misorientation distribution reveals intense peaks for low-angle grain bound-
aries (≈14 times the maximum in a random distribution); for higher angles (>15◦ ),
misorientations around 111 predominate. The plane distribution presents a
maximum for the {100} planes; but it shows more complex variations for a spe-
cific misorientation, as illustrated in case of θ = 5◦ [110] boundaries with the (1–10)
planes clearly favoured (Fig. 11.37). For all the large misorientations, the boundary
planes in magnesia are most often asymmetrical with one plane parallel to {100} in
one grain at least.
Other results on the distribution of grain boundaries as a function of their five
macroscopic parameters have also been obtained for aluminium [64]. The distribution
is relatively isotropic in commercially pure aluminium annealed such that an equiaxed
microstructure is achieved. However there is a clear tendency for grain boundaries
to terminate on low index planes.
In the analysis of the results, we must be aware that the serial sectioning technique
preferentially reveals boundary planes that are perpendicular to the analyzed surface.
When the sample has a strong axial uvw texture, the planes in zone with this
axis intersect the surface of analysis more frequently than inclined planes; a high
proportion of planes in zone with an axis may then constitute an artefact. To suppress
the ambiguity, it is necessary to proceed to another series of analyzes on sections
perpendicular to other directions.
To conclude, it is possible that the conjunction of a low coincidence index with a
dense plane leads to special properties of a grain boundary. A grain boundary tex-
ture established by considering the five geometrical parameters must then enable to
better approach the correlation between microstructure and polycrystal behaviour.
However, several observations are opposed to progress towards grain boundary engi-
neering among others:
Fig. 11.36 Projections of the grain boundary five parameter distribution in a magnesia polycristal: a
misorientation distribution (Rodrigues vectors) averaged on all the boundary planes. The different
triangles correspond to sections of the three-dimensional space perpendicular to the [001] axis
(r3 ); b grain boundary plane distribution averaged on all the misorientations, represented on a
stereographic projection with the (010), (110) and (111) planes marked by circles: black, grey and
white, respectively (see text) [63]
• Grain boundaries with same coincidence and same plane may possess very
different energies associated to different atomic structures resulting from local
relaxations. Furthermore, comparable energies of several boundaries do not nec-
essarily imply comparable answers to a given stimulus.
• The planes determined for establishing a texture are always mean planes, composed
of several facets that present specific properties, varying from one to the other.
An important question underlying these observations is obvious: Can we expect
back to the properties of a grain boundary ensemble knowing only the average
characteristics of the boundaries, being aware that only the structures at the atomic
11.5 Five Grain Boundary Macroscopic Parameters Distributions 393
Fig. 11.37 Observed grain boundary plane distribution for θ = 5◦ [110] boundaries. The popula-
tions, which are normalized and represented as multiples of a random distribution, are plotted on a
stereographic projection. The directions that correspond to pure tilt boundaries are also indicated
[63]
scale influence the intergranular behaviours? The following section outlines attempts
to circumvent this issue.
11.6 Grain Boundary Property Distributions
Grain boundary distributions, based on geometry, are established by auguring

different intergranular behaviours according to this geometry. But the link between
crystallographic parameters and grain boundary properties is far to be elucidated.
Any distribution directly based on a boundary property presents the great advantage
compared to the previous ones to have a real physical meaning, even though limited
to the property considered.
We may formally distinguish:
• Kinetic distributions based on a grain boundary transport property: diffusion, elec-
tric conductivity.
• Thermodynamic distributions based on the grain boundary energies: the answers
to thermal grooving, to wetting are indirect ways to account for energy
• Mechanical distributions: answers to cavitation under fatigue, to grain boundary
sliding, to fracture propagation.
The classification of grain boundaries according to their answers at a given
stimulus is often binary: the boundaries are simple resistant or not to corrosion,
fracture, cavitation.
We give hereafter only some results on grain boundary distribution explicitly
based on one of their properties. Other results, focusing only on the relations between
the responses of boundaries to a stimulus and their geometries, can be exploited to
classify the grain boundaries in an engineering perspective.
11.6.1 Grain Boundary Diffusivity Distribution
Expressions [11.6 and 11.7] are used to classify the grain boundaries in a polycrystal
according to their diffusivity [15]. The method rests on the disappearance time tD
at a given temperature TD of the extrinsic dislocation contrast, phenomenon linked
to the relaxation of the stresses associated to these dislocations. The process implies
dislocation motion by climb in the grain boundary, and is thus controlled by inter-
granular diffusion. Histograms of the activation energies QGB of the grain boundary
diffusion in a stainless steel submitted to various recrystallization thermal treatments
are established (Fig. 11.38) [65]. A change of the energy repartition occurs between
900 and 950 ◦ C, suggesting that a change of the boundary state is associated to grain
growth, but the processes allowing to interpret such an evolution are complex. We also
note a displacement of the energy distribution towards the low values after an ageing
treatment of austenitic steels with high carbon content (Fig. 11.39) [15]. This change
may be explained by carbide precipitation that induces chromium depletion at grain
boundaries and an increase of the intergranular diffusivity. This example well illus-
trates the interest of a distribution based on the boundary behaviours that accounts
for the real boundary microstructural states including crystallography, purity and
defects.
Practically, it is not necessary to know the reasons for which a given distribution of
the diffusivities occurs; only the overall grain boundary diffusion behaviour is taken
into account, whatever how it is dictated, by geometry and/or the chemistry. This
distribution is much more representative than that of the geometric characteristics
in view to trace the overall properties of a polycrystal. We may obtain, by different
methods, an average of the diffusivity distribution functions and get an intergranular
diffusion coefficient; the latter generally has a value with a satisfying order compared
to that determined by radiotracer measurements [66].
11.6.2 Grain Boundary Energy Distribution
The methods of thermal grooving [68] and wetting by liquid metal [69] of a material
enable to determine its relative grain boundary energies with respect to its free surface
energy from the measurements of the etching grooves according to relation (4.19).
11.6.2.1 Distribution Determined After Thermal Grooving
The relative grain boundary energy distributions, deduced from thermal grooving,
are established on nickel and austenitic steel slightly deformed by compression then
annealed at increasing temperatures (Fig. 11.40) [18]. The grain size augmentation,
resulting from annealing, is associated to a displacement of the interfacial energies
towards elevated values. This result is interpreted in terms of reactions between
11.6 Grain Boundary Property Distributions 395
Fig. 11.38 Histograms of the activation energies of grain boundary diffusion QGB (kJ·mol−1 ) in
stainless steel polycrystals that underwent various recrystallization treatments [65]
Fig. 11.39 Histograms of the activation energies of grain boundary diffusion in austenitic steels
with high carbon content: a after quenching from 1100 ◦ C; b after slow cooling and aging at
750 ◦ C [15]
Fig. 11.40 Histograms of

the relative grain bound-
ary energies γGB / γS after
annealing at 0.77, 0.8 and
0.9 Tm of polycrystals (Tm =
absolute melting temperature:
a 3 % deformed Nickel; b 6 %
deformed 316L steel [18]
boundaries, the high-energy boundaries being more mobile. This selection appears
in contradiction with the normal tendency of a material to minimize its energy under
annealing, but any analysis of the energetic balance must take into account the grain
size modification, thus the total boundary area.
11.6.2.2 Distribution Determined After Wetting
The measurement of the grain boundary etching groove angle resulting from partial
wetting by a liquid metal also enables to get the relative boundary energy with respect
to the solid–liquid interface energy considered as isotropic. The cumulated bound-
ary energy distributions so determined in case of two-dimensional (thin foils) and
three-dimensional (plates) zinc polycrystals, in contact with liquid gallium, are
reported in Fig. 11.41. More than 60 % of the boundaries have energies double to
that of the Zn/Ga interface and thus are susceptible to wetting [43].
11.6 Grain Boundary Property Distributions 397
Fig. 11.41 Cumulated

boundary energy distributions
deduced form the answers
to wetting by liquid gallium
of two-dimensional () and
three-dimensional () zinc
polycrystals [43]
Another mode of interfacial energy distribution, based on the distribution of the

dihedral angles formed by boundaries at triple junctions, is addressed in the next
section.
11.7 Triple Junction Distributions
The theoretical and experimental triple junction distributions until now rest on
the misorientations of the grain boundaries that form the junction. They present
the same limits that those evoked for the grain boundary distributions when the
boundary planes are neglected. However, they give a first overview on a polycrystal
organization, possibly analyzed in terms of percolation and, anyway, necessary to
engage in the way of grain boundary engineering. The triple junctions are mainly
distinguished according to the number of coincidence (or near coincidence) grain
boundaries implied in the junction [70]. This classification has the weakness inherent
to any selection only based on crystallography; in particular, it does not distinguish
between singular and vicinal grain boundaries and consider as general triple junction
any junction among three general boundaries. Nevertheless, it has a specific inter-
est for triple junction between = 3n boundaries. Another classification of triple
junctions rests on the distinction between grain boundaries based on their energy; it
gains in physical meaning, but it is also unsatisfying because the boundary behav-
iour is not only controlled by energy. Moreover, in the absence of a well-established
knowledge of the triple junction energy, we may only consider an energy criterion
associated with the boundaries that compose the junction; and we have seen that
the junction among three low-energy boundaries does not necessarily possess a low
energy. In this energetic classification, singular and vicinal triple junctions belong
to the class of special junctions according to the previous distinction [70] and gen-
eral triple junctions possess one general grain boundary, at least. The consideration
of singular triple junctions takes importance when applying the coincidence index
combination rule discussed below, as preliminary to any triple junction distribution.
11.7.1 Limit for Application of the Coincidence Index

Combination Rule
The coincidence index combination rule (10.1), also named Σ-product rule, controls
the connection of the grain boundaries at a triple junction. Theory and experiments
show that this rule has a major influence on the grain boundary network in a poly-
crystal. In so far as it applies at all the junctions, a non-random grain boundary
distribution may be analytically established. But this rule is strictly valid only for
singular junctions among three exact coincidence grain boundaries. In principle, for
a material of cubic symmetry, any grain boundary may be described in terms of exact
coincidence, through the inclusion of high index ; thus, the combination rule may
be applied at any junction. But generally the boundaries with index ≥ 29 are
not retained in the classification. A grain boundary is described as deviated from
the nearest coincidence (with ≤ 29) or, if the deviation becomes too large, it is
considered as a general boundary. Otherwise, in the materials of lower symmetry
(hexagonal, rhombohedral . . .), numerous boundaries do not correspond to exact
but approximate coincidence misorientations. The misorientations of three grain
boundaries (a, b and c) forming a triple junction are linked by a matrix relationship,
always satisfied:
Ma Mb = Mc (11.20)
Each of the grain boundaries being deviated of θi from the i coincidence
misorientation, the misorientation matrix Mi is equal to the product of the exact
coincidence matrix Mi by the deviation matrix i .
Equation (11.20) becomes:
Ma a · Mb b = Mc c (11.21)
By taking into account the relation that links the exact coincidence matrices (11.20):
Ma · Mb = Mc (11.22)
And by combining Eqs. (11.21) and (11.22), we obtain:
a = Mb c −1 −1
b Mb (11.23)
This equation indicates that a and c −1

b are similar matrices, and thus their traces
are equal:
tr (a ) = tr c −1
b (11.24)
By expressing the components of the angular deviation matrices in terms of angle/axis

pairs, by using the low angle approximation for each angle θi and by neglecting the
second order terms in the development, a relationship between the angular deviations
11.7 Triple Junction Distributions 399
may be established
θa2 = θb2 + θc2 + 2θb θc H (11.25)
with,
H = ub uc + vb vc + wb wc = cos ϕ (11.26)
[ui , vi , wi ] is the deviation axis from the nearest coincidence of the boundary i; ϕ,
the angle between these axes for the boundaries joints b and c, can take any value
between 0 and π, and thus −1 ≤ cos ϕ ≤ 1. The introduction of limited values of H
in Eq. (11.26) leads to a general relation between the angular deviation values [71]:
θlim ≤ θ1 + θ2 (11.27)
where θlim is the largest deviation among θa , θb or θc ; θ1 and θ2 are the
deviations of the other boundaries.
The so-called limit deviation rule (11.27) is always mathematically exact,
provided the use of a small angle approximation (θ < 20◦ ), and this whatever
the criterion of maximal deviation retained to classify a boundary as near coincident
[1–3]. Triple junctions are created for all the allowed permutations in Eq. (10.1) with
1 < ≤ 29, then two of the boundaries are authorized to randomly deviate from
the exact coincidence. Starting from the rotation matrices of these two boundaries,
the rotation matrix of the third one as well as its deviation from the coincidence are
deduced. The simulations of several hundred miles numerical examples show that
in most cases, the rule of limit deviations is observed [71]. If, theoretically, the rule
(11.27) is independent of the chosen selection criterion, such a criterion is however
involved, by reducing the number of possible combinations that obey rule (10.1). As
soon as we take into account a criterion of maximal deviation to coincidence for each
boundary constituting the junction, apparent violations appear in the limit deviation
rule and in that of index combination.
The derivation of the limit deviation rule uses a relation between the exact coinci-
dence matrices (11.22) that is not totally equivalent to Eq. (10.1). Indeed, for a pair of
matrices Ma and Mb , there is only one matrix Mc , thus only one value c ; on the
contrary, the combination rule suggests several possibilities for this index. The fact
that three index values obey the combination equation does not seem sufficient
for the limit deviation rule can be applied to the corresponding triple junction. It
must first be careful not to attribute a wrong coincidence to one of the boundaries of
the junction. Thus, there are violations to the limit deviation rule in the simulation
approach of triple junctions when the experimental deviations approach the maxima
allowed by Brandon [2]. We consider the triplet = 9, = 27a, = 3, common
in materials with low stacking fault energy, and we decide that one of the boundaries
displays an exact coincidence, for example = 9(θ = 0). The application of the
limit deviation rule imposes that the two other boundaries attached to = 9 deviate
from the same angular value from the exact from = 27a and = 3, respectively.
This condition is respected for the majority of the simulated examples, but aberra-
tions occur when the previous common deviation is large. If the numerical analysis
is then repeated by using the combination = 9, = 9, = 81d instead of the
erroneous = 9, = 27a, = 3, the limit deviation rule is strictly valid. The
confusion between one and the other combinations is even easier than the selection
criterion is more permissive. For example, using the Brandon criterion [2], a bound-
ary misorientated of 34◦ around 110 may be classified as = 27a with a deviation
of 2.42◦ (θmax = 2.88◦ ) or as = 9 with a deviation of 4.95◦ (θmax = 5◦ ). This
potential ambiguity is widely reduced with the use of another maximum deviation
criterion to the coincidence (Table 11.12), each boundary being more closely asso-
ciated with only one coincidence.
Table 11.12 Maximal authorized deviations from the coincidence misorientations according to
three different criteria [2–4] for the = 3n boundaries
θmax from the coincidence misorientation
Brandon [2] Ishida et Mac Lean [3] Palumbo et Aust [4]
3 8.66◦ 2.66◦ 5.97◦
9 5◦ 0.88◦ 2.40◦
27 2.88◦ 0.29◦ 0.96◦
However, even by referring to the Brandon criterion to classify the coincidence

grain boundaries, the cases of violations to the limit deviation rule are rare; they occur
for about 0.2 % of the simulated junctions. In all aberrant examples, Eq. (11.22) that
connects the exact matrices is not satisfied; it complies when we replace the erroneous
triplet by the correct combination; simultaneously, the rule of limit deviations is
respected. It is thus reasonable to assert that the limit deviation rule applies to any
set of grain boundaries that obey the coincidence index combination rule.
Nevertheless, apparent violations to the index combination rule may occur: this
is the case when the sum of the θ1 and θ2 deviations of two boundaries lead
to a θlim value according to (11.27) exceeds the maximal value θmax authorized
for the third boundary. More demanding is the coincidence criterion more amplified
are the breakings to the combination rule. For example, if the first two boundaries
in the triplet = 3, = 9, = 27 deviate from 0.8◦ from their respective
coincidence, that is authorized according to any of the criteria (see Table 11.12) and
if the third boundary deviates from 1◦ of = 27, the maximal deviation rule (11.27)
is respected, but the triple junction can be considered as = 3, = 9, = 27
only if the Brandon criterion is used to classify the boundaries. The fraction of triple
junctions that obey the combination rule is analytically calculated by considering,
for two of the boundaries, random deviations from the coincidence and inferior to
the maximal deviations authorized by one of the coincidence criteria [2–4]. The
proportions of combination of two = 3 and = 9 boundaries giving rise to
a = 27a boundary according to the Brandon criterion is about 16 %; it falls to
about 6 % if the Palumbo and Aust criterion is applied to select the near coincidence
boundaries. These calculations, generalized to other junctions and taking into account
the permutation of the three boundaries, enable to estimate the fraction of triple
junctions a − b − c such that the a b and c boundaries can legitimately be

classified as near coincidence ( ≤ 29) according to a selection criterion on the
form θmax = θ0 −p . Table 11.13 indicates these fractions for all the possible
triplets such that 3 ≤ < 29 with θ0 = 15◦ and p = −1/2 [2] or p = −5/6 [4].
Table 11.13 Calculated percentages of triple junctions a − b − c with 3 ≤ < 29 for which
the three boundaries a, b and c are legitimately classed as coincident according to the Brandon
criterion [2] or to the Palumbo and Aust criterion [4]
a − b − c Triple junction %
Brandon criterion Palumbo-Aust criterion
θmax = 15◦ −1/2 θmax = 15◦ −5/6
339 48 19
3 5 15 19 11
3 7 21 15 7
3 9 27 16 6
5 5 25 22 9
999 50 50
9 15 15 51 39
9 21 21 37 25
9 27 27 31 39
15 15 15 48 44
25 25 25 50 50
The analysis of Table 11.13 shows that, even in the best case, the coincidence index
combination rule (10.1), applied without precise information on the deviations θ , is
respected only for 50 % of the triple junction population. The percentages reported in
this table are valid only for a random distribution of the deviations from coincidence
and thus are not representative of any microstructure. In particular, the hypothesis of
random deviations is not justified for metals with low stacking fault energy where the
formation of annealing twins leads to the presence of = 3 boundaries extremely
close the exact coincidence. Similarly, a strong texture may generate grain boundaries
statistically very near a given coincidence. But, in any case, a naive application of the
coincidence index combination rule, without knowledge of the angular deviations,
can result in significant errors in the triple junction classification.
Finally, the combination rule is strictly respected only for the connection of three
exact coincidence grain boundaries. For the near coincidence boundaries, the two
rules (10.1) and (11.27) must be simultaneously considered when analyzing the triple
junctions in view of a classification. The application of the limit deviation rule may
lead to deviations that exceed those authorized by the retained selection criterion
[2–4]. In that case, only a fraction of the triple junctions classified as a − b − c
effectively corresponds to this index combination.
Although the previous discussion focuses on the necessity to apply the
coincidence index combination rule together with that of the limit deviations, most of
the published studies on the triple junction distributions have neglected this precau-
tion. Nevertheless, we report in the next section several works on the triple junction
distributions, which, while not perfect, demonstrate the importance of grain boundary
connections and open the way to a better description of the polycrystal.
11.7.2 Theoretical Approach of the Triple Junction Distribution
The -product rule (10.1) clearly indicates that the grain boundary connectivity
cannot be approached like a random percolation problem. Monte Carlo simulations,
used to construct physically realistic boundary networks, show that the non-random
character of the triple junction distribution is associated to the crystallographic con-
straints at each junction [72]. The correlations between junctions may be studied in
terms of network connectivity, then, related to the percolation threshold for grain
boundaries susceptible to damage (see Sect. 11.9). However, in a first time, the
approaches of the triple junction distributions ignore the combination rule (10.1).
The first triple junction distributions, established by simulation without considera-
tion of the crystallographic constraints at the junctions, are reported on the form of
histograms for f.c.c. materials (Fig. 11.42) [73]. They concern polycrystals displaying
or not a crystalline fibre texture. The near-coincidence grain boundaries ≤ 49 are
selected according to the Brandon criterion [2]. The junctions are denominated i-CSL
with i = 0, 1, 2 or 3 depending on the number of coincidence boundaries abutting
at the junction. Few 3-CSL triple junctions appear; their fraction increases slightly
in case of strong texture, but remains very inferior to those generally found in f.c.c.
materials susceptible of multiple twinning that present a high proportion of = 3n
Fig. 11.42 Triple junction

distributions in a f.c.c. material
with random oriented grains
and in a material displaying
a fibre texture 100, 110 or
111, strong (3◦ deviation) or
weak (15◦ deviation). The four
bars successively represent the
fractions of i-CSL junctions
with i = 3, 2, 1, 0, the number
of CSL boundaries at the
junction ( ≤ 49) [73]
Table 11.14 Effect of the texture on the distribution of the CSL boundaries
Sharpness of Type of triple Fibre texture Recrystallization texture
the texture junction 100 111 110 100 111 110
3◦ 0-CSL 9.1 15 15 0 0.2 7.7
3-CSL 22.1 13.9 6.7 98.7 82.7 26.4
5◦ 0-CSL 26.6 34.9 44 1.1 5 12.6
3-CSL 7.2 3.9 1.7 71.9 35.9 19.5
15◦ 0-CSL 61.4 66.9 67.9 36.7 48.8 48.1
3-CSL 0.6 0.3 0.2 4.4 1.4 1.6
The first column indicates the texture sharpness, the second the type of triple junctions: 0-CSL (no
CSL boundaries) or 3-CSL (with 3 CSL boundaries). The columns 3–5 correspond to a polycrystal
with an axial texture around 100, 111 and 110; the columns 6–8 correspond to a polycrystal
with a texture resulting from deformation of a single crystal followed by recrystallization [73]. For
comparison, the percentages of triple junctions in a random polycrystal are 69.6 % for 0-CSL and
0.16 % for 3-CSL [31]
boundaries. The percentage of 2-CSL triple junctions, implying two coincident grain
boundaries, is small in the absence of texture or in case of weak texture. The percent-
age of 1-CSL junction remains almost constant whatever the texture. The general
triple junctions (i = 0) dominate the triple junction population in non-textured or
weak-textured materials, but their number significantly decreases in the presence of
a sharp fibre texture. The frequencies of triple junctions with three (3-CSL) and zero
(0-CSL) boundaries in a non-textured material have been estimated to 0.50 and 58 %,
respectively. They are of the same order than the values determined by simulation in a
polycrystal constructed with Kelvin polyhedra [31]. Table 11.14 emphasizes the role
of the texture in the grain boundary connections at triple junctions simulated in two
cases: an axial texture that represents a fibre texture, the other one approximates the
texture obtained from a single crystal deformed then recrystallized. For each of them,
different texture acuities are considered, by decreasing order 3◦ , 5◦ and 15◦ [31].
The triple junction proportions are almost the same for a weak-textured
material and for a material displaying a random crystalline orientation distribution;
this supports the conclusion that the misorientation distributions are only sensi-
tive to strong texture components. The strong proportions of triple junctions among
three coincident grain boundaries, after deformation of a single crystal followed by
recrystallization, are explained by the small orientation differences between crystals
and the presence of a large number of low-angle boundaries.
The simulations also concern triple junctions between one-dimensional coinci-
dence (CAD) grain boundaries around 100, 110 and 111 axes. The CAD junc-
tions are classified according to the value of the index common to the three
boundaries (relation 11.5) i.e. = 3, = 4 and = 8, corresponding to a good
matching of the {111}, {200} and {220} planes, respectively [73]. The percentage of
CAD junctions with ≤ 8 significantly increases in case of strong fibre texture; it
can reach a value close 89 %, but decreases when > 8 (Fig. 11.43), in agreement
with the limitation of the CAD model to low-index grain boundaries in f.c.c. system.
Fig. 11.43 Simulated distributions of CAD triple junctions in f.c.c. polycrystals displaying a ran-
dom crystalline orientation distribution and a 100, 110 and 111 fibre texture. Only the good
matching of the {111} planes ( = 3), {200}( = 4) and {220}( = 8) increases in the presence
of a crystalline texture. For > 8 (value limited to 35), the percentage of CAD boundaries is small
and decreases with the texture [73]
More recent works take into account the combination rule at the triple junctions.
The boundary networks are often simulated by simply considering four types of grain
boundaries, those resulting from multiple twinning: = 3, = 9, = 27 and
random boundaries denominated R (for Random) with > 29. This simplification
pretty well corresponds to the microstructures of materials belonging to group I (see
Sect 11.3.2), composed by a majority of = 3n boundaries, (in particular = 3)
and for which the other coincident boundaries ( = 5, = 7, = 11. . .) are rarely
observed. It does not take into account the low-angle grain boundaries (close = 1)
that may appear by multiple twinning. We do not retain here the assumption of the
authors who attribute to the = 3n boundaries a resistant character to damage, and
to the other boundaries a propensity to be damaged [72]. We have seen that predicting
the behaviour of a boundary without knowledge of its plane is not realistic. Moreover,
we know that the energy of the = 3{111} boundary is very low compared to
that of = 9; approximating the behaviours of these two boundaries is therefore
questionable. Our focus here is only on the method to approach the triple junction
distribution.
If f is the proportion of = 3n grain boundaries and A the ratio between
the proportion f 3 of = 3 and the proportion f , the fractions of boundaries of
different types are then given by:
f3 = f A (11.28a)
f 9 = f (1 − A)A (11.28b)
f 27 = f (1 − A) 2
(11.28c)
fR = 1 − f (11.28d)
The four types of boundaries give rise to four types of triples junctions Ji , with
i = 0, 1, 2 or 3 depending on whether the junction contains 0, 1, 2 or 3 = 3n
boundaries. Figure 11.44 presents these four types of junctions with, for the junctions
J1 J2 and J3 , the possible combinations of the different grain boundary types.
Fig. 11.44 The four types of triple junctions Ji (i = 0, 1, 2, 3) and all the possible combinations
of the grain boundary types that satisfy the coincidence index combination rule. The boundaries
considered in the Monte Carlo simulations are limited to four types: = 3, = 9, = 27 and
general boundaries with > 29 [72]
If the triple junctions are randomly distributed in the microstructure, regardless of

the combination rule, the probabilities of forming each type of junctions are given by:
J0 = (1 − f )3 (11.29a)
J1 = 3 f (1 − f ) 2
(11.29b)
J2 = 3 f 2 (1 − f ) (11.29c)
J3 = f 3 (11.29d)
The random distributions of the different types of triple junctions Ji (i = 0, 1, 2, 3)

and those resulting from the crystallographic constraint linked to rule (3.1) imposed
to the microstructure in the simulations are represented on the form of curves
(Fig. 11.45). Some points corresponding to experimental distributions, measured on
Fig. 11.45 Triple junction distributions: analytically established from a totally random microstruc-
ture (----); simulated by using a Monte Carlo algorithm and by imposing a constraint to the grain
boundary network (—–); experimentally determined for copper and a nickel-based alloy by differ-
ent authors (circles, squares or triangles)—the imposed constraint A corresponds to the ratio of the
number of = 3 boundaries over the number of = 3n boundaries [72]
copper and Inconel polycrystals, are also reported in Fig. 11.45. For the experimental
data, the fraction of boundaries f as been determined from the general relation:
f = 1/3 (J1 + 2J2 + 3J3 ) (11.30)
The experimental results are not described by the relation (11.29) based on a ran-
dom distribution of the triple junctions; in particular, the percentage of junctions J2
is inferior to that predicted by Eq. (11.29c) while the junctions J1 and J3 appear more
frequently than expected by the relations (11.29b) and (11.29d). They are explained
on the basis of the crystallographic constraint contained in the combination rule: if
two coincidence boundaries meet together at a junction, the third boundary must
also be coincident. The results of the simulations taking into account the -product
rule (10.1) are given, for J1 , J2 and J3 , on the form of three curves corresponding
to different ratios A of the number of = 3 boundaries over the number of = 3n
boundaries (A = f 3 / f ). The augmentation of A leads to an increasing large devi-
ation from the random curve. It must be noted that the experimental points, coming
from different studies, are generally near a curve obtained for constrained junctions.
The junctions J0 being not directly affected by the rule (10.1), their simulated and
experimental distributions are close to the random character.
Finally, we must note that the proportions of the different triple junctions i-CSL
(i = 0, 1, 2 and 3) at the quadruple nodes are also determined by simulation, for
textured and non-textured materials. They are near those obtained at triple junctions
in the case of a random polycrystal with 70.1 % of 0-CSL junctions and 0 .20 % of
3-CSL junctions [31]. The number of triple junctions T and the number of quadruple
nodes Q in a polycrystal modelled by Kelvin polyhedral are related by T ≈ 1.5 Q;
this remains valid for textured materials.
11.7.3 Experimental Triple Junction Distributions
The microstructural topologies of polycrystals, included their triple junction distri-

butions, are especially studied in f.c.c. materials with low or medium stacking fault
energy. Indeed, in this type of materials, we may modify the grain boundary distribu-
tion through various thermo-mechanical proceedings, in view to improve properties
of the whole material. Modifications mainly result from multiple twinning linked to
the interactions between grain boundaries in the polycrystal.
The establishment of the triple junction distributions, after sequential thermo-
mechanical treatments, follows the determination of the grain boundary distributions
in OFE-copper and Inconel 600 [47]. Like in the simulation approach, four groups
of junctions are considered: i-CSL with i = 0, 1, 2 or 3 according to the number
of CSL boundaries ( ≤ 29) abutting to the junction. The results, similar for the
two materials, are reported in the case of OFE-copper that was compressed 30 %
several times, each deformation being followed by annealing at 400 ◦ C (Fig. 11.46).
A drastic drop in the percentage of general junctions associated with a large increase
of 3-CSL junctions occurs in the first stage of the treatment. The percentages of other
junctions are almost unchanged.
The experimental results concerning copper and Inconel are then compiled
and reported in function of the coincidence boundary percentage in the materials
(Fig. 11.47). They are compared to those of the analytical triple junction distribu-
tions, established by probability calculations [31] or resulting from simulations. A net
correlation appears on the curves drawn from calculations: an increase of the 3-CSL
triple junction fraction and a decrease of the number of 0-CSL junctions are associ-
ated to a coincidence boundary percentage augmentation. The experimental results
show similar tendencies for these two types of junctions. Otherwise, a disagreement
appears between the experimental and the analytical (or simulated) distributions for
the 2-CSL and the 1-CSL junctions.
Disparities between calculations and experiments may be due to an effect of
the texture of the material on the experimental distributions or, alternatively, reflect
the fact that analytical solutions do not contain information on the interconnection
Fig. 11.46 Triple junction

distribution in OFE copper
sequentially deformed and
recrystallized at 400 ◦ C. The
junctions composed to three,
two, one and zero coincidence
boundaries are represented
by circles, squares, diamonds
and triangles, respectively
[47]
Fig. 11.47 Theoretical triple

junction distributions estab-
lished from a probability
function (full line) [31] or by
simulation of 1000 isolated
junctions containing only
= 3n (n 0,1,2,3) bound-
aries (dotted lines) [47]. The
experimental distributions are
reported with the same sym-
bols than those of Fig. 11.46
between individual elements of the boundary network, This question of connectivity

will be approached in Sect. 11.8.
11.7.4 Grain Boundary Energy Distribution Starting from

the Dihedral Angle Distribution at Triple Junctions
The measurement of the dihedral angles in a polycrystal may provide information

on the relative energies of grain boundaries sharing a common triple junction. The
method requires two conditions:
Fig. 11.48 Geometry of a

triple junction showing the
true dihedral angles αi among
three boundary planes πi and
the planar dihedral angles
βi projected on a section
identified by its normal η
• The knowledge of the true dihedral angles α, while only their projections β are
determined on the sample surface (Fig. 11.48).
• The possibility to apply the Herring rule (10.3) by neglecting the torque term.
The dihedral angles measured on a sample section depend on the relative grain
boundary energies and on the triple line orientation with respect to the surface. If the
effect of the orientation is statistically random, any change in the distribution of the
planar dihedral angles βi may be related to the change of the true dihedral angles αi
(with i = 1, 2, 3) and thus to the energy distribution.
The relationships between the angles measured on a planar section and the true
dihedral angles (Fig. 11.48) are numerically analyzed starting from the assumption of
a normal energy distribution around an average value γm associated to a mean value
of the dihedral angles equal to 120◦ [74]. In that case, the variation of the interfacial
energy is defined by the values of the ratio of the standard deviation of the energy γ
over the average value γj . For ratios γ /γj superior to 0.4, the generated values
of γj (j = 1, 2, 3) are so different from each other that an equilibrium configuration
of the three boundaries cannot exist. In the range 0 ≤ γ /γJ ≤ 0.4 where
equilibrium may occur at triple junctions, the results of simulations show that the
standard deviation of the planar angle β is a monotonous function of the standard
γ deviation (Fig. 11.49). The minimal value of β equal to 22◦ , is observed for
γ = 0; it expresses the contribution of chance in the triple junction orientation with
respect to the planar section. It results that true variations of the energy values give rise
to angular deviations β higher than 22◦ . However, the β values slightly depend
from those of γ for ratio γ /γj ≤ 0.1. A linear dependence is observed for
higher values of this ratio. Given the assumption of normal energy distribution, a value
of γ /γj equal to 0.15 implies that approximately 62 % of the grain boundaries
(excluding twins) have energies between 0.85 and 1.15 times the average value γj
[74]. Experimentally, the values of the standard deviations of the dihedral angles vary
between 21◦ and 30◦ , depending on the thermo-mechanical history of the material.
The use of such a model, which only requires simple geometric measurements,
enables to explain the changes of the dihedral angle distribution in terms of changes
of the interfacial energy distribution.
Fig. 11.49 Curve deduced

from simulations showing
the evolution of the standard
deviation β (in degrees)
of the planar dihedral angles
in function of the relative
standard deviation of the grain
boundary energies γ /γj
in a polycrystal [74]
Starting again with the assumption of a normal grain boundary distribution in a

polycrystal γJ = N(γJ , γ), the relative width of this distribution w = γ /γJ
is determined from the distribution of the ratios λ = sin αi / sin αj (proportional to
γi / γJ ), where αi < αj are dihedral angles belonging to a same triple junction [75].
The number of α angles necessary to establish the relation is of the order of 100.
A normal energy distribution is revealed by a dihedral angle distribution that may
be also approximated by a Gaussian law (120◦ , δ◦ ) in the vicinity of the maximum.
The method is applied to zinc plates of thickness 600 μm (only the planar angles β
are measurable) and to thin sheets of thickness 60 μm (the values of the angles β
and α are then almost equal) [43]. When the grain boundaries are perpendicular to
the sample surface, as it is the case for thin sheets, the dihedral angle distribution is
narrower and better centred on 120◦ than in the case of a thicker plate (Fig. 11.50).
The calculated relative widths of the energy distributions are 0.15 for the sheet and
0.30 for the plate. By taking an average boundary energy in zinc equal to 308 mJ·m−2
at 160 ◦ C and a decrement of 0.4 mJ·m−2 per degree [76], the energies of most of
the boundaries vary between 200 and 400 mJ·m−2 .
Fig. 11.50 Histograms of the

dihedral angle distributions
at triple junctions in zinc
polycrystals on the form of
plates with a thickness equal
to 600 μm or of thin sheet
(e ≈ 60 μm) [43]
These approaches based on the triple junction configurations are only valid for
normal energy distributions in polycrystals, the energies being centred on an average
value. Moreover, the deviations from the average value are relatively small; this
authorizes the use of the simplified Herring rule. They are not appropriate when the
boundary distribution is far to be random, especially in the case of microstructures

resulting from multiple twinning or from a strong texture with a relatively large
number of low angle boundaries. They are also not valid when the boundary spatial
distribution in a polycrystal is heterogeneous. The existence of a local concentration
of a given type of grain boundaries in certain regions of the sample is expressed by
the notion of local grain boundary texture that is the subject on the next section.
11.8 Local Grain Boundary Texture
Akin to the overall texture, a real local grain boundary texture must be established
from the knowledge of the five macroscopic parameters of all the boundaries in a
given region of the microstructure. The difficulty to reach the orientations of a large
number of boundary planes with a sufficiently good accuracy generally reduces the
studies to those of the local misorientation distribution. The latter may permit to
better approach the collective behaviours of a polycrystal, always provided that the
knowledge about the links (statistically possible) between properties and misorien-
tation of a grain boundary is improved.
The experimental grain boundary misorientation distributions are
generally heterogeneous, in particular in the presence of a crystalline texture where
twinned islands or low-angle boundaries may appear (Fig. 11.21). The non-random
character of the spatial misorientation distribution is predicted and quantified by the
use of the orientation coherency function (OCF) [24]. If the value of this function
differs from 1, a correlation of the crystalline orientations exists; this results in local
gatherings of grain boundaries with the same misorientation. The formation of clus-
ters of grains, linked by grain boundaries sharing the same misorientation, raises the
question of the pertinent length in a polycrystal: the average grain size or the average
size of a grain cluster, sometimes called effective grain size? However, the presence
of grain clusters not necessarily implies the presence of grain boundary clusters. In
a first time, we must distinguish between the different types of clusters (of grains or
grain boundaries) that can exist in a crystalline microstructure.
11.8.1 Different Types of Clusters
The importance of local arrangements may be illustrated by considering two

morphologically identical microstructures of a material, possessing the same per-
centages of grains or grain boundaries of a given type but differently distributed.
Various types of clusters appear in these microstructures leading to different answers
to an external stimulus: grain clusters that gather grain boundaries of same misorien-
tation but without connection (Fig. 11.51) and real grain boundary clusters in which
the boundaries of same misorientation are locally connected the ones to the others
(Fig. 11.52). Elongated clusters of grains form a channel through which an intragran-
Fig. 11.51 Schematic repre-

sentation of two ensembles
of morphological identical
grains, possessing the same
fraction of grains with a
given orientation A (hachured
grains) a the grains A are ran-
domly distributed; b the grains
A form an elongated cluster
that crosses the polycrystal
(see text)
ular information may preferentially propagate (Fig. 11.51b). The grain boundaries
between the grains in a cluster have certainly neighbouring misorientations (low-
angle) but they are not connected. The grain boundary clusters may be nodular or
forming a chain. In case of nodular clusters of boundaries (Fig. 11.52b) and if the
latter have the same coincidence index, a correlation between the crystalline orienta-
tions necessarily exists, we have then a double cluster of grains and grain boundaries.
A cluster elongated in the form of chain of grain boundaries may form a continuous
path that crosses the sample (Fig. 11.52c).
Clusters of grains must play a role in the material properties that mainly depend
on the crystal structure (slip transmission. . .) and in those implying a transport across
the interfaces (electric resistivity…). Clusters of grain boundaries strongly influence
properties that are sensitive to propagation along grain boundaries like intergranular
damage (corrosion, fracture. . .). Thus, a weak region may form within the polycrystal
and constitute a link where an intergranular fracture may be initiated. Alternately,
a chain of permeable (resistant) grain boundaries may allow (impede) the transport
of a signal from one side of the sample to the other. In any case, the formation of
11.8 Local Grain Boundary Texture 413
Fig. 11.52 Various arrangements of grain boundaries in polycrystals of same morphology display-
ing the same percentage of boundaries of type S (bold line): a random distribution; b the boundaries
S form an equiaxed cluster; c the boundaries S form a continuous path (elongated cluster) allowing
a stimulus at the point X to follow the intergranular path until the point Y or Y
clusters (grains or grain boundaries), their shapes but also their spatial distributions
are important parameters to take into account for analyzing the material behaviours.
11.8.2 Observed Cluster Configurations
Clusters of grains and grain boundaries have already been observed at the mesoscopic
scale in textured materials, whatever their structure and their plastic properties (see
Sect. 11.3.4). We recall the presence of island twins in copper after primary recrys-
tallization [45], the formation of low-angle grain boundary clusters in a Fe-50 %
Ni alloy after grain growth, that are predicted by a value of the correlation function
different from 1 [46]. Clusters of cube-oriented grains associated to low-angle grain
boundaries are also detected in textured Al-Mg alloy [51].
Details of the connection between grains and/or grain boundaries have been stud-
ied by transmission electron microscopy. Small clusters of five {111} grains (dark
contrast) are revealed in thin foils (0.1–0.2 μm) of high purity aluminium deposited on
a single {111} crystal substrate of silicon. The clusters, with an equiaxed shape after
deposition, take the form of chains after annealing (chain-like clusters) (Fig. 11.53).
The grains are separated by low-angle or 111 CAD grain boundaries [77].
As-sintered alumina polycrystals doped with magnesia and yttria display an
equiaxed microstructure at the mesoscopic scale; however, observed at the micro-
scopic scale, certain regions gather low-angle grain boundaries, other regions present
grains with planes parallel to a basal plane, in one grain at least (Fig. 11.54) [78].
Clusters of grain boundaries of the same type are also detected in an alumina poly-
Fig. 11.53 Clusters of {111} grains in thin foils of high-purity aluminium: a equiaxed clusters in
as-deposited film; b Chain-like clusters in annealed film [77]
Fig. 11.54 Clusters of grain boundaries in an alumina polycrystal doped with magnesia and yttria
de joints: a clusters of low-angle grain boundaries; b clusters of boundaries with one plane parallel
to a basal (0001) plane in one grain at least (arrows) [78]
crystal doped with magnesia that has been deformed by compression: Fig. 11.55
shows a chain of general (non-coincident) grain boundaries that serves as path for
intergranular fracture [29].
The microstructures of the materials with a low stacking fault energy all present a
tendency to the formation of clusters of = 3n boundaries resulting from multiple
twinning (Fig. 11.56) [38]. The angular deviations of the = 3n boundaries with
respect to the exact coincidence misorientations are relatively small compared to the
deviations from the other coincidence boundaries.
A study, by electron backscattered diffraction (EBSD), of an Inconel alloy not
only describes but also quantifies the clusters constituted uniquely either by general
grain boundaries, or by coincidence boundaries with ≤ 29 and a minimal value of
the low-angle boundaries equal to 2◦ [79]. The samples undergo sequential thermo-
mechanical treatments analogous to those already evoked for the triple junction
analysis [48, 72]. The clusters are observed in two dimensions on sample sections;
Fig. 11.55 Electron micrograph showing the intergranular fracture propagation in an alumina
doped with magnesia, deformed by compression: the fracture follows the path of general boundaries
connected to each other on the form of chain-like cluster and becomes intragranular when meeting
the 2/3 coincidence grain boundary ( = 17b) [29]
Grain boundaries S Dq°

A/B 3 0.04
A/C 3 0.08
A/D 9 0.01
B/D 27b 0.04
B/E1 81a 0.11
C/D 3 0.08
C/E2 9 0.1
Fig. 11.56 Clusters of = 3n grain boundaries in 304-steel with the corresponding crystallo-
graphic parameters [38]
about 2500 grain boundaries and 800 triple junctions are analyzed for each case. To
facilitate the comparisons between microstructures, all the quantities are normalized
by reference to the mean linear intercept L that replaces the average grain size for
a given microstructure. An example of identification of grain boundary clusters is
presented in Fig. 11.57 [79].
The quantities that characterize a boundary cluster are issued from the percolation
theory [80]:
• The mass of a cluster, s, is defined as the total dimensionless length of boundaries
contained in the cluster. A mass close to unity represents an isolated boundary
with no neighbours of the same type, while a large mass (several tens or hundreds)
spans many grains; the cluster shape is not specified.
• The radius of gyration, Rg , of a boundary cluster indicates the average distance of
a boundary from its centre of mass defined by the vector r0 :

N
Rg2 = 1/N |ri − r0 |2 (11.31)
i=1
N is the number of boundaries or boundary segments in a cluster and ri is a vector

pointing to the position of the ith boundary or boundary segment:

N
r0 = 1/N ri (11.32)
i=1
Another interesting length scale is the maximum linear dimension of a cluster,

Dmax , which may govern the length of intergranular crack like that represented in
Fig. 11.55.
The previous quantities describe individual clusters. Average quantities
concerning the entire cluster population may be more representative of the microstruc-
ture. The weighted average cluster mass is given by:

s 2 ·n s
s
s = (11.33)
s·n s
s
where n s , the cluster mass distribution function, gives the number of clusters of size s
per unit area [80]. Similarly, the correlation length ξ is the weighted average diameter
of gyration:
2 Rg2 s 2 ·n s
s
ξ2 = 2 (11.34)
s ·n s
s
Finally, in a given sample, the cluster mass distribution gives the density of clusters
of mass s:
Fig. 11.57 Example of the

cluster identification process,
showing: a the full grain
boundary network, the mean
linear intercept L is reported;
b general grain boundaries are
selected; c an example of a
single interconnected random
boundary cluster [79]

m s = s·n s / s·n s (11.35)
s
The evolutions of the cluster mass distributions for coincident grain boundary
clusters (named special clusters) and for general boundary clusters are reported in
function of the cluster size in case of Inconel 600 recrystallized, then sequentially
deformed and annealed, each cycle being repeated four times (Fig. 11.58) [79]. The
density of large clusters of general boundaries, elevated in the initial state, decreases
while smaller clusters appear. The evolutions of the coincident boundary clusters are
complementary to those of the general clusters, with a size multiplied by ten after
four processing cycles.
These evolutions are presented in a simplified manner by considering the maximal
cluster mass and the weighted average mass (Fig. 11.59), then the maximum cluster
dimension and the correlation length (Fig. 11.60). In agreement with the observed
tendency of reduction of the general boundary cluster mass during grain boundary
engineering, the size of these clusters is reduced by about a factor of three.
The experimental qualitative or quantitative results on the boundary clusters are
limited. In addition to the binary classification and the non-consideration of the grain
boundary plane, the method to analyze clusters presents other shortcomings: The use
of bi-dimensional sections to describe a three-dimensional topology (apart from the
thin foils) and the use of finite observation areas. Artefacts of truncation of many
clusters occur that artificially increases the number of smaller clusters at the expense
of the larger ones. This effect generally diminishes if the investigation area increases.
For example, if 500–1000 grains are investigated, the truncation errors only concern
the very large clusters (mass higher than 100). The cluster examples, all observed
at the mesoscopic or microscopic scale, highlight microstructure in-homogeneities
that must also occur in industrial polycrystals. To extend the understanding of this
heterogeneity to macroscopic systems, simulation appears as the adequate approach.
Several works have developed in this direction, a certain number of which leads to
the concept of percolation (see Sect. 11.9).
11.8.3 Simulated Cluster Configurations
To approach the local grain boundary distribution by simulation, the boundaries have
been in a first step divided into two categories according to their misorientation,
superior or inferior to a fixed arbitrary value [81]. This binary classification may be
theoretically generalized to any number of distinct categories and does not present
a limitation of the methodology. In a first series of works, the method is applied
to two-dimensional polycrystals formed of regular hexagons that tile the plane and
to a microstructure displaying a fibre texture, the common 001 crystal axis being
perpendicular to the sample plane. In that case, only tilt grain boundaries are formed
with misorientation angles between 0◦ and 45◦ , due to the symmetry of the system.
The in-plane axes of each grain are assigned a random orientation with respect to
some fixed external coordinate system. The polycrystal is randomly generated; it
results a uniform distribution of the misorientation angles. The microstructure so
obtained is not in its lowest energy state, but is allowed to evolve such that the total
internal energy is minimized. The obtained stable microstructure is then submitted to
different stimuli: temperature, deformation, external field (electric or due to a gradient
Fig. 11.58 Quantitative description of the change of the grain boundary network of an Inconel alloy
during grain boundary engineering: a–e cluster mass distributions of the general grain boundaries
after 0, 1, 2, 3 and 4 cycles of processing, respectively; f–j complementary cluster mass distributions
of the special boundaries. Each size s reported on the x axes represents the upper bound of a size
range (the ranges are evenly spaced) [79]
Fig. 11.59 Changes in grain

boundary cluster masses
in function of the thermo-
mechanical treatment cycles
underwent by the material.
The maximum cluster size
and the weighted average mass
are analyzed for random and
special clusters. The general
clusters in the recrystallized
sample (cycle 0) have sizes
extended beyond the scan area,
so the data points represent a
lower bound [79]
Fig. 11.60 Changes in the

characteristic length scales
of grain boundary clusters
in function of the thermo-
mechanical processing cycle
underwent by the sample,
including the maximum clus-
ter dimension and the corre-
lation length for general and
special clusters [79]
in some quantity like temperature or defect concentration . . .) and the microstructure

evolution is modelled. The energy of the system is calculated, at each simulation
step, on the basis of the Read and Shockley formula (I.31) by summing on all the
grain boundaries; an interaction term is added, in case of an external stimulus. For a
detailed description of the methodology, the reader may refer to the joined articles
II and I by C. S. Nichols et al. [81].
The angle delineating low-angle and high-angle boundaries being fixed to 10◦ ,
the microstructure generated then relaxed is shown in Fig. 11.61. The grains related
by angles less than 10◦ form a very limited number of clusters composed of two to
Fig. 11.61 Microstructure

showing the hexagonal grains
and the clusters of low angle
grain boundaries (θ < 10◦ ) in
a two-dimensional polycrystal
randomly generated then
relaxed. The grains belonging
to a cluster are connected
by heavy lines through the
centres of the grains [81]
five grains. Note that, in fact, this approach considers clusters of grains with same
orientation separated by low-angle grain boundaries and not real grain boundary
clusters.
Evolutions of this microstructure under the effect of three temperatures, 75, 300
and 1000 K, are registered by taking values of the calculation parameters character-
istic of ceramics. The time scale for each calculation step is much larger than the
time scale of local atomic motions. The results, after 125 simulation steps, show a
clear augmentation of the low-angle grain boundaries and an increase of the cluster
size constituted by these boundaries (Fig. 11.62). The changes in the percentages
compared to those in the initial state are more marked after a treatment at 300 K, as
indicated on the boundary distribution histograms. The dotted line superimposed to
each histogram gives the misorientation distribution before evolution, corresponding
to the microstructure of Fig. 11.61. We must specify that the crystalline orientation
distribution, before and after evolution, does not present any preferential orientation.
The lower density of low-angle grain boundaries after annealing at 1000 K may be
attributed to the thermal fluctuations that allow the clusters to form and break up over
a shorter time scale than for lower temperatures.
By maintaining the temperature at 90 K, the effect of one-axial mechanical strain
on the microstructure evolution of the previous system is also simulated, for 5, 25 and
50 % deformation in tension and for 25 % deformation in compression (Fig. 11.63).
It is obvious that under tensile strains ε ≥ 25 %, the low-angle grain boundaries
form elongated clusters in the direction of the applied strain. The misorientation dis-
tributions only slightly differ with the deformation mode and with the strain level;
moreover, they are similar to those obtained in the absence of deformation. The over-
all grain boundary texture thus does not reflect the large microstructural difference
between a sample annealed at a given temperature and a sample deformed at the
same temperature. This example clearly supports the necessity to consider the local
texture of grain boundaries to capture their role in the material properties.
Fig. 11.62 Simulated microstructures of samples, for ceramic systems, allowed to evolve from
the initial configuration (Fig. 11.61) under the influence of a given temperature: a 90 K; b 300 K; c
1000 K. The results have been obtained after 125 steps of Monte Carlo simulation. The misorien-
tation distributions associated to each microstructure are shown in parallel [81]
In summary, the formation and the distribution of grain and/or grain boundary
clusters may be approached by simulation, applied to a simple microstructure, far
from the real microstructures. Real clusters are observed and their experimental char-
acteristics are analyzed. Due to the new techniques attached to the scanning electron
microscope, a large number of grain boundaries are examined, but the determina-
tion of the local arrangements remains tedious and has not the statistical character
achieved by simply establishing the percentages. Failing to know the local textures
in a material, and starting only from the calculated or measured grain boundary pro-
portions, the probability for a continuous path of boundaries of the same type to
Fig. 11.63 Simulated

microstructures of samples,
for ceramic systems, allowed
to evolve from the initial
configuration (Fig. 11.61)
under the influence of
various deformations at con-
stant temperature 90 K: a 5 %
in tension; b 25 % in tension;
c 50 % in tension; d 25 % in
compression. The results are
obtained after 125 time steps
[81]
cross a sample of a given geometry is predicted by the mathematical percolation the-

ory. Already, the characteristics of the previously observed clusters [79] or calculated
[81] are those of the percolation. The percolation concept applied to grain boundaries
and the possibility that it allows to approach some properties of the whole material
have logically their place in the study of grain boundary networks. However, a large
difference exists between the microstructure approach in terms of local texture and
that in terms of percolating clusters: the latter is a probabilistic approach that does
not generally take into account the possible correlations between grains and/or grain
boundaries in polycrystals.
11.9 Percolation Concept Applied to Grain Boundary

Networks
The percolation concept has been introduced by Hammersley in 1954 [82]; an acces-
sible presentation of the axioms and the basic notions may be found in the book
D. Stauffer [80]. It answers the question: how a connection may be established from
one end to the other of an ensemble of elements that are only partially and randomly
interconnected? The percolation works by all-or-nothing; this is a threshold phe-
nomenon. It postulates that a brutal transition occurs at the percolation threshold pc
for which an infinite cluster of connected elements appears (or disappears) when
the probability p of these elements increases (or decreases). Below this percola-
tion threshold, clusters of random shape appear with increasing grain sizes when
approaching pc . At the threshold pc , and by using a very large-size ensemble, a
continuous network of elements appears, extended to infinity in different directions
of space. This cluster, which is infinitely fragile at pc , strengthens above pc by pro-
gressively incorporating finite element clusters.
The percolation theory provides a statistical analysis of a transport phenomenon in
a disordered medium, the properties of which depending on the local microstructure.
Indeed, a transition threshold, that controls if the transport occurs or not, character-
izes the configuration of such a medium. Remember that the Latin word percolare
means filter and that the first application of percolation was the permeation of a
fluid in a porous medium. We may thus easily predict the interest of the percola-
tion approach for intergranular diffusion or for grain boundary wetting by a liquid.
We have chosen the latter phenomenon, subject of detailed studies [75, 83–85], to
briefly illustrate the possibilities of the method. More generally, any propagation of
a stimulus (sliding, fracture, corrosion…) in a grain boundary appears as a problem
that may be approached by the percolation theory.
11.9.1 Infinite Grain Boundary Network
We focus on grain boundaries as elements of a microstructural ensemble, the spa-

tial arrangement of which being quasi-impossible to describe, in particular in a
three-dimensional polycrystal. In a simplified model, the grain boundaries are divided
into two categories: those which have a poor resistance to a given solicitation called
weak boundaries and those that are resistant called strong boundaries. Then, we
address the question of the percentages of grain boundaries of a given type required
11.9 Percolation Concept Applied to Grain Boundary Networks 425
in order that a solicitation may propagate along or across the boundaries. If the per-
centage of certain boundaries increases, the probability p for the latter to forming
clusters of increasing sizes also increases. When this percentage reaches the perco-
lation threshold, the possibility to find a cluster that extends from one side of the
sample to the other is practically equal to 1, and this independently of the detailed
boundary distribution. The value of the threshold pc indicates the limit between
disjointed clusters and percolating clusters in an infinite network.
Two types of network percolation are generally described, implying either sites or
bonds in this network. Both are used to analyze boundary or grain boundary clusters.
• The bond percolation reports for a connection (open or connecting bond) or not
(closed or non-connecting bond) between two neighbouring sites (edges). A grain
boundary may be considered as a bond between two crystals or between two triple
junctions triples. If the connections are sufficient to form a continuous path, the
probability for transfer via these bonds is equal to unity; the system has reached its
percolation threshold pc that depends on the network geometry. In a crystalline lat-
tice, the value of this threshold regularly decreases when the coordination number
Z increases according to [86]:
d
Z · pc = (11.36)
d −1
with d, the Euclidian dimension of the medium.

• The site percolation, in which the vertices (bonds) rather than the edges (sites) are
declared to be open (full) or closed (empty). Each node of the lattice is randomly
occupied with a probability p or empty with a probability 1− p. The site percolation
may be considered as a complementary point of view to the bond percolation:
when an active site is near another active site, then there is a connecting bond
between them. It well applies to the connection between two crystals (sites) of
same orientation (considered as active sites) linked by a low-angle grain boundary
(connecting or open bond). The site percolation threshold also depends on the
coordination number but not so clearly than the bond percolation.
In the percolation problems, site and bond properties are generally simultaneously
involved. But we must take care that every bond model may be reformulated as site
model on a different lattice, but the converse is false.
The percolation theory can be applied for the grain boundary wetting in polycrys-
talline materials. The boundaries are separated into two distinct categories: wetted
grain boundaries with energy γGB > 2 γLS , considered as full (open) elements and
dry or non-wetted (closed) boundaries with energy γGB < 2 γLS , considered as
empty elements. γGB and γLS are the grain boundary energy and the energy of the
interface between the liquid and the solid, respectively. The wetting probability p
corresponds to the fraction of grain boundaries whose energy is higher than 2 γLS . If
the grain boundary energy is distributed according to a Gaussian law, N(γgb , γ)
(see Sect. 11.7.4), the fraction of wetted boundaries p can be estimated as the fraction
of the area under the distribution curve at the right of the vertical line γGB = 2 γLS
Fig. 11.64 Characteristic

curve of the grain bound-
ary energy distribution by
supposing that γLS is isotropic.
The wetting probability
is given by Q % (area in
grey/total area) [75]
[75]. An experimental proof of this estimate is provided by a wetting study of the

NaCl-H2 O system where the transparency of the samples allows to evaluate the frac-
tion of wetted grain boundaries to 30 %, in good agreement with the value deduced
form the energy spectrum for γGB = 75 mJ/m2 , γLS = 40 mJ/m2 and a relative
spectrum width w = γ /γGB = 0.3 (Fig. 11.64).
The topology adopted by the intergranular liquid thus reflects the high-energy
grain boundary network. The transport property of the liquid through the sample is
evaluated by referring to the percolation theory and by postulating the existence, in an
infinite system, of a critical concentration of wetted grain boundaries pc (percolation
threshold) for which an infinite cluster of connected boundaries appears ( p ≥ pc ).
11.9.1.1 Percolation Thresholds
The value of de pc depends on the topology of the material. Some typical values
are given in Table 11.15; they may be useful for grain boundary networks at two
dimensions (d = 2) in thin layers or in bi-dimensional polycrystals (thin plate with
a thickness inferior to the grain size) and at three dimensions (d = 3) in massive
polycrystals.
Table 11.15 Some values of the bond and site percolation thresholds for different lattices
Crystalline system d Z pc (bonds) pc (sites)
Hexagonal (honeycomb network) 2 3 0.653 0.696
Square 2 4 0.500 0.593
Triangular 2 6 0.347 0.500
Voronoï polygona 2 6 0.329 0.500
Diamond cubic 3 4 0.370 0.450
Body centred cubic 3 8 0.178 0.245
Face centred cubic 3 12 0.119 0.198
Kelvin polyhedra 3 14 0.096 0.154
Fig. 11.65 Fraction of wetted boundaries in function of the relative interfacial energy γLS
crit /γ
J
for different widths w of the grain boundary energy distribution [84]
In the grain boundary wetting approach, the parameters of the intergranular energy
distribution being known, the condition for infinite liquid propagation (fraction of
wetted boundaries equal to the percolation threshold) may be defined by a mini-
mum critical value of the interfacial energy γLS crit /γ [84]. An example is given
GB
in Fig. 11.65, that presents the fraction of wetted boundaries in function of the ration
γLS /γGB , for different widths w of the energy distribution. If all the grain bound-
aries in a network have the same energy (w = 0), the condition γLS crit /γ = 0.5
GB
is verified whatever the lattice topology. The value of the ratio γLS crit /γ cor-
GB
responding to the percolation threshold pc = 0.653, valid for a honeycomb bi-
dimensional (thin foil), is inferior to 0.5. Conversely, this ratio is superior to 0.5 for
a tri-dimensional network in a polycrystal with a coordination number Z = 10 and
a percolation threshold pc = 0.21. The difference between the values of γLS crit (two-
dimensional) and γLS (three-dimensional) increases with the width of the energy
crit
distribution [85]. As a result for the system, the wetting behaviour of samples two-
dimensional and three-dimensional must differ more than the grain boundary ener-
gies are more widely dispersed. At the same temperature, the liquid penetration may
propagate in a three-dimensional polycrystal and be stopped in a bi-dimensional foil.
The percolation thresholds are estimated for an ideal polycrystal composed of
identical polyhedral grains with 12 faces and 5 sides per face and with a random
grain boundary distribution. Different percolation sub-systems may be proposed
for a same structure depending on the wetted element. If the percolating elements
in a three-dimensional structure are the boundary planes and if the coordination
number Z is fixed to 10 (intermediary between Z = 8 for the b.c.c. system and
Z = 12 for the f.c.c. system), we use a site percolation threshold equal to 0.21. If the
percolating elements are the triple junctions, a bond percolation system in a diamond
cubic lattice (d = 3 et Z = 4) represents the connected network with a threshold
Table 11.16 Different types of grain boundary connectivity for a 3D polycrystalline microstructure
Elements forming The network Triple junctions (J) Grain boundaries Grain boundaries
Schematic drawing
Contact by Q Q+J J
Threshold pc 0.39 0.16 0.21
γLS
crit 0.67γGB 0.65γGB 0.62γGB
The liquid propagation through the selected elements forming the network is infinite only if the
interfacial energy is less than a critical value (see text). The contact points are indicated by Q for
quadruple points and J for triple junctions [85]
pc = 0.37. Finally, if the polycrystal is a thin foil with only one layer of grains in its
thickness (d = 2), each plane may be identified as a bond in a honeycomb network,
the percolation threshold reaches the value 0.65; the percolation path is substantially
shorter than in a three-dimensional system.
Different cases of grain boundary connectivity in a three-dimensional polycrys-
talline microstructure are presented in Table 11.16. The values of γLS crit are given for
a normal energy distribution, an average value γGB and a distribution width equal
to 0.3 [85].
The use of the percolation concept to practically approach grain boundary wetting
thus appears promising. However, several problems limit its application: the choice
of the parameter that discriminates between low and high energy seems somewhat
arbitrary, the polycrystal topology never corresponds to the reference one, the quasi-
inevitable existence of a material texture and of a non-null constraint, and above all
the finite size of the sample. The latter plays a major role in the permeability of a
boundary network. Generally, a displacement of the percolation threshold to a value
pc∗ > pc occurs for narrow samples and to a value pc∗ < pc for short samples.
11.9.1.2 Fractal Structure of Clusters
Near the percolation threshold, any property X, in particular the size of the largest
cluster, varies in average with p [86]:
X ∝ |p − pc |n (11.37)
where n is the critical exponent. A high value of n indicates that the property is
established with difficulty above pc .
The differences between various systems fade in the vicinity of pc ; all the systems
present a universal behaviour. Their properties no longer depend on their detailed
structure, but obey global laws called scaling laws. All the two-dimensional percola-
tion problems form a unique family; similarly all the three-dimensional percolation
problems form another family. When we are exactly at the percolation threshold,
the clusters constitute a fractal: each portion of a cluster is self-similar to the whole
cluster. The exponent n in relation (11.37) determines the fractal topology of the
percolation clusters characterized by a percolation correlation length ξ and by the
fractal Hausdorff dimension d [79, 87] that differs from the Euclidian dimension.
So, when we are near pc , d = 1.9 (instead of 2) for a bi-dimensional system and d
= 2.5 (instead of 3) for a three-dimensional system. For a two-dimensional system,
the number of points required to establish a reliable value of the fractal dimension
with an accuracy ±0.01 is estimated to 100.
11.9.2 Finite Grain Boundary Network
The percolation thresholds previously calculated imply an infinite system. In practice,

for a finite system like a grain boundary network in a real polycrystal, the percolation
threshold is not precisely defined. Any effective threshold value, numerically or
experimentally obtained, can be extrapolated to an infinite system only with great
caution. Indeed, the application of the percolation theory crucially depends on the
dimension of the system.
Generally, an effective percolation threshold p∗c is defined for a finite network of
length L by: ∗
p − p ∝ L−1/v (11.38)
c c
with ν, the critical index of correlation length, equal to 4/3 for bi-dimensional struc-
tures and 0.9 for three-dimensional structures [87].
11.9.3 Correlated Percolation
Numerous percolation problems arise in a non-random medium where the occupation

of one site or one bond may influence the occupation of its neighbours. A correlation
factor inherent to polycrystals is the misorientation θ between crystals that, inter
alia, influences the probability P of grain boundary wetting or the fraction of wet-
ted boundaries. In the wetting context, the values of the percolation thresholds for
a random θ angle distribution are compared to these values for a correlated system
of misorientations, obtained by simulation starting from the orientations of neigh-
bouring grains. The difference of the threshold values decreases when the coordina-
tion number increases. In case of thin foils, it reaches a maximal value close 0.02
Fig. 11.66 Displacement

of the percolation threshold
resulting form the corre-
lated character of the grain
boundary misorientations in a
honeycomb network [83]
(Fig. 11.66) [83]. If the dependence p(θ ) is known, the percolation properties of
textured materials may be modelled.
The polycrystals of low stacking fault energy materials constitute systems where
the grain boundaries are not randomly connected, with the presence of a large num-
ber of = 3n boundaries constrained to respect the combination rule (10.1) at
triple junctions. There are numerous percolation approaches of the properties of
these polycrystals; each of them starts with the hypothesis of a random distribu-
tion of the boundaries that are classified into two categories: either susceptible or
resistant to the propagation of a process (corrosion, diffusion and intergranular frac-
ture). The resistant category is supposed to be mainly composed by the = 3n
boundaries. This assumption is questionable because only the coherent twins = 3
{111} display resistance properties really different from those of the other bound-
aries. Despite this remark, the approach by Schuh et al. [72] deserves to be reported:
Monte Carlo simulation is used to construct a two-dimensional boundary networks
by maintaining the distinction between two boundary categories and by adding the
crystallographic constraint of the combination rule at triple junctions. The constraint
is defined by the ratio A = f 3 / f where f 3 is the frequency of = 3 boundaries
and f that of the other coincidence boundaries, mainly = 3n . The bond percola-
tion threshold is considered as associated to the percentage of resistant boundaries
in a two-dimensional hexagonal network: pc = f c = 1 − 0.653 = 0.347 [87].
Thus, there is an infinite network of susceptible boundaries (general boundaries with
> 29) if p < 0.347; this network is destroyed when the percentage of resistant
boundaries increases ( p > 0.347), impeding the transport propagation. The perco-
lation is studied on hexagonal grids with 100 grains on each side, involving about
3.104 boundaries. For given input values of f and A, several hundred lattices are
constructed and the fraction of networks containing a percolating cluster of sus-
ceptible boundaries is calculated. Figure 11.67 shows the evolution of the ratio
in function of the fraction of resistant boundaries f The percolation threshold for
a random grain boundary distribution (bold line) is equal to 0.347, in agreement
Fig. 11.67 Probability for susceptible grain boundaries in a hexagonal network with 100 grains
by side to percolate in function of the percentage of resistant boundaries f The bold line illus-
trates the results for non-constrained boundaries, they agree with those obtained by the classical
random percolation theory. The fine lines show the displacement of the percolation threshold when
crystallographic constraints are imposed to the system [72]
with the results of the classical percolation theory; it is displaced towards higher
values when a crystallographic constraint is imposed: pc ≈ 0.5 for A = 0.67 and
pc ≈ 0.55 for A = 0.5; the percentage of = 3 boundaries then represents then
half the coincident boundaries. It must be noted that the sense of these displacements
is in agreement with the previous tendency [83], but the amplitude is larger. The
calculated thresholds for infinite systems are abrupt; on the contrary, each curve
in function of f exhibits some spread around the calculated values; this is a finite-
size effect. The application of the product rule (10.1) and thus the existence of
local constraints have a substantially impact on the percolation behaviour of a grain
boundary network [72]. In all cases, the percentage of resistant boundaries required to
break a continuous path of boundaries susceptible to some damage must be superior
( pc ≥ 0.5) to the two-dimensional bond percolation threshold ( pc = f c = 0.347).
The previous results are limited to two-dimensional networks, but scale laws exist
that relate the standard percolation problems in two and three dimensions [87]. These
laws are applicable only if the microstructure, here the grain boundary distribution,
is isotropic. This is often the case of the twin-dominated microstructures for which
experimental studies detect a clear tendency to isotropy. From an engineering point of
view, the three-dimensional boundary networks of materials susceptible to multiple
twinning must be considered with a more demanding criterion than that suggested by
the percolation theory. If a factor influences the element distribution, for example a
texture in the case of a polycrystal, the percolation is said directed. A stress field may
impede or enhance intergranular wetting in certain directions; it results an anisotropy
of the percolation threshold that may lead to a transition from a three-dimensional
geometry to a two-dimensional geometry implying an increase of pc .
In summary, the percolation theory is strictly probabilistic and does not rest on any
correlation in the spatial arrangement of the grain boundaries. As soon as we want to
apply this percolation model to the relation between the boundary distribution and a
property of the polycrystal, we face several complications:
• Solutions are obtained for ideal networks (identical Kelvin polyhedra, for example);
this is rarely the case for the grain organization.
• In its simplest, the percolation model rests on a binary description of the microstruc-
ture with two categories of grain boundaries, susceptible (weak) or resistant
(strong). In practice, one can consider more nuanced responses of the bound-
aries to any solicitation, depending on a more complex way of their geometric
parameters.
• In fact, weak(strong) boundaries may be preferentially associated to other weak
(strong) boundaries; a small proportion of boundaries of the same type is then
sufficient to form a continuous network.
• The condition for a connected network of weak boundaries to be dispersed in
the whole sample may lead to an overestimation of the boundary resistance to a
stimulus. To make the material vulnerable, it is just sufficient that a cluster of weak
boundaries exceeds a critical dimension. At the extreme, we know that breaking
a ceramic depends on the existence of a weak link according to Weibull statistics;
similarly, the existence of a small fraction of weak boundaries may be sufficient
to cause intergranular fracture; it is then invalid to consider propagation through
a continuous percolation path.
• The percolation threshold may differ according to the type of grain boundary dam-
age: one-dimensional path for corrosion, two-dimensional path for intergranular
fracture.
Furthermore, the fractal considerations at the percolation threshold are mainly

applied to the shape and the size of the clusters, thus to geometrical properties. Pure
theory of percolation is well ahead of the applications. Two types of difficulty can
be reported:
• It is rare to perform experiment consistent with theoretical conditions under which

the microscopic scale, where transmission occurs between network elements, is
actually very small compared to the macroscopic scale, that of the entire network.
• The dynamic aspects are difficult to characterize in practice.
However, despite all these restrictions, the percolation theory provides a math-
ematical tool that may be used to predict some macroscopic properties of poly-
crystalline materials when the connectivity of one of the microstructural elements
controls the behaviour of the whole: grain boundary diffusion at relatively low tem-
perature (when the volume diffusion may be neglected, intergranular corrosion, grain
boundary wetting by a liquid, intergranular fracture . . . All the remarks (advantages
and limits) that accompany the percolation approach of grain boundary may be
extended to any transport phenomenon (diffusion, electric conductivity . . .) and to
any propagation of a solicitation via the boundaries. We have seen in particular that
the propagation of an intergranular fracture (Fig. 11.55) depends on the type of grain
boundary and that the connection of certain boundaries, more susceptible to fracture
than the others, may lead to the existence of a continuous crack path. In the absence
of a detailed description of the boundary network, an approximation of the criti-
cal percentage of grain boundaries susceptible to propagate a given process enables
to predict the polycrystal behaviour. In that sense, the approach of grain boundary
wetting by the percolation theory constitutes a useful example to go towards grain
References
1. K.T. Aust, J.W. Rutter, Trans TMS-AIME 215, 119 (1959)

2. D.G. Brandon, Acta Metall. 14, 1479 (1966)
3. Y. Ishida, M. Mc.Lean, Phil. Mag. 27, 1125 (1973)
4. G. Palumbo, K.T. Aust, Acta Metall. Mater. 38, 2343 (1990)
5. H Kokawa, T. Watanabe, S. Karashima, Phil. Mag. A44, 1239 (1981)
6. M. Déchamps, F. Barbier, A. Marrouche, Acta Metall. 35, 101 (1987)
7. V. Randle, Acta Mater. 46, 1451480 (1997)
8. D. Wolf, J. Phys. 46, C4–197 (1985)
11. V. Paidar, Phil. Mag. A66, 41 (1992)
12. S. Lartigue, L. Priester, J. Phys. 49, C5–451 (1988)
13. T. Watanabe, Res. Mech. 11, 284 (1984)
14. L. Priester, Mater. Sci. Eng. A 309–310, 430 (2001)
16. V. Traskine, Terra Abstracts 1, 190 (1995) (A VOIR)
17. P. Lin, G. Palumbo, U. Erb, K.T. Aust, Scr. Metall. Mater. 33, 1387 (1995)
18. W. Przetakiewicz, K.J. Kurzydlovski, M.W. Grabski, Mater. Sci. Technol. 2, 106 (1986)
19. L.C. Lim, R. Raj, Acta Metall. 32, 1183 (1984)
20. L. Priester, Rev. Phys. Appl. 24, 419 (1989)
22. H.J. Bunge, Texture Analysis in Materials Science (Butterworthhs, London, 1982)
23. F. Haessner, J. Pospiech, K. Sztwiertnia, Mater. Sci. Eng. 57, 1 (1983)
24. B.L. Adams, P.R. Morris, T.T. Wang, K.S. Willden, S.I. Wright, Acta Metall. 35, 2935 (1987)
25. J.K. Mackenzie, Biometrika 45, 45 (1958)
26. J.K. Mackenzie, Acta Metall. 12, 223 (1964)
27. D.H. Warington, N Boon, Acta Metall. 23, 599 (1975)
28. D.H. Warrington, J. Phys. C4–36, 87 (1975)
29. S. Lartigue, L. Priester, J. Am. Ceram. Soc. 71, 430 (1988)
30. J.K. Mackenzie, M.J. Thompson, Biometrika 44, 205 (1957)
31. A. Garbacz, M.W. Grabski, Acta Metall. Mater. 41, 469–475 (1993)
32. V.Y. Gertsman, K. Tangri, R.Z. Valiev, Acta Metall. Mater. 42, 1785 (1994)
33. L.K. Fionova, Mater. Chem. Phys. 37, 201 (1994)
34. Yu. Lisovski, L.K. Fionova, Interface Sci. 4, 119 (1996)
35. M. Lannoo, P. Friedel, Atomic and Electronic Structure of Surfaces (Spinger, Berlin, 1991),
p. 71
36. G. Palumbo, K.T. Aust, Recrystallization 90, ed. by T. Chandra (AIME, Warrendale, 1990),
p. 101
37. G. Herrmann, H. Gleiter, G. Bäro, Acta Metall. 24, 353 (1976)
38. V. Yu Gertsman, K. Tangri, Phil. Mag. A64, 1319 (1991)

39. V. Randle, Acta Mater. 47, 4187 (1999)
40. M.W. Grabski, J. Phys. 46, C4–567 (1985)
41. L.S. Shvindlerman, B.B. Straumal, Acta Metall. 33, 1735 (1985)
42. T. Watanabé, N. Yoshikawa, S. Karashima, Conf Tokyo, November 1981
43. P. Volovitch, V. Traskine, T. Baudin, L. Barralier, Interface Sci. 10, 303 (2002)
44. T. Watanabe, S. Tsurekawa, Acta Mater. 47, 4171 (1999)
45. O.V. Mishin, G. Gottstein, Mater. Sci. Eng. A249, 71 (1998)
46. F. Caleyo, T. Baudin, R. Penelle, V. Venegas, Mater. Sci. Eng. A298, 227 (2001)
47. M. Kumar, W.E. King, A.J. Schwartz, Acta Mater. 48, 2081 (2000)
48. W.E. King, A.J. Schwartz, Scr. Mater. 38, 449 (1998)
49. V.Y. Gerstman, J.A. Szpunar, Scr. Mater. 38, 1399 (1998)
50. T. Shibayanagi, H. Takatani, S. Hori, Mater. Sci. Forum 94–96, 495 (1991)
51. K. Matsumoto, T. Shibayanagi, Y. Umakoshi, Acta Mater. 45, 439 (1997)
52. D.J. Jensen, Mater. Sci. Eng. A234, 762 (1997)
53. T. Watanabe, K.I. Arai, K. Yoshimi, H. Oikawa, Phil. Mag. Lett. 59, 47 (1989)
54. J. Harase, Can. Metall. Q. 34, 185 (1995)
55. R. Shimizu, J. Harase, Acta Metall. 37, 1241 (1989)
56. D.E. Fleet, P.I. Welch, P.R. Howell, Acta Metall. 26, 1479 (1978)
57. V. Randle, Acta Metall. Mater. 42, 1769 (1994)
58. V. Randle, P. Davies, B. Hulm, Phil. Mag. A 79, 305 (1999)
59. D. Wolf, K.L. Merkle, in Materials Interfaces: Atomic Level Structure and Properties, ed. by
D. Wolf, S. Yip (Chapman & Hall, London, 1992), p. 87
60. A. Garg, W.A.T. Clark, J.P. Hirth, Phil. Mag. 59A, 479 (1989)
61. B.L. Adams, Metall. Trans. 17A, 2199 (1986)
62. J.E. Hilliard, Trans. Am. Inst. Min. Engrs. 224, 1201 (1963)
63. D.M. Saylor, A. Morawiec, G.S. Rohrer, Acta Mater. 51, 3663 (2003)
64. D.M. Saylor, B.S. El Dasher, A.D. Rollett, G.S. Rohrer, Acta Mater. 52, 3649 (2004)
65. W.A. Swiatnicki, M.W. Grabski, Mater. Sci. Eng. 100, 85 (1988)
66. W.A. Swiatnicki, W. Lojkowski, M.W. Grabski, Acta Metall. 4, 599 (1986)
68. W.W. Mullins, Trans. AIME 218, 373 (1960)
69. C.S. Smith, Trans. Metall. Soc. AIME 175, 15 (1948)
70. E.G. Doni, G.L. Bleris, Phys. Stat. Sol. 110, 393 (1988)
71. M. Frary, C.A. Schuh, Acta Mater. 51, 3731 (2003)
72. C.A. Schuh, R.W. Minich, M. Kumar, Phil. Mag. 83, 711 (2003)
73. P. Fortier, K.T. Aust, W.A. Miller, Acta Metall. Mater. 43, 339 (1995)
74. K.J. Kurzydlowski, Mater. Charact. 26, 57 (1991)
75. V. Yu Traskine, Z.N. Skvortsova, Colloid J. 59, 767 (1997)
76. A. Passerone, N. Eustathopoulos, P. Desre, J. Less Common Met. 52, 37 (1977)
77. L. Fionova, O. Konokenko, V. Matveev, L. Priester, S. Lartigue, F. Dupau, Interface Sci. 1,
207 (1993)
78. S. Lartigue, L. Priester, F. Dupau, P. Gruffel, P. Carry, Mater. Sci. Eng. A164, 211 (1993)
79. C.A. Schuh, M. Kumar, W.E. King, Acta Mater. 51, 687 (2003)
80. D. Stauffer, Introduction to Percolation Theory (Taylor and Francis, London, 1985)
81. C.S. Nichols, R.F. Cook, D.R. Clarke, D.A. Smith, Acta Metall. Mater. 39(1757), 1667 (1991)
82. J.M. Hammersley, K.W. Morton, J. Roy. Stat. Soc. B 16, 76 (1954)
83. V. Traskine, P. Volovitch, P. Protsenko, Y. Kutcherinenko, S. Botbhenkov, Proc. HTC-2000,
Japan Welding Research Institute
84. P.M. Volovitch, P. Protsenko, Z.N. Skvortsova, L. Barrallier, V. Yu Traskine, Colloid J. (Kol-
loidnyi Zhurnal) 64, 306 (2002)
85. P. Volovitch, V. Traskine, L. Barrallier, Z. Metallkd. 95, 4 (2004)
86. S. Kirkpatrick, Rev. Mod. Phys. 45, 574 (1973)
87. A. Aharony, D. Stauffer, Introduction to Percolation Theory (Taylor and Francis, London,
1992)
Epilogue
Invitation Au Voyage
Là, tout n’est qu’ordre et beauté,
Luxe, calme et volupté.
(Charles Baudelaire)
Call for Elsewhere
There, all is order and beauty,
Luxury, calm and voluptuousness.
(Charles Baudelaire)
We have not reached the end of our route Grain Boundaries, but nonetheless we
are far away from our first step. The latter took place in a familiar territory,
described in many books, though rather barren with its elements of
bicrystallography, constraints at two levels, intrinsic and extrinsic, its structural
units, but rich territory of all its specificities. In a second stage, we went through
unstable areas where the properties of each individual depend heavily on
interactions with its neighbours and we tried to understand the resulting changes.
In the end, we took risks by journeying into areas where collective behaviour takes
precedence over individual behaviour; there the law of networks prevails.
Other domains have not been explored, those of properties; but we have made
some inroads into some of them. We also provided the luggage and sufficient
information to undertake the journey.
We hope to have filled the underlying objective in all stages of the work which
is to encourage our readers to journey towards a kind of Eldorado:
• Where the paths of the different explorers in the Grain boundaries domain
converge,
• Where the internal barriers to the world of materials vanish
One can be a scientific and also dream…

DOI: 10.1007/978-94-007-4969-6, Springer Science+Business Media Dordrecht 2013
Recommended Reading
Books Dedicated on Interfaces (and Dislocations)
D. Mc Lean, Grain Boundaries in Metals (Clarendon Press, Oxford, 1957)

J.P. Hirth, J. Lothe,Theory of Dislocations (Mc Graw Hill, New York, 1968)
H. Hu (ed.), The Nature and Behavior of Grain Boundaries (Plenum Press, New York, 1972)
Ecole d’Été de Métallurgie Physique Gassin 1973, Interfaces et Surfaces en Métallurgie, (Trans
Tech Publications, Switzerland, 1975)
G.A. Chadwick, D.A. Smith, Grain Boundary Structure and Properties (Academic Press,
London, 1976)
W.C. Johnson, J.M. Blakely (eds.), Interfacial Segregation (Pub. ASM, Metals Park, 1977)
R.W. Balluffi (ed.),Grain Boundary Structure and Kinetics( Pub. ASM, Metals Park, 1980)
Structure et Propriétés des Joints Intergranulaires, J. de Physique, C6–43 (1982)
M. Lannoo,Les joints de grains dans les matériaux, Les Editions de Physique (1985)
D. Wolf, S. Yip, Materials Interfaces (Chapman and Hall, London, 1992)
A.P.Sutton, R.W. Balluffi, Interfaces in Crystalline Materials (Clarendon Press, Oxford, 1995)
V. Randle, The Role of the Coincidencce Site Lattice in Grain Boundary Engineering (The
Institute of Metals, London, 1996)
G. Saindrenan, R. Le Gall, F. Christien,Endommagement interfacial des métaux – Ségrégation
interfaciale et conséquences (Ellipses, Paris, 2002)
Books Dedicated to Methods
P.B. Hirsch, A. Howie, M.J. Whelan, R.B. Nicholson, D.W. Pashley, Electron Microscopy of
Thin Crystals (Butterworths, London, 1965)
A.K. Head, P. Humble, L.M. Clarebrough, A.J. Morton, C.T. Forwood, Computed Electron
Micrographs and Defect Identification (North Holland, Netherlands, 1973)
M.H. Loretto, R.E. Smallman,Defect Analysis in Electron Microscopy (Chapman and Hall,
London, 1975)

438 Recommended Reading
J.C.H. Spence,Experimental High-Resolution Microscopy (Clarendon Press, Oxford, 1981)

D. David, R. Caplain, Méthodes usuelles de caractérisation des surfaces (Eyrolles, Paris, 1988)
J. Ruste et J.-F. Bresse (eds.), Les nouvelles techniques de micro et nano-analyse (EDP Sciences,
France, 1995)
H.O. Kirchner, L.P. Kubin, V. Pontikis (ed.), Computer Simulation in Materials Science, NATO
ASI (E series, vol.308, ) (Kluwer Academic Press, Dordrecht, 1996)
Ecole thématique du CNRS, Microscopie des Défauts Cristallins (St Pierre d’Oléron, mai, 2001)
J.-L. Pouchou, (ed.),L’analyse EBSD – Principes et Applications, (EDP Sciences, France, 2004)
Index
A D
Atomic structure of grain boundary, 77, 144, Decomposition of dislocations, 141, 142, 250,
152, 161, 182, 191, 219 251, 256, 257, 290, 291
Dichromatic complex, 9–11, 52
Dichromatic pattern, 10, 12
B Disclination, 46, 49, 90, 91, 103, 104, 113,
Bicrystallography, 3, 8, 23, 27, 244, 307 273, 275–277, 280, 281, 319, 326–
Burgers vector density, 29, 30, 43, 138, 329, 331, 333
145, 269, 270 Dislocation accommodation (or intergranular
stress relaxation), 2, 141, 253, 257,
269, 270, 283, 290, 292, 297–299
C Dislocation density tensor, 30, 280, 331
Cavitation, 269, 305, 344, 393 Disconnection, 189, 191, 223
Classification of grain boundaries, 23, 122,
167, 342, 393
Cluster (of grains or grain boundaries), 411 E
C.n.i.d. = cell of non-identical displacements, Effective interplanar spacing, 121, 298, 299,
11, 12, 25, 57, 59, 113, 115, 121, 342
145, 295, 297 Elastic stress fields at grain boundaries, 45
Coincidence Electronic state density, 151
approximated coincidence, 14, 18 Electronic structure of grain boundary, 192, 194
concept of coincidence and coincidence Enrichment factor, 154–156, 158, 192, 211
lattice, 12, 14, 19 Equivalent rotation (or description), 6–8, 22,
one-dimensional coincidence or plane 35, 37, 309, 318, 347
matching model, 16–17
planar or bi-dimensional coincidence, 16
Coincidence index combination rule, 397, 399, F
400, 405 Faceting, 25, 27, 66, 72, 94, 109, 112, 127,
Combination of dislocations 166, 176, 224, 342
Compensation temperature, 173
Complexion, 148, 201–205, 344
Contact angle, 226, 228 G
Correlation function or correlation degree, Grain boundary
345, 346, 368, 412 amorphous grain boundary, 296

440 Index
G (cont.) Grain boundary sliding, 96, 147, 165, 276,

asymmetrical tilt grain boundary, 24, 71, 300, 343
75, 76, 121, 125, 126 Grain boundary step, 190
CAD = one-dimensional coincidence
grain boundary, 40, 349, 365, 403
CSL = coincidence grain boundary, 27, I
37, 40, 140, 249, 338, 364, 382, 415 Image force, 150, 241–245, 255
delimiting grain boundary, 55, 56, 144 Interface operation = isometry, 3–5, 23
favoured grain boundary, 52, 53, 55, 59, Interfacial (or intergranular) energy, 23, 83,
84, 85 99–108, 110, 112, 113, 115, 117,
general grain boundary, 24, 53, 122, 206, 118, 121, 123, 145, 152, 189,
239, 246, 249, 270, 273, 293, 295, 195–197, 218, 221, 224, 273,
297, 298, 361, 384, 397 348, 355, 380, 317, 329, 342,
quasi-periodic (or quasi-crystalline) grain 409, 426, 427
boundary, 64, 145, 270, 292, 296, Interfacial tension, 99, 108, 109, 226, 234,
297 311, 312, 325
singular grain boundary, 122, 177, 184, Intergranular corrosion, 337, 432
246, 273, 274, 307, 312 Intergranular diffusion coefficient, 201, 206,
symmetrical tilt grain boundary, 13, 24, 26, 208, 217, 236, 271, 272, 290,
27, 32, 42, 43, 45, 46, 61, 65, 67, 71, 298–300, 343, 394
75, 76, 84, 86, 123, 124, 126, 172, Intergranular dislocation
175, 251, 253, 264, 265, 273, 274, coherency/anti-coherency, 31
284, 285, 311 extrinsic, 11, 35, 40, 41, 82, 138, 142, 144,
twist grain boundary, 24, 26, 27, 32, 34, 36, 239, 257, 269, 275, 277, 292
38, 39, 46, 49, 52, 66, 67, 76, 79, 89, intrinsic
96, 102, 115, 116, 139, 168, primary intrinsic, 33, 34
179–181, 249, 285, 297 secondary intrinsic, 37, 140, 274
three-dimensional grain boundary, 68, 233, misfit, 19, 65, 116, 149, 176, 180, 197, 199,
235 221–223, 226
vicinal grain boundary, 109, 122, 127, 174, partial, 44
222, 239, 249, 274, 397 Intergranular fracture, 194, 339, 345, 379, 411,
Grain boundary Distribution 413, 415, 429, 432
coincidence grain boundary distribution Intergranular precipitation, 206
(CGBD), 338 Intergranular segregation
distribution of the five macroscopic equilibrium segregation, 147, 148, 152,
parameters, 338 165, 205, 207–210, 212, 213
grain boundary character distribution non-equilibrium segregation, 147, 148,
(GBCD), 27, 118, 338, 351 207–210, 212
grain boundary energy distribution, 394, Intergranular wetting, 204, 205, 431
395, 408, 426, 427 Invariance (invariant), 3, 8
grain boundary misorientation distribution
(GBMD), 338
grain boundary plane distribution, 338, L
380, 383, 387, 390–392 Lattice, 12, 14–16, 18, 22, 341
triple junction distribution, 379, 397, 401, coincidence lattice (CSL), 12, 14–16, 18,
402, 407, 408 22, 341
Grain boundary Engineering, 174, 204, 213, decomposition lattice, 61, 62
300, 305, 366, 370, 397, 418 DSC lattice, 8, 10, 14, 15, 19, 37, 38, 84,
Grain boundary melting, 95–98, 131, 354 113, 140, 250, 251, 257, 306, 328,
Grain boundary plane 331, 379
median, 25, 31, 52, 57, 66, 68, 71, 72, 100, O-lattice, 19–22, 33, 34, 36
102, 171, 183 O2-lattice, 19, 22, 23, 37, 38, 41
mean (or average), 8, 23, 24, 25, 32, 127, Local atomic relaxation at grain boundaries,
144, 297, 300, 314 12, 389
Index 441
Localization (delocalization) of the S

dislocation core Segregation energy (or free enthalpy), 149,
Low energy criteria, 121 153, 158, 161, 164, 170, 182, 183,
186, 187, 196, 197, 199, 208
Structural unit, 49–56, 58–70, 72–79, 81–87,
M 89–91, 93, 94, 98, 102, 103, 113,
Macroscopic degree of freedom 121, 123, 125, 129, 130, 138, 139,
(or parameter), 5, 23, 52, 100, 144, 159, 165, 166, 170, 177, 183,
122, 254, 296 189, 190, 192, 200, 251, 253, 263,
Microscopic degree of freedom (or parameter), 274, 276, 319–323, 327
5, 12, 49, 113, 287 c-Surface, 11, 113, 115, 145, 295
Misorientation, 301, 340, 344, 345, 347, 353, Symmetry operation, 3, 4, 6, 11, 43, 44,
354, 379 307, 318
P T
Phase transition (or phase transformation) Thermal etching, 109, 325, 326, 344
at grain boundaries, 93, 201 Transmission of dislocations, 254
Percolation, 339, 370, 397, 401, 414, Tricrystal, 333
420, 422–432 Tricrystallography, 307
Point defects, 80, 135–137, 147, 195, 205, 212 Triple junction, 320, 399, 401, 402, 406, 407,
428
Triple junction energy, 322, 325, 397
Q Texture of grain boundaries, 421
Quaternion, 7, 8
W
R Wigner-Seitz cell, 33
Rigid body translation, 3, 4, 6, 11, 44, 55, 59, Wulff construction, 106, 226–230, 233, 235
67, 78, 94, 113, 114, 145, 146, 314,
327, 329, 330
Rodrigues vector, 7, 8, 390, 391

(Louisette Priester (Auth.) ) Grain Boundaries

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Louisette Priester (Auth.) ) Grain Boundaries

Uploaded by

Copyright:

Available Formats

Springer Series in Materials Science

For further volumes:

Library of Congress Control Number: 2012952016

Ó Springer Science+Business Media Dordrecht 2013

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Most current studies tend to present the knowledge on interfaces in crystalline

Then I found again the metal with a multiscale approach, by simultaneously

Part I From Grain Boundary Order to Disorder

1 Geometrical Order of Grain Boundaries . . . . . . . . . . . . . . . . . . . 3

2 Mechanical Stress Order of Grain Boundaries. . . . . . . . . . . . . . . 29

2.3.3 Interest and Limit of the Intrinsic

3 Atomic Order of Grain Boundaries. . . . . . . . . . . . . . . . . . . . . . . 49

4 Grain Boundary Order/Disorder and Energy . . . . . . . . . . . . ... 93

4.2 Interfacial Energy: Thermodynamic Aspects

Part II From Ideal to Real Grain Boundary

5 Defects in the Grain Boundary Structure . . . . . . . . . . . . . . . . . . 135

6 Grain Boundary Segregation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

6.2 Thermodynamic Approaches of Equilibrium Segregation. .... 151

7 Precipitation at Grain Boundaries. . . . . . . . . . . . . . . . . . . . . . . . 217

7.3 Generalised Wulff Construction for Nucleus Equilibrium

8 Interactions Between Dislocations and Grain Boundaries . . . . . . . 241

9 Intergranular Stress Relaxation . . . . . . . . . . . . . . . . . ........ 269

9.2 Evolution of Extrinsic Dislocation Stress Fields . . . . . . . . . . . 277

Part III From the Free to the Constrained Grain Boundary

10 The Triple Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 305

11 Grain Boundary Network: Grain Boundary Texture . . . . . . . . . . 337

11.2.1Crystal Orientation and Grain Boundary

Recommended Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437

CAD Coincidence Axis Direction

ODF Orientation Distribution Function

1. W.T. Read, W. Shockley, Phys. Rev. 78, 275 (1950).

The physical invariances that rest on the group structure seem

1.1 Grain Boundary Geometry

x II = (A/τ )GI x I (1.1)

L. Priester, Grain Boundaries, Springer Series in Materials Science 172, 3

(A /τ ) = (A/τ ) · GI = αGI (1.2)

GII = αGI α−1 and GI = α−1 GII α (1.3 bis)

In the previous equations, A is generally a complex transformation that implies a

1.1.1 Crystallographical Parameters

The construction of a grain boundary is schematized on Fig. 1.1.

1.1.2 Equivalent Rotations: Disorientation

1.1.3 Rodrigues Vector and Quaternions

The four components e1 , e2 , e3 and e4 are related by: e1 2 + e2 2 + e3 2 + e4 2 = 1.

1.2.1 General Methodology: From the Dichromatic Complex

Table 1.2 The different steps DICHROMATIC COMPLEX

1.2.1.1 Dichromatic Complex

The dichromatic complex is formed by the crystals I and II misorientated and

1.2.1.2 Ideal Bicrystal

A bicrystal is formed by cutting the dichromatic complex by a plane, then by rejecting

1.2.1.3 Real Bicrystal: Real Grain Boundary

Fig. 1.3 Cell of non-identical

1.2.2 Dichromatic Pattern: Coincidence Site Lattice (CSL)

CSL = I = TI ∩ TII (1.12)

The coincidence index varies discontinuously with the misorientation angle θ

Table 1.3 List of certain θ ◦ 110

Fig. 1.4 Evolution of the

It is always possible to define exact coincidence relationships between two crystals

1.2.3 Translation Lattice of the Bicrystal (DSC Lattice)

Starting with two interpenetrated lattices in coincidence relationship, let us trans-

DSC = U = TI ∪ TII (1.16)

1.2.4 Extension of the Coincidence Notion

1.2.4.1 Planar or Bi-dimensional (2D) Coincidence

Table 1.3 List of certain θ ◦ 110

Most often a high-angle grain boundary presents an angular deviation θ around

If the favoured = 5 [001] (210) boundary may be formed by units B or B