Dislocation Cross-Slip in Face-Centered Cubic Solid

Solution Alloys

THÈSE NO 8383 (2018)

PRÉSENTÉE LE 12 FÉVRIER 2018
À LA FACULTÉ DES SCIENCES ET TECHNIQUES DE L'INGÉNIEUR
LABORATOIRE DE MODÉLISATION MÉCANIQUE MULTI-ÉCHELLE
PROGRAMME DOCTORAL EN MÉCANIQUE

ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE

POUR L'OBTENTION DU GRADE DE DOCTEUR ÈS SCIENCES

PAR

Wolfram Georg NÖHRING

acceptée sur proposition du jury:

Prof. J. Botsis, président du jury

Prof. W. Curtin, directeur de thèse
Prof. D. Rodney, rapporteur
Prof. K. Albe, rapporteur
Dr P. Derlet, rapporteur

Suisse
2018
Whence, in order to discover ores, I think it better to abandon the way of bestial
and fearless men and to choose the way of using the signs that are exhibited to us
through the benignity of Nature, founded on truth and approved by all experts
because of their experience, which, as is evident, does not consist of words or
promises of incomprehensible and vain things.
— Vannoccio Biringuccio, De la pirotechnia (1540)
Acknowledgements
This work was carried out from 2014 to the time of writing at the École polytechnique fédérale
de Lausanne (EPFL), in the Laboratory for Multiscale Mechanics Modeling (LAMMM). Support
was provided by the European Research Council through the Advanced Grant “Predictive
Computational Metallurgy”, ERC Grant agreement No. 339081 – PreCoMet, and by EPFL
through the use of the facilities of its Scientific IT and Application Support Center.
Before introducing the subject matter, I would like to address a few words of thanks. First and
foremost, I would like to thank my thesis advisor, Prof. William A. Curtin, for his support. He
always encouraged me to focus on the important parts of physical problems and to keep the
larger context in mind. His guidance was very helpful for someone who tends to get bogged
down in details. Moreover, I am thankful for his kind and patient leadership style, and for the
opportunity to visit several interesting conferences and workshops abroad.
I also thank the members of the jury for agreeing to participate in my PhD exam. Furthermore,
I would like to thank the current and former members of LAMMM: Fabio Pavia, Kris Baker,
Mike Francis, Phil Moseley, Aitor Luque, Céline Varvenne, Wu Zhaoxuan, Varun Rajan,
Francesco Maresca, Binglun Yin, Nikolaos Bouklas, Till Junge, Satish Rao, Ben Szajewski,
Mostafa Khosrownejad, Ali Tehranchi, Max Hodapp, Predrag Andric, Shankha Nag, Rasool
Ahmad, Yi Hu, Eleanor Mak, Vladimir Dorodnitsyn and Florian Maurin; as well as a number of
visiting scholars: Ronald Miller, Derek Warner, Ryo Kobayashi, Maryam Ghazisaeidi, Jun Song
and Daniel Mulvihill. I enjoyed the friendly atmosphere in the lab and the lively discussions.
Moreover, I owe a debt of gratitude to the secretary of LAMMM, Géraldine Palaj, for her
assistance with numerous administrative issues.
I fondly recall the time when Céline, Kris, Aitor and others would supply LAMMM with cake
almost every week. I am also grateful to Kris for making me aware of the String method, and
to Céline and Aitor for their work on the average atom method. Both methods became
indispensable for my work on cross-slip. Furthermore, I am indebted to Aitor for his kind
support. I also want to thank Wu Zhaoxuan for many interesting discussions about, and
assistance with, atomistic modeling, simulation of dislocations, and technical problems. I
thank Satish, who has studied cross-slip extensively himself, for his advice and interesting
discussions. Ben introduced me to hiking in Switzerland, which I enjoy a lot. Mike was the
“master of ceremony” of many lively discussions. I thank Mostafa, who had entered the PhD
program before me, for helpful advice regarding my next steps. Moreover, I thank Ali for his
advice and our tremendous discussions. Never forget: you wont find the global minimum. I
thank Max, Predrag and Aitor for our many meetings. I wonder how many lemonade bottles
we emptied over the course of three years. I fondly recall several hikes and many inspiring
conversations with Varun and Max, who share my sense of humor. In this spirit, I wish Max
much success with the further development of dislocation jokes. Finally, I thank Francesco,
Till and Nikolaos for their sage advice. A special thanks goes to Till for translating the abstract

i
Acknowledgements

of this thesis into French.

I also thank my friends from the “Stammtisch”, in particular Simon Schütz, with whom I
departed on many recreational expeditions into the alps. Without him as planner and
cartographer, I would not have went on some hikes and been lost on others. Moreover, I want
to thank my friends from Germany for their support. I cannot name them all here, however, I
want to thank Simon for our Skype chats, Felix for many exhilarating messages, and Benni and
Kathrin for visiting me thrice. Furthermore, I thank Johannes Joel Möller for proofreading the
manuscript and basically for his friendship since we were “little” students. Last but not least, I
want to thank my family. Without their support this work would not have been possible.

Lausanne, December 5, 2017 W. G. N.

ii
Abstract
The mechanical strength of metals depends on their resistance against various microscopic
deformation processes. In ductile metals, the most important process is shearing of the crystal
lattice by dislocations. One of the fundamental aspects of dislocation motion is cross-slip of
screw dislocations, the process by which they change their glide plane. In Face-Centered
Cubic (FCC) metals, cross-slip is supposed to play a role in dislocation structuring, work
hardening, recovery, fatigue, etc. Most prior studies on cross-slip in FCC metals focused on
pure metals. There have been few studies of solute effects on cross-slip, which are important
for engineering alloys. Here, the effects of substitutional solutes are studied using atomistic
simulations and statistical modeling.
In the first part of the thesis, the mechanism and energy of cross-slip of short (40 Burgers
vectors long) dislocations in Ni-Al, Al-Mg and Cu-Ni alloys are determined using atomistic
calculations. These calculations are carried out with real random alloys and with “average”
alloys, where the real atom types are replaced by a single average type. By comparison, it
is shown that cross-slip is controlled by fluctuations in the solute concentration, i.e. the
activation energy for cross-slip is a distributed variable with a large variance around the mean
value. The latter changes only little with concentration. Most importantly, activation energies
that are significantly lower than the mean value can be observed in random alloys. A linear
correlation between the activation energy and the energy difference between the state of the
dislocation before and after cross-slip is observed. An analytical, parameter-free model of
this energy difference is developed, which takes random changes in solute-dislocation and
solute-solute binding energies into account. Thus, it is possible to predict the distribution of
activation energies for nucleation of cross-slip.
In the second part, cross-slip of long (102 –103 Burgers vectors) dislocations is studied using a
random walk model. Cross-slip is seen as a discrete process, where one Burgers vector long
subsegments of the dislocation cross-slip one after another. Associated with each step is a
random energy due to random changes in solute binding energies, as well as a deterministic
energy change due to constriction formation and stress effects. The random walk model
allows the calculation of the activation energy distribution for arbitrary dislocation lengths
and stresses. Cross-slip of long dislocations is unlikely at zero stress, due to increasing
frequency of high activation energies with increasing length. However, an external stress
eliminates these high barriers. Cross-slip then becomes a weakest-link problem. Like in the
case of short dislocations, activation energies that are significantly lower than average-alloy
estimates can be observed in real random alloys.

Keywords: screw dislocation, cross-slip, face-centered cubic crystals, solid solution, atomistic
simulation, transition path calculation, random walk, modeling

iii
Zusammenfassung
Die Festigkeit der Metalle ist abhängig von ihrem Widerstand gegen verschiedene
mikroskopische Verformungsmechanismen. Der wichtigste Mechanismus in duktilen
Metallen ist die Abscherung des Kristallgitters durch Versetzungen. Ein wichtiger Aspekt der
Versetzungsbewegung ist die Quergleitung von Schraubenversetzungen, also die Änderung
ihrer Gleitebene. Im Falle der kubisch-flächenzentrierten (kfz-) Metalle nimmt man an,
dass die Quergleitung bei der Bildung von Versetzungsstrukturen, der Kaltverfestigung,
der Ermüdung, etc. eine Rolle spielt. Bei früheren Untersuchungen von Quergleitung in
kfz-Metallen standen zumeist die reinen Metalle im Mittelpunkt. Es liegen nur wenige
Studien zum Einfluss von Fremdatomen auf die Quergleitung vor. Dieser spielt in
technischen Legierungen eine wichtige Rolle. In der vorliegenden Arbeit wird der Einfluss
von substitutionellen Fremdatomen mittels atomistischer Simulationen und statistischer
Modellierung untersucht.
Im ersten Teil der Arbeit werden der Mechanismus und die Energie der Quergleitung kurzer
(40 Burgersvektoren langer) Versetzungen in Ni-Al, Al-Mg und Cu-Ni mit atomistischen
Simulationen bestimmt. Diese Berechnungen werden mit echten und “gemittelten”
Mischkristallen durchgeführt. Bei letzteren werden die verschiedenen echten Atomtypen
durch einen einzigen durchschnittlichen Typ ersetzt. Im Vergleich zeigt sich, dass die
Quergleitung stark von Schwankungen in der Konzentration der Fremdatome abhängt, das
heißt die Aktivierungsenergie der Quergleitung ist eine Zufallsvariable mit erheblicher
Streuung um den Mittelwert. Letzterer ändert sich nur geringfügig mit der Konzentration.
Am wichtigsten ist jedoch, dass in den echten Mischkristallen Aktivierungsenergien
vorkommen, die erheblich unterhalb des Mittelwertes liegen. Es wird festgestellt, dass die
Aktivierungsenergie linear mit der Energiedifferenz zwischen Anfangs- und Endzustand der
Versetzung korreliert. Ein analytisches, parameterfreies Modell dieser Energiedifferenz wird
vorgestellt. Erfasst werden Änderungen der Bindungsenergien zwischen Fremdatomen und
Versetzungen, sowie der Bindungsenergie von Fremdatom-Paaren. Somit ist es möglich, die
Verteilung der Aktivierungsenergie für die Nukleation der Quergleitung vorherzusagen.
Im zweiten Teil der Arbeit wird die Quergleitung langer (102 –103 Burgersvektoren)
Versetzungen mit einem Random-Walk-Modell untersucht. Die Quergleitung wird als
diskreter Prozess aufgefasst, bei dem einen Burgersvektor lange Subsegmente der
Versetzung nacheinander quergleiten. Verbunden mit jedem Schritt ist sowohl eine zufällige
Energieänderung, welche zufälligen Änderungen der Bindungsenergien geschuldet ist als auch
eine deterministische Energieänderung, aufgrund von Spannungseffekten und der Bildung
einer Einschnürung. Das Random-Walk-Modell erlaubt die Berechnung der Verteilung
der Aktivierungsenergie für beliebige Versetzungslängen und äußere Spannungen. Im
spannungsfreien Zustand ist die Quergleitung langer Versetzungen wegen der zunehmenden
Häufigkeit hoher Aktivierungsenergien unwahrscheinlich. Äußere Spannungen eliminieren

v
Acknowledgements

diese Barrieren jedoch. Die Quergleitung wird dann zu einem Problem des schwächsten Glieds.
Wie im Falle der kurzen Versetzungen werden in echten Mischkristallen Aktivierungsenergien
beobachtet, die weit unterhalb des mit einem gemittelten Mischkristall bestimmten Wertes
liegen.

Stichwörter: Schraubenversetzung, Quergleitung, kubisch-flächenzentrierter Kristall,

Mischkristall, atomistische Simulation, Reaktionspfad-Berechnung, Zufallsbewegung,
Modellierung

vi
Résumé
La résistance mécanique de métaux dépends de leur résistance à différents processus de
déformation microscopiques. Dans le cas des métaux ductiles, le processus le plus important
est le cisaillement du réseau cristallin par des dislocations. Le glissement dévié forme un
aspect fondamental du mouvement de dislocations. C’est le mécanisme par lequel elles
changent de plan de glissement. Le glissement dévié est censé jouer un rôle dans la
structuration des réseaux de dislocations, l’écrouissage, le rétablissement, la fatigue, e.t.c.,
pour les métaux à réseau cubique à faces centrées (CFC). La plupart des études existantes du
glissement dévié en métaux CFC se concentre sur les métaux pures. Seulement quelques peu
de travaux ont étudié les effets de solutés sur le glissement dévié, effets qui sont importants
pour les alliages utilisés en ingénierie. Ici, l’effet des solutés substitutionnels est étudié en
utilisant des simulations atomiques et la modélisation statistique.
La première partie de cette thèse détermine, en utilisant des calculs atomiques, le mécanisme
et l’énergie du glissement dévié de dislocations courtes (40 vecteurs Burgers de longs) dans
des alliages Ni-Al, Al-Mg et Cu-Ni. Ces calculs sont exécutés avec des vrais alliages aléatoires et
des alliages "moyens", dans lesquels les vrais types d’atomes ont été remplacés par un seul
type moyen. En comparant les résultats, on peut montrer que le glissement dévié est contrôlé
par des fluctuations de la concentration de solutés, c’est-à-dire que l’énergie d’activation du
glissement dévié est une variable distribuée qui a une forte variance autour de sa valeur
moyenne. Cette moyenne ne dépend que faiblement de la concentration. Surtout, on peut
observer des énergies d’activation qui sont significativement plus basses que la moyenne dans
les alliages aléatoires. On observe une corrélation linéaire entre l’énergie d’activation et la
différence d’énergie entre l’état d’une dislocation avant et après un glissement dévié. Un
modèle analytique et sans paramètres de cette différence d’énergie est développé. Il tient
compte des variations aléatoires des énergies de liaisons dislocation-soluté et soluté-soluté.
Ainsi, il est possible de prédire la distribution des énergies d’activation pour la nucléation de
glissement dévié.
La deuxième partie étudie le glissement dévié de longues dislocations (de 100 à 1000
vecteurs Burgers de long) en utilisant un modèle de marche aléatoire. Le glissement est
interprété comme un processus discret, où des sous-segments de dislocation d’une longueur
d’un vecteur Burgers sont déviés l’un après l’autre. Chaque pas est associé à une énergie
aléatoire due aux variations aléatoires d’énergie de liaison de solutés, ainsi qu’une variation
déterministe d’énergie due à la formation d’une constriction et des effets de contrainte.
Le modèle de marche aléatoire permet de calculer la distribution d’énergie d’activation
pour de longueurs de dislocations et états de contrainte arbitraires. Le glissement dévié est
peu probable à contrainte nulle à cause de la fréquence élevée dés énergies d’activations
hautes pour les dislocations longues. Par contre, une contrainte appliquée élimine ces hautes
barrières. Le glissement dévié devient un problème de maillon faible. Comme dans le cas des

vii
Acknowledgements

dislocations courtes, on peut observer des énergies d’activation significativement plus basses
que dans les alliages moyens dans les vrais alliages aléatoires.

Mots clés : dislocation vis, glissement dévié, cristaux à réseau cubique à faces centrées,
solution solide, calcul atomique, calcul te chemin de transition, marche aléatoire,
modélisation

viii
Contents
Acknowledgements i

Abstract (English/Français/Deutsch) iii

List of Figures xi

List of Tables xiii

List of Symbols xv

Introduction 1

1 Theoretical Background 5
1.1 Dislocations in Face-Centered Cubic Crystals . . . . . . . . . . . . . . . . . . 5
1.2 Cross-Slip in Face-Centered Cubic Crystals . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Friedel-Escaig Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 Continuum Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.3 Atomistic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.4 Other Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.5 Cross-Slip at Heterogeneities . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.6 Alloying Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Transition State Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2 Methods 17
2.1 Atomistic Calculations with Classical Potentials . . . . . . . . . . . . . . . . . 17
2.1.1 Embedded Atom Method . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.2 Average Atom Approximation . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Selected Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Calculating the Transition Path of Cross-Slip . . . . . . . . . . . . . . . . . . . 20
2.3.1 Initial and Final States . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 String Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.3 Initial Guess for the Path . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.4 Application of an Escaig Stress . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Solute Binding Energies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3 Cross-Slip of Short Dislocations 27

3.1 Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

ix
Contents

3.2.1 Average Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2.2 True Random Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Preliminary Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Analytic Model for the Statistical Distribution of ΔE act . . . . . . . . . . . . . . 35
3.4.1 General model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.2 Solute-Dislocation Contribution . . . . . . . . . . . . . . . . . . . . . . 35
3.4.3 Solute-Solute Contribution . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.4 Results and Comparison to Simulations . . . . . . . . . . . . . . . . . . 40
3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4 Cross-Slip of Long Dislocations 45

4.1 Random Energy Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Random Walk Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2.1 Deterministic Energy Change . . . . . . . . . . . . . . . . . . . . . . . 50
4.2.2 Stress Effects in the Random Walk Model . . . . . . . . . . . . . . . . . 51
4.3 Validation Against Atomistic Simulations . . . . . . . . . . . . . . . . . . . . . 54
4.3.1 Zero Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3.2 Non-zero Escaig Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.4 Cross-Slip of Long Dislocations . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4.1 Zero Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4.2 Non-zero Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4.3 Weakest-Link Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4.4 Predictions Based on Zero Stress Data . . . . . . . . . . . . . . . . . . . 62
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

5 Summary 71

6 Outlook 73

A Appendix 77
A.1 Calculation of Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
A.1.1 Average Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
A.1.2 Random Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
A.2 Nudged Elastic Band Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
A.3 Convergence in String Method Calculations . . . . . . . . . . . . . . . . . . . . 81
A.4 System Size Dependence of ΔE act in Atomistic Calculations . . . . . . . . . . . 86
A.5 Partially Cross-Slipped Dislocations . . . . . . . . . . . . . . . . . . . . . . . . 89
A.6 Extended Solute Pair Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
A.6.1 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
A.7 E det under Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
A.8 Correlation between ΔE act and ΔE end . . . . . . . . . . . . . . . . . . . . . . . 99

Bibliography 111

Acronyms 113

Curriculum Vitae 115

x
List of Figures
1 Introduction to dislocations and cross-slip . . . . . . . . . . . . . . . . . . . . 2

1.1 Dislocations in FCC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.2 Friedel-Escaig cross-slip mechanism . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Schöck-Seeger-Wolf and Fleischer cross-slip mechanisms . . . . . . . . . . . . 11
1.4 Model potential energy surface . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.1 Setup for transition path calculations . . . . . . . . . . . . . . . . . . . . . . . 20

2.2 Transition path calculation on a 2D model landscape . . . . . . . . . . . . . . 23
2.3 Fixed atoms in dislocation core . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.1 Typical states during cross-slip in average Ni-15 at.% Al . . . . . . . . . . . . . 29

3.2 Cross-slip transition paths in average alloys . . . . . . . . . . . . . . . . . . . . 29
3.3 Typical cross-slip transition states in random alloys . . . . . . . . . . . . . . . 30
3.4 Transition paths in Ni+15 at.% Al random alloys . . . . . . . . . . . . . . . . . 31
3.5 Activation energies ΔE act of cross-slip in random and average alloys . . . . . . 31
3.6 Cross-slip activation energy ΔE act vs. end state energy difference ΔE end . . . . . 34
3.7 Reference frames for measuring relative solute/dislocation positions . . . . . . 36
3.8 Cutting of first-nearest neighbor solute pairs . . . . . . . . . . . . . . . . . . . 39
3.9 Observed and predicted standard deviation of ΔE end . . . . . . . . . . . . . . . 41

4.1 Correlation between pair loss/formation . . . . . . . . . . . . . . . . . . . . . 47

4.2 Observed and predicted standard deviation od ΔE end (extended solute pair model) 48
4.3 Dislocation partitioned into 1b long segments . . . . . . . . . . . . . . . . . . 49
4.4 Step selection in the random walk . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.5 Energy as a function of the length of the cross-slipped segment . . . . . . . . . 52
4.6 Distributions of ΔE act for 40b long dislocations (random walk & atomistics) . . 55
4.7 Frequency of the location of the maximum as a function of step . . . . . . . . 56
4.8 Correlation of ΔE act and ΔE end in random walk calculations . . . . . . . . . . . 57
4.9 Distributions of ΔE act for different random alloys and Nb = 40. . .1000 . . . . . . 59
4.10 ΔE act in random Cu+33 at.% Ni for Nb = 40, 200, 1000 . . . . . . . . . . . . . . 59
4.11 ΔE act for a 500b long dislocation in Cu+33 at.% Ni at stresses 0–200 MPa . . . . 60
4.12 ΔE act for a 500b long dislocation in random alloys at (0, 10, 200) MPa stress . . . 61
4.13 Weakest-link scaling under Schmid stress . . . . . . . . . . . . . . . . . . . . . 63
4.14 Weakest-link scaling under Escaig stress . . . . . . . . . . . . . . . . . . . . . . 64
4.15 Weakest-link scaling: predictions based on zero stress data (Cu+33 at.% Ni) . . 65
4.16 Weakest-link scaling: predictions based on zero stress data (Al+06 at.% Mg) . . 66
4.17 Weakest-link scaling: predictions based on zero stress data (Ni+15 at.% Al) . . . 66

xi
List of Figures

A.1 Relative difference between average and random alloy properties . . . . . . . . 79

A.2 Relative change of ΔE act per iteration . . . . . . . . . . . . . . . . . . . . . . . 83
A.3 Perpendicular force on transition path in String method calculations . . . . . . 84
A.4 Relative energy and perpendicular force on transition path . . . . . . . . . . . 84
A.5 Difference between String and NEB estimates for ΔE act . . . . . . . . . . . . . 85
A.6 Extreme cases of differences between NEB and String paths . . . . . . . . . . . 85
A.7 Cross-slip transition path vs. radius of the configuration . . . . . . . . . . . . . 87
A.8 Cross-slip transition path vs. length of the configuration . . . . . . . . . . . . . 87
A.9 Cross-slip transition path vs. radius at 300 MPa Escaig stress . . . . . . . . . . . 88
A.10 Partially cross-slipped dislocation . . . . . . . . . . . . . . . . . . . . . . . . . 89
A.11 Unit cell used in the derivation of the extended solute pair model . . . . . . . . 91
A.12 Energy during cross-slip at 150 MPa Escaig stress as a function of step . . . . . 98
A.13 ΔE act,avg as a function of Escaig stress . . . . . . . . . . . . . . . . . . . . . . . 99
A.14 ΔE act of pure Ni as a function of Schmid stress . . . . . . . . . . . . . . . . . . 99
A.15 ΔE act -ΔE end -correlation in random walk and atomistics (additional data) . . . 100

xii
List of Tables
1.1 Zero-stress cross-slip energy according to Püschl and Schöck . . . . . . . . . . 10
1.2 Reduction of cross-slip activation energy at heterogeneities . . . . . . . . . . . 13

2.1 FIRE parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.1 Neighbor distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.2 Average alloy activation energy and standard deviation of ΔE end per step . . . . 54

A.1 Basic properties of Ni-Al average and random alloys . . . . . . . . . . . . . . . 77

A.2 Basic properties of Cu-Ni average and random alloys . . . . . . . . . . . . . . . 78
A.3 Basic properties of Al-Mg average and random alloys . . . . . . . . . . . . . . . 78
A.4 Equilibrium segment length ζ ∗ of partially cross-slipped dislocations . . . . . . 90

xiii
List of Symbols
|...| Euclidean norm of . . .

γSF Stable stacking fault energy

γeff,g Effective stacking fault energy on the glide plane

γeff,c Effective stacking fault energy on the cross-slip plane

ΔE act Activation energy

ΔE act,avg Activation energy of cross-slip in an average alloy

ΔE end Energy difference between a dislocation lying on the cross-slip and glide plane

ΔE end,s-s Contribution to ΔE end from differences in solute-solute binding energies

ΔE end,s-s,XY Contribution to ΔE end from cutting/forming of X -Y solute pairs

ΔE end,s-d Contribution to ΔE end from differences in solute-dislocation binding energies

ΔE end,s-d,X Contribution to ΔE end,s-d from X -solutes

Δt Timestep

ΔVi Relative energy of image i in a transition path calculation

ξ Dislocation line direction

τglide
Esc
Escaig stress on the glide plane

τcross
Esc
Escaig stress on the cross-slip plane

τglide
Sch
Schmid stress on the glide plane

τcross
Sch
Schmid stress on the cross-slip plane

ζcsn Characteristic length of cross-slip nucleation

a Lattice parameter

b Magnitude of the Burgers vector of a 1/2a 110 dislocation

bp Magnitude of the Burgers vector of a 1/6a 112 Shockley partial dislocation

xv
List of Tables

d Equilibrium dislocation dissociation width

dg Dissociation width on the glide plane

dc s Dissociation width on the cross-slip plane

Ec Energy of a single constriction

E det Deterministic energy in random walk

Esc
E det Contribution to E det from Escaig stresses
Sch
E det Contribution to E det from a Schmid stress on the cross-slip plane

F N -dimensional vector of atomic forces

Fi⊥ Perpendicular force on image i

kB Boltzmann constant

N Number of atoms

Nb Dislocation length in Burgers vectors

Nc Constriction length in Burgers vectors

Nd Width of stacking fault in number of unit cells

NT Number of atom types

NS Number of images on the transition path

Nζ Length of stacking fault in number of unit cells

R 3N -dimensional vector of atomic positions

sˆ Normalized reaction coordinate

Us-s,XY Vector of solute-solute binding energies

Un ,XY Binding energy of an n-th order pair of X and Y atoms

xvi
Introduction
Alloys are the most important class of materials for load bearing structural components in
machines, vehicles, aircraft, etc. The purpose of alloying is to achieve superior mechanical
properties relative to those of the corresponding pure metal. One of the underlying principles
is that alloying elements influence the microscopic deformation mechanisms that govern
the macroscopic mechanical behavior. A thorough understanding of this influence is
indispensable for designing new, improved alloys.

In ductile metals, movement of dislocations is the most important atomic-scale deformation

mechanism. Dislocations are line-like lattice defects, whose movement shears off one part of
the crystal with respect to the other. Due to distortion of the lattice, dislocations create a stress
field, through which they interact with solute atoms. These interactions and their effects on
dislocation motion must be understood in order to make alloy strength predictable.

The principle of dislocations is that the crystal is not sheared off at once, by a relative
displacement of the two parts over the whole dividing surface. Such a displacement would
require very high applied stresses [41]. Instead, the shear is localized. Only a small portion of
the crystal is sheared off initially. The line separating sheared and not yet sheared parts of the
crystal is the dislocation line, see Fig. 1a) and b). The dislocation is therefore a line-like
lattice defect. As it moves through the lattice, a greater portion of the crystal is sheared off.
Dislocations are characterized by the line direction ξ and the Burgers vector b, which indicates
the relative displacement of the two parts of the crystal. Based on the relative orientation
between ξ and b, two principal types of dislocations can be distinguished, with different
constraints concerning their movement, see Fig. 1a). Edge dislocations with ξ · b = 0 can
move conservatively, i.e. without addition or subtraction of mass, only on the lattice plane that
is spanned by ξ and the b. Screw dislocations with ξ  b, on the other hand, can move on any
lattice plane that contains b. The process by which dislocations change between planes is
referred to as cross-slip.

Cross-slip is important in several ways. A new dislocation source can be generated by double
cross-slip [84]. On the other hand, cross-slip can reduce the total dislocation length because it
enables mutual annihilation of pairs of opposite dislocations on nearby planes [63]. These two
processes are illustrated in Fig. 1b). Furthermore, cross-slip enables dislocations to overcome
obstacles, such as precipitates [51, 128]. Recently, the importance of cross-slip for dislocation
network formation has been demonstrated using computer simulations [54, 159].

This thesis deals with cross-slip in substitutional alloys of Face-Centered Cubic (FCC) metals.

1
List of Tables

Figure 1: a) Differences between edge and screw dislocations; the line direction ξ and the
Burgers vector b of an edge dislocation are perpendicular and it can move only on the lattice
plane spanned by these vectors (red); screw dislocations with ξ × b = 0 can move on any plane
that contains b; cross-slip is the process by which they change between planes; b) edge and
screw dislocations in a simple cubic lattice; c) cross-slip as a dislocation source or sink
mechanism; left: screw part (red) of a Frank-Read source [40] cross-slips twice and generates a
source on an adjacent plane; right: screw parts (red) of two dislocation loops on adjacent
planes cross-slip towards each other and mutually annihilate; d) in FCC metals, cross-slip is
complicated by the splitting of perfect dislocations into Shockley partial dislocations, which
have mixed edge/screw character and create a stacking fault

In FCC metals, cross-slip has been invoked to explain observed dislocation network structures,
like sheets [55] or persistent slip bands [34], which are formed during cyclical deformation.
Mutual annihilation of screw dislocations by cross-slip may explain why the microstructure in
the first stage of FCC single crystal deformation consists mainly of edge dislocation dipoles [5,
33, 132]. Furthermore, mutual annihilation by cross-slip has been incorporated into models of
work hardening and dynamic recovery [61, 62, 87] and creep at intermediate homologous
temperatures [24, 42, 76, 97]. A more controversial issue is whether cross-slip can also explain
the transition between the second and third stage of deformation, from linear to parabolic
hardening, see e.g. Refs. [3, 32, 66, 125] and Ref. [14, p. 154]. Finally, cross-slip has also been

2
 List of Tables

invoked to explain the copper-brass texture transition [22, 68, 69, 129].

FCC cross-slip is a thermally activated process with an energy barrier. The cause for this energy
barrier is the splitting of complete dislocations, whose Burgers vector is a translation vector of
the lattice, into pairs of Shockley partial dislocations, see Fig. 1c). The partial dislocations have
mixed character and therefore cannot readily cross-slip, but must merge again or undergo
some other transformation first. Additionally, a stacking fault, i.e. two atomic planes with
Hexagonal Close-Packed (HCP) coordination, is created between the partial dislocations.

Different cross-slip mechanism have been discussed in the literature [39, 43, 125] and the
required energy or stress has been calculated using elastic or atomistic models [27, 31, 32, 78,
99, 101, 102, 112, 117, 124, 153, 154], see also Ref. [100] or [14, Ch. 5]. The most frequently
discussed mechanism for cross-slip in FCC is the Friedel-Escaig [32, 43] (F-E) mechanism,
where cross-slip starts with formation of a point constriction. More recently, cross-slip at
heterogeneities, such as jogs or forest dislocations, has been subject of research [103, 104,
106–109, 142]. However, most studies to date have considered cross-slip in pure metals. There
are only few dedicated studies of alloying effects on cross-slip. Previous studies considered
specific alloys [25] and solute arrangements [151, 152]. In other models, solute effects were
reduced to a simple friction term acting on the dislocation [53], or it was assumed that solutes
are fully segregated to the dislocation, which is unlikely during deformation at moderate
temperatures. In any case, prior studies considered only solute effects on cross-slip nucleation
in short (~101 Burgers vectors) dislocations segments. To the best of our knowledge, cross-slip
of long (102 –103 Burgers vectors) dislocations has not been studied before.

This thesis is motivated by the lack of comprehensive studies of solute effects on dislocation
cross-slip in FCC alloys. The following important questions are addressed in this work.

• What is the effect of substitutional solute atoms on the cross-slip activation energy?
• How does the activation energy change with solute concentration?
• How to model solute effects on cross-slip analytically?
• How to predict the activation energy as a function of matrix and solute properties, and
concentration?
• How does cross-slip of long dislocations differ from cross-slip of shorter ones? How does
the barrier scale with nucleus size?
• How to predict the activation energy distribution for cross-slip of long dislocations?

To answer these questions, the activation energy of cross-slip was calculated for various solid
solutions using atomistic methods. Guided by these calculations, which cover various possible
solute effects, we developed statistical models of the distribution of activation energies
for cross-slip nucleation in short (40b ) dislocations, and full cross-slip of long (102 –103b )
dislocations. We restrict our attention to the important case of random distributions of
substitutional solutes, because in many alloys of practical interest, a fraction of the alloying
elements remains in solution. A random distribution of solute atoms is a fair assumption
when there is no strong preference for short range ordering, and at intermediate temperatures,
where segregation is slow. Finally, this work is also relevant for High Entropy Alloys [162],
which are highly-concentrated, multi-component random solid solutions.

3
List of Tables

This thesis is structured as follows. Ch. 1 gives a more detailed introduction into the relevant
topics. In addition, the existing literature on cross-slip in FCC is reviewed. The atomistic
simulation methods used to study cross-slip are explained in Ch. 2. Ch. 3 deals with nucleation
of cross-slip in short (40b ) dislocation segments. The activation energy for cross-slip in
various Ni-Al, Al-Mg and Cu-Ni alloys is determined using atomistic simulations and a model
of this distribution is proposed. It is shown that cross-slip is controlled by fluctuations in the
solute concentration and that the activation energy in random alloys can be much lower than
expected based on models of average alloying effects. Cross-slip of long dislocations is studied
in Ch. 4. A random walk model is developed, which allows to predict the cross-slip activation
energy distribution for dislocations of arbitrary length. The results are summarized in Ch. 5.
Finally, open questions and directions for future research are discussed in Ch. 6.

4
1 Theoretical Background

Disclaimer

Parts of this chapter are adapted from the following article, with permission from the
coauthor and the publisher.
W. G. Nöhring and W. A. Curtin. “Dislocation Cross-Slip in FCC Solid Solution
Alloys”. In: Acta Materialia 128 (Apr. 2017), pp. 135–148. ISSN: 13596454. DOI:
10.1016/j.actamat.2017.02.027

This chapter gives an overview of the theoretical background for the current work. In the first
section, important properties of dislocations in FCC crystals are reviewed. For more details
about FCC dislocations and dislocation theory in general, the reader is referred to Ref. [1].
Sec. 1.2 deals with cross-slip in FCC. The Friedel-Escaig (F-E) mechanism is explained and
other proposed mechanism are reviewed. Furthermore, we review approaches to calculate the
required energy or stress, as well as the literature on cross-slip at heterogeneities and alloying
effects. Finally, Sec. 1.3 introduces some elements of transition state theory which are relevant
for this work.

1.1 Dislocations in Face-Centered Cubic Crystals

Dislocation theory predicts that slip preferentially takes place on close packed planes and
along close-packed directions. In the face-centered cubic (FCC) lattice, these are the families
of {111} planes and 110 directions, respectively. Indeed, the twelve possible pairings
among these families are the most important slip systems in FCC. The corresponding perfect
dislocations have Burgers vector a /2 110 .

However, perfect dislocations are typically unstable. It is energetically favorable for them to
split up into pairs of Shockley [46] partial dislocations with Burgers vector of type a /6 112 . A
simple argument to this effect is Frank’s rule [85, see Frank’s comment]: Since the energy of a
dislocation depends on the square of the magnitude of its Burgers vector, it will tend to split up
if this value is greater than the sum of the magnitudes of the resulting partial Burgers vectors.
√
The magnitude b of a a /2 110 dislocations is a/ 2, whereas the magnitude b p of the Burgers

5
Chapter 1. Theoretical Background

√
vector of a Schockley partial dislocation is a/ 6. Since b 2 > 2b p 2 , a /2 110 dislocations are
expected to dissociate. Perfect and partial dislocations in FCC are sketched in Fig. 1.1.

Figure 1.1: Dissociation of perfect dislocations into Shockley partial dislocations in FCC;
left: perfect and partial dislocations in a cylindrical body; red: plane across which the
displacement is discontinuous (cut plane); green: stacking fault; right: displacements of atoms
in the two lattice planes above (dark gray) and below (light gray) the red plane in the left figure;
arrows indicate the displacement with respect to the dislocation-free reference configuration;
a stacking fault (green) is created between the two partial dislocations

Each a /2 110 dislocation can split up into two different pairs of a /6 112 partial dislocations,
which correspond to different {111} slip planes. For example, a /2[110] can split up into
a /6[121] and a /6[211] on the (111) plane, or into a /6[211] and a /6[121] on the (111) plane.
The Burgers vector of a Shockley partial dislocation is not a full translation vector of the FCC
lattice. As a consequence, the stacking sequence of {111} lattice planes, which is A-B-C-A-B-C
in FCC, changes to A-B-C-B-C-A in the region between the two partial dislocations. Note that
B-C-B-C is the {111} stacking sequence of the hexagonal close-packed HCP lattice. Thus,
there are two planes of HCP coordinated atoms, an intrinsic stacking fault. Associated with
this fault is a surface tension (energy per area) γSF , which is typically on the order of several
tens to hundreds of mJ m−2 in FCC metals.

γSF is one of the factors that control the equilibrium separation d between the partial
dislocation cores. Two other factors are the interaction between the elastic fields of the
dislocations, and, if applicable, their interaction with an external stress field. For analysis of
the fields, it is useful to consider the edge and screw components of the partial dislocation
separately. Within the framework of isotropic elasticity, such separation is always possible
because there is no interaction between pure edge and screw dislocations. Within anisotropic
elasticity, separation is possible in special cases, see p. 360 in [1] and references therein.
Among them is the present case of interest, where the perfect dislocation is a pure screw

6
 1.2. Cross-Slip in Face-Centered Cubic Crystals

dislocation. The Burgers vectors of the corresponding partial dislocations have the same screw
components, but oppositely aligned edge components. The former repel each other, whereas
the latter attract each other. An external stress that interacts with the screw components will
cause a bow-out. This kind of stress is referred to as Schmid1 stress. An external stress that
interacts with the edge components, on the other hand, will change the splitting width and is
referred to as Escaig2 stress.

The equilibrium splitting width d can be obtained by requiring that the forces resulting from
the stacking fault and the internal and external fields are balanced. Within the framework of
isotropic3 elasticity, the solution for an originally pure screw dislocation is [12, 114]
 
1 μb 2 1
d =  √  1− , (1.1)
γSF + τe b / 2/ 3 8π 3 (1 − ν )

where μ is the shear modulus and ν is Poisson’s ratio. Within anisotropic elasticity, the solution
is [1, p. 361-362]
 
1 1 b2
d =  √  K s − Ke , (1.2)
γSF + τe b / 2/ 3 8π 3

where
    1/2
K s = C 44C 55 − C 16 ,
     1/2
 C¯  − C 
1 C     C 55 11 12
(1.3)
Ke = 2 + 22

C¯11 + C 12 
 C¯  + C  + 2C 
 ,
3 C¯11 C 22 11 12 55

are the energy prefactors of the edge and screw components of an a /6 112 partial dislocation,
respectively, and the primed coefficients are functions of the cubic elastic constants C 11 , C 12 ,
and C 44 ,
  
C 11 = C 11 C 12 = C 12 C 55 = C 44
 1  1  1
C 22 = C 11 + H C 23 = C 12 − H C 44 = C 44 − H (1.4)
2 2 2

  C
C¯11 = C 11 22 H = 2C 44 + C 12 − C 11 .

1.2 Cross-Slip in Face-Centered Cubic Crystals

Dissociation of perfect dislocations into Shockley partial dislocations complicates cross-slip.
The two partials lie in only one of the four preferred {111} planes and have mixed screw-edge
character. Thus, cross-slip requires recombination of the partial dislocations or some other
core transformation. Over the years, several cross-slip mechanisms and methods for
calculating the required energy have been proposed, see Püschl’s review [100] and chapter 5 in
Caillard and Martin’s book [14]. The present study focuses on a mechanism proposed by
1. Named after Erich Schmid (1896–1983)
2. Named after Bertrand Escaig

7
Chapter 1. Theoretical Background

Friedel [43]. The associated activation energy was first calculated by Escaig [31, 32], using a
line tension model. Therefore, this mechanism is referred to as the Friedel-Escaig (F-E)
mechanism. Atomistic calculations of the cross-slip transition path in Cu [112, 143], Ni [143],
and Al [57] indicate that the F-E mechanism is the operative mechanism for cross-slip if the
dislocation line is sufficiently long.

Cross-slip can nucleate at heterogeneities like jogs or forest dislocations. Such

“heterogeneous” cross-slip has received considerable attention in recent years, because the
associated activation energy may be significantly lower than that of homogeneously nucleated
cross-slip. Here, we are concerned with only one type of heterogeneity, namely solute atoms.
Solute effects on cross-slip involving additional heterogeneities will be left for further study.

The rest of this section is structured as follows. The F-E mechanism is explained in the next
subsection. In Sec. 1.2.2 and Sec. 1.2.3, continuum and atomistic models for calculation of
the cross-slip activation energy are reviewed. Other mechanisms for homogeneous and
heterogeneous cross-slip are discussed briefly in Sec. 1.2.4 and Sec. 1.2.5. The section
concludes with a review of the literature about alloying effects on FCC cross-slip.

1.2.1 Friedel-Escaig Mechanism

Fig. 1.2 shows the Friedel-Escaig cross-slip mechanism. The perfect screw dislocation is
initially dissociated into a pair of Shockley partial dislocations on the (111) glide plane (Fig.
1.2a)). The first step of cross-slip is the formation of a point constriction (Fig. 1.2b)). Starting
from there, the dislocation re-dissociates into a pair of Shockley partial dislocations on the
(111) cross-slip plane (Fig. 1.2c)). Since the dislocation line must be closed, the left and right
end of the cross-slipped segment are constricted, hence there are now two point constrictions
in total. Subsequently, these constrictions move laterally and the cross-slipped segment grows.

Figure 1.2: Friedel-Escaig cross-slip mechanism

8
 1.2. Cross-Slip in Face-Centered Cubic Crystals

Note that the partial dislocations are close to edge orientation near the left constriction, and
close to screw orientation near the right one. Atomistic models show that the screw-like
constriction has a negative energy relative to the fully dissociated state, thus it could form
spontaneously at a free surface [102, 110].

1.2.2 Continuum Models

Cross-slip costs energy. Initially, energy must be spent to change the splitting width. Some of it
is regained later, when the dislocation re-dissociates on the cross-slip plane. Other energy
changes are caused by dislocation curvature. First of all, the total line length increases, and
therefore the total energy as well. Furthermore, the local character of the dislocation changes,
which affects the energy per length, the interaction energy between the partial dislocation,
and their self-interaction energy. The sum of these energy changes is positive, thus an energy
barrier must be overcome to cross-slip. The required energy must be supplied by an external
stress or thermal fluctuations. Stroh [133] calculated the energy of a point constriction using
an isotropic elastic line tension model. Similarly, Escaig [31, 32] used a line tension model to
calculate the activation energy of cross-slip. The derivation of this model can be found
elsewhere [14, 31, 32, 60]. Here, only some important assumptions are summarized.

For the purpose of calculating the interaction energy between infinitesimal segments with the
same z -coordinate (see Fig. 1.2), it is assumed that both segments are straight. The interaction
energy between segments with different z -coordinate is neglected, as is self-interaction. The
only way in which curvature enters the calculation is through a line tension term, which
penalizes the increase in length. Furthermore, three ill-defined parameters appear in the
calculation: an inner and an outer core cutoff radius, and a minimum splitting width. The
latter is the distance below which the two partial dislocations are considered to have
combined into a perfect screw dislocation.

These approximations make the F-E model quantitatively inaccurate. However, it provides
some qualitative information. At zero applied stress, the barrier energy tends to two times the
constriction energy. According to Stroh4 , the latter scales as
   
μb 2 1 R
Ec ∝ d 1− log , (1.5)
8π 3 (1 − ν ) r0

where r 0 and R are the inner and outer core cutoff radius, respectively. For a discussion of
stress effects, Escaig and Schmid components must be considered separately. An Escaig stress
on the glide plane has a stronger effect on the barrier than an Escaig stress on the cross-slip
plane. A Schmid stress on the cross-slip plane is even less effective. A Schmid stress on the
glide plane can be ignored, since it causes only normal slip. These differences in effectiveness
of the stress components have recently been confirmed by direct atomistic calculations of
cross-slip under stress [60].

Using Escaig’s low stress approximation [32], Bonneville et al. [10] calculated a barrier energy
of 1.6 ± 0.2 eV for Cu. Saada [117] obtained (single) constriction energies of 0.7 eV, 0.55 eV and
0.32 eV in Cu for minimum dissociation widths of 2b , 1b and b /2. For other metals, a similarly
4. The dependence on Poisson’s ratio, which was neglected by Stroh, has been included here.

9
Chapter 1. Theoretical Background

strong dependence on this parameter was observed. Thus, Saada argues that the F-E model
permits only order of magnitude estimates.

Several groups developed more elaborate models, which avoid some of the aforementioned
approximations. In the model by Duesbery et al. [27], dislocation self-interactions are
calculated explicitly using isotropic elasticity, thus eliminating the line tension approximation.
For this purpose, the dislocation lines are discretized into piecewise linear curves. The
estimated barrier energy of Cu is5 3.7 eV. Püschl and Schöck [99] also described the
dislocations as piecewise linear curves, but calculated segment interactions using anisotropic
elasticity. In addition, the authors physically motivated their choice of minimum splitting
width. It was chosen such that the energy required to reach this width equals the dissociation
energy calculated with a Peierls-Nabarro [86, 95] model of the slip distribution in the
dislocation core. According to this model, the constriction energy, which is approximately half
the activation energy of cross-slip, follows the trend [100]
 
Ec −2 d d
≈ 1.2 × 10 log 2 . (1.6)
Ks b 3 b b

Using elastic constants from Ref. [1], and the experimentally measured splitting widths listed
in Saada’s publication [117], see Refs. [16, 17, 56, 82, 123], one obtains the barrier energies
listed in Tab. 1.1. Finally, the arguably most elaborate continuum model to date was proposed
by Ramírez et al. [101]. Here, the dislocation lines are approximated as cubic splines, thus
avoiding discontinuities at the nodes where segments meet. The screw dislocation is actually
modeled as a group of fractional dislocations, whose interactions are computed using Cai’s
[13] isotropic nonsingular solution for the elastic fields. Like in a Peierls-Nabarro model, the
fractional dislocations are subject to a lattice restoring force, which in this case is obtained
from ab-initio calculations. For Cu, Ramírez et al. obtain a barrier energy of 1.43 eV.

Table 1.1: Energy required for cross-slip at zero stress according to Püschl and Schöck [99]

Element Ni Al Cu Ag Au
Energy (eV) 1.0 0.2 1.6 1.8 0.8

1.2.3 Atomistic Models

In the past two decades, atomistic calculations have increasingly been used to study cross-slip.
Rasmussen et al. [110, 113] were the first to estimate the energy barrier using an atomistic
technique. By separately calculating the formation energy of the two constrictions involved in
the process, they arrived at an estimate of 2.7 eV for Cu. Using a similar technique, Rao et al.
[102] obtained a value of 4.85 eV and 2.35 eV for two Ni potentials [148]. Note that the first
potential underestimates γSF . This difference shows that an appropriate interatomic potential
must be used to obtain an accurate energy estimate. Based on Eq. 1.1 and Eq. 1.5, one would
require that the potential gives the correct lattice parameter, elastic constants, and stacking
fault energy. Püschl [121] has argued that the whole γ -surface should be correct. Calculations
with an anisotropic-elastic Peierls-Nabarro model by Szajewski et al. [137] indicate that the
5. Note that the authors use the Voigt average [147] of μ for calculation, which is an upper bound.

10
 1.2. Cross-Slip in Face-Centered Cubic Crystals

dissociation energy is mainly determined by the difference between γSF and γT , the maximum
energy for a stacking fault displacement along the 110 direction.

Several groups have used atomistic transition path calculations to determine the cross-slip
activation energy and pathway. The cross-slip mechanism does not need to be known a priori.
Nudged Elastic Band (NEB) [59, 81] calculations with Cu [112, 143], Ni [143], and Al [57]
showed that the Friedel-Escaig mechanism is the operative mechanism for cross-slip if the
dislocation line is sufficiently long.

1.2.4 Other Mechanisms

Besides the F-E mechanism, two other mechanisms for homogeneous cross-slip have been
discussed in the literature. Schöck and Seeger [120, 124] assumed that the dislocation first
constricts over a certain length, before it bows out into the cross-slip plane, see Fig. 1.3a). In
contrast to the F-E mechanism it does not redissociate immediately6 . The predicted activation
energies for Al and Cu are 1.05 eV and 10 eV, respectively [124]. Wolf [153, 154] calculated the
activation energy for the case where cross-slip by this mechanism takes place at the tip of a
dislocation pile-up. For a pile-up of 20 dislocations under low external stress (τ /μ ≈ 3 × 10−4 ),
the activation energy is approximately 0.6 eV in Al and 2.4 eV in Cu7 . Considering that the
Schöck-Seeger-Wolf mechanism has a high activation barrier if there are no pile-ups, and that
the F-E mechanism is typically observed in transition path calculations, the latter seems to be
the more likely mechanism for homogeneous cross-slip in FCC. However, a combination of
the two mechanisms has recently been observed in molecular dynamics simulations [92].

Figure 1.3: a) Schöck-Seeger-Wolf cross-slip mechanism; b) Fleischer mechanism

6. Seeger et al. [126] argued that a process with immediate redissociation would require strongly correlated
movement of atoms, and would therefore have low statistical weight.
7. Using Equ.
 5.1 and Equ. 5.3 in Ref. [153], with material parameters from Ref. [1], and an effective shear
modulus μ = (C 11 − C 12 )C 44 /2, where C 11 , C 12 and C 44 are the cubic elastic constants.

11
Chapter 1. Theoretical Background

A constriction-free mechanism was proposed by Fleischer [39], see Fig. 1.3. He suggested
that the leading Shockley partial dislocation first decomposes into another Shockley partial
dislocation on the cross-slip plane and a sessile dislocation with Burgers vector of type
a /3 010 or a /6 101 , depending on whether the cross-slip plane makes an obtuse or acute
angle with the glide plane. In a second step, the trailing Shockley partial dislocation on the
glide plane merges with the sessile dislocation to form the trailing dislocation on the cross-slip
plane. Fleischer argued that according to Frank’s [85] rule, the product of the first step should
have a lower energy than a perfect screw dislocation, hence this mechanism should be
preferred to cross-slip involving a line constriction. Marcinkowski et al. [78] calculated
the required energy and the critical stress for both mechanisms using a two-dimensional
isotropic-elastic model. The critical stress of Fleischer cross-slip is found to be on the
order of Gigapascals and to decrease with decreasing stacking fault energy. In a metal with
γSF = 20 mJ m−2 at an applied stress of 500 MPa, the required energies per dislocation length
−1 −1
are 0.27 eV Å and 0.2 eV Å for cross-slip onto an obtuse and acute plane, respectively. This
energy seems prohibitively high. Howeever, recent calculations of the critical stress with a
multiscale technique by Xu et al. [160] indicate that the critical stress of Fleischer’s mechanism
is only few tens of MPa higher than that of the F-E mechanism. Fleischer cross-slip is
sometimes observed in molecular dynamics simulations of deformation [7, 79, 90]. Moreover,
cross-slip without full constriction is observed in atomistic transition path calculation when
the dislocation line is very short (few b ) [26, 57].

1.2.5 Cross-Slip at Heterogeneities

Early on, it was proposed that cross-slip could occur at a heterogeneity in the dislocation line.
Stroh already [133] suggested that jogs could facilitate cross-slip. Mechanisms for cross-slip
at dissociated superjogs and dislocation intersections were proposed by Hirsch [50] and
Washburn [150], respectively. More recently, heterogeneous cross-slip mechanisms were
studied using atomistic methods. Generally, these calculations indicate that heterogeneities
facilitate cross-slip. Rasmussen et al. [113] demonstrated that cross-slip can be spontaneous if
a screw-like constriction is formed at the surface of the crystal. Vegge et al. [142] simulated
cross-slip at jogs and found that the energy was reduced by a factor of four. In a series of
papers, Rao and coworkers considered cross-slip at mildly attractive intersections with forest
dislocations in Ni [103, 104] and Cu [107], at mildly repulsive dislocation intersections in Ni,
Cu, and Ni3 Al [109], at surfaces in Ni and Cu [106], and at jogs [108]. In all cases, the activation
energy was significantly less than in the homogeneous case, see Tab. 1.2. Thus, cross-slip at
heterogeneities should be taken into account when discussing the overall rate of cross-slip.
See Ref. [54] for an example where such mechanisms have been incorporated into Discrete
Dislocation Dynamics calculations.

1.2.6 Alloying Effects

In most studies of cross-slip to date, the metal was assumed to be pure. Extension to alloys
is primarily considered by assuming that the effects of alloying are limited to changing
the relevant average material properties entering the F-E model. In particular, alloying is
considered mainly to change the stacking fault energy γSF , with decreases in γSF leading to

12
 1.3. Transition State Theory

Table 1.2: Factors by which the activation energy for cross-slip is reduced at heterogeneities,
compared to homogeneous cross-slip; “—” = spontaneous cross-slip

heterogeneity material factor reference

jog Cu 4 [142]
jog Cu 4–5 [108]
surface Cu — [113]
surface Cu, Ni 20–30 [106]
dislocation intersections Ni 2–5 [103]
dislocation intersections Cu, Ni 2.5–6 (Cu), 3–20 (Ni) [107]
dislocation intersections Cu, Ni, Ni3 Al — [106]

wider dissociation of the initial partial dislocations, and thus a higher energy for forming the
constriction. A more accurate dimensionless measure for the dissociation width, and hence
the tendency for cross-slip, is γSF /μb , the ratio of stacking fault energy to shear modulus and
Burgers vector magnitude [19]. However, this reduction to a single parameter is a simplistic
model of the alloying effect that misses important mechanisms, as we will see. Surprisingly,
there have been few dedicated studies of alloying effects on FCC cross-slip to date.

Hong and Laird [53] estimated the propensity for cross-slip based on a model of the energy
that is required to recombine a straight dissociated dislocation. The effect of solutes is to
create a frictional force that impedes recombination, and therefore cross-slip. However, this
model ignores the cross-slip mechanism. Furthermore, it is not immediately clear why solutes
would always impede cross-slip. It seems equally possible that they locally constrict the
dislocation, which would facilitate cross-slip. Andrews et al. [2] developed an isotropic-elastic
model to compute the energy to form a Stroh constriction using a line tension model, but with
the assumption of solute segregation to the dissociated dislocation. The assumption of
segregation leads to strong pinning of the initial dislocation, leading to very high energy
barriers for cross-slip (e.g. in Cu-Zn, from approximately 1.64 eV at 0.0025 at.% Zn to ca.
10.8 eV at 20 at.% Zn). This assumption is probably not valid for considering materials being
deformed steadily at normal strain rates and moderate temperatures, where solute diffusion
is too slow to cause significant segregation near dislocations that are temporarily pinned
at obstacles. An atomistic study of cross-slip in solid solutions without segregation was
conducted by Du et al. [25], who calculated the activation energy for cross-slip in Ni-2 at.% Al
and Ni-10 at.% Al, using the Nudged Elastic Band [81] method. The activation energy increased
by 0.2 eV if an Al-Al solute pair was formed in the process, due to strong Al-Al near-neighbor
repulsion in Ni-Al; this result will echo our broader findings below. A similar methodology was
used by Wen et al. [151, 152] for Ni with very ordered arrangements of interstitial H solutes.
Overall, there are no systematic studies of cross-slip in realistic FCC solid solution alloys that
would reveal the clear effects of alloying on this important process.

1.3 Transition State Theory

In chemistry and physics, one is often interested in the rate r (events/second) at which a
system of atoms undergoes a change of configuration, which could be a chemical reaction, or

13
Chapter 1. Theoretical Background

the movement of a defect in a crystal. Dealing with such problems is the subject of reaction
rate theory and, more specifically, transition state theory [37]. Here, we are interested in
dislocation cross-slip, which is a particular change of configuration of the atoms in a defective
crystal. The purpose of this section is to introduce some basic rate expressions and to motivate
calculation of the activation energy ΔE act for cross-slip, which is a central quantity in this thesis.
For a more elaborate discussion, with emphasis on rates of dislocation motion, see Ref. [122].

Frequently, the rate is described using the Arrhenius equation [4],

 
ΔE act
r (T ) = A exp − , (1.7)
kB T

where kB is Boltzmann’s constant, T is temperature, and A is a constant. ΔE act is the maximum

energy encountered during transformation from the initial into the final state, relative to the
energy of the initial state. The exponential term is typically interpreted as the likelihood of
overcoming the energy barrier, and A as the attempt frequency. Frequently, it is assumed that
A is on the order of the Debye frequency νD , i.e. ~1013 s-1 . Molecular dynamics simulations of
screw dipole annihilation under stress indicate that the rate of cross-slip of short (30b [144,
145] or 28b [83]) and longer (200b [92]) dislocations can be described with Eq. 1.7. However,
Vegge et al. [144, 145] measured A =2 × 1015 s−1 . The high frequency compared to the Debye
frequency was explained with translational symmetry along the dislocation line.

The Arrhenius equation is an empirical relation. In transition state theory, the rate is derived
from the motion of the atoms using statistical mechanics. It is useful to introduce some
elements of this theory, which will reappear in Sec. 2.3.2, where a method for calculating ΔE act
is explained. For a detailed explanation of the theory see Ref. [122] or Ref. [48]. Note that
we consider classical systems; quantum effects are ignored. The state of a system of N
atoms is described by 3N positions and 3N velocities. This state can be envisioned as a 6N
dimensional vector in phase space. If one ignores the exact values of the velocities, it is
sufficient to consider the 3N dimensional vector R of atomic coordinates in configuration
space. The potential energy of the system is a single-valued function of R. Hence, the
configuration is a point on a (3N − 1)-dimensional potential energy surface (PES). This surface
consists of basins of attraction around local minima, which are separated by saddle points, see
Fig. 1.4. In thermal equilibrium, the system spends most of the time moving around randomly
in a particular basin. From time to time, velocity fluctuations lead to crossing of a saddle point
and the system moves to the next basin. Two local minima A and B may correspond to two
physically significant states, e.g. a system with a dislocation before and after cross-slip. We
henceforth refer to these minima as initial and final state. The most likely path for the
transition is the Minimum Energy Path (MEP) between A and B . Invoking the picture of a
mountainous landscape, this is the path through the valley, and across the point P on the
saddle with the lowest elevation. Using the harmonic approximation for the vibration of
atoms, Vineyard [146] derived the rate of atomic jump processes in solids as
3N   
j =1 νj (V (P ) − V (A ))
r = 3N −1 exp − , (1.8)
j =1 νj kB T

where ν1 . . . νN and ν1 . . . νN −1 are the real-valued8 normal frequencies for vibration around A
and P , respectively, and V (A ) and V (P ) are the values of the potential energy at A and P .

14
 1.3. Transition State Theory

V (P ) − V (A ) = ΔE act , hence this equation is similar to Eq. 1.7. For cross-slip, Vegge et al. [145]
derived
  1/2
ω0 M ω0
A=ζ , (1.9)
2 π 2 πkB T

where ζ is the length of the dislocation, M is a mass associated with a constriction, and ω0 is a
frequency for oscillations in the width of a straight dislocation. A linear dependence of the rate
of cross-slip on ζ has been observed by Oren et al. [92].

Figure 1.4: Two-dimensional model Potential Energy Surface [59] with two minima separated
by a saddle point; dashed line: approximate minimum energy path

8. At P , the normal mode in the direction of the path has a negative eigenvalue and an imaginary frequency. This
is the missing frequency in the product in the denominator of Eq. 1.8

15
2 Methods

Having reviewed the theoretical background, we can now turn to the simulation methods used
in this thesis. In Sec. 2.1, we briefly review the interatomic potential models that were used in
this work. The alloy systems for which simulations were carried out are introduced in Sec. 2.2.
In Sec. 2.3, we explain the transition path calculations. In particular, we explain the setup, the
calculation method, and how to perform such calculations at finite stress. Finally, in Sec. 2.4,
we explain how solute-dislocation and solute-solute binding energies were calculated.

2.1 Atomistic Calculations with Classical Potentials

A typical configuration for calculating the cross-slip transition path consists of approximately
105 atoms. It is not feasible to simulate such a large number of atoms with first principles
methods, therefore energies and forces were calculated using classical potentials. The
Embedded Atom Method (EAM) [20] is sufficient for FCC metals and was therefore used
throughout this work. All simulations were performed with the LAMMPS [96] software package.

A solid solution can be modeled in two ways. A straightforward way is to assign to the atomic
sites a random distribution of atom types that is consistent with the desired concentration.
Alternatively, a single “average” atom type can be used. The corresponding “average alloy”
ideally has the same average properties as the real random alloy, but is free from concentration
fluctuations. The EAM and the corresponding first-order average-atom approximation [131,
141] are discussed in the following.

2.1.1 Embedded Atom Method

In the EAM, the energy of each atom is the sum of pairwise interaction energies and an
embedding energy, which depends on the local electron density. Consider an alloy with NT
distinct atom types and a total number of N atoms. Let X and Y refer to atom types, and i and
j to atom sites. Denote distances between sites by r i j . Furthermore, let {si ,X } be occupation

17
Chapter 2. Methods

variables, where si ,X = 1 if the atom at site i is type X and zero otherwise. The total energy is

N 
NT
1 
N TN
E ({ si ,X }) = si ,X FX ( ρi ) + si ,X s j ,Y VXY (r i j ), (2.1)
i X
2 i ,j i X ,Y

where

N 
NT
ρi = s j ,X ρX (r i j ), (2.2)
j i X

and where VXY (r i j ) is the pair potential function, FX ( ρi ) the embedding energy functional, ρi
the total electron density at site i , and ρX (r i j ) the contribution of an X -atom at site j to ρi .

Note that all VXY (r i j ) and ρX (r i j ) are radially symmetric. Thus, the EAM represents a
reasonable approximation for materials without preferred bonding angles, such as the FCC
metals.

2.1.2 Average Atom Approximation

Random alloys can be modeled using the average atom approximation. The central idea is to
replace the true atom types with a single fictitious “average” atom type (A-atom), which gives
approximately the same average energy. The A-atom EAM potential was derived by Smith et al.
[131] and Varvenne et al. [141].

Averaging Eq. 2.1 over occupation variables si ,X , Varvenne et al. obtain the preliminary result

N 
NT
1 
N N T

E  = c X FX ( ρi ) + c X cY VXY (r i j ), (2.3)

i X
2 i ,j i X ,Y

where c X is the concentration of type X and . . .  indicates averaging. This equation is valid
under the assumption that the si ,X are uncorrelated in the random alloy, i.e. si ,X s j ,Y  = c X cY .

Eq. 2.3 is simplified further by expanding the embedding energy in a Taylor series around the
average electron density. Averaging eliminates the first-order term in the series, hence
FX ( ρi ) = FX ( ρi ) + O( ρi −  ρi )2 . Neglecting the second order term leads to a simple
expression for the average energy,

N
1
N
E  = FA ( ρi ) + VAA (r i j ), (2.4)
i
2 i ,j i

where

NT
FA ( ρi ) = c X FX ( ρi ),
X

NT
VAA (r i j ) = c X cY VXY (r i j ), (2.5)
X ,Y


N 
NT
 ρi  = c X ρX (r i j ).
j i X

18
 2.2. Selected Alloys

Thus, the EAM potential functions of the A-atom are the concentration-weighted averages of
the corresponding pure element functions. In addition, one takes the weighted average of the
pure element masses mX to obtain the mass m A of the A-atom, which is required in molecular
dynamics simulations:


NT
mA = c X mX . (2.6)
X

Varvenne et al. [141] showed that the A-atom accurately predicts average properties of true
random alloys, for example the lattice parameter, the elastic constants, and the stacking fault
energy. Furthermore, it facilitates the calculation of solute-dislocation binding energies,
which would be cumbersome using a direct approach. However, the A-atom cannot predict
properties that are controlled by solute fluctuations, e.g. the Peierls stress.

The study by Varvenne et al. was limited to zero temperature. Recently, Nöhring and Curtin
[89] verified the A-atom approach for finite temperature calculations. The lattice parameter,
the elastic constants and the Helmholtz free energy of equiatomic FeNiCr and Ni+15 at.% Al
were calculated at temperatures between 0 K and 700 K. It was shown that the A-atom predicts
the mean lattice parameter and the elastic constants of true random alloys accurately, to
within a few percent. The difference in vibrational entropy was at most 0.05kB /atom. Hence,
the A-atom is also useful for approximating true random alloys at finite temperature.

2.2 Selected Alloys

Our purpose is to demonstrate the effect of alloying on cross-slip across a range of FCC alloys
by using model alloy materials over a wide range of solute concentrations. Here, we consider
alloys of Al+(2, 6, 10, 14, 18, 22) at.% Mg, Ni+(2, 4, 8, 10, 12, 15) at.% Al and Cu+(10, 22, 33, 68,
79, 90) at.% Ni as described by the EAM potentials for Al-Mg [74], Ni-Al [98], and Cu-Ni [91].
While these model materials may exceed the solute concentrations of the corresponding real
alloy materials (e.g. Al-Mg is limited to ~5 at.% Mg and Ni-Al is limited to ~13 at.% Al),
the chosen systems serve to cover a range of situations that span real materials (e.g. low
and high stacking fault energies; large and small solute misfit volumes; weak and strong
solute-solute interaction energies). Results will show that alloying can be significant even at
low concentrations (e.g. Al+2 at.% Mg). Furthermore, the main new features of the present
study are expected to be relevant to the emerging classes of FCC High Entropy Alloys (HEA)
[162], which are essentially highly concentrated multicomponent solid solutions.

Some basic properties of average and random Ni-Al, Cu-Ni and Al-Mg alloys with the
aforementioned compositions are listed in Tab. A.1, Tab. A.2 and Tab. A.2 in App. A.1. The
tables show the lattice parameter, the cubic elastic constants, the stacking fault energy and
the measured screw dissociation width. Methods for calculating these properties are also
explained in App. A.1.

19
Chapter 2. Methods

2.3 Calculating the Transition Path of Cross-Slip

There are a number of atomistic methods for determining the transition path of a microscopic
process, such as the Nudged Elastic Band method [59, 81] or the String method [30]. They
provide information about the mechanism and activation energy of the process. Here, this
kind of calculation was used to study cross-slip in random and average alloys. The setup and
execution of these calculations is explained below. Sec. 2.3.1 deals with the procedure for
generating initial and final states. The main method, a variant of the String method, is
explained in Sec. 2.3.2. The section closes with a discussion of transition path calculations at
finite Escaig stress.

2.3.1 Initial and Final States

The atomic configurations for transition path calculations were cylinders with axis parallel to
the [101] crystal direction, see Fig. 2.1. Periodic boundary conditions were applied along this
direction. The cylinders were subdivided into a shell and a core region (not to be confused
with the dislocation core). Atoms in the shell were kept fixed. Its width was larger than two
times the cutoff radius of the potential, hence atoms in the core were √ not influenced by the
free surface. If not stated otherwise, the diameter of the core was 10 3a and the length of the
cylinder was 40b . These dimensions were chosen based on a parameter study with pure Ni, see
App. A.4. The cylinders contained a screw dislocation with complete Burgers vector a /2[101]
and line direction [101], which dissociates into Shockley partial dislocations a /6[211] and
a /6[112] on the (111) plane, or into a /6[211] and a /6[112] on the (111) plane. In transition
path calculations, it was assumed that the former pair is the initial state, and the latter the final
state. Accordingly, the (111) and (111) planes are called glide and cross-slip plane, respectively.

Figure 2.1: Setup for transition path calculations.

20
 2.3. Calculating the Transition Path of Cross-Slip

Generating initial and final states is complicated by the need for consistent cut planes. The cut
plane of a dislocation is the plane across which displacements are discontinuous, see Fig.
1.1. If the dissociated dislocations were created by directly inserting the Shockley partial
dislocations, then each of them would have its own cut plane, and the initial and final state
would be inconsistent. The transition path between them would include extra displacements
to transform one cut into the other. Thus, both states were generated by inserting the
complete a /2[101] dislocation with cut plane (111). In one copy of the pristine configuration
the dislocation was inserted at the center of the cylinder. During a subsequent energy
minimization, it would dissociate either on the glide or on the cross-slip plane. In another copy
the center of the dislocation was shifted slightly in the [121] or [121] direction, which would
result
√ in dissociation on the complementary plane. The shift was typically less than 2–3% of
a / 3, the distance between {111} planes. However, the required shift was not known a priori,
therefore several trials were typically required. In all cases, the perfect dislocation was inserted
by displacing all atoms according to Stroh’s [134] anisotropic-elastic solution. The energy was
minimized using the conjugate gradient (CG) algorithm [49] and the FIRE algorithm [8] with
the parameters in Tab. 2.1. Only atoms in the interior were moved. The minimization was
considered converged when the norm of the 3N -dimensional force vector fell below 10−6 –10−8
−1
eV Å . However, this threshold could not be achieved in some calculations with Cu-Ni and
Al-Mg. The force norm would initially decrease but then start to oscillate around a value above
the threshold. These oscillations were interpreted as an effect of potential tabulation, since
they occured more frequently in Al-Mg, where the potential is relatively coarsely discretized
−1
(500 points). The maximum observed residual force was 4 × 10−4 eV Å in calculations with
−
Al-Mg, and 7 × 10−6 eV Å in calculations with Cu-Ni. However, examination of the energies
1

and forces for a number of cases indicates that the total energy decreases by no more than
−1 −1
1 × 10−5 eV if the force norm is reduced from 10−3 –10−4 eV Å to below 1 × 10−6 eV Å , hence
this level of convergence is sufficient for our purposes.

Table 2.1: FIRE parameters; see Ref. [8] for reference.

parameter N min finc fdec fα αstart Δtmax Δt (initial)

value 20 1.1 0.5 0.99 0.25 2 fs 1 fs

The aforementioned method for obtaining the complementary dissociated state by shifting the
center of the dislocation relies on the fact that the two pairs of Shockley partial dislocations are
symmetrically equivalent and therefore indistinguishable in the chosen simulation cell if the
metal is pure. Thus, a minute perturbation suffices to bias the complete dislocation towards
dissociation on a particular plane. In a true random alloy, however, there are additional
perturbations from solute atoms, which may strongly favor dissociation on one plane. Thus,
larger center shifts and more trials may be necessary to obtain the complementary state, and
the process would be more cumbersome and harder to automatize. In order to avoid this
problem, average alloy initial and final states were used as templates for generating the
corresponding random alloy configurations. After insertion of the dislocations, the average
alloys were converted into true random alloys by replacing the A-atoms with a random
distribution of true atoms types that was consistent with the desired concentration. The
distribution was generated by defining a true atom type for each atom, such that the correct

21
Chapter 2. Methods

concentrations were achieved, and then randomly shuffling1 the type-to-atom mapping.

However, recall that the average alloy approximation is not exact. Both lattice parameter and
elastic constants deviate slightly from the corresponding mean random alloy values, see
Fig. A.1. When the average alloy configurations are converted into random alloys, there is
hence an error in the simulation cell size and the displacement field of the dislocation,
which was calculated using average alloy material properties. The error due to the lattice
parameter difference was partially corrected by scaling the configurations by the ratio of
lattice parameters. The error caused by the difference in elastic constants could not be
corrected, but is expected to be small. Note that the maximum relative difference between
average and random alloy elastic constants is typically on the order of a few percent. In the
Ni-Al system it is 6.1 % in C 11 at 15 at.% Al. In Al-Mg, the maximum differences are 6.0 % in C 11
and 7.5 % in C 44 at 22 at.% Mg, see Fig. A.1 in App. A.1. In Cu-Ni, the relative difference is less
than 1 % for all components, at all concentrations. Anyway, it will be shown that changes
of average material properties have a negligible effect on the cross-slip activation energy,
compared to solute concentration fluctuations.

After changing atom types and scaling the configurations, the potential energy of the core
atoms was minimized using CG and FIRE. During minimization, the dislocation can adapt to
the random solute environment. In some cases, it may lower its energy by slipping away from
the center, towards a favorable solute fluctuation. Such slip is undesirable, since it results in
an additional energy change, which is not associated with cross-slip. This contribution
comes from movement against image forces, which, in turn, result from the fixed boundary
conditions. It was avoided as follows. The random alloy configurations were discarded
when the amount of slip in the initial or final state exceeded a threshold value of 2 Å. The
displacement of the dislocation was measured by calculating the average position of the
non-FCC √ atoms in the core as obtained from a Common Neighbor Analysis (CNA) [52] with
cutoff ( 0.125 + 0.5)a . In some cases, the displacement criterion was applied only to the
initial state. The dislocation was then allowed to slip after cross-slip, but transition paths were
later truncated by selecting a new final state, where the dislocation had slipped less than the
threshold value.

2.3.2 String Method

The goal of the calculation is to find a Minimum Energy Path (MEP) between initial and final
state through 3N -dimensional configuration space. The MEP fulfills the condition that the
force component F⊥ perpendicular to the path is zero everywhere. The activation energy ΔE act
is the maximum energy along this path. In practice, one works with a discrete approximation
of the MEP. Thus, the path is discretized into a chain of NS states or “images”, see Fig. 2.2.
Typically, one demands that the images are evenly spaced along the path or close to maxima. A
standard method for calculating transition paths, which is implemented in LAMMPS, is the
Nudged Elastic Band (NEB) [59, 81] method, see App. A.2. However, this method was observed
to be unstable in preliminary calculations. Frequently, the length of the path would grow out
of bounds2 , and intermediate states would achieve extremely high energies, of up to several

1. Using numpy.random.RandomState.shuffle from the Python-Numpy package [149], which implements the
Fisher-Yates [38] shuffle.

22
 2.3. Calculating the Transition Path of Cross-Slip

hundreds or thousands of electron volts. A modified version of the Simplified and Improved
String [29] method turned out to be more stable and was used instead. The method was
implemented using the Python interface to LAMMPS.

Figure 2.2: Transition path calculation on a 2D model landscape z (x , y ); dots represent images
and vectors indicate direction and magnitude of the local gradient; the unconverged path (red)
has perpendicular force components; the converged path (black) has zero perpendicular force

Consider an initial guess for MEP, i.e. the collection of 3N -dimensional vectors Ri (i = 1 . . . NS )
that store the atomic coordinates. In the modified String method as described in Ref. [29], each
image moves in the direction of its gradient on the (3N − 1)-dimensional PES, i.e.
 i = −∇Vi ,
R (i = 1 . . . NS ), (2.7)
where R  i is the 3N -dimensional vector of atomic velocities and ∇Vi is the gradient of the
potential energy of atom i . Integrating Eq. 2.7 moves the images towards a local minimum of
the PES, which also reduces Fi⊥ , and thus moves them closer to the MEP. However, without
further constraints, the images would eventually end up in the local minimum. Thus,
after each integration step, they are moved back to their ideal position along the path by
interpolation. Reaction coordinates sˆi are introduced to specify the positions,
si
sˆi = ,
s NS
where (2.8)
s 1 = 0,
si = si −1 + | Ri∗ − Ri∗−1 | (i = 2. . .NS ),
and where Ri∗ is the vector of atomic coordinates of image i after the integration step. In
this thesis, an equal arc-length parameterization is used throughout, i.e. the ideal reaction
coordinates are sˆi ,ideal = (i − 1)/(NS − 1). After the integration step, sˆi  sˆi ,ideal . The correct
reaction coordinates are enforced by interpolation of the atomic coordinates. If sˆi ,ideal lies in
the interval [sˆl , sˆr ], then, using linear interpolation,
sˆi ,ideal − sˆl  ∗ 
Ri = Rl∗ + Rr − Rl∗ . (2.9)
sˆr − sˆl
2. According to Kolsbjerg et al. [65], the reason for divergence is that the NEB “spring forces”, which should keep
the images equispaced, only penalize differences between segment lengths, but not total path length.

23
Chapter 2. Methods

Under ideal conditions, the MEP is a stationary state of the dynamics given by Eq. 2.7 in
conjunction with the interpolation scheme [29]. An image that is already on the path moves
only downhill along the path in the integration step, and the interpolation step brings it back
to its ideal position.

Here, the String method was implemented using the Python interface to LAMMPS and some
modifications were made to the original scheme. Atoms were moved by performing energy
minimization in LAMMPS. The parameterization was checked only every 100 iterations of the
minimizer. The atomic coordinates were then extracted and interpolated linearly as in Eq. 2.9.
We refer to these outer iterations as re-parameterizations. The FIRE [8] or QUICKMIN [59, 127]
algorithms were used in the minimization. The dynamics are then not those in Eq. 2.7, but
modified Newtonian dynamics, where the velocity and force vector are mixed to obtain a
velocity with a stronger downhill component. In addition, velocities are zeroed whenever the
dot product of velocity and momentum is positive, i.e. whenever a state is moving uphill on
the PES. Therefore, the motion of states lying on the path is biased to stay on the path, and it is
expected that the MEP is still a stationary state. The convergence of our String method
calculations of MEPs for cross-slip is discussed in App. A.3.

2.3.3 Initial Guess for the Path

If not mentioned otherwise, the initial guess for the path Ri (i = 1 . . . NS ) was created by
linearly interpolating between the initial and final state R1 and RNS , respectively. This is the
standard method, also often used for NEB [59]. However, there is no proof that the converged
path obtained with this initial guess is the path with the lowest possible ΔE act .

In some cases, average alloy calculations were started with a guess for the path that is closer to
a F-E path than the default guess. First, a guess for the transition state of the F-E mechanism
was generated. A 20b long piece of the initial state and a 20b long piece of the final state were
combined to form a complete 40b long configuration. Smooth constrictions were created at
the interface between the two parts by linearly interpolating over a length of 15 Å along z
between the atomic coordinates in the initial and undissociated state, and over another 15 Å
between the coordinates in the undissociated and final state. The initial guess for the path was
created by linearly interpolating between atomic positions in the initial, transition and final
state.

2.3.4 Application of an Escaig Stress

Stress effects on cross-slip were studied by performing transition path calculations with
stressed configurations. We make the usual distinction between resolved stresses that couple
to the edge parts of the partial dislocations (Escaig stresses), and resolved stresses that couple
to the screw parts (Schmid stresses). The former change the splitting width and the latter
cause either regular slip or bow-out. Consider the coordinate system of Fig. 2.1. By resolving
the applied stress tensor σi j (i , j = 1, 2, 3; corresponding to the x -, y - and z -directions) onto
the slip planes and partial Burgers vector components, one obtains the following Escaig and

24
 2.4. Solute Binding Energies

Schmid stresses on the glide and cross-slip planes, respectively:

τglide
Esc
= σ12
τglide
Sch
= −σ23
1 √  (2.10)
τcross
Esc
= 7σ12 + 2 2 (σ11 − σ22 )
9
1 √ 
τcross
Sch
= 2 2σ13 − σ23 .
9
Using Hooke’s law, one can find a strain tensor εi j (i , j = 1, 2, 3) that yields a pure Escaig stress
τglide
Esc
on the glide plane (τglide
Sch
= τcross
Sch
= τcross
Esc
= 0), for example

ε 11 = −ε 22 − ε 33
7 1 Esc
ε 22 = √ τglide
8 2 C 44
3 (C 11 − C 12 − 2C 44 ) Esc
ε 33 = √ τglide (2.11)
16 2 (C 11 − C 12 ) C 44
 
√ 4
γ12 = 2ε 12 = 8 ε 22 − ε 33
7
γ23 = γ13 = 0,

where γi j (i , j = 1, 2, 3, with i  j ) are engineering shear strains. For a pure Escaig stress τcross
Esc

on the cross-slip plane (τglide = τcross = τglide = 0), one may impose
Sch Sch Esc

ε 11 = −ε 22 − ε 33
9
1 Esc
ε 22 = − √ τcross
8 2 C 44
3 (C 11 − C 12 − 2C 44 ) Esc (2.12)
ε 33 = √ τcross
16 2 C 44 (C 11 − C 12 )
√
γ12 = 2ε 12 = − 8ε 33
γ23 = γ13 = 0.

The desired strain components were imposed on all atoms. When applying a strain ε 33 , the
simulation cell was deformed as well, since the z -direction is periodic. The energy was then
minimized using CG and FIRE. Strained random alloy configurations were prepared by scaling
the corresponding average alloy configurations and subsequently replacing the A-atoms by
real atom types, as described in Sec. 2.3.1. Afterwards, the energy was minimized using CG and
FIRE.

2.4 Solute Binding Energies

The binding energy of a solute (s) at position (x , y , z ) relative to a defect is the energy
associated with moving the solute from infinity to (x , y , z ) , i.e.

ΔE s-defect = E s-defect − E s − E defect − 2E pure , (2.13)

25
Chapter 2. Methods

where E s−defect is the energy of the atomic configuration with the solute and the defect, E s and
E defect are the energies with an isolated solute and defect, respectively, and E pure is the energy
of the pure configuration.

Solute-dislocation binding energies were calculated as follows. The √ basic configuration was a
periodic cylinder like the one shown in Fig. 2.1. The radius was 20 3a , which is twice the value
used in transition path calculation. The length of the cylinder was only 6b , because only
individual solutes are studied and so the cell need only be slightly longer than the cutoff radius
of the interatomic potential. The configuration was made of A-atoms. A screw dislocation was
inserted on the initial plane and relaxed, as described in Sec. 2.3.1. One row of atoms in the
core of each partial dislocation was then fixed to prevent the dislocation from slipping towards
or away from the solute during subsequent calculations. The selected rows were the rows
of non-FCC coordinated core atoms with the highest or lowest x -coordinate, see Fig. 2.3.
Typically, these rows were symmetry-equivalent under two-fold rotation about the z -axis
through the center of the dislocation. For each of the 5330 unique atomic sites in this geometry,
the solute-dislocation binding energy was computed by inserting the desired solute atom into
the desired site (removing the A-atom occupying the site), followed by fully minimizing of the
−1
energy using CG and FIRE until the norm of the force vector fell below 1 × 10−6 eV Å , or after
200 000 iterations, and taking the difference between the potential energy after and before
insertion of the solute minus the solute binding energy in the perfect A-atom FCC lattice.

Figure 2.3: Core of a dissociated screw dislocation in Cu+33 at.% Ni, indicating the rows of
atoms that are held fixed.

Solute-solute binding energies were computed using a cube of A-atom material aligned along
100 with edge lengths 40a and fully periodic boundary conditions. A neighbor pair of
A-atoms was then replaced by the desired real solute pair and the energy was minimized with
CG and FIRE. The minimization was stopped when the norm of the force vector fell below
−1
1 × 10−8 eV Å , or after 20 000 or 30 000 iterations. The solute-solute binding energy is then
the potential energy of system containing the solute pair minus twice the energy of a single
solute in the perfect A-atom FCC lattice.

26
3 Cross-Slip of Short Dislocations

Disclaimer

This chapter is adapted from the following article, with permission from the coauthor
and the publisher.
W. G. Nöhring and W. A. Curtin. “Dislocation Cross-Slip in FCC Solid Solution
Alloys”. In: Acta Materialia 128 (Apr. 2017), pp. 135–148. ISSN: 13596454. DOI:
10.1016/j.actamat.2017.02.027

This chapter deals with the activation energy ΔE act of cross-slip of short (40b) dislocation
segments. We present atomistic calculations of ΔE act in average and true random alloys. These
calculations show that ΔE act is a random variable with large fluctuations around the mean
value. Therefore, average-alloy models are not useful for judging the ease of cross-slip in
alloys, and a statistical model is required. We observe that ΔE act correlates with the energy
difference ΔE end between the initial and final state. Thus, the distribution of ΔE act can be
predicted based on knowledge of the distribution of ΔE end . An analytical model of the
standard deviation σ [ΔE end ] is developed. The chapter is structured as follows. The atomistic
calculations are briefly summarized in Sec. 3.1. The results are presented in Sec. 3.2 and
discussed in Sec. 3.3. An analytical model of the distribution of ΔE act is derived and validated
in Sec. 3.4. Finally, implications of the results are discussed in Sec. 3.5.

3.1 Calculations
Using the methods explained in Sec. 2.3, we calculated transition paths of cross-slip in average
and true random alloys of Al+(2, 6, 10, 14, 18, 22) at.% Mg, Ni+(2, 4, 8, 10, 12, 15) at.% Al and
Cu+(10, 22, 33, 68, 79, 90) at.% Ni. For each alloy and concentration, we performed one
calculation with the corresponding average alloy and 20 calculations with random alloys. The
20 samples had a random solute distribution corresponding to the specified concentration,
but were selected from a larger set, based on the displacement of the dislocation away from
the center of the cell, as described in Sec. 2.3.1. In the case of Cu-Ni, the displacement
criterion was applied both to the initial and final state, hence the dislocation was close to the
center in both states. In the case of Ni-Al and Al-Mg, the criterion was only applied to the

27
Chapter 3. Cross-Slip of Short Dislocations

initial state, but transition paths were later truncated by selecting a new final state, where the
dislocation had slipped less than the threshold value. The transition paths consisted of 32
images each. Initial guesses were generated by linear interpolation between the coordinates in
the initial and final states.

Here, we consider short dislocations. The length of the cylindrical configuration sketched in
Fig. 2.1 was 40b . This length was chosen after studying the dependence of the transition path
and energy in pure Ni, see App. A.4. The length 40b is the shortest length at which the
cross-slip barrier becomes length-independent, i.e. at which there is very limited interaction
between the two constrictions formed during cross-slip nucleation. For example, the energy at
40b is 1.72 eV and is only 0.02 eV larger at 160b . This length is therefore a characteristic length
of cross-slip nucleation denoted by ζcsn . Different alloys may have different ζcsn because of
their different elastic properties. For simplicity, the same length of ζcsn = 40b was also used for
all other materials, which is ex-post-facto determined to be a good assumption.

The String method as described in Sec. 2.3.2 was used to find the true minimum energy
paths between initial and final state. Iterations were stopped when the two-norm of the
displacement of each intermediate state relative to its position in the previous iteration
was below 10−3 Å, or, if this threshold was not reached, after a total number of 300
re-parameterizations. However, this threshold was reached in all average alloy calculations.
−1
Furthermore, the norm of F⊥ at the transition state was less than 2.3 × 10−4 eV Å , and the
−
absolute maximum was 3.2 × 10−4 eV Å , in Cu+79 at.% Ni. Convergence was slightly worse
1

in random alloy calculations. In Cu-Ni, 35 out of 120 calculations stopped at 300 iterations, in
Al-Mg 36 and in Ni-Al 25. The average value of | F⊥ | at the transition state was highest in Al-Mg;
however, the influence of the stopping criterion was negligible. In calculations with less than
−1
300 iterations, the average force was 7.1 × 10−4 eV Å , and in those which stopped at 300
− −1
iterations, it was 8.1 × 10−4 eV Å . The absolute maximum force was 3.4 × 10−2 eV Å , in a
1

calculation with Ni+15 at.% Al. See App. A.3 for a more detailed discussion of the convergence.

3.2 Results
3.2.1 Average Alloys

Our calculations yield two important pieces of information. First, they reveal the cross-slip
mechanism to be the F-E mechanism. Fig. 3.1 shows the initial, transition and final state of
cross-slip in Ni-15 at.% Al. The corresponding states in the other average alloys are similar.
Second, they reveal the energy barrier ΔE act , which is the maximum energy encountered along
the transition path.

Fig. 3.2 shows representative paths for the three average alloy systems, one particular
composition for each binary alloy; results for all other average-alloy systems are similar. The
paths are characterized by a steep increase and decrease in energy near the start and end,
respectively, and a region of small energy change near the center. The steep increase and
decrease in energy correspond to constriction formation and annihilation, respectively, while
the plateau in the center corresponds to moving apart of the constrictions. Note that the initial
and final state have the same energy, because they are symmetrically equivalent in the

28
 3.2. Results

te
te ta te
l sta ons sta
tia iti al
in
i ns
n
tra

Figure 3.1: Typical states during dislocation cross-slip in average Ni+15 at.% Al. Atoms are
colored according to their common neighbor value [52] (red=HCP, blue=BCC, white=other).
Atoms in FCC coordination are not shown. OVITO [135] was used for visualization.

homogeneous A-atom materials.

Figure 3.2: Relative energy along the cross-slip transition path in several average alloy systems.
The reaction coordinate is the normalized displacement between the initial and final state, see
Eq. 2.8. Note that a particular value does not correspond to the same configuration (e.g. same
length on cross-slip plane) for two different paths. Thus, the shape of the paths should be
compared only qualitatively.

The maximum of each curve is the activation energy ΔE act . As expected, it correlates with the
normalized stacking fault energy γSF/μb , which is shown in the figure. Al+22 at.% Mg has the
highest γSF/μb and the lowest ΔE act , whereas Ni+15 at.% Al has the lowest γSF/μb and the highest
ΔE act . The variation of ΔE act in each alloy system is comparatively small, see Fig. 3.5. For
Cu-Ni, the maximum change, relative to ΔE act of Cu, is −0.08 eV, at 79% Ni. In Al-Mg, the
maximum change is −0.15 eV, at 18% Mg and in Ni-Al, it is 0.14 eV, at 15% Al. As will be shown
below, these changes are small compared to changes of ΔE act due to compositional disorder in
true random alloys.

3.2.2 True Random Alloys

Like the average-alloy calculations, the true random alloy calculations predict the F-E
mechanism to be the operative cross-slip mechanism. Figs. 3.3a) shows an example of F-E
cross-slip in Ni+15 at.% Al. Fig. 3.3a) and b) show the transition states from the paths with the

29
Chapter 3. Cross-Slip of Short Dislocations

highest and lowest ΔE act in Ni+15 at.% Al (3.09 and 0.86 eV). Note that the configurations are
almost completely dissociated on the cross-slip and glide plane, respectively.

Figure 3.3: Typical states determined by a transition path calculation for random Ni+15 at.% Al:
a) states from a path with ΔE act = 1.07 eV; b) and c) transition states from the paths with the
highest and lowest ΔE act in Ni+15 at.% Al (3.09 and 0.86 eV). Atoms are colored according
to their common neighbor value [52] (red=HCP, blue=BCC, white=other). Atoms in FCC
coordination are not shown. Ovito [135] was used for visualization.

The random alloy calculations predict the same F-E mechanism as found in the average alloy,
but they show very different energy profiles. As an example, the transition paths of all the
random samples of the Ni+15 at.% Al alloys are shown in Fig. 3.4, with the final state points
being corrected as described earlier. The paths are colored according to their activation energy
and the transition state is marked with a circle. In contrast to the average alloy calculations,
initial and final states in true random alloys have different energies, because the solute field
around the dislocation is different on the glide and cross-slip plane. Most importantly, ΔE act
varies significantly from sample to sample.

Note that for paths with high ΔE act , the transition state tends to be closer to the final state. The
transition states of the paths with the highest and lowest ΔE act have the highest and lowest
reaction coordinate, in agreement with the observation that they are almost completely
dissociated on the final and initial planes, respectively. This trend agrees with Hammond’s [45]
postulate, which states that the transition state structure is similar to the initial state when
ΔE act is low.

Fig. 3.5 shows ΔE act as a function of concentration. In every alloy system and at every
concentration, the distribution of the random alloy energies is much larger than the change of
the average alloy energy relative to that of the pure metal. Even in the dilute limit, the lowest
value of ΔE act of the random alloy is typically several tenths of an electron volt lower than ΔE act
of the corresponding average alloy. The significant statistical distribution of ΔE act is the most
important basic result of the random alloy calculations. In the following sections, we will

30
 3.2. Results

relative energy (eV)

1

−1

−2

0.0 0.5 1.0

reaction coordinate

Figure 3.4: Transition paths for specific samples in the true random Ni+15 at.% Al alloy, colored
according to ΔE act ; transition states are indicated by circles.

discuss its implications and develop a model for predicting typical values of ΔE act .

Figure 3.5: Cross-slip activation energies of the 20 individual samples of random alloys for (a)
Ni-Al, (b) Cu-Ni, and (c) Al-Mg; also shown are the average alloy values at each composition; in
(c) blue markers indicate samples with two nuclei; green markers indicate samples with a
partially cross-slipped state with negative relative energy

Finally, we note that random Al-Mg exhibits two pecularities. In 25 out of 120 calculations,
cross-slip nucleated at two sites, and these two nuclei later coalesced to form the
fully-cross-slipped segment. These cases are marked green in Fig. 3.5c). Furthermore, 17
samples have an intermediate minimum that corresponds to a partially cross-slipped state

31
Chapter 3. Cross-Slip of Short Dislocations

and has a negative relative energy. Five of those samples also form two nuclei. The deepest
minimum has a relative energy of −0.39 eV. These cases are marked blue in Fig. 3.5c). Note that
the barrier for spreading of the cross-slipped segment is then higher and more important than
the barrier taken from the initial state. However, for consistency with the other results we draw
the latter.

3.3 Preliminary Discussion

The most striking feature of the random alloys is the large scatter of ΔE act around the average
value. This scatter is caused by fluctuations in the solute concentration, which are excluded a
priori in the average alloys and in any models that consider only average material properties.
As a consequence, ΔE act in real random alloys may be significantly lower than the average
value ΔE act,avg . This is important because cross-slip is a thermally activated process (see Eq.
1.7), so that small decreases in ΔE act lead to exponentially faster cross-slip rates. Along a long
dislocation line, regions where the concentration fluctuations give low values of ΔE act will thus
dominate the rate of cross-slip nucleation. The fluctuations occur around the average ΔE act , so
the average ΔE act sets an initial scale, but the large fluctuations lead to significant lowering of
ΔE act in many cases, so that cross-slip nucleation is controlled by concentration fluctuations.

Accordingly, cross-slip in random alloys must be treated as a statistical problem and the effects
of solute fluctuations must be accounted for. A simple model for typical values of ΔE act in
random alloys will be derived in the following section. Here, we continue with analysis of the
simulation data. The activation energy ΔE act and the difference in initial and final energies,
ΔE end , are strongly correlated, as shown in Fig. 3.6. If ΔE end is low (more negative), then ΔE act
is smaller. Across all concentrations for a given alloy system, the correlation is accurately
captured by a linear function,

ΔE act = α ΔE end + ΔE ref , (3.1)

where ΔE ref is the activation energy at ΔE end = 0 and α ≈ 0.5 is a constant. The value of ΔE ref
is just slightly lower than the average ΔE act of the corresponding average alloys shown in Fig.
3.5; the variations of the average alloy ΔE act with concentration are small so that aggregating
the data across all concentrations is acceptable. A linear correlation between ΔE act and ΔE end
is sometimes observed for chemical reactions, where it is referred to as the Bell-Evans-Polanyi
[6, 36] (BEP) principle, see also [21]. The BEP principle is usually applied for simple chemical
reactions of the type AB + C → AC + B , and the slope α = 0.5 emerges for simple triangular,
symmetric transition paths.

Considering the correlation shown in Fig. 3.6, we can express ΔE act in terms of ΔE end by taking
ΔE ref = ΔE act,avg so that

ΔE act = ΔE act,avg + 0.5ΔE end . (3.2)

Since the initial and final state energies are controlled by random solute concentration
fluctuations, as in solute strengthening of random alloys, particular values are random
variables chosen from a normal distribution [72]. The quantity ΔE end is then also a random
variable with standard deviation σ [ΔE end ]. Assuming that ΔE end follows a normal distribution,

32
 3.3. Preliminary Discussion

we can write the probability distribution of ΔE act as

 2
1 ΔE act − ΔE act,avg
P [ΔE act ] =
exp − . (3.3)
1
πσ [ΔE end ]2
1
2
σ [ΔE end ]2
2

The statistical distribution of ΔE act is thus controlled by the distribution of ΔE end . In the
next section, we show that ΔE end can be calculated analytically, thus leading to an analytic
description for the statistical distribution of ΔE act .

33
Chapter 3. Cross-Slip of Short Dislocations

Figure 3.6: Cross-slip activation energy ΔE act vs. end state energy difference ΔE end , showing a
linear correlation, for (a) Ni-Al, (b) Cu-Ni, and (c) Al-Mg; in (c) blue markers indicate samples
with two nuclei; green markers indicate samples with a partially cross-slipped state with
negative relative energy

34
 3.4. Analytic Model for the Statistical Distribution of ΔE act

3.4 Analytic Model for the Statistical Distribution of ΔE act

3.4.1 General model

Due to the correlation between ΔE act and ΔE end , as shown in Fig. 3.6 and discussed in the
previous section, the determination of the distribution of ΔE act of the random alloy can be
reduced to the problem of determining the distribution of ΔE end . Recall that ΔE end is the
difference in total energies between the final and initial states of the cross-slip. We will
characterize the distribution of ΔE end in terms of its standard deviation.

The end states differ in energy because, in a given distribution of random solutes, the
solute-dislocation and the solute-solute interaction energies are different for the two
dislocation configurations (initial and final). Since the correlation shows, ΔE act ≈ ΔE act,avg
when ΔE end ≈ 0, we are concerned with determination of the standard deviation of ΔE end in a
given alloy system at a given alloy concentration. The analysis proceeds by considering a
fixed specific random distribution of solutes, as indicated in Fig. 3.7, computing the energy
difference of the initial and final states for this particular random distribution, and then
performing an analytical statistical averaging over all possible random distributions to obtain
the standard deviation of ΔE end . There are two separate contributions to ΔE end : (i) a
contribution from changes in the solute/dislocation interaction energy and (ii) a contribution
from changes in solute/solute interaction energy due to the fact that the change in position of
the stacking fault alters the relative positions of solutes with respect to each other, especially
those that are immediately on either side of the initial or final stacking fault plane. These two
contributions in energy lead to two independent contributions to the standard deviation of
ΔE end , which we denote as σ [ΔE end,s-d ] and σ [ΔE end,s-s ]. The total standard deviation for
ΔE end is then computed as

σ [ΔE end ]2 = σ [ΔE end,s-d ]2 + σ [ΔE end,s-s ]2 . (3.4)

3.4.2 Solute-Dislocation Contribution

The contribution σ [ΔE end,s-d ] to the standard deviation of ΔE end due to solute-dislocation
binding energies can be modeled by adapting the solute strengthening model of Leyson and
Curtin [71, 72]. Consider a slice of material parallel to the dislocation line and having a
thickness of one period along the line direction, as shown in Fig. 3.7a). We label the atomic
sites by their {x , y } positions in the plane of the figure as {x i , y j }, with the origin at the center
of the dislocation structure, and x the direction along the glide plane. Similarly, let z be the
direction along the dislocation line and label the sites by their z position as zk . Define an
equivalent coordinate system with respect to the cross-slipped dislocation, with positions
denoted by (x i, y j, zk ). When the dislocation cross-slips, the relative position of the atom at
(xi , y j , zk ) changes to (x i, y j, zk ). If we denote the solute/dislocation binding energy of a
X -solute as Us-d,X (x i , y j , zk ), then the change in total solute/dislocation binding energy is the
sum over all sites given by [71]

  
ΔE end,s-d,X = si j k Us-d,X (x i, y j, zk ) − Us-d,X (x i , y j , zk ) , (3.5)
i jk

35
Chapter 3. Cross-Slip of Short Dislocations

where si j k is the occupation variable with si j k = 1 if a solute exists at site (i , j , k ) and si j k = 0

otherwise.
a)

cross-
stacking fault slip

b)

Figure 3.7: a) Reference frames for measuring the relative solute/dislocation positions before
and after cross-slip; note that the solute distribution across sites (i , j ) does not change. b) The
result of cross-slip is the same as a mirroring of the dislocation at (101) (dashed line), leading
to simplifications in the analysis.

If the solute concentration is high, the corresponding average alloy should be used as matrix
material, so that the solute-dislocation binding energies are the binding energies in an average
solute environment, and with respect to the dislocation structure in that environment [140].

Eq. 3.5 gives the total energy change for a unit dislocation segment. We are interested in
typical fluctuations for a dislocation with finite length ζcsn , which corresponds to Nb = ζcsn/b
slices along z . For this line length, there are Nb sites with the same in-plane coordinate (x , y ),
and solutes occupying any of these sites have the same solute/dislocation interaction energy.
The typical energy change resulting from cross-slip of this length of dislocation is the standard
deviation given by

 
σ [ΔE end,s-d,X (ζcsn )] = ΔE end,s-d,X (ζcsn )2 − ΔE end,s-d,X (ζcsn ) 2 , (3.6)

where · denotes an average over all values of the occupation variables. The derivation of
σ [ΔE end,s-d,X (ζcsn )] is analogous to the derivation of the standard deviation of the energy
change during regular slip in a random alloy, see [72]. The following assumptions are made (i)
the si j k are uncorrelated Bernoulli random variables, (ii) Us-d,X (x i , y j , zk ) depends only on the
atom at site (i , j , k ), and, as noted above, (iii) Us-d,X (x i , y j , zk ) is only a function of x i and y j , i.e.

36
 3.4. Analytic Model for the Statistical Distribution of ΔE act

Us-d,X (x i , y j , zk )→Us-d,X (x i , y j ) due to periodicity along z . We do not repeat the statistical

analysis here, which involves standard but extensive manipulations. The final result is

ζcsn   2
σ [ΔE end,s-d,X ] = c X (1 − c X ) Us-d,X (x i, y j) − Us-d,X (x i , y j ) , (3.7)
b
ij

where c X is the concentration of X -solutes. The total value of σ [ΔE end,s-d ] for an alloy with NT
components is obtained using the rule for the standard deviation of a sum of normally
distributed random variables,


NT
σ [ΔE end,s-d ] = (σ [ΔE end,s-d,X ])2 . (3.8)
X

Finally, note that in the idealized picture of Fig. 3.7a), cross-slip is geometrically equivalent to
performing a mirror operation of the random solute distribution across the (101) plane, as
shown schematically in Fig. 3.7c). This operation is described by the affine transformation
   √   
xi 1 1 8 xi
= √ . (3.9)
y j 3 8 −1 yj

Due to this symmetry, the only inputs to the theory are the binding energies Us-d,X (x i , y j ) in
the initial state since the binding energy Us-d,X (x i, y j) at site (i , j ) is the binding energy at the
symmetry-equivalent site in the undeformed lattice. The number of required calculations
can be further reduced by exploiting the fact that the configuration has two-fold rotation
symmetry about a line parallel to the z axis through the center. Thus, binding energies need
only be calculated for the atoms in the upper half of the x y -plane.

3.4.3 Solute-Solute Contribution

The second contribution to ΔE end comes from changes in direct solute-solute binding
energies. For example, solute pairs (first, second, . . . neighbors) may be formed or destroyed
along the stacking fault ribbon when it is annihilated on the glide plane and/or reformed on
the cross-slip plane. Du et al. [25] have shown that this effect is important for cross-slip in
Ni-Al, and Rodary et al. [115] have shown that it is important for the related problem of solute
strengthening, due to the strong repulsion of Al-Al near-neighbor pairs.

Here, we restrict our attention to first-nearest neighbor pairs. The contribution to ΔE end from
formation and annihilation of pairs of X solutes is

ΔE end,s-s = Us-s,X X ,1 ΔN s-s , (3.10)

where Us-s,X X ,1 is the solute-solute binding energy, and ΔN s-s is the dimensionless net change
in the number of solute pairs due to cross-slip. The contribution to σ [ΔE end ] can then be
estimated as

σ [ΔE end,s-s,X X ] = Us-s,X X ,1 σ [ΔN s-s ] . (3.11)

37
Chapter 3. Cross-Slip of Short Dislocations

Like in the case of solute-dislocation binding energies, the total contribution for an alloy with
NT components is obtained using the rule for the standard deviation of a sum of normally
distributed random variables,


NT
σ [ΔE end,s-s ] = (σ [ΔE end,s-s,X X ])2 , (3.12)
X

where we have neglected unlike pairs of atoms.

σ [ΔN s-s ] in Eq. 3.11 can be approximated using a simple model of solute pair formation and
annihilation in the stacking fault of a single dislocation. We make the simplifying assumption
that the dislocation is straight and the stacking fault is an ideal stacking fault everywhere. In
reality, the dislocations may be curved and the atomic displacement close to the partial
dislocation cores that delimit the fault is not exactly the displacement corresponding to a
perfect fault. Furthermore, we assume that the solute distribution is random and uncorrelated,
so that we can replace occupation numbers by the average solute concentration.

Fig. 3.8 shows the two atomic layers in a section of a stacking fault. Atoms in the lower layer are
gray. A crystallographic unit cell is shaded red. In a stacking fault ribbon of length ζ along
√
[101], the dislocation line direction, and width d along [121], there are ζ/b × d/b 3 such cells. In
each cell, there are two atoms in the upper layer. Each atom exchanges one nearest neighbor
in the lower plane when the fault is formed or annihilated.
√
Let N ζ ≡ 2ζ/b and Nd ≡ d/b 3. Label the atoms in the upper layer by i = 1 . . . N ζ along x  [101]
and by j = 1 . . . Nd along y  [121]. The net change in solute-solute pairs when the fault in Fig.
3.8 is formed or destroyed is:


Nl 
Nd
 
ΔN s-s,sf = si j si j ,2 − si j ,1 , (3.13)
i j

where si j , si j ,1 and si j ,2 are the occupation numbers of the atom at site (i , j ) and the two
nearest neighbors in the lower plane that are exchanged in the process. si j is one if there is a
solute at site (i , j ), and zero otherwise.
  The same applies to si j ,1 and si j ,2 . For convenience, let
 
u i j ≡ si j ,2 − si j ,1 . Note that u pq = c X − c X = 0 ∀ p , q .
 
Calculating the standard deviation of ΔN s-s,sf requires ΔN s-s,sf  2 and ΔN s-s,sf 2 . The former is
zero, because


Nl 
Nd
  
ΔN s-s,sf  = si j si j ,2 − si j ,1
i j


Nl 
Nd
     
= si j si j ,2 − si j ,1
i j


Nl 
Nd
= c X (c X − c X )
i j

= 0.

38
 3.4. Analytic Model for the Statistical Distribution of ΔE act

1 2

stacking fault

Figure 3.8: Lower images: layer of atoms perpendicular to the dislocation line; a
nearest-neighbor pair of solutes (blue) in the stacking fault (red) is destroyed as the dislocation
moves from the glide to the cross-slip plane; top: view from top onto the stacking fault; gray
atoms are in the lower layer of the fault; per unit cell in the fault (shaded red), there are three
atoms in the upper layer of the fault (white), however, two are shared with neighboring cells
and thus are counted as one half; during cross-slip, each atom in the upper layer exchanges
one nearest neighbor atom in the lower layer

Furthermore,

 2

Nl 
Nd
ΔN s-s,sf =
2
si j u i j
i j


Nl 
Nd
 2 
Nl 
Nd 
Nd 
Nl 
Nl 
Nd
= si j u i j + si j u i j si k u i k + si j u i j sk j u k j
i j i j k j i k i j


Nl 
Nl 
Nd 
Nd
+ si j u i j sk l u k l .
i k i j l j

Averaging eliminates all but the first term, therefore:

Nd   
  Nl 
ΔN s-s,sf 2 = si2j u i2j . (3.14)
i j

39
Chapter 3. Cross-Slip of Short Dislocations

si j , si j ,1 and si j ,2 are Bernoulli random variables, hence:

   
u i2j = si2j ,2 − 2si j ,2 si j ,1 + si2j ,1
     
= si2j ,2 − 2 si j ,2 si j ,1 + si2j ,1
= c X − 2c X2 + c X
= 2c X (1 − c X ).

Thus

  
Nl 
Nd
ΔN s-s,sf 2
= 2c X2 (1 − c X )
i j

= 2c X2 (1
− c X )Nl Nd
ζd
= 4c X2 (1 − c X ) √ .
b2 3

Assuming that pair formation and destruction on the glide and cross-slip plane is independent,
the variance of the net pair change due to cross-slip is:
  
Var [ΔN s-s ] = 2 ΔN s-s,sf 2 − ΔN s-s,sf  2 . (3.15)

The standard deviation is therefore:


ζd
σ [ΔN s-s ] = 8c 2 (1 − c X ) √ . (3.16)
b2 3

3.4.4 Results and Comparison to Simulations

In order to verify our model, we compared estimates of σ [ΔE end ] based on Eq. 3.4, Eq. 3.8 and
Eq. 3.12 to results from direct atomistic calculations. In the case of Al-Mg and Ni-Al, only Mg
and Al solutes, respectively, were taken in the account. However, in the case of Cu-Ni, both Cu
and Ni solutes were considered. The average stacking fault width d , which is required for Eq.
3.16, was determined using the Dislocation Extraction Algorithm [136] as implemented in
OVITO [135], by taking the average distance between the extracted partial core coordinates in
the corresponding average alloy.

Recall that our explicit cross-slip transition state computations were performed on only 20
samples. To obtain accurate estimates for the statistical distribution (standard deviation) of
the end state differences ΔE end , we have generated 100 sets of initial and final states for each
alloy composition. In these cases, we impose the constraint on off-center displacements (see
Sec. 2.3.1) for both initial and final states. In the case of Cu-Ni, the 20 sets of data from the
transition path calculations were added, because there the constraint was active in both states,
and thus this data was equivalent to the newly generated data.

The predicted and measured values for σ [ΔE end ] are compared in Fig. 3.9, and the two
individual contributions from solute-dislocation and solute-solute interactions are also

40
 3.4. Analytic Model for the Statistical Distribution of ΔE act

a) Ni-Al b) Cu-Ni

c) Al-Mg

solute-dislocation contribution
solute-solute contribution

Figure 3.9: Observed (filled symbols) and predicted (open symbols) standard deviation of the
end state energy difference ΔE end in (a) Ni-Al, (b) Cu-Ni, and (c) Al-Mg. The individual
contributions to the overall standard deviation, due to solute-dislocation (star symbols) and
solute-solute interactions (triangle symbols) are also shown.

shown. For Ni-Al, the prediction is satisfactory, with a maximum error of 0.1 eV at
10 at.% Al. For Ni-Al, the solute-solute interaction dominates the distribution. For Al-Mg, the
prediction is also satisfactory up to 10 at.% Mg, with the solute/dislocation interaction
dominating. At higher concentrations, the predicted standard deviation saturates while the
measured value continues to increase. This deviation is due to the increasing importance of
non-near-neighbor solute-solute interactions which, though individually small, are very
numerous and so contribute non-negligibly to the standard deviation at higher solute
concentrations. These interactions can be included in the analysis as a generalization of the
near-neighbor pair analysis. An extended model will be presented in Ch. 4. Finally, in
Cu-Ni, the prediction is accurate at the lowest Cu and Ni concentrations but then deviates

41
Chapter 3. Cross-Slip of Short Dislocations

more significantly at intermediate concentrations. The reason for these deviations is that
the solute/dislocation and solute/solute interaction energies in Cu-Ni are generally quite
small in magnitude, and hence the neglect of all of the small but much larger number of
further-neighbor solute-solute interactions is not accurate; again this can be computed and
the agreement with simulations improved, but we do not present the details here.

Summarizing the analysis in this section, we have developed an analytic prediction for the
standard deviation of the end-state energy difference ΔE end in terms only of measurable
solute/dislocation and solute/solute interaction energies. Explicitly combining the above
results for X -solutes, the analysis yields

ζcsn   2
σ [ΔE end,X ]2 = c X (1 − c X ) Us-d,X (x i, y j) − Us-d,X (x i , y j )
b
ij
  (3.17)
ζcsnd
+ Us-s,
2
X X ,1 8c X2 (1 − cX ) √ .
b2 3

This result, while validated against simulations for the current set of alloys and concentrations,
is completely general. It can thus be applied to obtain the statistical distribution of ΔE end
relevant for the subsequent computation of cross-slip activation barriers for any other FCC
alloys, whether existing or proposed alloys.

3.5 Discussion
We have shown that (i) the cross-slip activation energy ΔE act in random FCC solid solutions is
a stochastic variable due to the explicit variations in solute positions relative to the dislocation
undergoing cross-slip, with large variations around the average cross-slip barrier; (ii) there
is a linear correlation between ΔE act and the difference ΔE end between the initial and
final cross-slip states; (iii) the standard deviation of ΔE end can be computed analytically
as a function of alloy concentration, solute/dislocation interaction energies, solute-solute
interaction energies, for the characteristic cross-slip nucleation length ζcsn ; and therefore (iv)
we have an analytic model for the distribution of cross-slip activation energies in random FCC
alloys. We now discuss various implications of these results.

First, average material properties are insufficient for predicting the rate of cross-slip in a real
alloy. As noted in Sec. 3.3, low values of ΔE act caused by random solute fluctuations will
control the rate. In particular, an average parameter like the normalized stacking fault energy
γSF/μb is not a useful measure for the ease of cross-slip in random alloys. Another consequence
of the strong influence of random fluctuations on the activation barrier is that the probability
of cross-slip nucleation in an alloy depends more strongly on the length of the dislocation.
Here, we must clearly distinguish between nucleation, i.e. cross-slip of a short (length ζcsn )
segment, and lateral growth of this segment. A dislocation line with a length of ζ consists of
N = ζ/ζcsn segments of length ζcsn . A long dislocation with large N will sample many different
local solute configurations and hence many local cross-slip nucleation environments and is
therefore likely to contain segments with cross-slip energy barriers that are much lower than
the average value. Among the N segments, the segment with the lowest activation barrier

42
 3.5. Discussion

satisfies, on average,
∫ ∗
ΔE act
  1
P (ΔE act )d ΔE act = . (3.18)
−∞ N
∗ can
where P(ΔE act ) is the Gaussian distribution of activation energies Eq. 3.3. Thus, ΔE act
be obtained from the inverse normal cumulative distribution function. An asymptotic
approximation for large N [23] yields
     
∗ σ N2 N2
ΔE act ≈ ΔE act,avg − log − log log . (3.19)
2 2π 2π

The weak dependence of the smallest activation energy on the number of segments N
allows for reasonable estimates of the barrier over a wide range of lengths. Typical
dislocation densities in metals range from well-annealed materials with dislocation density
ρ = 1011 –1012 m−2 to highly deformed materials with ρ = 1014 m−2 . The corresponding
dislocation segment lengths are ζ ∝ 1/√ ρ and thus vary from 100 nm to 3000 nm. For
ζcsn = 40b and ρ ≈ 1012 m−2 , the minimum activation barriers for such segment lengths are
ΔE act ≈ ΔE act,avg − 1.54σ [ΔE end ]. For a typical Al+2 at.% Mg alloy, this activation energy is
only 0.42 eV. For a typical Ni+10 at.% Al alloy (model matrix for Ni-Al superalloys), the
operative barrier would be 0.55 eV. For a Cu+33 at.% Ni alloy, the operative barrier would be
1.06 eV. All of these energy barriers are significantly below the average energy barrier of the
same alloy, and correspond to very large increases in cross-slip rate at moderate temperatures
(for instance, see Eq. 1.7 with kT = 0.0254 eV at T = 300 K).

The barriers for cross-slip nucleation can therefore be very low in Ni-Al and Al-Mg. However, a
nucleus of length 40b must subsequently expand along the dislocation line by lateral motion
of the two constrictions in order to grow the cross-slip region. Expansion involves sequentially
overcoming smaller energy changes of the order of σ [ΔE end ( ζ = 1b )] for the lateral advance of
the cross-slipped segment by one Burgers vector. These energy changes can be positive or
negative, depending on whether the glide or cross-slip plane is energetically favorable at each
site. For a long dislocation line, the growing nucleus will eventually encounter regions where
there is a long sequence of small positive energy changes, and this sequence can create a
significant barrier against further lateral expansion of the cross-slip segment nucleation. On
the other hand, if the nucleated segment can expand sufficiently before encountering such a
barrier, it can still activate the myriad processes associated with cross-slip. That is, extension
of the cross-slip across the entire screw-oriented portion of the line length is not necessary to
create cross-slip processes in the alloy. The analysis of characteristic cross-slip segment
lengths, and barriers, is a statistical problem that will be addressed in the next chapter. The
discussion here serves to identify the problem, and to emphasize that while cross-slip is
not driven the average barrier, neither is it driven completely by the single lowest barrier
(weak-link behavior).

43
4 Cross-Slip of Long Dislocations

In the previous chapter, we demonstrated that cross-slip nucleation, which occurs over scales
of ~40b , is strongly influenced by fluctuations in the spatial distribution of solutes over the
scale of the critical nucleus. However, full cross-slip of a long (~103b ) dislocations requires that
the nucleated segment expand laterally out to longer lengths. One the one hand, it becomes
more likely to find easy spots for cross-slip nucleation. On the other hand, unfavorable
binding energy fluctuations could accumulate and block cross-slip of the whole dislocation.
Another interesting scenario for long dislocations is discussed in App. A.5, where we consider
the possibility that partially cross-slipped configurations are equilibrium states.

Thus, cross-slip of long dislocations is more complicated. We are interested in the distribution
of activation energies for lengths of ~103b , which are important for understanding real alloys.
However, it is not clear whether the BEP principle applies at these lengths so that Eq. 3.2 can
be used to predict the distribution of ΔE act . At the same time, it is infeasible to study the
behavior at the necessary lengths using atomistic simulations.

In this chapter, we present a new model that does not rely on Eq. 3.2, and that can predict the
activation energy distributions for dislocations of arbitrary lengths, at zero and nonzero stress.
Cross-slip is seen as a random walk, where each step corresponds to the cross-slip of a
1b long segment of the dislocation. In this process, deterministic and random energies
are accumulated. The former reflect constriction formation and stress effects, and can be
calculated from zero-stress parameters alone. To predict the random energies more accurately,
the model presented in Sec. 3.4 is extended to account for solute pairs of first to eight order.

The rest of the chapter is structured as follows. In the first section, we present the extended
model of σ [ΔE end ]. In Sec. 4.2 we then introduce the random walk model. In Sec. 4.3, we
validate the random walk model predictions of the activation barrier by comparisons with fully
atomistic results at the short nucleation length of 40b for zero stress and non-zero stresses. In
Sec. 4.4, we then use the validated random walk model to study long dislocations, which
cannot be treated using atomistic calculations, again for zero stress and non-zero stresses.
Finally, in Sec. 4.5, we summarize the broad general findings and discuss their implications for
cross-slip processes in metal alloys.

45
Chapter 4. Cross-Slip of Long Dislocations

4.1 Random Energy Change

In Ch. 3, we derived an analytical, parameter-free model of the standard deviation of ΔE end .
However, only first-nearest neighbor pairs of solutes were considered in the solute-solute part.
The standard deviation of solute-solute binding energies is then approximately the energy of
such a pair times the standard deviation of the change in number of pairs when the stacking
fault is destroyed on the glide plane and regenerated on the cross-slip plane. This model is
sufficient for Ni-Al and Al-Mg at concentrations up to ca. 15 at.% Al and Mg, respectively. In
Cu-Ni alloys with higher concentrations of the secondary atom (up to 33 at.%), the standard
deviation was previously underestimated. We now present an extended model, which includes
up to eight-nearest neighbor pairs, see Tab. 4.1 for the corresponding distances. In the
derivation, we consider pair changes for all atoms in a unit cell below the stacking fault, see Fig.
4.1. Thus, certain correlations between pair cutting processes are taken into account. For
example, the uppermost atom is initially a second-nearest neighbor of atoms 1 and 2. After a
displacement a /6[112], which creates the fault, the upper atom is a third-nearest neighbor of
atom 1 and a first-nearest neighbor of atom 2. Thus, there is a correlation between the
neighbor changes of atoms 1 and 2. Moreover, for each atom individually, there is a correlation
between second- and first or between second- and third-nearest neighbor pair changes. In
addition to higher order pairs and correlations, we also take pairs of unlike atom types into
account. The derivation of σ [ΔE end,s-s ] is along the lines of the derivation in Sec. 3.4.3, but
much more involved, hence we explain it in App. A.6 and present only the final result here.

Let Us-s,XY ,n be the binding energy of a n th order pair of solutes X and Y . Write the pair

energies of different order as a vector Us-s,XY , i.e. Us-s,XY = (Us-s,XY ,1 , Us-s,XY ,2 , . . . , Us-s,XY ,8 ),
where the superscript  indicates transposition. In A.6, it is shown that the contribution to
σ [ΔE end,s-s ] from cutting/formation of X -Y pairs can be written using vector-matrix notation,

ζd 1
σ [ΔE end,s-s,XY ] = √ c X cY
b 2 3 (1 + δXY )
 1/2 (4.1)

× Us-s,XY [C1 + C2 (c X + cY ) − (C1 + 2C2 ) c X cY ] Us-s,XY ,

where c X and cY are the average concentrations of atom types X and Y , respectively; and C1
and C2 are matrices of constant coefficients, see Eq. A.22. Note that the form above accounts
for the fact that pairs XY and Y X are equivalent. Thus, the total standard deviation of the
solute-dislocation binding energy for an alloy with NT atom types is the double sum

NT
σ [ΔE end,s-s ] = (σ [ΔE end,s-s,XY ])2 . (4.2)
X,
Y ≥X

The contribution of solute-dislocation binding energies to σ [ΔE end ] is modeled as before

using Eq. 3.7 and Eq. 3.8, and σ [ΔE end ] is given by Eq. 3.4.

To verify the model, we compared predictions for σ [ΔE end ] in Ni-Al, Al-Mg and Cu-Ni alloys to
results from direct atomistic calculations. All solute-solute pairs, including mixed pairs, and all
solute-dislocation interactions were considered in the analytical calculation. The binding
energies were calculated as described in Sec. 2.4, using the average-atom approximation for
the matrix.

46
 4.1. Random Energy Change

Figure 4.1: Schematic of correlation between pair loss and formation during formation of a
stacking fault; left: before before formation of the fault; one unit cell of atoms below the fault
and a single atom above the fault are shown; the green plane indicates the center of the fault
along z ; the fault is produced by displacing all atoms above the green plane by a /6[112];
initially, the atom above the plane is a second-nearest neighbor of atoms 1 and 2; after
formation of the fault, it is a third-nearest neighbor of atom 1 and a first-nearest neighbor of
atom 2

Table 4.1: Neighbor distances

neighbor order
1 2
3 √
4
5 √
6
7 8
distance (a ) 1/2 1 3/2 2 5 /2 3 7/2 2

The predictions are compared to the results of the atomistic calculations described in Ch. 3.
Recall that the dislocation length ζ was 40b , and the number of realizations was 100 (Ni-Al and
Al-Mg alloys) or 120 (Cu-Ni alloys). In addition, new atomistic transition path calculations
were carried out during development of the random √ walk√model. In these calculations, the
inner radius of the cylinder was increased from 10 3a to 13 3a , in order to be consistent with
simulations of cross-slip under stress, where a larger radius is required to avoid image effects
due to increased splitting of the dislocation. The calculations with applied stress are explained
further below. In addition, the criterion limiting slip of the dislocation away from the center,
see Sec. 2.3.1, was applied both to the initial and final state in these calculations. The number
of realizations was 100 for each concentration.

Predicted and observed values are compared in Fig. 4.2. Black bars show the atomistic data
from Ch. 3. Gray bars show the new atomistic data. In Ni-Al alloys (Fig. 4.2a)) first nearest
neighbor pairs dominate; including higher order pairs has little effect. However, such pairs are
important in Al-Mg (Fig. 4.2b)) at higher concentrations and in the Cu-Ni alloys (Fig. 4.2c)). In
the latter case, the higher order pair model correctly predicts the peak at ca. 68 at.% Ni, which
is not seen if only first-nearest neighbor pairs are taken into account. Note that predictions
including up to fifth nearest neighbors are shown in Fig. 4.2. Contributions from sixth to
eighth order pairs are typically less than 0.01 eV. The maximum absolute difference between a
prediction with first to fifth neighbor pairs and the corresponding atomistic result is −0.084 eV,
in Al+14 at.% Mg.

47
Chapter 4. Cross-Slip of Long Dislocations

Figure 4.2: Standard deviation of ΔE end for a 40b long dislocation. Colored bars: predictions
made with Eq. 3.4 and the extended solute pair model; the different colors indicate the
maximum order of neighbor pairs included in the calculation. Black bars: observations from
atomistic calculations reported in Ch. 3; the number of realizations 100 in Ni-Al and Al-Mg,
and 120 in Cu-Ni. Gray bars: observations from additional atomistic calculations discussed in
this chapter (100 realizations); ΔE end also includes solute-dislocation binding energies.

48
 4.2. Random Walk Model

4.2 Random Walk Model

ΔE end depends on the length of the cross-slipped segment. As the segment grows, more and
more solute binding energy differences are accumulated. The energy barrier for cross-slip is
then the sum of these energies, which is a statistically distributed variable, plus a deterministic
energy corresponding to constriction formation. The accumulation of energies can be
simulated using a random walk model, where cross-slip of a subsegment is a step and the
associated random change in energy is the step length.

Consider a screw dislocation with a length of Nb times the Burgers vector magnitude b .
Partition the dislocation into Nb unit segments of length 1b . We view cross-slip as the discrete
process of moving the dislocation unit segment by unit segment from the glide to the cross-slip
plane, as shown schematically in Fig. 4.3b). Each unit segment is either fully dissociated on the
glide plane or fully dissociated on the cross-slip plane, and the current cross-slip segment
consists of a contiguous set of N flip flipped unit segments each of length b . We ignore the shape
of the constrictions joining segments on the glide and cross-slip planes. The constriction
energy must be determined with some other method and merely enters as input parameter, as
we shall see. Cross-slip can nucleate in any of the Nb unit segments. However, we assume that
there can be only one nucleus. On each step after the first, this nucleus grows by attaching the
left or right neighboring segment on the glide plane. Associated with each step is a random
change in binding energies, which is obtained from the model in Sec. 4.1. Before simulation of
the cross-slip process, a random energy change is drawn for each unit segment from a normal
distribution with standard deviation σ [ΔE end ] (Eq. 3.4) and mean zero. Note that ζ = 1b in Eq.
3.7 and Eq. 4.1, since the segments are one Burgers vector long. During the initial growth of the
cross-slip nucleus (N flip < Nc where Nc b is the total length of the two constrictions), the
energy cost to form a constriction must also be introduced, as described below.

Figure 4.3: a) Cross-slip transition state in a Ni-15at.% Al average alloy, from an atomistic
transition path calculations reported in Ch. 3; only atoms that don’t have regular FCC
coordination are shown; gray: segment on the glide plane; blue: cross-slip plane; atoms
visualized using OVITO [135]. b) Simplified picture of the process, as used in the random walk
model; the dislocation is partitioned into Nb segments of 1b length; cross-slip is the discrete
process of moving the segments one by one from the glide to the cross-slip plane.

On each step, the unit segment (left or right) chosen for the next flip is the unit segment with
the lower energy change, see Fig. 4.4. For a given starting point, the total accumulated energy
(constriction contributions plus random contributions) after N flip flips is computed as a
function of the total length N flipb of the cross-slipped segment. The process is continued until

49
Chapter 4. Cross-Slip of Long Dislocations

N flip = Nb , i.e. until the entire dislocation has cross-slipped. The activation barrier for the
entire cross-slip is the maximum energy encountered over the entire process. The activation
barrier for the given segment is then the minimum of the activation barriers over all Nb
possible starting points for cross-slip nucleation. In the presence of Escaig stresses or Schmid
stresses, additional deterministic energies are added to account for these effects, as described
below.

rst nucleated segment (randomly chosen)

random energy

(...)

Figure 4.4: Step selection in the random walk; cross-slip can nucleate in any of the 1b
segments; subsequently, however, low-energy steps are preferred

4.2.1 Deterministic Energy Change

In addition to the random change of solute-dislocation binding energies, there is a

deterministic energy change due to constriction formation. In pure metals at zero stress,

50
 4.2. Random Walk Model

the activation energy of F-E-cross-slip is approximately the formation energy of two point
constrictions, i.e. ΔE act,avg ≈ 2Ec [99]. This total energy is distributed over a physical length
Nc b ; that is, the full two-point-constriction energy builds gradually over some physical length,
as seen in directly atomistic models of the transition path. To capture the energetics of the
constriction formation, we proceed as follows.

The first unit segment on the cross-slip plane cannot nucleate without the formation of a
full constriction on the glide plane. Thus, there is an additional energy cost for the first
flip of Ec = ΔE act,avg /2. The two half-constrictions on the cross-slip plane then grow as
the cross-slipped segment becomes longer. At a length Nc b , the half-constrictions are
well-separated and the total energy cost of forming the constrictions has been accumulated.
To model this, we assume that the additional constriction energy per flip decreases linearly as
the segment length grows from b to Nc b . The total energy thus increases quadratically. At flip i
of the nucleation process, the total deterministic energy is thus given as
 
⎧
⎪ 1 i (1 − i + 2Nc )
⎨ +
⎪ ΔE act,avg i ∈ [1, Nc ]
E det (i ) = 2 2Nc (Nc + 1) . (4.3)
⎪
⎪ ΔE
⎩ act,avg i ∈ (Nc , Nb − Nc + 1)

For simulations, we use periodic boundary conditions to eliminate end effects. Thus, when
the cross-slip process is nearly complete, the constrictions should annihilate, and the
energy regained. Thus, the deterministic energy is modified as follows for i in the range
[Nb − Nc + 1, Nb ):
 
(i − Nb + Nc ) (1 + i − Nb + Nc )
E det (i ) = ΔE act,avg 1− , (4.4)
2Nc (Nc + 1)

and E det (Nb ) = 0. Fig. 4.5 shows the energy vs. cross-slipped length from an atomistic
transition path calculation for Ni+15 at.% Al as compared to the deterministic energy of Eq.
4.3 using the atomistic activation energy and Nc = 10. The length of the cross-slipped
segment in the atomistic calculation was measured using the dislocation analysis (DXA)
algorithm implemented in Ovito [135]. Atomistically, the constriction forms at zero length, but
this energy must be assigned to the first unit segment of length b . The use of ΔE act,avg /2
for the first flip underestimates the atomistic energy by approximately 0.19 eV on the first
step. Larger errors are seen in the region of constriction annihilation. This may explain
later results in which there are small (< 0.1 eV) differences between the median values of
activation energy distributions from atomistic and random walk calculations, but the model of
annihilation is only of importance for comparing to the simulation results at short total
lengths (Nb = 40). Overall, the model for constriction formation and annihilation is in good
qualitative agreement with the atomistic simulation, and is fully satisfactory for the purposes
of this work.

4.2.2 Stress Effects in the Random Walk Model

Escaig stresses that lower the energy of the final state relative to the initial state will facilitate
cross-slip. Net Schmid stresses acting on the cross-slip plane will also facilitate cross-slip. Both
types of stresses will also help to overcome very high activation barriers that can arise in the

51
Chapter 4. Cross-Slip of Long Dislocations

1.5
relative energy (eV)

E det
1.0

0.5
atomistic data

0.0
0 5 10 15 20 25 30 35 40
cross-slipped length (Burgers vector)

Figure 4.5: Energy as a function of the length of the cross-slipped segment from an atomistic
transition path calculation with average Ni+15 at.% Al and the corresponding approximate
profile E det used in random walk calculations; the first few points at zero length in the
atomistic profile correspond to formation of the constriction on the glide plane.

random walk under zero stress. The work done by such external stresses is deterministic and
hence can be included in the random walk model by modifying the deterministic energy
contribution in Eq. 4.3.

We first address Escaig stresses, which couple to the edge components of the partial
dislocations of the dissociated screw dislocation in FCC materials. These stresses change the
width of the partial separation. Let τglide
Esc
and τcross
Esc
be the Escaig stresses on the glide and
cross-slip plane, respectively. These stresses have two effects. First, they do work on the
system when the splitting width changes from its equilibrium value on the glide plane to its
equilibrium value on the cross-slip plane. These deterministic energies can be computed in
atomistic simulations, but models using anisotropic elasticity are more general. When
Esc
comparing with atomistic simulations, the atomistic result for E det was used. In other cases,
1
an anisotropic-elastic estimate was used,
   
b3 Ke γeff,g
Esc
E det =− Ks − log , (4.5)
8π 3 γeff,c
where Ke and K s are the energy prefactors of the edge and screw dislocation, respectively, see
Eq. 1.3; and γeff,g and γeff,c are the effective stacking fault energies on the glide and cross-slip
plane,
  
b τglide
Esc
b τcross
Esc
γeff,g = γSF + √ , γeff,c = γSF − √ . (4.6)
2 3 2 3

Second, the Escaig stresses change the constriction energy. According to Püschl and Schöck
[99, 100] Ec ∝ log (2d /b ) d /b , thus we scale the energy of the constriction on the glide plane by
1. Equ. 5.7 in Ref. [14] with dM = d and splitting widths from anisotropic elasticity [1, Equ. 13.149b]

52
 4.2. Random Walk Model

the ratio (log (2d ∗ /b ) /log (2d /b ))(d ∗ /d ), where d ∗ is the stressed equilibrium dissociation
width. Using the anisotropic-elastic solution for d ∗ , Eq. 1.2, this ratio can be written as
  
γSF log (K s − Ke /3) b / 4πγeff,g
fc = . (4.7)
γeff,g log ((K s − Ke /3) b /(4πγSF ))

Taking into account the effect of τcross

Esc
on the constriction on the cross-slip plane is more
complicated, because the stress also changes the constriction size, and hence Nc . Thus,
we neglect this energy contribution. Nevertheless, we obtain a good agreement between
predicted activation energies and atomistic data, see App. A.7.

Out of the two Schmid stress components, only the one on the cross-slip plane needs to be
considered. A Schmid stress on the glide plane simply causes the dislocation to slip away. The
segment on the cross-slip plane, however, is pinned by the two constrictions and is therefore
expected to bow out. The work done by bowing out can then also drive lateral expansion of the
cross-slipped segment. We assume that the bow-out takes the shape of a circular arc with
radius R , arc-length s and swept area A . Within isotropic linear elasticity, the total energy of
bowing out can be described analytically [156]. For a size ib of cross-slipped segment, the total
energy due to bow-out is
Sch
E det (i ) = T s (i ) − τcross
Sch
bA (i ),
where
   
−1 ib
s (i ) = 2R sin − ib ,
2R
   2 (4.8)
−1 ib ib
A (i ) = R sin
2
1− ,
2R 2R
T
R =
τcross
Sch
b

and T is an effective dislocation line tension representing the cost of creating additional
dislocation line length.

The value of T is guided by the work of Kang et al. [60], who compared the stress dependence
of the cross-slip activation energy in the Friedel-Escaig model and in atomistic simulations.
Comparing the shape of the cross-slipped segment in the two models, the authors suggested
T = α μb 2 /2, where α is a parameter in the range [0.1, 0.6], and μ is an effective isotropic shear
modulus. Following the authors and Scattergood and Bacon [119], we use

(C 11 − C 12 ) C 44
μ ≡ Ks = . (4.9)
2

Kang et al. used a value of 0.45 for α . However, the FE-model overestimated the
activation energy. Even after applying a constant shift of −0.7 eV, the energy was slightly
√ (2∝ 10 MPa) Schmid
2
too high at high stresses. Thus, we use a smaller line tension of
1/2 × 0.45μ (b / 3) /2 = 0.075μb 2 /2, which is slightly below the lower bound suggested by
Kang et al. The activation energies predicted using this line tension in Eq. 4.8 are closer to their
atomistic results than the corresponding predictions of the FE-model, see Fig. A.14. Additional

53
Chapter 4. Cross-Slip of Long Dislocations

random energy fluctuations during slip from the straight to the bowed-out configuration due
to solute-dislocation interactions (essentially solute strengthening) are neglected.

In total, in the presence of Escaig and/or cross-slip Schmid stress, the deterministic energy
profile for the random walk model is
 
⎧
⎪ fc i (1 − i + 2Nc )
⎪
⎪ + ΔE act,avg + i E det
Esc
+ E det
Sch
(i ) i ∈ [1, Nc ]
⎨ 2
⎪ 2Nc (Nc + 1)
E det (i ) =   . (4.10)
⎪
⎪ fc + 1
⎪
⎪ ΔE act,avg + i E det + E det (i )
Esc Sch
i > Nc
⎩ 2

Under periodic boundary conditions for comparison to atomistic simulations, the constriction
is annihilated in the manner used at zero stress; this is only relevant for pure Escaig stresses.

4.3 Validation Against Atomistic Simulations

Here, we compare predictions of the random walk model against full atomistic transition state
simulations at Nb = 40 using periodic boundary conditions, for both zero stress and applied
Escaig stresses. The results demonstrate that the random walk model captures all major
features of the simulations with very good quantitative accuracy, justifying its application to
long dislocation lines (Nb  40) in the next section.

Note that√the atomistic calculations

√ were carried out with slightly larger configurations (inner
radius 13 3a instead of 10 3a ), in order to minimize boundary effects when an Escaig stress
is applied on the cross-slip plane, see App. A.4. ΔE act,avg was recalculated with the slightly
larger geometry as well, although the differences are negligible (< 0.03 eV). ΔE act,avg and the
standard deviaton of the random energy per step are listed in Tab. 4.2.

Table 4.2: Average-alloy cross-slip activation energy ΔE act,avg and standard deviation of the
random energy per 1b step, as calculated with Eq. 3.4

alloy ΔE act,avg (eV) σ [ΔE end ] |ξ=1b (eV)

Ni+02%Al 1.73 0.033
Ni+15%Al 1.89 0.189
Al+02%Mg 0.73 0.032
Al+06%Mg 0.69 0.054
Al+22%Mg 0.63 0.088
Cu+22%Ni 1.65 0.047
Cu+33%Ni 1.64 0.062
Cu+79%Ni 1.51 0.054

4.3.1 Zero Stress

At zero Escaig stresses, a total of 100 realizations were used to generate the atomistic results
and a total of 10000 realizations were used in the (much faster) random walk calculations. Fig.

54
 4.3. Validation Against Atomistic Simulations

4.6 shows a box plot of the activation energy distributions at zero stress for a number of alloys.
The line in each box indicates the median, while the lower and upper edges show the first and
third quartile, respectively.

Figure 4.6: Distributions of cross-slip activation energies in solid solution alloys for Nb = 40
with periodic boundary conditions, from atomistic (A) and random walk (R) simulations; the
upper and lower edge of each box represent the third and second quartile of the data, i.e. 50%
of all values lie in this range; the horizontal line indicates the median. The number of
realizations in atomistic calculations was 100, except in the case of Cu+22 at.% Ni and
Cu+79 at.% Ni at finite stress where only 50 realizations were used; 10000 random walk
realizations are used in all cases.

Atomistic and random walk results are typically similarly distributed. The difference in
median values is always less than 0.07 eV (in Ni+15 at.% Al) and the largest relative difference is
12% (in Al+22 at.% Mg). The largest relative difference in variance is 78% (in Ni+02 at.% Al) but
the actual difference is very small (0.007 eV higher than the atomistic value). The similitude of
the distributions is assessed by the two-sided Kolmogorov-Smirnov (K-S) [64, 130] statistic.
The Null hypothesis that the samples are drawn from the same continuous distribution can
only be rejected at significance level 0.05 in the case of Al+02 at.% Mg and Ni+02 at.% Mg. If
the random walk samples are shifted to eliminate the difference in median values, the Null
hypothesis cannot be rejected in any case. Thus, the random walk model typically yields
distributions very similar to those obtained from the atomistic calculations.

Small differences in median values are likely caused by deviations of E det from the true
deterministic energy profile. For example, in the case of Ni+02 at.% Al, E det overestimates the
true energy at the end of constriction formation and during annihilation, see Fig. 4.5. However,
the maximum of the random walk frequently occurs in these regions, see Fig. 4.7. Thus, the
random walk tends to yield slightly higher energies, see Fig. 4.6. For completeness, we note
that in the case of Ni+15 at.% Al, where σ [ΔE end ] is higher, the maximum frequently occurs on
the first and last few steps, see Fig. 4.7.

55
Chapter 4. Cross-Slip of Long Dislocations

frequency of maxima at step

0.10

0.08
Ni+02%Al
0.06
Ni+15%Al
0.04

0.02

0.00
0 10 20 30 40
step

Figure 4.7: Frequency of the location of the maximum as a function of step

Furthermore, the random walk model predicts the correlation between the activation
energy ΔE act and the end-state difference ΔE end that was empirically found in the atomistic
simulations at length 40b . For example, Fig. 4.8 shows ΔE act versus ΔE end from the atomistic
and random walk calculations for Ni+15 at.% Al. Linear regression of both data sets yields
slopes of 0.5 and 0.54, respectively. Slopes of 0.53–0.54 are obtained from the random walk
calculations in other materials, see Fig. A.15 in App. A.8. Therefore, the random walk model,
with only the average barrier height and solute-solute and solute-dislocation interaction
energies as inputs, reproduces a major empirical feature of the atomistic simulations.

In fact, the correlation is not much different from the correlation between the maximum and
location of a regular normal random walk, where there is no preference for small steps, and no
preference for paths with low maximum. A joint plot for such a walk is shown in Fig. 4.8b). A
linear fit to the data yields a slope of 0.51. However, the trends are different in the region of
positive and negative ΔE end . In the region of positive ΔE end , ΔE act is bounded from below by
ΔE end , since, by construction, the maximum cannot be less than the final location of the walk.
In the region of negative ΔE end , the slope decreases, an effect which can also be seen in the
random walk data of Fig. 4.8a). This difference between the regions of positive and negative
ΔE end becomes very clear when the number of steps is increased. Fig. 4.8c) shows random
walk energies for Nb = 1000. In this case, using ΔE act ≈ ΔE act,avg + 0.5ΔE end to predict typical
values of ΔE act would be very inadequate.

4.3.2 Non-zero Escaig Stresses

Using the same atomistic methods as in our previous work, transition state calculations were
carried out for Escaig stresses of τglide
Esc
= 150 MPa and τcross
Esc
= 150 MPa for several alloys. For
Cu+22 at.% Ni and Cu+79 at.% Ni, only 50 realizations were simulated. In applying the random
Esc
walk model, the value of E det was taken from the average of the end-state difference obtained
in the atomistic simulations. The results are shown in Fig. 4.6. The difference in median values
is typically small, with a maximum relative difference of 10% (0.14 eV) in Cu+33 at.% Ni
with stress on the cross-slip plane. The widths of the distributions are typically similar.
However, in Al+06 at.% Mg there is a comparatively large relative difference in the variances,
−68% and −78% for stresses on the glide and cross-slip plane, respectively, but only small

56
 4.3. Validation Against Atomistic Simulations

6 6
a) Ni+15 at.% Al b) Gaussian random walk
5 ΔE act,avg = 1.89 eV 5 ΔE act,avg = 0 eV
σ [ΔE end ] |ξ=1b = 0.19 eV σ [ΔE end ] |ξ=1b = 0.19 eV
4 4
random walk:
ΔE act (eV)

ΔE act (eV)
3 slope = 0.54 3
random walk:
2 2 slope = 0.51
atomistics:
slope = 0.50
1 1

d
en
ΔE
=
0 0

t
ac
ΔE
−1 −1
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
ΔE end (eV) ΔE end (eV)

c) Ni+15 at.% Al, Nb = 1000

20

15
ΔE act (eV)

10

0
−20 −10 0 10 20
ΔE end (eV)

Figure 4.8: Joint plots of the distributions of activation energies ΔE act and end state energy
differences ΔE end ; a) comparison of random walk and atomistic data for Ni+15 at.% Al and
Nb = 40; hexagonal bins: random walk data, color indicates frequency, see marginal
distributions; "+": atomistic data; lines: linear regression; b) distribution for an ideal random
walk without drift, i.e. E det = 0, and no selection of low-energy steps or low-energy paths; c)
random walk data (our model) for Ni+15 at.% Al and Nb = 1000 (no periodic boundary
conditions)

absolute differences (−0.008 eV and −0.012 eV). The K-S statistical analysis shows that the Null
hypothesis that the samples are drawn from the same continuous distribution can be rejected
at significance 0.05 for all alloys and stresses except in the case of Cu+79 at.% Ni. However, if
the difference in median values is subtracted, then the hypothesis can only be rejected in the
case of Al+06 at.% Mg. Thus, most of the error comes from the difference in median values,
which is likely a consequence of errors in the deterministic energy profile, as discussed in
Sec. 4.3.

57
Chapter 4. Cross-Slip of Long Dislocations

4.4 Cross-Slip of Long Dislocations

Useful cross-slip in real materials requires the cross slip of segments much longer than 40b ,
although 40b is the critical nucleation size in elemental metals. Having validated the random
walk model at atomistically-accessible sizes, we can now apply it to much larger lengths
relevant to real materials. Simulations were performed at lengths Nb = 40, 100, 200, 500, 1000
with no periodic boundary conditions, so the constrictions were not annihilated at the end of
the process, and results effectively corresponding to cross-slip of a slightly longer segment or a
segment pinned by obstacles.

4.4.1 Zero Stress

For zero applied stresses, Fig. 4.9 shows distributions of ΔE act for lengths between 40b and
1000b . A clear trend is seen: with increasing length, the frequency of high activation energies
increases. For the third quartile, we observe a trend ΔE act,avg + σ [ΔE end ] x Nb /2 with x = 1.2.
The scaling of higher percentile levels follows the same form, but with larger x . On the other
hand, the lower tail of the distribution does not change significantly. The decrease of the first
quartile is much less pronounced than the increase of the third quartile. Accordingly, the
median values increase by several tenths of an electron volt, in the case of Ni+15 at.% Al
even by more than one electron volt. The minimum barrier, however, barely changes. The
increasing frequency of high activation energies means that cross-slip of long dislocations is
much harder and less frequent in solid solutions. However, this result pertains only to zero
stress. As shown in the next section, the high barriers are easily overcome by fairly modest
applied stresses, leading to much lower activation energies.

Histograms of the energy distributions of Cu+33 at.% Ni, for Nb = 40, 200, 1000 clearly show
the tail of high ΔE act that develops with increasing length, see Fig. 4.10.The inset plot shows for
each step in the Nb = 200 calculation the frequency with which the maximum occured on this
step, and the median energy of the corresponding maxima. The distribution of the maxima
between the peaks is reminiscent of the arcsine distribution expected for Wiener processes
[70]. The low energy paths are those where the maximum occurs early on. Conversely, high
energy paths are those where the maximum occurs near the end. Accordingly, these paths are
particularly sensitive to the work done by an external stress in extending or bowing out the
dislocation on the cross-slip plane.

4.4.2 Non-zero Stress

Schmid and Escaig stresses in the range 1–200 MPa were applied to dislocations with length
Nb = 500. The constrictions were not annihilated at the end of the process and no periodic
boundary conditions were used. In contrast to the calculations with short segments in
Esc
Sec. 4.3.2, we did not use the atomistic value for E det , but rather the elastic estimate according
to Eq. 4.5.

Fig. 4.11 shows energy distributions for Cu+33 at.% Ni. The other alloys exhibit similar
behavior, see Fig. 4.12. Applied stresses tend to eliminate the high energy barriers. The
additional work done by the applied stresses as the cross-slip segment length increases are

58
 4.4. Cross-Slip of Long Dislocations

Figure 4.9: Distributions of ΔE act for different alloys and dislocation lengths Nb ; the upper and
lower edge of each box represent the third and second quartile of the data, i.e. 50% of all values
lie in this range; the horizontal line indicates the median; the dot marks the minimum value

Figure 4.10: Histograms of activation energy distributions in Cu+33 at.% Ni for three
dislocation lengths Nb (sample size 10000, no periodic boundaries); inset: distribution of the
step at which the maximum occurs (histogram, left axis) and median energy of those maxima
that occur on a particular step (points, right axis)

59
Chapter 4. Cross-Slip of Long Dislocations

more than sufficient to overcome the slowly accumulating random energies that cause the
large barriers. A Schmid stress on the cross-slip plane is particularly effective, with the third
quartile being reduced by 1.1 eV upon increasing the Schmid stress from 1 MPa to 5 MPa. Such
low Schmid stress could easily be reached in real materials due to long range stresses and stress
fluctuations caused by the local dislocation network. Escaig stresses, by comparison, are less
effective in that higher stresses in the range of 25–50 MPa are required to eliminate the high
activation barriers. Nonetheless, such stress levels remain moderate. Note that the observed
stress effects are more complicated than in elastic or atomistic models for pure metals. Usually,
τglide
Esc
is expected to be the most effective stress for reducing the activation energy while τcross
Sch
is
supposed to be least effective [60]. Fig. 4.11 shows that this is only true at higher stresses.

Figure 4.11: Activation energy distributions in Cu+33 at.% Ni with Nb = 500 for different
Escaig and Schmid stresses; the upper and lower edge of each box represent the third and
second quartile of the data, i.e. 50% of all values lie in this range; the horizontal line indicates
the median; the dots mark the minimum and maximum value of each dataset

The characteristic stress required to eliminate high activation energies can be estimated as
follows. As noted above, the maximum barrier is unlikely to occur at the end of the random
walk if the magnitude of work performed by the external stress after Nb steps is likely to exceed
the accumulated random energy contributions. As shown in Fig.  4.9, a majority of the
activation energies have an energy of less than ΔE act,avg + σ [ΔE end ] 1.2 Nb /2. Thus, the work

should exceed σ [ΔE end ] 1.2 Nb /2, and limit stresses can be obtained by solving

Nb E det (τglide , τcross
Esc Esc Esc
) + E det (τcross , Nb ) + σ [ΔE end ] 1.2 Nb /2 = 0.
Sch Sch
(4.11)

The energy change of the constriction on the glide plane due to τglide
Esc
has been ignored. It only
contributes an offset to the total energy and thus has no influence on the position of the
maximum. Assuming pure Escaig and Schmid stresses, one obtains characteristic stresses of

60
 4.4. Cross-Slip of Long Dislocations

4
ΔE act (eV)

0
Ni+02%Al Ni+15%Al Al+06%Mg Cu+79%Ni
alloy

Figure 4.12: Activation energy distributions in solid solution alloys for Nb = 500 and high or
low Escaig and Schmid stresses; the upper and lower edge of each box represent the third and
second quartile of the data, i.e. 50% of all values lie in this range; the horizontal line indicates
the median;

τglide
Esc
= 21.3 MPa, τglide
Esc
= 20.7 MPa, and τcross
Sch
= 2.1 MPa for Cu+33 at.% Ni at Nb = 500,
consistent with the results in Fig. 4.11.

In summary, moderate applied stresses that act to facilitate cross-slip are sufficient to
eliminate high barriers and drive the entire distribution of activation barriers to much lower
values than the mean value at short lengths.

4.4.3 Weakest-Link Scaling

The previous observations motivate us to treat cross-slip under stress as a weakest-link

problem. At sufficiently large stress, most maxima occur early in the random walk. Thus, the
activation energy distribution at a long length Nb can be inferred from the corresponding
distribution for a much shorter length N ref by accounting for the number n = Nb /N ref of
approximately statistically-independent segments of size N ref at which cross-slip can initiate.
The cumulative distribution P (ΔE act , Nb , τ ) for length Nb at the given stress τ should then obey

P (ΔE act , Nb , τ ) = 1 − (1 − P (ΔE act , N ref , τ ))(Nb /Nref ) , (4.12)

where P (ΔE act , N ref , τ ) is the cumulative distribution function of activation energies at N ref at
the given stress. To test this hypothesis, we carried out further random walk calculations with
lengths Nb = 5000 and Nb = 10000. We then used the random walk data from calculations at
Nb = 500 as reference, i.e. N ref = 500. Calculations were carried out for Cu+33 at.% Ni,

61
Chapter 4. Cross-Slip of Long Dislocations

Al+06 at.% Mg and Ni+15 at.% Al.

Fig. 4.13 shows the cumulative distribution functions for the cross-slip activation energy
for 5 MPa Schmid stress on the cross-slip plane, and also 50 MPa for Cu+33 at.% Ni, for
Nb = 500, Nb = 5000, and Nb = 10000. The figure also shows the predictions for Nb = 5000
and Nb = 10000 as obtained from weak-link scaling. The weak-link scaling is very accurate,
within fractions of a percent, in all cases. The agreement is achieved because, even at the lower
stress level, all large barriers have been eliminated within the length Nb = 500.

Fig. 4.14 shows the cumulative distribution functions for the cross-slip activation energy at
both 5 MPa Escaig stress, for Cu+33 at.% Ni and Ni+15 at.% Al. As can be expected from Fig.
4.11 and Fig. 4.12, the weakest-link prediction is not accurate at this low stress because the
largest barriers can be encountered at lengths larger than the reference length Nb = 500. Thus,
higher activation energies are underestimated. The error is smaller in Cu+33 at.% Ni because
the characteristic stress calculated with Eq. 4.11 is only τcross
Esc
= 6.6 MPa for Nb = 5000 and
τcross = 4.7 MPa for Nb = 10000, close to 5 MPa. Nevertheless, the differences in median values
Esc

are only −0.12 eV and −0.08 eV, respectively, and so the weak-link scaling captures much of the
shift of the activation barriers to lower values, but overestimates the shift.

At 50 MPa Escaig stress, the prediction of weak-link scaling is excellent. At this stress level,
all large barriers are overcome well within the length of Nb = 500, making the weak-link
model applicable. This is consistent with the estimates for the characteristic stresses of
τcross
Esc
= 20.8 MPa and 14.8 MPa for Nb = 5000 and Nb = 10000, respectively, in Ni+15 at.% Al.

4.4.4 Predictions Based on Zero Stress Data

At sufficiently high stresses, Eq. 4.12 allows for the prediction of the activation energy
distributions of very long dislocations based on knowledge of the distribution at length N ref .
However, the reference distribution needs to be calculated at the given applied stress, and this
distribution is not readily accessible. Making predictions would be easier if the reference
distribution at zero stress could be used, because this distribution is computable from the
underlying solute/dislocation and solute/solute interaction energies. At zero stress, however,
N ref must then be small so that the reference distribution does not have a large tail at high
energy barriers. This suggests using the distribution at the fundamental cross-slip length of
Nb ≈ 40. This length is also accessible by atomistic simulations if suitable potentials are
available.

The effect of stress can then be approximated by shifting the reference distribution to lower
energies, according to the deterministic change of ΔE act,avg . However, not all energy
levels should be shifted by the same amount. Low barriers typically correspond to paths
where the energy barrier occurs early. These barriers are relatively insensitive to the
work done by the stress in extending or bowing out the dislocation on the cross-slip
plane. Conversely, energy barriers equal to ΔE act,avg or higher likely correspond to paths
where the maximum occurs later, and these see the full stress effect. To approximate this
difference in sensitivity, we make the shift a linear function of ΔE act for ΔE act < ΔE act,avg , i.e.

62
 4.4. Cross-Slip of Long Dislocations

Cu + 33% Ni, τcSch = 5 MPa Cu + 33% Ni, τcSch = 50 MPa

cumulative frequency
1.0 a) 1.0 b)

0.5 0.5

Nb = 500
0.0 0.0
0.75 1.00 1.25 1.50 0.75 1.00 1.25 1.50

Al + 6% Mg, τcSch = 5 MPa Ni + 15% Al, τcSch = 5 MPa

cumulative frequency

1.0 c) 1.0 d)

0.5 0.5

0.0 0.0
0.1 0.2 0.3 0.4 0.5 0.2 0.4 0.6 0.8 1.0 1.2
ΔE act (eV) ΔE act (eV)

Figure 4.13: Testing the validity of the weakest-link scaling relation (Eq. 4.12); lines: cumulative
distributions of ΔE act at 5 or 50 MPa Schmid stress on the cross-slip plane, in different alloys
and for different lengths Nb ; red: distribution for Nb = 500; green: distributions for Nb = 5000
and Nb = 10000; red and green data were calculated with the random walk model; dashed
lines: predictions using Eq. 4.12 for Nb = 5000 and Nb = 10000, based on the data for Nb = 500

P (ΔE act , N ref , τ ) → P (ΔE act − ΔE shift , N ref , 0) with

⎧     
⎪
⎪ f −1 ΔE act,avg (τ ) fc + 1
⎨ c
⎪ ΔE act,avg + − ΔE act ΔE act < ΔE act,avg
ΔE shift = 2 ΔE act,avg 2 , (4.13)
⎪
⎪
⎪ ΔE act,avg (τ ) − ΔE act,avg ΔE act ≥ΔE act,avg
⎩
where ΔE act,avg (τ ) is the average-alloy activation energy at stress τ , i.e. the maximum of
E det . The first term for ΔE act < ΔE act,avg is the contribution from the change of energy of
the constriction on the glide plane, which applies to all paths, and the second term is the
contribution from the work on the cross-slip plane, which has only limited effect for paths with
low ΔE act , as explained above. Note that paths with ΔE act ≥ ΔE act,avg see the full stress effect.

Using the activation energy distribution of 40b long lines as reference data, we predicted
the corresponding activation energy distributions at a dislocation length of 1000b under
Schmid or Escaig stresses of 50, 200 and 600 MPa. Note that Nb = 1000 corresponds to a
dislocation density of roughly 1013 m−2 in the alloys considered here, typical of the range
during deformation of engineering alloys. The highest stress level is not necessarily relevant
for real alloys. For example, reported stresses in compression tests with Al+3 wt.% Mg at room
temperature are below 300 MPa [75]. Cu-Ni single crystals deform in room-temperature
tensile tests under resolved shear stresses not higher than 140 MPa [93]. Nevertheless, it is

63
Chapter 4. Cross-Slip of Long Dislocations

Cu + 33% Ni, τcEsc = 5 MPa Cu + 33% Ni, τcEsc = 50 MPa

cumulative frequency
1.0 a) 1.0 b)

0.5 0.5

0.0 0.0
0.75 1.00 1.25 1.50 0.75 1.00 1.25 1.50

Ni + 15% Al, τcEsc = 5 MPa Ni + 15% Al, τcEsc = 50 MPa

cumulative frequency

1.0 c) 1.0 d)

0.5 0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 1.2 0.2 0.4 0.6 0.8 1.0 1.2
ΔE act (eV) ΔE act (eV)

Figure 4.14: Cumulative distributions of ΔE act at 5 or 50 MPa Escaig stress on the cross-slip
plane, in different alloys and for different lengths Nb ; red: distribution for Nb = 500; green:
distributions for Nb = 5000 and Nb = 10000; red and green data were calculated with the
random walk model; dashed lines: predictions using Eq. 4.12 for Nb = 5000 and Nb = 10000,
based on the data for Nb = 500

interesting to check the validity of our approach in the limit of very high stress.

In the following, we discuss predictions for Cu+33 at.% Ni, Al+06 at.% Mg and Ni+15 at.% Al.
Note that these alloys have increasing ratios of σ [ΔE end ] /ΔE act,avg , meaning that random
fluctuations become more important relative to the deterministic energy. In Fig. 4.15, the
predictions for Cu+33 at.% Ni and Nb = 1000b are compared to the results of full random walk
calculations and the average-alloy activation energies that one would expect based on E det .
The different stress types τcrossEsc
, τcross
Sch
and τglide
Esc
are considered separately in subfigures a), b)
and c). We see that the predictions (dashed lines) are close to the direct results (solid lines) at
all stress levels (colors). In some cases, for example at τcross
Esc
= 200 MPa, ΔE act is overestimated
by roughly 0.1 eV. Furthermore, the predicted distributions tend to be slightly broader at
200 MPa and 600 MPa. However, these errors are small compared to the error one would make
by using only the average-alloy activation energy (shown as the cross symbols in the figures).
At the median value, for example, the error of the average alloy is typically four to five times
larger than the error of the weakest-link prediction. When predicting cross-slip rates, this error
would be further amplified by the exponential weighting of ΔE act in Eq. 1.7.

In Al+06 at.% Mg, the predictions are satisfactory as well, see Fig. 4.16. The error is typically
less than 0.1 eV. At τglide
Esc
= 50 MPa and τcross
Esc
= 50 MPa, the highest 10–15% of energies are

64
 4.4. Cross-Slip of Long Dislocations

underestimated, because some high barriers have not been eliminated by the stress in the
direct calculations. However, the other 90–85% of energies are predicted accurately, and it are
the these lower energies which are more important, due to the exponential weighting in Eq.
1.7.

The mismatch between prediction and direct calculations at 50 MPa Escaig stress is even
stronger in Ni+15 at.% Al, see Fig. 4.17. ΔE act is underestimated above the median value.
However, this error is not surprising, since 50 MPa is just above the limit stresses of
τglide
Esc
= 48 MPa and τcross
Esc
= 46 MPa according to Eq. 4.11. The prediction is nevertheless useful,
since it indicates that a significant number of paths have an activation energy of around 0.7 eV,
hence thermally activated cross-slip is plausible. Based on the average-alloy value of 1.89 eV,
one would perhaps rule it out. At the other stress levels, the prediction is much more accurate.

1.0 a) τcEsc 1.0 b) τcSch

cumulative frequency

cumulative frequency
0.5 0.5

0.0 0.0
0.0 0.4 0.8 1.2 1.6 0.0 0.4 0.8 1.2 1.6
ΔE act (eV) ΔE act (eV)
1.0 c) τgEsc
cumulative frequency

Cu + 33% Ni
RW with Nb = 1000
predicted with Nb = 40
0.5
ΔE act,avg at τ

50 MPa
0.0 colors: 200 MPa
0.0 0.4 0.8 1.2 1.6 600 MPa
ΔE act (eV)

Figure 4.15: Cumulative distributions of cross-slip activation energies in Cu+33 at.% Ni

under stress; dashed lines are predictions made with Eq. 4.12, based on shifted zero-stress
distributions for Nb = 40

65
Chapter 4. Cross-Slip of Long Dislocations

1.0 a) τcEsc 1.0 b) τcSch

cumulative frequency

cumulative frequency
0.5 0.5

0.0 0.0
−0.2 0.0 0.2 0.4 0.6 −0.2 0.0 0.2 0.4 0.6
ΔE act (eV) ΔE act (eV)
1.0 c) τgEsc
cumulative frequency

Al + 6% Mg
RW with Nb = 1000
predicted with Nb = 40
0.5
ΔE act,avg at τ

50 MPa
0.0 colors: 200 MPa
−0.2 0.0 0.2 0.4 0.6 600 MPa
ΔE act (eV)

Figure 4.16: Cumulative distributions of ΔE act in Al+06 at.% Mg under stress; dashed lines are
predictions made with Eq. 4.12, based on shifted zero-stress distributions for Nb = 40

1.0 a) τcEsc 1.0 b) τcSch

cumulative frequency

cumulative frequency

0.5 0.5

0.0 0.0
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
ΔE act (eV) ΔE act (eV)
1.0 c) τgEsc
cumulative frequency

Ni + 15% Al
RW with Nb = 1000
predicted with Nb = 40
0.5
ΔE act,avg at τ

50 MPa
0.0 colors: 200 MPa
0.0 0.5 1.0 1.5 600 MPa
ΔE act (eV)

Figure 4.17: Cumulative distributions of ΔE act in Ni+15 at.% Al under stress; dashed lines are
predictions made with Eq. 4.12, based on shifted zero-stress distributions for Nb = 40. The
simulated data (solid lines) shows that a 50 MPa Escaig stress is not sufficient to eliminate all
high energy barriers. The prediction according to Eq. 4.12 cannot predict high barriers, hence
there is a discrepancy.

66
 4.5. Discussion

4.5 Discussion
We have demonstrated that (i) the standard deviation σ [ΔE end ] of random energy fluctuations
caused by cross-slip in random alloys can be predicted accurately based on solute/dislocation
and solute/solute interaction energies; (ii) the random walk model captures the accumulation
of energy fluctuations during cross-slip over long lengths, such that (iii) the distribution of the
cross-slip activation energies ΔE act can be computed for any dislocation length, including the
range of practical interest (~103b ), where direct atomistic calculations are infeasible. We find
that such long dislocations frequently have very high ΔE act , on the order of several electron
volts, but that moderate stresses (100 –101 MPa), especially Schmid stresses, eliminate these
barriers and lead to activation barriers that are much lower than the average value. We have
then shown that cross-slip becomes a weakest-link problem, enabling computation of barriers
for long lengths in terms of barriers computed at shorter lengths. We now discuss implications
of these results and some further aspects of our model.

First of all, complete cross-slip of long dislocations (103b ) is unlikely at zero stress, because of
the increasing frequency of high activation energies. Due to the exponential weighting of ΔE act
in Eq. 1.7, these barriers are exceedingly unlikely to be overcome by thermal activation alone.
Shorter subsegments could perhaps cross-slip, but equilibrium lengths of such segments are
expected to be on the order of 102 –103b , see App. A.5, and are therefore unlikely to be reached.
However, when these high barriers are eliminated by an external Escaig or Schmid stress, then
ΔE act is typically much lower than the value ΔE act,avg (τ ) that one would expect based on
an average-alloy model. Thus, cross-slip is much faster than expected. For example, in
Cu+33 at.% Ni with a 50 MPa Escaig stress on the glide plane, the median activation energy for
a 1000b long dislocation is 1.14 eV, whereas the average-alloy estimate is 1.59 eV. Assuming
constant prefactors in Eq. 1.7, the rate of cross-slip in the random alloy is roughly 7 orders of
magnitude faster than expected based on an average-alloy model. In Al+02 at.% Mg and
Ni+15 at.% Al, the rate is increased by ca. 4 and 16 orders of magnitude, respectively.

Under stress, cross-slip becomes a weakest-link problem because the cross-slipped segment
expands spontaneously after nucleation. The distribution of ΔE act depends on (i) the
distribution of ΔE act for cross-slip nucleation, which takes place over a length much shorter
than the total length, and (ii) how many times this distribution is sampled over the total length.
We observed that even small Schmid stresses on the cross-slip plane, on the order of a few
MPa, eliminate high barriers and make cross-slip a weakest-link problem. Thus, cross-slip of
moving dislocations is likely always a weakest-link problem, since it implies a Schmid stress
τglide
Sch
on the glide plane. Except in high symmetry cases, the applied stress will likely have
nonzero Schmid components on both planes, hence the conditions for weakest-link scaling
are likely to be fulfilled. The observed critical Schmid stresses for weakest-link scaling are on
the order of a few MPa, and therefore comparable to the Peierls stress in pure metals.

In alloys, one additionally has to consider solute strengthening. Recall that we have neglected
energy fluctuations during bowing-out, i.e. we have neglected solute strengthening on the
cross-slip plane. However, if we assume that the external stress is sufficiently high to make
dislocations move on the glide plane, then it is, as before, likely that critical stresses on the
cross-slip plane are sufficient, so that a bow-out can be formed.

67
Chapter 4. Cross-Slip of Long Dislocations

Internal stresses have to be considered as well. For example, the critical stress for weakest-link
scaling can be overcome easily in a pile-up of n dislocations against an obstacle, where the
external stress is amplified by a factor equal n [18]. Pile-ups were proposed early on as likely
sites for cross-slip [125]. Dislocations that do not reside in pile-ups nevertheless experience
stresses from other dislocations, e.g. forest dislocations. A particularly interesting scenario is
cross-slip of a dislocation in a screw dipole, which can lead to annihilation (see Refs. [108, 111,
145] for atomistic simulations of this process). Here, one would expect large attractive stresses.

The easier/faster cross-slip implied by our results here would have several consequences for
plastic deformation behavior. Screw dipole annihilation plays a role in fatigue and may
determine the saturation stress [11], hence our observations may help to interpret fatigue of
alloys. Due to the faster annihilation by cross-slip one should expect larger minimum stable
dipole heights as in pure metals, and hence lower saturation stresses. Discrete [54, 77] and
continuum dislocation dynamics calculations [157–159] have demonstrated the importance of
cross-slip for dislocation microstructure formation. A cellular microstructure is only observed
if cross-slip is activated in such calculations. Thus, one would expect to see a well-developed
cellular microstructure in deformed random alloys. Moreover, this microstructure should form
more rapidly with strain than in pure metals. Considering the results of Xia and coworkers
[158], one would also expect higher yield points and an increased hardening rate. However, it
might be difficult to distinguish these effects from other solute strengthening effects.

The case of Ni+15 at.% shows that low values of ΔE act can be realized even in alloys with high
ΔE act,avg if σ [ΔE end ] is high and moderate stresses are applied. In High Entropy Alloys [162],
where σ [ΔE end ] can be expected to be high due to compositional disorder, cross-slip should
therefore be relatively easy. Indeed, Rao et al. [105] recently observed spontaneous cross-slip
in atomistic simulations of regular dislocation slip in a Co40 Fe16.67 Ni36.67 Ti16.67 random alloy at
300 K, even though ΔE act,avg is 4.6–4.9 eV. The effective value of ΔE act was estimated to be
0.3 eV. Liu et al. [73] observed that twins in Al0.1 CoCrFeNi were formed by a mechanism which
involves cross-slip of dislocations that were piled up against a stacking fault. These studies
show the potential importance of cross-slip in HEAs. Somewhat conflicting in this sense is the
study of Cao et al. [15], who performed creep tests with a Alx CoCrFeNi (x=0.15, 0.60) HEA and
determined activation volumes of 32–52b 3 , which are consistent with a cross-slip mechanism
[102], but also high activation energies (3.5–4 eV). Furthermore, Otto and coworkers [94]
observed planar slip in CoCrFeMnNi, which seems to be in conflict with our suggestion of
easy cross-slip. The authors attributed planar slip to short-range order (SRO). Gerold and
Karnthaler [44] argued that SRO causes planar slip because leading dislocations in a group
destroy this order, resulting in glide plane softening and a preference for further slip on this
plane. Thus, it seems, that SRO would not increase the energy barrier for cross-slip per se. The
exact effect of SRO on ΔE act is unclear, but one may speculate that correlations in the solute
distributions would reduce σ [ΔE act ]. Recall that we assumed a completely random solute
distribution. Our model also contradicts a model of planar slip by Hong and Laird [53].
They argued that solutes create a friction stress on the partial dislocations, which impedes
constriction and hence cross-slip. The friction stress increases with atomic size misfit and
solute content, hence it would seem that solute addition should generally impede cross-slip.
However, we have shown that one needs to consider solute binding energy fluctuations, which
can be positive or negative and thus impede or facilitate cross-slip.

68
 4.5. Discussion

A few other aspects of our model merit discussion. Some approximations were made when
modeling the deterministic energy E det (i ) (i = 1. . .Nb ). The current model is appealing
because of its simplicity. The energy profile is reduced to two variables, the energy of a single
constriction and the cross-slipped length at which the constrictions on the cross-slip plane are
approximately fully formed. Both can be determined using atomistic or elastic models.
We obtained a satisfactory agreement between E det and the atomistic data by choosing a
formation length of 10b . Indeed, the relative energy has almost reached its final value of
ΔE act,avg at this point, see Fig. 4.5. Strictly, one would have to choose a value of 20b , i.e. half of
the total cell length of 40b , which we identified in our previous publication [88] as the
periodic length in Ni at which ΔE act,avg becomes independent of length. However, it is easy to
see that the quadratic energy profile would then underestimate the relative energy more
strongly in the region of constriction formation. Better approximations of the energy-length
profile could be constructed using a different functional form. For example, in Püschl
and Schöck’s anisotropic-elastic model, the energy of the two half-constrictions on the
cross-slip plane varies with cross-slipped length L and dissociation width d approximately as
2/π arctan(pL /d )2Ec , where 2Ec is the energy of two well-separated half-constrictions, and p
is a numerical factor in the range 0.8–1.4. The model of E det under stress could be improved by
including the effect of τcross
Esc
on the constriction on the cross-slip plane. This could be done by
simultaneously scaling the length and the energy of this constriction.

We have also assumed that the ideal Friedel-Escaig mechanism is the operative cross-slip
mechanism. This means that the segments on the glide and cross-slip plane are joined by
point constrictions. Oren et al. [92] have recently observed a variation of this mechanism in
molecular dynamics calculations, where fully constricted segments are formed in the process.
Analysis of our atomistic configurations (at zero stress) with the Dislocation Analysis (DXA)
algorithm implemented in OVITO [135, 136], shows that such segments occur frequently. In
random alloys, lengths of roughly 3b are frequently seen before the transition state. At the
transition state, however, the fully constricted length is typically only around 1b . Since the
length at the transition state is typically short, it is a fair assumption to use the activation
energy of ideal Friedel-Escaig cross-slip for ΔE act,avg .

Three more subtle assumptions were made when modeling the random energy. First, by
assigning a random energy change to 1b segments, we associate solute-dislocation binding
energies with such short segments. In reality, however, the binding energy quantifies the
interaction between the solute and an infinitely long dislocation. The energy has contributions
from atoms whose distance to the solute atom along the dislocation line is greater than
1b . However, the interaction energy drops of rather quickly, due to the 1/r 2 scaling of the
solute stress field. Moreover, a sufficiently long line is realized after cross-slip of a few 1b
segments. These are the likely reasons why our approximation does not have a significant
effect on the results. Another approximation was made in the derivation of σ [ΔE end ]. The
contributions to ΔE end from solute-dislocation and solute-solute binding are sums of energies
times occupation variables, which are Bernoulli random variables. Hence, these contributions
are discrete random variables. Here, we assume that they are normally distributed, even for 1b
segments. In the case of the contribution from solute-solute binding, this approximation is
expected to deteriorate when the number of contributing pairs is low, in a similar way as the
approximation of a binomial distribution B (n , p ) with a normal distribution deteriorates at
low n and p . The slight overestimation of σ [ΔE act ] of Ni+02 at.% Al is perhaps an indication of

69
Chapter 4. Cross-Slip of Long Dislocations

this problem. Furthermore, we have calculated solute-dislocation binding energies using the
dislocation core structure at zero stress. Thus, our model of σ [ΔE end,s-d ] does not account for
a change of dissociation width d under stress. Since the highest contributions to ΔE end,s-d
come from the immediate core region of the partial dislocations, the error of approximation
should be small, but one must be careful in alloys where there is significant interaction
between solutes and the stacking fault, and at high Escaig stresses. Similarly, σ [ΔE end,s-s ]
was calculated using the zero-stress value of d . However, here one could simply use a
stress-dependent value.

70
5 Summary

In this thesis, dislocation cross-slip in Face-Centered Cubic solid solutions has been studied
using a combination of atomistic simulations and statistical modeling. Atomistic calculations
of cross-slip transition paths in various Ni-Al, Al-Mg and Cu-Ni solutions have shown that
cross-slip is controlled by solute fluctuations, i.e. the activation energy ΔE act for cross-slip is a
stochastic variable with a large variance. The change of the mean value with concentration is
small compared to the variance. Therefore, it is not sufficient to consider only average alloying
effects, e.g. a change of the mean stacking fault energy, when estimating the propensity for
cross-slip in alloys. At moderate stresses, typical values of ΔE act are much lower than the
average value ΔE act,avg , hence cross-slip is much faster than expected.

Two statistical models of the activation energy distribution have been presented. For short
(40b ) dislocations, a linear correlation ΔE act ≈ ΔE act,avg + 1/2ΔE end holds, where ΔE act,avg is
the activation energy in the average alloy, and ΔE end is the energy difference between the fully
dissociated dislocation on the glide and cross-slip plane, respectively. Using this correlation, it
is possible to estimate the distribution of ΔE act based on the distribution of ΔE end , which is
approximately normal with mean zero and standard deviation σ [ΔE end ]. The origin of ΔE end
are changes in solute-dislocation and solute-solute binding energies. The latter comes from
cutting or formation of solute pairs, which occurs when the stacking fault of the dislocation is
destroyed on the glide plane and re-generated on the cross-slip plane. In the first part of the
thesis, we presented a parameter-free, analytical model of σ [ΔE end ], which, however, is
limited to first nearest neighbor pairs of solutes. Nevertheless, the atomistic and analytical
calculations indicate that ΔE act for cross-slip nucleation can be much lower than expected
based on average-alloy models.

The activation energy for cross-slip of dislocations with a length of hundreds or thousands of
Burgers vectors was discussed in the second part of the thesis. At such lengths, transition
path calculations become cumbersome or infeasible. Therefore, a random walk model was
developed, which allows to predict the activation energy distribution based on knowledge
of solute binding energies, the activation energy ΔE act and constriction size Nc in the
corresponding average alloy, and basic material properties like the elastic constants and the
stacking fault energy. Furthermore, the model accounts for stress effects. The dislocation is
partitioned into 1b long segments, and cross-slip is the discrete process of moving them one
by one from the glide plane to the cross-slip plane. Each step causes a (i) deterministic energy

71
Chapter 5. Summary

change, due to constriction formation and the work of an external stress; and (ii) a random
change of ΔE end . The deterministic energy depends on ΔE act,avg and the typical length of the
double-constriction that is formed in the process. The work of the external stress can be
calculated based on zero-stress material parameters. An extended model for the random
fluctuations in ΔE end was presented, which accounts for first to eighth nearest neighbor pairs
of solutes and correlations between loss and formation of pairs of different order. The random
walk model accurately predicts the distributions of ΔE act for 40b long dislocations and the
correlation between ΔE act and ΔE end that was seen in this case. Cross-slip of long dislocations
(102 –103 b) is unlikely at zero stress. High values of ΔE act , on the order of several electron
volt, become more frequent with increasing length, whereas the lowest observed activation
energies change only little. High values of ΔE act are associated with cross-slip paths where the
maximum occurs late, and these barrier can be eliminated by an applied stress. Thus,
cross-slip becomes a weakest link problem at sufficiently high stresses. The governing energy
barriers are those for nucleation of a short cross-slipped segment, hence the distribution of
ΔE act for a long dislocation can be obtained by scaling the distribution for a shorter one. At
sufficiently high stresses, the distribution may even be inferred from the distribution for a 40b
long segment at zero stress.

72
6 Outlook

The current study has shed light on important aspects of dislocation cross-slip in FCC solid
solution alloys and enabled prediction of the corresponding activation energies ΔE act at
different levels of approximation. Our results may help other researchers to interpret
simulations and experiments where unexpected cross-slip is observed. For example, Rao et al.
[105] recently observed cross-slip in a molecular dynamics simulation of regular slip in a High
Entropy Alloy (HEA). According to average-alloy models, cross-slip is exceedingly unlikely
because ΔE act,avg is 4.6 eV–4.9 eV. Rao et al. explained the low effective barriers with solute
concentration fluctuations, in agreement with the results of this thesis.

Considering the observation of Rao et al. , it would be interesting to estimate the range of ΔE act
for HEAs that are discussed in the literature. In general, σ [ΔE end ] should be large in such
highly concentrated multi-component alloys. Accordingly, σ [ΔE act ] should be large as well. A
challenge is that interatomic potentials for these alloys are typically not available. However,
ΔE act,avg can be estimated using anisotropic-elastic models, and the solute binding energies
that enter in our model of σ [ΔE end ] can be calculated with density functional theory, see for
example Refs. [71, 72, 161]. Thus, it is in principle possible to calculate the range of ΔE act with
the random walk model. The results would help to understand trends in dislocation network
formation in this class of alloys.

Moreover, we expect that the insights from this thesis will help to improve models of plastic
deformation of FCC alloys. For example, our results may help to improve the description of
cross-slip in Discrete Dislocation Dynamics (DDD). In the following, we outline how the
existing approach for pure metals could be modified in order to enable simulation of solid
solution alloys and HEAs. In the approach of Kubin et al. [67], screw dislocation segments can
cross-slip on each timestep with a probability

  
l τμ − τIII Vact
P step = A δt exp , (6.1)
l0 kB T

where A is a numerical coefficient, l the segment length, l 0 a reference length, δt the timestep,
τμ the stress acting on the dislocation, τIII the critical resolved shear stress at the onset of
stage III, and V the activation volume of cross-slip. Note that without the δt this equation
corresponds to the rate of cross-slip. In the approach of Hussein et al. [54], the rate is

73
Chapter 6. Outlook

calculated as
  
l  − E c − Vact τ Esc
glide
− τ Esc
cross "
r = ωA exp #, (6.2)
l0 kB T #
! $
where ωA is an attempt frequency. A segment cross-slips if r >= 1. Eq. 6.1 and Eq. 6.2
are appropriate for pure metals, where the activation energy for cross-slip is deterministic.
Furthermore, it is reasonable to assume that the total cross-slip rate depends on segment
length, since cross-slip can nucleate anywhere along the line with the same ΔE act , and
nucleation events are presumably uncorrelated if they occur far away from each other.
Analytical considerations [145] also indicate such a dependence. In random solid solutions the
situation is different and the equations need to be modified. ΔE act should be used in the
enumerator in the exponential. However, ΔE act is now a random variable. A new value of
ΔE act should be drawn for each screw segment whenever it moves, since the random solute
environment changes. The distribution of ΔE act , which depends on stress and dislocation
length, could be obtained by scaling a zero-stress reference distribution according to Eq. 4.12.
This reference distribution, in turn, needs to be calculated only once, using the random walk
model. A technical hurdle is to implement efficient scaling and sampling of the distributions.
The program must also check whether Eq. 4.11 is fulfilled1 , so that weakest-link scaling can be
used. Note that the probability now depends on length through ΔE act and Eq. 4.11. The
pre-exponential factor, on the other hand, should be independent of length. Due to the rules
for path selection in our model, there is a unique cross-slip path for each segment in its
current solute environment. Thus, the term (l /l 0 ) should be removed in Eq. 6.1 and Eq. 6.2.
The issue of the pre-exponential factor will be discussed further below.

In addition to DDD, our results may improve work hardening laws that account for the
evolution of the dislocation density during loading. Some models incorporate dislocation
annihilation by cross-slip of opposite dislocation [35, 62, 75, 87]. The rate of annihilation thus
depends on the rate of cross-slip. In a pure metal with a single ΔE act , both rates could
be written like in Eq. 1.7. In solid solutions, calculating the total cross-slip rate is more
complicated. Since ΔE act follows a distribution which depends on dislocation length, one
would have to do a double integral over length and energy to get the total rate of cross-slip. For
this purpose, it may be necessary to parameterize the reference distribution.

An important task for future work is to test our model experimentally. This is a challenging
task, since cross-slip occurs alongside other dislocation processes during deformation of real
materials. An indirect test would be to predict the stress-strain curves of alloys at different
compositions and temperatures using appropriately revised work hardening models, and to
compare the results to experimental measurements. This test is somewhat ambiguous, since
one would also test additional assumptions in the work hardening model. Nevertheless,
failure of these predictions would cast doubt on our model for the distribution of ΔE act . A
more direct test would be to observe individual dislocations and dislocation networks in
deformed solid solution alloys, using an electron microscope. The alloy should have a high
value of σ [ΔE end ], but also a high value of ΔE act,avg . The latter should be so high that thermally
activated cross-slip is unlikely if it is not assisted by binding energy fluctuations. If it can

1. “=” should then be replaced by ≤ in Eq. 4.11.

74
be ruled out that nucleation took place at a heterogeneity — which might be difficult or
impossible, especially for jog-assisted cross-slip — then observation of cross-slip in the bulk
would corroborate our suggestion that fluctuations in the binding energy may lower ΔE act .
Furthermore, observation of cross-slip traces on the surface and well-formed cell structures in
the interior would indicate that cross-slip is active [157, 158]. The alloy should not exhibit
short range order (SRO), which must be ruled out as a cause for slip planarity and absence of a
cell structure. Similarly, our model could be used to make a prediction about the transition
between cellular and planar dislocation structures as a function of alloy composition, which
could then be tested experimentally. For example, Hong and Laird [53] tried to explain
the transition in CuZn and CuAl with their cross-slip model. Finally, it might be useful
to determine the apparent activation energy and activation volume of cross-slip in solid
solutions using the technique of Bonneville and Escaig [9, 10].

Having discussed possible applications of the model and experimental validation, we now
turn to further aspects of cross-slip that should be adressed in future work. First of all, the
pre-exponential factor of the cross-slip rate (A in Eq. 1.7) should be studied. This factor can be
interpreted as an attempt frequency for cross-slip. It would be desirable to have an analytical
solution for this frequency in solid solution alloys. Vegge et al. [145] derived an expression
for pure metals (Eq. 1.9). However, their solution is invalid here, since it is based on the
assumption that a displacement of the critical nucleus along the line direction costs zero
energy. In a solid solution, there would be a random change of binding energies, hence this
assumption is violated. Thus, an analytical solution for the pre-exponential factor in solid
solutions is an open problem. Furthermore, the validity of harmonic transition state theory
(HTST) for describing the rate of cross-slip should be examined. There could be anharmonic
effects, for example from thermal expansion and softening. They have been shown to increase
the rate of dislocation nucleation [116] and glide through a solute field [118], respectively. One
could perhaps improve the rate estimate by accounting for the temperature dependence of the
material properties that appear in our model, for example the elastic constants and the
stacking fault energy. Furthermore, it should be checked whether A and ΔE act are correlated
according to the Meyer-Neldel [80] compensation rule. Low values of ΔE act would then be
compensated by low values of A . One may speculate that there is a small compensation
effect due to the correlation between nucleus size and activation energy. We have seen that
cross-slip paths with low ΔE act tend to be those where the maximum occurs early on, at a short
cross-slipped length. The vibrational modes at the transition state should depend on the size
of the critical nucleus, which would then establish a link between the attempt frequency and
ΔE act . However, this effect is likely small, because the size of the critical nucleus does not vary
by orders of magnitude at moderate stresses. To address the question, the cross-slip rate
should be calculated directly using atomistic simulations, for example with the method of
Vegge et al. [145] and Oren et al. [92]. They subjected a screw dipole to stress and measured the
waiting time for annihilation by cross-slip at different temperatures. A and ΔE act can then be
obtained from an Arrhenius plot, i.e. a plot of ln r = A − ΔE act /kB T as a function of 1/T .
Alternatively, the rate could be estimated using the Finite Temperature String Method [28, 139]
or transition path sampling techniques like the Transition Interface Sampling Method [138].

Moreover, the effects of short range order (SRO) on the distribution of ΔE act need to be studied.
As speculated in Sec. 4.5, SRO could cause a reduction of σ [ΔE end ], thus ΔE act would be more
narrowly distributed around ΔE act,avg , and cross-slip would become less frequent. This

75
Chapter 6. Outlook

supposition could be tested by calculating minimum energy paths for cross-slip in solute
environments with SRO. The simulations would be very similar to those described in Sec. 2.3,
except that correlations would have to be introduced in the solute distributions. However,
deriving an analytical model would be more difficult. Occupation variables for different sites
around the dislocation are now correlated, hence products of these variables cannot simply be
replaced by powers of solute concentrations when taking averages.

In addition to SRO, it would be useful to study the effect of solute segregation. It is likely that
ΔE act increases, since the cross-slipping dislocation leaves a favorable solute environment.
Such an effect is seen in the study of Andrews et al. [2], who calculated the energy for
constriction formation in a segregated solute field using a line tension model. The constriction
energy increases with solute content. It would be interesting to supplement this study with
atomistic minimum energy path calculations of cross-slip in a segregated alloy. Moreover, it
would be interesting to study the effect of segregation on cross-slip at heterogeneities, for
example at intersections with forest dislocations. Segregation should be stronger around
immobile dislocations, thus cross-slip at intersections could potentially be more affected than
cross-slip of mobile dislocations in the bulk.

Another point for further study is the role of other point defects like interstitials and vacancies.
In the case of interstitials, one would expect a stronger interaction with the cross-slipping
dislocation due to stronger distortion of the lattice by the interstitial. On the other hand, the
concentration of interstitials is typically lower. For example, the maximum concentration
of C in austenitic Fe is about 2%. Thus, the strength of interstitial effects on cross-slip
is not immediately clear. Like in the case of SRO and segregation, the first step towards
understanding this effect would be to perform minimum energy path calculations with
corresponding alloys. In particular, it would be interesting to perform calculations with
hydrogen atoms, which are very mobile and can therefore segregate to the dislocation on the
glide plane. For the reasons outlined above, it is likely that they impede cross-slip. The effect of
vacancies, on the other hand, can perhaps be studied using a modified version of the solute
binding energy model that was used here, where vacancies are effectively treated like solute
atoms. Like in the case of solute atoms, segregation would complicate the analysis.

In summary, there are numerous points for future research. The results of this thesis could
be used to improve models of FCC strength. Moreover, our model should be tested
experimentally and the pre-exponential factor of the cross-slip rate should be studied. Finally,
it would be interesting to study the effects of other point defects, SRO, segregation, as well as
solute effects on heterogeneous cross-slip.

76
A Appendix

A.1 Calculation of Basic Properties

Basic properties of average and true random alloys of Ni-Al, Cu-Ni and Al-Mg are listed in
Tab. A.1, Tab. A.2 and Tab. A.3. Fig. A.1 shows the relative differences between average alloy and
random alloy values, except for the difference in lattice parameter, which is smaller than 0.1%
in all cases. The methods for calculating the properties are explained in Sec. A.1.1 and
Sec. A.1.2.

Table A.1: Basic properties of average and true random alloys of Ni-Al; solute concentration c
in at.%; screw dislocation dissociation width d in Å; intrinsic stacking fault energy γSF in
mJ m−2 ; cubic elastic constants C 11 , C 12 and C 44 in GPa; A/R: average/true random alloy value

a d γSF C 11 C 12 C 44
c(Al) A R A R A R A R A R A R
0 – 3.520 – 10 – 134.4 – 241.3 – 150.8 – 127.3
2 3.521 3.521 10 10 127.0 126.6 237.0 236.9 150.8 151.1 127.1 127.1
4 3.522 3.522 10 10 119.4 120.0 233.2 232.3 150.9 151.3 126.7 126.7
8 3.526 3.525 10 11 103.6 112.6 227.4 223.7 151.0 151.8 125.7 125.7
10 3.529 3.527 10 11 95.6 100.6 225.5 219.6 151.0 152.0 125.1 125.1
12 3.532 3.530 10 11 87.7 93.9 224.5 216.0 151.0 152.1 124.4 124.4
15 3.538 3.535 12 11 75.8 77.7 224.2 211.3 150.8 152.2 123.4 123.2

A.1.1 Average Alloys

The average-alloy lattice parameter was determined by locating the minimum of the cohesive
energy curve for a single FCC unit cell with periodic boundary conditions in all directions. For
this purpose, the Sequential Least Squares Programming algorithm in PYTHON-SCIPY [58] was
used. The potential energy was calculated at each iteration of the minimizer by calling
LAMMPS through its Python interface.

The cubic elastic constants C 11 , C 12 and C 44 were calculated using the Elastic script that is
shipped with LAMMPS. The configuration was 5 × 5 × 5 FCC unit cells large and periodic
boundary were applied in all directions. The script calculates all 21 elastic constants

77
Appendix A. Appendix

Table A.2: Basic properties of average and true random alloys of Cu-Ni; solute concentration c
in at.%; screw dislocation dissociation width d in Å; intrinsic stacking fault energy γSF in
mJ m−2 ; cubic elastic constants C 11 , C 12 and C 44 in GPa; A/R: average/true random alloy value

a d γSF C 11 C 12 C 44
c(Al) A R A R A R A R A R A R
0 – 3.615 – 14 – 44.8 – 170.2 – 122.6 – 75.8
10 3.612 3.612 14 13 48.5 49.0 175.2 175.5 123.7 124.1 80.0 80.1
22 3.607 3.606 13 13 54.5 56.1 182.2 182.7 125.7 126.3 85.3 85.3
33 3.599 3.598 13 12 61.5 64.5 189.4 189.9 127.9 128.5 90.4 90.4
68 3.566 3.564 10 11 93.2 97.6 227.6 227.0 148.4 148.3 107.3 107.0
79 3.552 3.551 10 10 105.3 108.7 234.3 234.3 148.9 149.5 112.7 112.5
90 3.537 3.536 10 10 116.9 118.6 240.9 240.8 148.8 149.1 118.9 118.7
100 – 3.520 – 10 – 125.6 – 247.0 – 148.0 – 125.5

Table A.3: Basic properties of average and true random alloys of Al-Mg; solute concentration c
in at.%; screw dislocation dissociation width d in Å; intrinsic stacking fault energy γSF in
mJ m−2 ; cubic elastic constants C 11 , C 12 and C 44 in GPa; A/R: average/true random alloy value

a d γSF C 11 C 12 C 44
c(Al) A R A R A R A R A R A R
0 – 4.032 – 11 – 101.3 – 118.0 – 62.2 – 36.7
2 4.042 4.041 10 10 95.4 97.6 114.5 114.2 60.4 61.2 34.4 34.2
6 4.062 4.060 10 9 86.7 89.2 106.3 105.6 57.0 58.1 29.8 29.5
10 4.083 4.080 10 9 81.1 85.8 99.4 98.3 54.1 55.2 25.5 25.4
14 4.105 4.101 9 9 78.9 85.8 94.1 93.1 51.8 53.0 21.6 22.1
18 4.127 4.123 9 9 78.9 88.4 91.0 89.7 50.4 51.8 18.4 19.9
22 4.149 4.145 9 9 81.4 89.9 92.7 87.5 51.3 51.1 17.2 18.6

for a general anisotropic material by applying appropriate strains and measuring the
resulting stresses. C 11 , C 12 and C 44 were calculated as the mean values of the corresponding
symmetry-equivalent components.

The average screw dislocation splitting width d was determined using the Dislocation
Extraction Algorithm (DXA) [136] as implemented in OVITO [135], by taking the average
distance between the extracted partial core coordinates in the average-alloy configurations
from Ch. 3.

The intrinsic stacking fault energy γSF was calculated as follows. The basic configuration was a
rectangular cuboid which was elongated in z direction. The x , y and z -directions of the
laboratory coordinate system were parallel to the [110], [112] and [111] directions of the
crystal. The length of the configuration in x and y was more√ than eight times the cutoff radius
of the interatomic potential. The length along z was 24 3a . Periodic boundary conditions
were applied along x and y , and free boundary conditions along z . In a first step, the normal
stresses along x and y were reduced below 0.5 bar using the LAMMPS command fix boxrelax
in conjunction with the CG minimizer. The box relaxation was reset every 1000 steps of the
minimizer, but at most 50 times. After pressure relaxation, the energy of the system was

78
 A.1. Calculation of Basic Properties

difference with respect to random alloy (%)

difference with respect to random alloy (%)

a) Ni-Al 12 b) Cu-Ni
6
10
4
8
2
6
0
4
−2
−4 2

−6 0

−8 −2
−10 −4
−12 −6
−14 −8
2 4 8 10 12 15 10 22 33 68 79 90
Al concentration (at.%) Ni concentration (at.%)
8
difference with respect to random alloy (%)

c) Al-Mg
6

2 d
0 γSF
C 11
−2
C 12
−4 C 44
−6
−8
−10
2 6 10 14 18 22
Mg concentration (at.%)

Figure A.1: Difference between the properties of average alloy and the corresponding true
random alloys, relative to the value of the true random alloy; a) Ni-Al alloys; b) Cu-Ni alloys; c)
Al-Mg alloys

minimized at fixed box size using FIRE. Iterations were stopped when the two-norm of the
−1 −1
3N -dimensional vector of atomic forces F fell below 1 × 10−6 eV Å or 1 × 10−8 eV Å , or
after 20000 iterations. A stacking fault was generated by displacing the upper half (along z ) of
the configuration by a /6[211]. Typically, the displacements were applied incrementally to
allow simultaneous estimation of the unstable stacking fault energy. Thereafter, the energy of
the system was minimized using FIRE. The atoms were constrained to move along the z
direction only by setting the x - and y -components of their force vectors to zero.

79
Appendix A. Appendix

A.1.2 Random Alloys

The configuration for calculating the random alloy lattice parameter was a 30 × 30 × 30 FCC
unit cells large cube, which was prepared using the lattice parameter of the corresponding
average alloy. The atom types were replaced randomly by the corresponding true atom
types according to the desired concentrations. Three realizations were prepared for each
concentration. To determine the mean lattice lattice parameter of the random alloy, the
pressure was relaxed independently along x , y and z using the LAMMPS command fix
boxrelax in conjunction with CG minimization. The mean lattice parameter is then the mean
edge length of the cube divided by 30. The pressure tolerance was 0.5 bar. However, this
threshold could not always be reached. In the case of Al-Mg alloys, the largest residual
pressure was 10.4 bar, in one calculation with Al+18 at.% Mg. At all other concentration levels,
the mean absolute residual pressure of the three corresponding calculations was less than
1.6 bar. In the case of Ni-Al and Cu-Ni, the mean absolute pressure was less than 3.3 bar and
0.5 bar, respectively.

The configurations from the lattice parameter calculations were also used for determining C 11 ,
C 12 and C 44 . This was done by imposing a strain and measuring the resulting stress, following
the description in [89]. For C 11 and C 12 , a uniaxial strain was applied along y . Switching from
Voigt to tensor notation, the response is

σ22 = C 2222 ε 22 ,
σ11 = C 1122 ε 22 , (A.1)
σ33 = C 2233 ε 22 .

Simulations were performed for six strain levels, compressive as well as tensile. The absolute
values of strain were | ε 22 | = 1 × 10−5 , 5 × 10−5 , 1 × 10−4 . After deforming the configuration,
the energy of the system was minimized using FIRE. The elastic constants in Eq. A.1 were
determined by linearly fitting the stress-strain data. C 12 was taken as the mean value of C 1122
and C 2233 . To determine C 44 , a simple shear γ12 was applied, which causes a stress

σ12 = C 1212 ε 12 . (A.2)

Positive and negative shears were applied, with | γ12 | = 1 × 10−5 , 5 × 10−5 , 1 × 10−4 . The
minimization that was performed in both types of calculations after each strain increment was
−1
stopped when the two-norm of the force vector fell below 1 × 10−6 eV Å . However, in the
−1
case of Al-Mg a rather high tolerance of 2 × 10−3 eV Å was used, hence the corresponding
results are more uncertain. Note that random alloy elastic constants where not used in any of
the calculations in Ch. 3 and Ch. 4. They are calculated here in order to be able to estimate the
error of the average-alloy approximation.

The splitting width d was determined using (DXA), like in the average alloys. The values listed
in the tables above are the mean values of the different true random alloy realizations with the
corresponding concentration.

Finally, γSF was calculated like in the case of the average alloys. The configuration was
constructed using the corresponding A-atom lattice parameter. Before relaxation of the
pressure in x and y , the configuration was converted into a true random alloy by assigning a

80
 A.2. Nudged Elastic Band Method

random distribution of real atom types according to the desired composition. In the case of
−1
Al-Mg, the force tolerance was increased to 1 × 10−4 eV Å , due to the convergence problems
discussed in Sec. 2.3.1.

A.2 Nudged Elastic Band Method

In the Nudged Elastic Band (NEB) [59, 81] method, the 3N -dimensional force vector Fi acting
on image i is calculated as the sum of components Fi and Fi⊥ parallel and perpendicular to the
transition path, respectively. Fi enforces that the images maintain a particular spacing. Fi⊥ is
the force derived from the potential energy minus the component parallel to the path. The
formulas for the three forces are
Fi = Fi⊥ + Fi ,
Fi = (| Ri +1 − Ri |ki − | Ri − Ri −1 |ki −1 ) τ̂i , (A.3)
⊥
Fi = −∇V (Ri ) + [∇V (Ri ) · τ̂i ] τ̂i ,

where | . . . | indicates the vector norm, Ri is the 3N -dimensional vector of atomic positions, ki
is a constant, ∇V (Ri ) is the gradient of the potential energy, and τi is the tangent to the path in
3N -dimension space. The hat indicates that it is normalized i.e. τ̂i = τi /| τi | . Note that Fi⊥
goes to zero if the images lie on the path. Fi goes to zero if the images achieve a specific
spacing, which depends on the constants ki . If a single constant k is used, i.e. ki = k ∀i , then
equidistant spacing of the images is enforced.

LAMMPS implements the so-called improved tangent estimate [47], where τi is either the
forward difference τ̂i+ = Ri +1 − Ri , the backward difference τ̂i− = Ri − Ri −1 , or a mixture
thereof, depending on the energies Vi ,
⎧
⎪ τ̂i+ Vi +1 > Vi > Vi −1
⎪
⎪
⎪
⎪
⎨ τ̂ −
⎪ Vi −1 > Vi > Vi +1
τi = i
, (A.4)
⎪
⎪
+
τ̂i ΔVimax
τ̂ − ΔV min
+ i Vi > Vi +1 > Vi −1 ∨ Vi +1 > Vi −1 > Vi
⎪
⎪ i
⎪
⎪ τ̂ ΔVi
+ min −
+ τ̂i ΔVi max
Vi > Vi −1 > Vi +1 ∨ Vi −1 > Vi +1 > Vi
⎩ i
where
Vi max = max (|Vi +1 − Vi | , |Vi −1 − Vi |),
(A.5)
Vi max = min (|Vi +1 − Vi | , |Vi −1 − Vi |).

A.3 Convergence in String Method Calculations

In the String method the minimum energy path between initial and final state is determined
iteratively. Ideally, the calculation converges to a unique path with vanishing perpendicular
forces. That is, for each image i , the 3N -dimensional vector Fi of atomic forces is parallel to
the path. In practice, the calculation may end before this state is reached. Recall that the
calculations in Ch. 3 and Ch. 4 were stopped when the Euclidean norm of the maximum
displacement of any image fell below 10−3 Å, or after 300 or 600 String iterations. In the
following, we check whether the calculations are converged by examining the change of

81
Appendix A. Appendix

energy during String iterations and the residual perpendicular forces. Finally, we check how
much the results depend on the calculation method.

Fig. A.2 shows the relative change of activation energy ΔE act per outer iteration1 in calculations
with Al+22 at.% Mg from Ch. 4. Towards the end of each calculation, ΔE act changes only by
fractions of a percent. In one case, a change of 0.5% occurs in the last iteration. In all other
String method calculations reported in Ch. 4, the change in the last iteration is smaller than
this value. Hence, it seems unlikely that continuing the String calculation for many more
iterations would yield a substantially different result.

Next, we examine the residual perpendicular forces. The perpendicular force of image i is

Fi⊥ = Fi − [Fi · τ̂i ] τ̂i , (A.6)

where τ̂i is the tangent to the path at the position of image i in 3N -dimensional space. We use
the following tangent definition, which is similar to the “improved” tangent definition [47] that
is frequently used in NEB calculations, see Sec. A.2,

⎧
⎪ τ̂i+ Vi +1 > Vi > Vi −1
⎪
⎪
⎪
⎪
⎨ τ̂ −
⎪ Vi −1 > Vi > Vi +1
τi = i
. (A.7)
⎪
⎪
+ −
τ̂i + τ̂i Vi > Vi +1 > Vi −1 ∨ Vi +1 > Vi −1 > Vi
⎪
⎪
⎪
⎪ τ̂ + + τ̂i− Vi > Vi −1 > Vi +1 ∨ Vi −1 > Vi +1 > Vi
⎩ i
Fig. A.3 shows the distributions of the Euclidean norm of Fi⊥ from the calculations reported
in Ch. 4. Note that the forces of the first and the last image were excluded because they
correspond to local minima with close to zero total force. The median norm is lower than
10−2 eV/Å, both in calculations with average and random alloys. The norm tends to be
lower in average alloy calculations compared to random alloy calculations, which is perhaps
not surprising, considering the smoother potential energy landscape of the homogeneous
average alloy systems. The maximum values are on the order of 10−1 eV/Å. These values are
magnitudes higher than the norm of the residual force that one would commonly tolerate in
regular energy minimizations (10−6 –10−6 eV/Å). However, such large values mainly occur in
the region of constriction formation and annihilation, see Fig. A.4, where the norm is drawn as
a function of reaction coordinate for two average alloys. The residual forces are much lower at
the transition state, where the activation energy is measured.

Additionally, NEB calculations were carried out to check how much the results depend on the
algorithm for finding minimum energy paths. The Ni+15 at.% Al random alloy configurations
from Ch. 3 were used as test cases. Two sets of NEB calculations were performed. In the first
case, the default initial guess for the path, i.e. a linear interpolation between initial and
final state, was used in the calculations. In the second case, the output paths from the
corresponding String calculations were used as input. Fig. A.5 shows the differences between
the activation energies calculated with NEB and String method. They are typically less than
−1
1 × 10−2 eV Å . In a few cases, the difference is on the order of 0.1 eV. It tends to be smaller
when the output of the String calculation is used as initial guess. The maximum difference is
0.17 eV, in a calculation with Ni+12 at.% Al. Here, cross-slip nucleates at a different position
1. i.e. iterations on which the String was reparameterized

82
 A.3. Convergence in String Method Calculations

along the line. Nevertheless, the transition paths are similar, see Fig. A.6a). Significantly
different transition paths are obtained in one calculation for Ni+15 at.% Al, see Fig. A.6b). Here,
the NEB calculation predicts a path with more local minima. However, closer inspection revals
that the initial part of the path corresponds to a failed cross-slip attempt, where a nucleus is
formed and disappears again. The second part of the path is qualitatively similar to the paths
obtained with the String method or the NEB method with String input. In particular, the
transition state in this section of the path is very similar to the transition states in the other
calculations. These observations demonstrate that our calculations are robust against a
change of method. Furthermore, they indicate that continuing the String calculations until all
| Fi⊥ | (i = 1 . . . NS ) are below 10−2 eV/Å would have little effect.

We conclude that the calculated paths are not fully converged minimum energy paths.
However, they are sufficiently well converged for our purposes. Continuing the calculations is
unlikely to lead to significantly different values of ΔE act . The expected changes are very small
compared to the overall level of ΔE act and random fluctuations of ΔE act in random alloys.
Furthermore, we have demonstrated that our results are robust against a change of calculation
method.

100
Al+22% Mg
random alloys
average alloy
10−2
− ΔE act )/ΔE act

10−4
(ΔE act
(i )

10−6

10−8
0 100 200 300 400 500 600
String iteration

(i )
Figure A.2: Change of activation energy ΔE act as a function of String iteration i , relative to the
final value ΔE act ; data from calculations with Al+22 at.% Mg reported in Ch. 4; note that some
calculations terminated due to reaching the maximum number of 600 String iterations; in
random Al+22 at.% Mg, a change of 0.5% in the last iteration is seen in one case; this is the
maximum final value across all alloy datasets

83
Appendix A. Appendix

100
images 1 . . . Ns − 1
transition state
10−1
| Fi⊥ | (eV/Å)

10−2

10−3

10−4
Al

Al

Ni

Ni
M

M
2%

5%

3%

9%
2%

6%

2%
+0

+1

+3

+7
+0

+0

+2
Ni

Ni

Cu

Cu
Al

Al

Al

Figure A.3: Distributions of the magnitudes of the force vectors Fi⊥ ; the magnitude is a
measure of convergence, since Fi⊥ is zero for images lying on the MEP; data for random alloys
from calculations reported in Ch. 4; blue: distribution for all images except the initial and final
one, which correspond to local minima (3000 = 100 × 30 data points per alloy); orange:
magnitude of Fi⊥max at the transition state (i = i max ); 100 values per alloy; the upper and lower
edge of each box represent the third and second quartile of the data, i.e. 50% of all values lie in
this range; the horizontal line indicates the median; points indicate maximum values;

2.0
10−1
Ni+15%Al
1.5 10−2
relative energy (eV)

10−3
| Fi⊥ | (eV/Å)

1.0
10−4
Al+22%Mg

0.5 10−5

10−6
0.0
0.0 0.2 0.4 0.6 0.8 1.0
reaction coordinate

Figure A.4: Relative energy (solid lines, left y-axis) and magnitude of F⊥ (dashed lines, right
y-axis) as a function of reaction coordinate; data from calculations with average alloys of
Ni+15 at.% Al (teal) and Al+22 at.% Mg (magenta); high perpendicular forces are seen in the
region of constriction formation; data from calculations reported in Ch. 4

84
 A.3. Convergence in String Method Calculations

Figure A.5: Magnitude of difference between Nudged Elastic Band (NEB) and String method
estimates of ΔE act ; gray and white indicate the initial guess used in the NEB calculation; white:
using the same initial guess as in the String method calculation; gray: output of String method
calculation used as initial guess; the upper and lower edge of each box represent the third and
second quartile of the data, i.e. 50% of all values lie in this range; the horizontal line indicates
the median; points indicate maximum values; String method data from calculations with Ni-Al
alloys reported in Ch. 3; 20 String method calculations were performed for each concentration;
however, some NEB calculations with Ni+12 at.% Al and Ni+15 at.% Al diverged; thus the
sample size of gray and white datasets is 16 and 19 at 12% Al, and 16 and 17 at 15% Al

Figure A.6: Transition paths from String and NEB calculations; orange: output of String
calculation used as initial guess for NEB calculation; violet: default initial guess; note that the
NEB/String paths overlap with the String paths; two extreme cases are shown: a) paths from
calculations with a particular Ni+12 at.% Al random alloy realization; here, the difference in
ΔE act between NEB/default and String calculations is larger than in other calculations with
Ni-Al random alloys; cross-slip nucleates at a different position in the NEB calculation;
nevertheless, the paths are qualitatively similar; b) significant qualitative difference between
NEB/default and String paths in a calculation with Ni+15 at.% Al; the initial part of the NEB
path represents a failed cross-slip attempt, where a nucleus is formed and disappears; the
subsequent portion of the path is qualitatively similar to String and NEB/String paths.

85
Appendix A. Appendix

A.4 System Size Dependence of ΔE act in Atomistic Calculations

The atomic configurations used in cross-slip transition path calculations must be large
enough to minimize boundary effects. Two dimensions have to taken into account, the inner
radius of the cylinder and its length along the periodic direction. If the radius is too small, the
width of the dislocation may be far from the value in an infinite medium, due to interactions
with the fixed shell. If the cylinder is too short, self-interaction of the two constrictions across
the periodic boundary may influence ΔE act .

To determine the required system size, transition path calculations were performed for
different radii and periodic lengths. The simulations were carried out with pure Ni, using
the potential of Ref. [98]. Preliminary calculations showed that ΔE act of this material is
comparable to ΔE act of Cu and Ni as given by the potential of Ref. [91], and higher than ΔE act of
Al as given by the potential of Rev. [74]. Thus, it was assumed that the dimensions found here
would also be sufficient for the other materials.

The radius was varied between 8 and 22b . 16 images were used in the calculation. Fig. A.7a)
and b) show the transition paths and the corresponding activation energies, respectively. ΔE act
decreases with increasing radius. However, at 12b and larger, √ the difference with respect to the
value at 22b is less than 0.1 eV. In practice, a radius
√ of 10 3a was used, which is slightly larger
than the largest radius tested here. ΔE act at 10 3a is 1.724 eV, and thus only 0.005 eV smaller
than the value at 22b .

To study the effect of length, transition path calculations were carried out with 5b to 160b long
configurations. Two sets of calculations were carried out. In the case, the length was varied
between 5b and 60b , but the number of images was kept constant at 16. In a second set,
30–160b long dislocations were simulated and the number of images was increased with
increasing length, in order to avoid loss of resolution of the path. The resulting transition paths
and activation energies are shown in Fig. A.8a) and b), respectively. ΔE act increases with length
until it reaches a value of roughly 1.7 eV. At the largest length of 160b , cross-slip nucleates at
two positions along the dislocation line when using the standard initial guess for the path.
Thus, the transition path is a combination of two regular cross-slip transition paths and the
activation energy is approximately twice the value for a single nucleus. However, if the
calculation is started with an improved initial guess for the path, see Sec. 2.3.3, then a regular
cross-slip transition path is obtained. For lengths shorter than 15b , the dislocation does
not cross-slip by the Friedel-Escaig mechanism. Instead of forming point constrictions, it
constricts over the full length before changing slip plane. The length at which ΔE act becomes
independent of length is approximately 40b . Between 40b and 160b , ΔE act increases by less
than 0.03 eV. Thus, a length of 40b was used in the calculations reported in Ch. 3 and Ch. 4.

In calculations at nonzero Escaig stress on the cross-slip plane, a larger configuration may be
required to compensate for widening of the dislocation. A more limited parameter test was
carried out to determine the required inner radius of the configuration. Transition path
calculations were performed
√ with
√ average Ni+15 at.% Al at 300 MPa stress. The radius was
varied in the range 10 3a –17.5 3a . ΔE act increases slightly with radius,
√ see Fig. A.9. Between
√ √
12.5 a and 17.5 a , the increase is only 0.01 eV, hence a radius of 13 3a was used in practice.

86
 A.4. System Size Dependence of ΔE act in Atomistic Calculations

Figure A.7: Cross-slip transition path data for pure Ni as a function of the inner radius of the
cylindrical atomistic configuration; a) transition paths for different radii; b) activation energy
as a function of radius; 16 images were used in the calculations

Figure A.8: Cross-slip transition path data for pure Ni as a function of the periodic length of the
atomistic configuration; a) transition paths for different lengths; green lines: transition paths
for a 160b long configuration; dashed line: result from a calculation using the default initial
guess for the path (interpolation between initial and final state); two cross-slip nuclei are
formed here; solid line: result from a calculation with an improved initial guess; the number of
images was 16 in calculations with 5–60b length and 256 in calculations with 160b length; b)
activation energy as a function of length; the data point from the calculation with 160b length
and two nuclei is not shown; gray line: calculations with length 5–60b and 16 images; black
line: calculations for 30b , 40b , 80b and 160b length; the number of images was 48, 64, 128 and
256; inset: zoom on the range 40b –160b

87
Appendix A. Appendix

1.5 1.56
relative energy (eV)

1.0
1.55
√

ΔE act (eV)
0.5
radius / 3a
0.0 10.0 1.54
−0.5 12.5
15.0
−1.0 17.5
1.53

−1.5
0.0 0.2 0.4 0.6 0.8 1.0 10.0 12.5 √ 15.0 17.5
reaction coordinate radius / 3a

Figure A.9: Influence of the inner radius of the configuration at finite stress; a) transition paths
and b) activation energies of Ni+15 at.% Al at 300 MPa Escaig stress

88
 A.5. Partially Cross-Slipped Dislocations

A.5 Partially Cross-Slipped Dislocations

The large fluctuations in solute binding energies raise an interesting possibility. Should we
expect to observe partially cross-slipped dislocations in equilibrium? Consider the dislocation

dissociation on ...
cross-slip plane glide plane

*
L

Figure A.10: Partially cross-slipped dislocation

of total length L shown in Fig. A.10. Some segments are dissociated on the cross-slip plane,
others on the glide plane. Random fluctuations in solute-dislocation and solute-solute
binding energies are unbiased and may promote dissociation on either plane. Thus, we should
expect that about half of the segments are dissociated on the cross-slip plane. To find the
equilibrium segment length ζ ∗ , we minimize the relative energy of the partially cross-slipped
configuration (relative to that of a dislocation lying fully on the glide plane). For each segment
on the cross-slip plane, the energy increases by the energy of two isolated constrictions, which
is approximately equal to ΔE act,avg . On the other hand, a segment will only cross-slip if there
are favorable solute fluctuations that promote dissociation on the cross-slip plane. The energy
gained by cross-slip is typically one standard deviation of ΔE end , which grows with the square
root of the segment length ζ , hence

L ζ L
ΔE = − σ [ΔE end ] |ζ =1b +ΔE act,avg ,
2ζ b 2ζ
(A.8)
ΔE 1 1
=− √ σ [ΔE end ] |ζ =1b + ΔE act,avg ,
L 2 bζ 2ζ

where σ [ΔE end ] |ζ =1b is the standard deviation of ΔE end for a 1b long segment. The length ζ ∗
for which ΔE is minimum can easily be found by requiring ∂(E /L )/∂ ζ = 0. The result is
 2
ζ ∗ = 4b ΔE act,avg /σ [ΔE end ] |ζ =1b . (A.9)

Using the extended model for σ [ΔE end ] that was presented in Sec. 4.1, we find that ζ ∗ is
in the range 202b to 11264b for the alloys considered here, see Tab. A.4. At dislocation
densities of 1012 to 1013 m−2 , typical values for L are on the order of 103b . Thus, it seems
that partially cross-slipped states could be observed in extreme cases with low values of
ΔE act,avg /σ [ΔE end ] |ξ=1b .

89
Appendix A. Appendix

Table A.4: Equilibrium segment length ζ ∗ of partially cross-slipped dislocations according to

Eq. A.9.

alloy ζ ∗ /b
Ni+02%Al 11264
Ni+15%Al 402
Al+02%Mg 2112
Al+06%Mg 659
Al+22%Mg 202
Cu+22%Ni 4847
Cu+33%Ni 2843
Cu+79%Ni 3102

A.6 Extended Solute Pair Model

Our goal is to calculate the standard deviation of the change in solute-solute binding energies
due to formation and destruction of solute pairs in the stacking fault ribbon during cross-slip.
The corresponding model from Sec. 3.4.3 can be improved by taking into account higher order
neighbor pairs and correlation between pair changes. For this purpose, we write the change
ΔE end,s-s,sf,XY in solute-solute binding energy during stacking fault formation in a very general
form. Recall that a stacking fault consists of two {111} planes with hexagonal-closed packed
coordination. In our derivation, we take atoms below the fault, including those in the lower
plane of the fault, as fixed. The stacking fault is generated by displacing the atoms above,
including those in the upper plane of the fault, by the Burgers vector of a Shockley partial
dislocation. Consider a unit cell in the lower region, as shown in Fig. A.11. The upper plane of
the cell (green atoms) corresponds to the lower plane of the fault. The fault can be created by
√ fault of length ζ and width d , there are
displacing the atoms above by a /6[112]2 . For a stacking
N ζ × Nd such cells, where N ζ = ζ /b and Nd = d /b / 3. To get the total change in solute-solute
binding energy, we need to sum contributions from all atomic sites in these cells. Even lower
planes do not need to be considered if we restrict ourselves to eight nearest neighbor pairs.
The corresponding neighbor distances are listed in Tab. 4.1.

Label the unit cells with indices i j , where i = 1 . . . N ζ and j = 1 . . . Nd . Let Ωlo,ij be the set of
sites in the unit cell i j . There are three or five sites in each {111} plane of the unit cell. Some
however, lie on a face or in a corner of the cell and are therefore shared with neighboring unit
cells. There are only two full sites per plane and eight per unit cell. To generate Ωlo,ij , one could
either associate eight sites with each cell in a consistent way, or take all sites and introduce
extra occupation variables or weighting factors to account for the fact that some sites are
associated with multiple unit cells. We choose the latter method, because it allows to take into
account more correlations between pair changes within each unit cell. Furthermore, let Ωhi be
the set of sites above the fault, including the upper plane of the fault. Assign occupation
variables sX ,i j k , where sX ,i j k = 1 if there is a X -solute at site i j k , and sX ,i j k = 0 otherwise. The

2. Slightly different results would be obtained with a displacement a /6[121], i.e. parallel to x . The directions
[112] and [121] are not equivalent in our model. As we shall see, we consider only neighbor changes for sites in a
single unit cell as shown in Fig. A.11. With this particular choice of sites, the threefold rotation symmetry in the
(111) plane (the plane of the paper) is lost, hence [112] and [121] are not equivalent. However, a /6[112] is the
correct displacement corresponding to a Shockley partial of a 1/2a [101] screw dislocation.

90
 A.6. Extended Solute Pair Model

stacking fault
displacement

Figure A.11: Unit cell used in the derivation of the change in the number of solute pairs during
formation of an intrinsic stacking fault; the cell is below the fault and seen from above. Green
atoms represent the lower plane of the fault. Atoms in the upper plane of the fault would be on
top of the blue atoms before formation of the fault. To form the fault, all atoms above the green
atoms are displaced by a /6[112].

sX ,i j k are Bernoulli random variables with probability and expectation value equal to the
concentration c X of type X .

The total change in solute-solute binding energies for pairs formed by X -atoms in Ωlo,ij and
Y -atoms in Ωhi is the sum of contributions from all unit cells, i.e.

 
N ζ Nd

ΔE end,s-s,sf,XY = ΔE end,s-s,sf,XY ,i j . (A.10)

i j

To calculate ΔE end,s-s,XY ,i j , one sums the changes in solute-solute binding energies for each
site in Ωlo,ij . Accounting for pairs up to order N max ,

 
N max 
ΔE end,s-s,sf,XY ,i j = sX ,i j k w k Us-s,XY ,l sY ,i j m Δ(i , j , k , m , l ), (A.11)
k ∈Ωlo,ij l m ∈Ωhi

where Us-s,XY ,l is the binding energy of an l -th nearest neighbor XY pair. The w k are extra
occupation variables, which account for the fact that face and corner sites are shared by
multiple unit cells, hence we would have associated only one face or corner site in each plane
and unit cell with this cell. They are Bernoulli random variables with probability 1 for sites
completely inside the cell, 1/2 for face sites, and 1/4 for corner sites. Δ(i , j , k , m , l ) indicates if
a l -th order neighbor pair is formed, lost, or preserved, i.e.

Δ(i , j , k , m , l ) = [a (i , j , k , m , l ) − b (i , j , k , m , l )] , (A.12)

where b (i , j , k , m , l ) = 1 if the sites are l -th nearest neighbors before formation of the fault,
and zero otherwise; and a (i , j , k , m , l ) = 1 if they are l -th nearest neighbors after formation of
the fault, and zero otherwise.

The variance of ΔE end,s-s,sf,XY is

 
Var [ΔE end,s-s,sf,XY ] = ΔE end,s-s,sf,XY 2 − (ΔE end,s-s,sf,XY )2 , (A.13)

91
Appendix A. Appendix

where . . .  indicates the average with respect to sX ,i j k and sY ,i j k . However, a direct calculation
would be intractable due to the large number of terms involved. We simplify the problem by
making Δ(i , j , k , m , l ) independent of i and j . Neighbor changes are analyzed only once, for all
atoms in the unit cell shown in Fig. A.11, i.e.
Δ(i , j , k , m , l ) ≡ Δc (k , m , l ) = [a c (k , m , l ) − b c (k , m , l )] , (A.14)
where a superscript c is used to refer to the reference cell. Thus, we consider only subsets
Ωclo ⊂ Ωlo,ij and Ωchi ⊂ Ωhi . By using Δc (k , m , l ) instead of Δ(i , j , k , m , l ) we still capture some
correlations. For example, we account for exchange of neighbor sites between the sites in the
reference unit cell. However, we ignore exchanges between sites in different unit cells with
different i and j . The numbers Δc (k , m , l ) were determined by finding the neighbors of sites in
a unit cell as shown in Fig. A.11 before and after insertion of the fault. The crystals were
generated with LAMMPS and the distance-based selection could be done conveniently using
OVITO [135].

We can now write for the second term in Eq. A.13, noting that sX ,i j k and sY ,i j m are independent
in a random solid solution,
 Nζ N 2
 d
 
(ΔE end,s-s,sf,XY )2 = ΔE end,s-s,sf,XY ,i j
i j

⎡ ⎤2
⎢ ζ      
N Nd
 N max
 c ⎥
= ⎢⎢ sX ,i j k w k Us-s,XY ,l sY ,i j m Δ (k , m , l )⎥⎥ (A.15)
⎢ i j k ∈Ωc l ∈Ωchi ⎥
⎣ lo
l
⎦
⎡    ⎤ 2
⎢ N max
⎥
= ⎢⎢N ζ Nd c X cY wk Us-s,XY ,l Δc (k , m , l )⎥⎥
⎢ k ∈Ωclo l ∈Ωchi ⎥
⎣ l
⎦
Define
 
N max 
β ({w k }) ≡ wk Us-s,XY ,l Δc (k , m , l ), (A.16)
k ∈Ωclo l l ∈Ωchi

then
  2  2
ΔN s-s,sf
loss
= N ζ Nd c X cY β ({w k }) . (A.17)

The first term in Eq. A.13 involves the square of a sum and can be expanded as follows. For the
sake of brevity we write αi j ≡ ΔE end,s-s,sf,XY ,i j . Thus
+  Nζ N 2,
   d

ΔE end,s-s,sf,
2
XY = αi j
i j
+ Nζ N , + Nζ Nζ N N ,
 d  d d

= αi2j + αi j αpq (A.18)

i j i p i j q j
+ Nζ Nζ N , + Nζ N N ,
 d  d d

+ αi j αp j + αi j αi q .
i p i j i j q j

92
 A.6. Extended Solute Pair Model

The variables αi j , αpq , αp j and αi q are independent, hence

N ζ Nd  
     
N ζ N ζ Nd Nd
 
ΔE end,s-s,sf,
2
XY = αi2j + αi j αpq
i j i p i j q j
(A.19)
Nζ
Nd Nζ
   
Nd 
Nζ Nd
  
+ αi j αp j + αi j αi q .
i p i j i j q j

For the last three terms, one obtains

 
N ζ N ζ Nd Nd
   
αi j αpq = N ζ N ζ − 1 Nd (Nd − 1) (c X cY β ({w k }))2 ,
i p i j q j


NζNd

Nζ
   
αi j αp j = N ζ Nd N ζ − 1 (c X cY β ({w k }))2 , (A.20)
i p i j

 
N ζ Nd Nd
 
αi j αi q = N ζ Nd (Nd − 1) (c X cY β ({w k }))2 .
i j q j

The first term in Eq. A.19 is more cumbersome to resolve, because αi2j is a sum of terms that
may contain squares of occupation variables, e.g. sX2 ,i j k sY ,i j m sY ,i j s . Since the occupation
variables are Bernoulli random variables, the averages of their squares are the same as the
averages of the variables themselves, i.e. c X and cY . Each such product contains at least two
independent variables, thus the averaged products have the form c X v cY w with v + w ∈ [2, 3, 4].
The correct products of concentrations can be substituted for the products of occupation
variables using a computer algebra program, e.g. MATHEMATICA [155]. Corresponding
MATHEMATICA code is presented in App. A.6.1.

The variance of ΔE end,s-s,XY , the change in solute-solute binding energies during cross-slip, is
approximately two times the variance of ΔE end,s-s,sf,XY . Combining Eq. A.13, Eq. A.20, Eq. A.19
and Eq. A.17, one obtains the preliminary result
 N ζ Nd 

  
Var [ΔE end,s-s,XY ] = 2 N ζ Nd (c X cY β ({w k }))2 + αi2j . (A.21)
i j

Both terms on the right hand side still contain the extra occupation variables w k . They must
be averaged over. Recall that the w k are Bernoulli random variables, with probabilities 1
for center sites, 1/4 for corner sites, and 1/2 for sites on the face of the unit cell. Since
they are Bernoulli variables, powers of w k are replaced by the corresponding probabilities.
Again, a computer algebra program can be used to make these substitutions in Eq. A.21, see
App. A.6.1. The final result for Var [ΔE end,s-s,XY ] is quite long. However, all terms are of the form
AUs-s,XY ,l Us-s,XY ,m c X v cY w , where A is an integer, v , w = 1, 2 and l , m = 1, . . . , N max . Thus, the
variance can be written using matrix-vector notation, as in Eq. 4.1,
ζd 1
Var [ΔE end,s-s,XY ] = √ c X cY
3 b2 (1 + δXY ) (4.1)

× Us-s,XY [C1 + C2 (c X + cY ) − (C1 + 2C2 ) c X cY ] Us-s,XY ,

93
Appendix A. Appendix

with coefficient matrices

2 −4 0 0 0 0 0 0
 0 "
6 −8 0 0 0 0 0 #
#
0 0 14 0 −8 0 0 0 #
#
0 0 0 6 0 0 0 0 #
C1 = 8 #,
0 0 0 0 24 −8 −8 0 #
#
0 0 0 0 0 9 0 0 #
#
#
0 0 0 0 0 0 38 0 #
! 0 0 0 0 0 0 0 6 $ (A.22)
−16 47 5 5 −16 3 −1 −1 "
 0 −62 86 −12 37 −16 −14 2 ##
0 0 34 349 673 157 1309 275 ##
1 0 0 0 97 602 114 1280 290 ##
C2 = .
2 0 0 0 0 426 231 2468 652 ##
0 0 0 0 0 −48 603 78 ##
0 0 0 0 0 0 2424 1146 ##
! 0 0 0 0 0 0 0 96 $

If only first nearest neighbor pairs of like solutes (X = Y , c = c X = cY ) are considered, one
obtains
ζd
Var [ΔN s-s ] = 8c 2 (1 − c ) √ , (A.23)
b2 3

which is equivalent to our previous solution for this special case, Eq. 3.16.

A.6.1 Implementation

In the following, we present Wolfram Mathematica (Version 11) [155] code for writing Eq. A.11
and substituting powers of occupation variables by concentrations.

Suppose the information about pair changes is stored in a text file pairs.txt with four
columns and multiple rows. Each row corresponds to one pair that is lost or gained. The first
column indicates the order (l = 1 . . . N max ) of the pair, the second the sign (−1=loss, +1=gain),
the third the unique identifier of the site in the unit cell below the fault, and the fourth the
unique identifier of the site above the fault. If pairs up to the eighth order are considered, then
eleven sites in the unit cell below the fault need to be taken into account. Number those sites
from one to eleven and let 1, 6, and 11 be center sites with w k  = 1 in Eq. A.21. Furthermore,
let 2–5 be face sites ( w k  = 1/2), and 7–10 corner sites ( w k = 1/4 ). We consider the more
general case where the atoms above the fault have a different type than those in the unit cell
below.

First, load the data and write the sum Eq. A.11. The pair energies are indicated by symbols Ei
(i = 1 . . . 8). The occupation variables for sites in the unit cell are written as pi, where i is the
site number. Similarly, the occupation variables for sites above the fault are written as si.
Symbols wi indicate the extra occupation variables w k in Eq. A.21.

94
 A.6. Extended Solute Pair Model

1 (*Load the information and write the sum*)

2 Numbers = ReadList["pairs.txt",Table[Number,4]];
3 Formatter[x_]:=ToExpression[
4 ToString[
5 StringForm[
6 "E``*``*p``*s``*w``",x[[1]],x[[2]],x[[3]],x[[4]],x[[3]]
7 ]
8 ]
9 ]
10 (*We could ignore pairs of higher order*)
11 (*Restrictions={
12 E2\[Rule]0,E3\[Rule]0,E4\[Rule]0,E5\[Rule]0,
13 E6\[Rule]0,E7\[Rule]0,E8\[Rule]0
14 };*)
15 Restrictions={};
16 EnergyChange=Total[Map[Formatter,Numbers]]/.Restrictions;

In the following, we define a module which substitutes powers of the occupation variables pi
and si by the corresponding concentrations cp and cs, i.e. c X and cY in Eq. 4.1. Furthermore,
powers of wk are replaced by the appropriate site weight. Finally, Var [ΔE end,s-s,XY ] is derived.

1 (*Expression for matching powers of the symbol s *)

2 (*see https://mathematica.stackexchange.com/a/75295*)
3 MatchPowersOf=s_Symbol/;StringMatchQ[SymbolName[Unevaluated@s],#]&;
4 CalcVarianceUnlike[EnergyChange_]:=Module[
5 {
6 AverageOfSquare, SquareOfAverage,
7 AllTermsA,AllTermsB,
8 ListOfAllTermsA,ListOfAllTermsB,
9 ResultA,ResultB,
10 Uniquep,Uniques,
11 TmpList,TmpList2,TmpList3,
12 EnergyProduct,
13 Uniquews,
14 SiteWeights,
15 },
16 (*Average values of the extra occupation variables for *)
17 (*center, face and corner sites. Here: 1,6,11=center sites, *)
18 (*2-5=face sites, 7-10=corner sites*)
19 SiteWeights={1,1/2,1/2,1/2,1/2,1,1/4,1/4,1/4,1/4,1};
20 (*Calculate the average of the square of the pair change*)
21 AllTermsA=Expand[EnergyChange^2];
22 ListOfAllTermsA=List@@AllTermsA;
23 ResultA={};
24 (*Consider individual terms of the sum, which are products of*)
25 (*pair energies and mixed powers of occupation variables. *)

95
 Appendix A. Appendix

26 (*Substitute the corresponding concentration for any power of*)

27 (*an occupation variable, since it is a Bernoulli random variable.*)
28 For[i=1,i<= Length[ListOfAllTermsA],i++,
29 TmpList=List@@ListOfAllTermsA[[i]];
30 (*Number of unique s-occupation variables*)
31 Uniques=Length[Cases[TmpList,MatchPowersOf["s*"]^_.]];
32 (*Number of unique p-occupation variables in the product*)
33 Uniquep=Length[Cases[TmpList,MatchPowersOf["p*"]^_.]];
34 (*Number of unique w-occupation variables*)
35 Uniquews={};
36 For[j=1,j<=11,j++,
37 AppendTo[
38 Uniquews,
39 Length[
40 Cases[
41 TmpList,
42 MatchPowersOf[ToString[StringForm["w``",j]]]^_.
43 ]
44 ]
45 ];
46 ];
47 (*Get product of pair energies by deleting occupation variables from list*)
48 TmpList2=DeleteCases[TmpList,MatchPowersOf["s*"]^_.];
49 TmpList3=DeleteCases[TmpList2,MatchPowersOf["w*"]^_.];
50 EnergyProduct=Times@@DeleteCases[TmpList3,MatchPowersOf["p*"]^_.];
51 For[j=1,j<=11,j++,
52 EnergyProduct=EnergyProduct*SiteWeights[[j]]^Uniquews[[j]];
53 ];
54 AppendTo[ResultA,
55 EnergyProduct*(
56 Nl*Nd*cp^Uniquep*cs^Uniques +
57 Nl*Nd*(Nd-1)*cp^2*cs^2 +
58 Nl*Nd*(Nl-1)*cp^2*cs^2 +
59 Nl*Nd*(Nd-1)*(Nl-1)*cp^2*cs^2
60 )
61 ];
62 ]; (*end of loop over sum terms*)
63 AverageOfSquare=Collect[FullSimplify[Total[ResultA]],cp];
64 (*Calculate the square of the average of the pair change*)
65 AllTermsB=Expand[EnergyChange];
66 ListOfAllTermsB=List@@AllTermsB;
67 ResultB={};
68 For[i=1,i<= Length[ListOfAllTermsB],i++,
69 TmpList=List@@ListOfAllTermsB[[i]];
70 Uniques=Length[Cases[TmpList,MatchPowersOf["s*"]^_.]];
71 Uniquep=Length[Cases[TmpList,MatchPowersOf["p*"]^_.]];

96
 A.6. Extended Solute Pair Model

72 TmpList2=DeleteCases[TmpList,MatchPowersOf["s*"]^_.];
73 EnergyProduct=Times@@DeleteCases[TmpList2,MatchPowersOf["p*"]^_.];
74 AppendTo[ResultB,EnergyProduct*(Nl*Nd*cp^Uniquep*cs^Uniques)];
75 ];
76 (*Replace site weights AFTER taking the square!*)
77 SquareOfAverage=Expand[Total[ResultB]^2];
78 ResultB={};
79 ListOfAllTermsB=List@@SquareOfAverage;
80 For[i=1,i<= Length[ListOfAllTermsB],i++,
81 TmpList=List@@ListOfAllTermsB[[i]];
82 Uniquews={};
83 For[j=1,j<=11,j++,
84 AppendTo[
85 Uniquews,
86 Length[
87 Cases[
88 TmpList,
89 MatchPowersOf[ToString[StringForm["w``",j]]]^_.
90 ]
91 ]
92 ];
93 ];
94 EnergyProduct=Times@@DeleteCases[TmpList,MatchPowersOf["w*"]^_.];
95 For[j=1,j<=11,j++,
96 EnergyProduct=EnergyProduct*SiteWeights[[j]]^Uniquews[[j]];
97 ];
98 AppendTo[ResultB,EnergyProduct];
99 ]; (*end of loop over sum terms*)
100 SquareOfAverage=Collect[FullSimplify[Total[ResultB]],cp];
101 FullSimplify[2*(AverageOfSquare-SquareOfAverage)]
102 ] (*end of module*)

Finally, the variance for unlike and like pairs can be calculated as follows.

1 (*Variance for unlike pairs, where atoms in the unit cell have type p and*)
2 (*atoms above the fault have type s*)
3 VarUnlike=CalcVarianceUnlike[EnergyChange];
4 (*Adding the variance of the opposite pair*)
5 Swap1=VarUnlike/.{cp->tp,cs->ts};
6 Swap2=Swap1/.{ts->cp,tp->cs};
7 VarUnlikeSwapped=FullSimplify[Swap2];
8 VarUnlikeCombined=FullSimplify[VarUnlike+VarUnlikeSwapped]
9 (*Variance of like pairs of solutes with concentration c*)
10 VarLike=(1/2*VarUnlikeCombined)/.{cp->c,cs->c}

97
Appendix A. Appendix

A.7 E det under Stress

0.5 1.5
1.5 a) Ni+02%Al b) Al+06%Mg c) Cu+33%Ni
relative energy (eV)

1.0
1.0 τgEsc = 150 MPa 0.0
atomistics 0.5
0.5 E det
τcEsc = 150 MPa −0.5 0.0
0.0
atomistics
E det −1.0 −0.5
−0.5
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
step step step

Figure A.12: Relative energy during cross-slip under 150 MPa Escaig stress as a function of
step, i.e. length of the cross-slipped segment in Burgers vectors; solid lines: results from
atomistic transition path calculations; dashed lines: determinisitic energy E det for random
walk calculations (Eq. 4.10).

Fig. A.12 compares E det according to Eq. 4.10 with Nc = 10, and τglide
Esc
= 150 MPa or
τcross
Esc
= 150 M P a to results from atomistic calculations. The cross-slipped length in
the atomistic calculations was measured using the dislocation analysis (DXA) algorithm
implemented in Ovito [135, 136]. The best match is obtained in the case of Ni-15 at.% Al. In
Al-06 at.% Mg and Cu-33 at.% Ni, E det underestimates the relative energy, especially near the
center of the curve. In the case of Cu-33 at.% Ni, the error is larger when the Escaig stress is
applied on the cross-slip plane, which is a consequence of neglecting the stress effect on the
constriction on this plane. However, the difference between activation energies obtained from
the E det curve and from atomistic calculations is typically less than 0.1 eV for Escaig stresses up
to 300 MPa, see Fig. A.13.

Our model of the effect of a Schmid stress on the cross-slip plane can be verified by
comparing activation energies predicted with Eq. 4.10 to results from atomistic transition path
calculations reported by Kang et al. [60], see Fig. A.14. We also show their predictions based on
the Friedel-Escaig model. Note that the material is pure Ni. When evaluating Eq. 4.10 to
find the activation energy, we used the material parameters reported in their publication
(a = 3.52 Å, C 11 = 244 GPa, C 12 = 149 GPa, C 44 = 119 GPa, ΔE act,avg ≈ 2.27 eV). The activation
energies predicted using the stress-dependent deterministic energy profile are only about
0.06 eV higher than the atomistic energies. Qualitatively and quantitatively, a better match is
obtained than with the Friedel-Escaig model.

98
 A.8. Correlation between ΔE act and ΔE end

a) Ni+02%Al 0.7 b) Al+06%Mg

activation energy (eV)

activation energy (eV)

1.7
0.6
1.6
0.5
1.5
0.4
1.4
0.3
1.3
0 100 200 0 100 200
Escaig stress (MPa) Escaig stress (MPa)
c) Cu+33%Ni
activation energy (eV)

1.6
glide plane
1.5
atomistics
1.4 E det
1.3 cross-slip plane
atomistics
1.2 E det
1.1

0 100 200
Escaig stress (MPa)

Figure A.13: Cross-slip activation energies as a function of Escaig stress on the glide (Yellow) or
cross-slip plane (Violet); solid lines: results from atomistic transition path calculations; dashed
lines: maximum of the determinisitic energy E det for random walk calculations (Eq. 4.10)

2.3

2.2
activation energy (eV)

FE energy −0.7 eV
2.1 (Kang, Yin & Cai)
atomistics
2.0 (Kang, Yin & Cai)
E det
1.9

1.8

0 200 400 600 800 1000

τcSch (MPa)

Figure A.14: Cross-slip activation energies in pure Ni as a function of Schmid stress on the
cross-slip plane. Red: atomistic data reported by Kang et al. [60]; gray: values calculated by
Kang et al. using their implementation of the Friedel-Escaig [32, 43] model, shifted by −0.7 eV;
blue: maximum of the determinisitic energy E det for random walk calculations (Eq. 4.10,
evaluated with the material properties reported by Kang et al.

A.8 Correlation between ΔE act and ΔE end

Fig. A.15 shows the correlations between the distributions of ΔE act and ΔE end from the
atomistic and random walk calculations reported in Sec. 4.3.1 (Nb =40, zero stress). Linear

99
Appendix A. Appendix

regression yields a slope of 0.53–0.54 for the random walk data. In case of the atomistic
calculations, the slope shows more variation. The mean value across all alloys is 0.5, however.

2.0
Ni+02at.% Al Ni+15at.% Al
4

random walk random walk

1.8 fit: 1.69 eV + 0.53ΔE end fit: 1.76 eV + 0.54ΔE end
3

2
1.6
atomistics atomistics
fit: 1.66 eV + 0.46ΔE end 1 fit: 1.75 eV + 0.50ΔE end
p -value: 8.09 × 10−25 p -value: 7.59 × 10−25
1.4
−0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 −4 −2 0 2 4
1.2
Al+02at.% Mg Al+06at.% Mg
1.2
1.0
1.0 random walk random walk
fit: 0.70 eV + 0.54ΔE end fit: 0.64 eV + 0.53ΔE end
0.8
0.8
0.6
0.6
atomistics 0.4 atomistics
fit: 0.66 eV + 0.57ΔE end fit: 0.64 eV + 0.43ΔE end
0.4
p -value: 2.07 × 10−24 p -value: 3.11 × 10−16
0.2
ΔE act (eV)

−1.0 −0.5 0.0 0.5 1.0 −1.0 −0.5 0.0 0.5 1.0
2.0 2.2
Al+22at.% Mg Cu+22at.% Ni

2.0
1.5 random walk random walk
fit: 0.58 eV + 0.53ΔE end fit: 1.59 eV + 0.54ΔE end
1.8
1.0
1.6

0.5 atomistics atomistics

1.4
fit: 0.52 eV + 0.54ΔE end fit: 1.56 eV + 0.54ΔE end
p -value: 1.21 × 10−28 p -value: 8.55 × 10−22
0.0 1.2
−2 −1 0 1 2 −1.0 −0.5 0.0 0.5 1.0
Cu+33at.% Ni Cu+79at.% Ni
2.25
2.00
2.00 random walk random walk
fit: 1.57 eV + 0.54ΔE end 1.75 fit: 1.44 eV + 0.54ΔE end
1.75
1.50
1.50
1.25
1.25 atomistics atomistics
fit: 1.54 eV + 0.44ΔE end 1.00 fit: 1.38 eV + 0.55ΔE end
1.00 p -value: 2.05 × 10−19 p -value: 1.12 × 10−32
−1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5
ΔE end (eV)

Figure A.15: Correlation between ΔE act and ΔE end , in atomistic (violet, crosses) and random
walk (green, hexagonal bins) calculations from Sec. 4.3.1 (Nb =40, zero stress)

100
Bibliography
[1] P. M. Anderson, J. P. Hirth, and J. Lothe. Theory of Dislocations. 4th ed. New York:
Cambridge University Press, 2017. ISBN: 978-0-521-86436-7.
[2] S. D. Andrews, H. Sehitoglu, and I. Karaman. “Constriction Energy in the Presence of a
Solute Field”. In: Journal of Applied Physics 87.5 (2000), p. 2194. ISSN: 00218979. DOI:
10.1063/1.372161.
[3] A. S. Argon. “A New Model of Dynamic Recovery for Stage III of Pure FCC Metals
without Cross Slip”. In: International Journal of Materials Research 100.4 (2009),
pp. 488–493. DOI: 10.3139/146.110059.
[4] S. Arrhenius. “Über die Reaktionsgeschwindigkeit bei der Inversion von Rohrzucker
durch Säuren”. In: Zeitschrift für physikalische Chemie 4.6 (1889), pp. 226–248.
[5] Z. S. Basinski. “Work Hardening: Surface Effects. Dislocation Distribution in Deformed
Copper Single Crystals”. In: Discussions of the Faraday Society 38 (1964), pp. 93–102.
DOI: 10.1039/DF9643800093.
[6] R. P. Bell. “The Theory of Reactions Involving Proton Transfers”. In: Proceedings of the
Royal Society of London. A 154 (1936), pp. 414–429. DOI: 10.1098/rspa.1936.0060.
[7] E. Bitzek et al. “Dislocation Cross-Slip in Nanocrystalline FCC Metals”. In:
Physical Review Letters 100.23 (June 2008). ISSN: 0031-9007, 1079-7114. DOI:
10.1103/PhysRevLett.100.235501.
[8] E. Bitzek et al. “Structural Relaxation Made Simple”. In: Physical Review Letters 97.17
(Oct. 2006). ISSN: 0031-9007, 1079-7114. DOI: 10.1103/PhysRevLett.97.170201.
[9] J. Bonneville and B. Escaig. “Cross-Slipping Process and the Stress-Orientation
Dependence in Pure Copper”. In: Acta Metallurgica 27 (1979), pp. 1477–1486. DOI:
10.1016/0001-6160(79)90170-6.
[10] J. Bonneville, B. Escaig, and J. Martin. “A Study of Cross-Slip Activation
Parameters in Pure Copper”. In: Acta Metallurgica 36.8 (1988), pp. 1989–2002. DOI:
10.1016/0001-6160(88)90301-X.
[11] L. M. Brown. “Toward a Sound Understanding of Dislocation Plasticity”. In:
Metallurgical Transactions A 22.8 (1991), pp. 1693–1708. DOI: 10.1007/BF02646493.
[12] W. Cai and W. D. Nix. Imperfections in Crystalline Solids. 1st ed. Cambridge, UK:
Cambridge University Press, 2016. ISBN: 978-1-107-12313-7.

101
Bibliography

[13] W. Cai et al. “A Non-Singular Continuum Theory of Dislocations”. In: Journal of the
Mechanics and Physics of Solids 54.3 (Mar. 2006), pp. 561–587. ISSN: 00225096. DOI:
10.1016/j.jmps.2005.09.005.
[14] D. Caillard and J. Martin. Thermally Activated Mechanisms in Crystal Plasticity. Ed. by
R. Cahn. 1st. Vol. 8. Pergamon Materials Series. Amsterdam: Pergamon, 2003.
[15] T. Cao et al. “The Influence of Al Elements on the Structure and the Creep Behavior of
AlxCoCrFeNi High Entropy Alloys”. In: Materials Letters 164 (Feb. 2016), pp. 344–347.
ISSN: 0167577X. DOI: 10.1016/j.matlet.2015.11.016.
[16] C. Carter and S. Holmes. “The Stacking Fault Energy of Nickel”. In: Philosophical
Magazine 35.5 (1977), pp. 1161–1172. DOI: 10.1080/14786437708232942.
[17] D. Cockayne, M. Jenkins, and I. Ray. “The Measurement of Stacking-Fault Energies
of Pure Face-Centred Cubic Metals”. In: Philosophical Magazine 24.192 (1971),
pp. 1383–1392. DOI: 10.1080/14786437108217419.
[18] A. H. Cottrell. “Theory of Dislocations”. In: Progress in Metal Physics 1 (1949).
[19] P. Coulomb. “Comment on Graphs Relating Some Property to Stacking Fault Energy”.
In: Scripta Metallurgica 15.7 (1981), pp. 769–770. DOI: 10.1016/0036-9748(81)90017-X.
[20] M. S. Daw and M. Baskes. “Embedded-Atom Method: Derivation and Application to
Impurities, Surfaces, and Other Defects in Metals”. In: Physical Review B 29.12 (1984),
pp. 6443–6453. DOI: 10.1103/PhysRevB.29.6443.
[21] K. A. Dill and S. Bromberg. Molecular Driving Forces. 2nd. London, New York: Garland
Science, 2011. ISBN: 978-0-8153-4430-8.
[22] I. L. Dillamore and W. Roberts. “Rolling Textures in F.C.C. and B.C.C. Metals”. In: Acta
Metallurgica 12.3 (1964), pp. 281–293. DOI: 10.1016/0001-6160(64)90204-4.
[23] D. E. Dominici. “The Inverse of the Cumulative Standard Normal Probability Function”.
In: Integral Transforms and Special Functions 14.4 (Aug. 2003), pp. 281–292. ISSN:
1065-2469, 1476-8291. DOI: 10.1080/1065246031000081698.
[24] J. Dorn and N. Jaffe. “Effect of Temperature on the Creep of Polycrystalline Aluminum
by the Cross-Slip Mechanism”. In: Transactions of the Metallurgical Society of AIME 221
(1961), pp. 229–233.
[25] J.-P. Du, C.-Y. Wang, and T. Yu. “Cross-Slip Process in Model Ni(Al) Solid Solution: An
Embedded-Atom Method Study”. In: Computational Materials Science 91 (Aug. 2014),
pp. 192–199. ISSN: 09270256. DOI: 10.1016/j.commatsci.2014.04.063.
[26] M. S. Duesbery. “Dislocation Motion, Constriction and Cross-Slip in FCC Metals”. In:
Modelling and Simulation in Materials Science and Engineering 6.1 (1998), p. 35. DOI:
10.1088/0965-0393/6/1/005.
[27] M. S. Duesbery, N. P. Louat, and K. Sadananda. “The Mechanics and Energetics
of Cross-Slip”. In: Acta metallurgica et materialia 40.1 (1992), pp. 149–158. DOI:
10.1016/0956-7151(92)90208-V.
[28] W. E, W. Ren, and E. Vanden-Eijnden. “Finite Temperature String Method for the Study
of Rare Events”. In: The Journal of Physical Chemistry B 109.14 (2005), pp. 6688–6693.
ISSN: 1520-6106, 1520-5207. DOI: 10.1021/jp0455430.

102
 Bibliography

[29] W. E, W. Ren, and E. Vanden-Eijnden. “Simplified and Improved String Method for
Computing the Minimum Energy Paths in Barrier-Crossing Events”. In: The Journal of
Chemical Physics 126.16 (2007), p. 164103. ISSN: 00219606. DOI: 10.1063/1.2720838.
[30] W. E, W. Ren, and E. Vanden-Eijnden. “String Method for the Study of Rare
Events”. In: Physical Review B 66.5 (Aug. 2002). ISSN: 0163-1829, 1095-3795. DOI:
10.1103/PhysRevB.66.052301.
[31] B. Escaig. “Contribution à l’étude des déviations dans la structure cubique faces
centrées, cubic centrée, et hexagonale compacte”. PhD. Paris: Université de Paris, 1968.
[32] B. Escaig. “Sur le glissement dévié des dislocations dans la structure cubique à faces
centrées”. In: Journal de Physique 29.2-3 (1968), pp. 225–239. ISSN: 0302-0738. DOI:
10.1051/jphys:01968002902-3022500.
[33] U. Essmann. “Elektronenmikroskopische Untersuchung der Versetzungsanordnung
in plastisch verformten Kupfereinkristallen”. In: Acta Metallurgica 12.12 (1964),
pp. 1468–1470. DOI: 10.1016/0001-6160(64)90142-7.
[34] U. Essmann and H. Mughrabi. “Annihilation of Dislocations during Tensile and Cyclic
Deformation and Limits of Dislocation Densities”. In: Philosophical Magazine A 40.6
(1979), pp. 731–756. DOI: 10.1080/01418617908234871.
[35] Y. Estrin et al. “A Dislocation-Based Model for All Hardening Stages in Large
Strain Deformation”. In: Acta Materialia 46.15 (1998), pp. 5509–5522. DOI:
10.1016/S1359-6454(98)00196-7.
[36] M. G. Evans and M. Polanyi. “Further Considerations on the Thermodynamics of
Chemical Equilibria and Reaction Rates”. In: Transactions of the Faraday Society 32
(1936), pp. 1333–1360. DOI: 10.1039/TF9363201333.
[37] H. Eyring. “The Activated Complex in Chemical Reactions”. In: The Journal of
Chemical Physics 3.2 (Feb. 1935), pp. 107–115. ISSN: 0021-9606, 1089-7690. DOI:
10.1063/1.1749604.
[38] R. A. Fisher and F. Yates. Statistical Tables for Biological, Agricultural and Medical
Research. 3rd ed. London: Oliver & Boyd, 1948.
[39] R. L. Fleischer. “Cross Slip of Extended Dislocations”. In: Acta Metallurgica 7.2 (1959),
pp. 134–135. DOI: 10.1016/0001-6160(59)90122-1.
[40] F. C. Frank and W. T. Read Jr. “Multiplication Processes for Slow Moving Dislocations”.
In: Physical Review 79.4 (1950), p. 722.
[41] J. A. Frenkel. “Zur Theorie der Elastizitätsgrenze und der Festigkeit kristallinischer
Körper”. In: Zeitschrift für Physik 37.7-8 (1926), pp. 572–609. DOI: 10.1007/BF01397292.
[42] J. Friedel. “Sur le fluage par déviation”. In: Revue de Physique Appliquée 12.10 (1977),
pp. 1649–1654. ISSN: 0035-1687. DOI: 10.1051/rphysap:0197700120100164900.
[43] J. Friedel. “Regarding Seeger’s Paper on Work Hardening”. In: Dislocations and
Mechanical Properties of Crystals. Ed. by J. Fisher et al. John Wiley & Sons, Inc., 1957,
pp. 330–332.
[44] V. Gerold and H. P. Karnthaler. “On the Origin of Planar Slip in F.C.C. Alloys”. In: Acta
Metallurgica 37.8 (1989), pp. 2177–2183. DOI: 10.1016/0001-6160(89)90143-0.

103
Bibliography

[45] G. S. Hammond. “A Correlation of Reaction Rates”. In: Journal of the American

Chemical Society 77.2 (1955), pp. 334–338. DOI: 10.1021/ja01607a027.
[46] R. Heidenreich and W. Shockley. “Study of Slip in Aluminium Crystals by Electron
Microscope and Electron Diffraction Methods”. In: Report of a Conference on Strength
of Solids. London: Physical Society, 1948, pp. 57–75.
[47] G. Henkelman and H. Josson. “Improved Tangent Estimate in the Nudged Elastic Band
Method for Finding Minimum Energy Paths and Saddle Points”. In: The Journal of
Chemical Physics 113.22 (2000), p. 9978. ISSN: 00219606. DOI: 10.1063/1.1323224.
[48] N. E. Henriksen and F. Y. Hansen. Theories of Molecular Reaction Dynamics. 1st ed.
Oxford: Oxford University Press, 2008. ISBN: 978-0-19-920386-4.
[49] M. R. Hestenes and E. Stiefel. “Methods of Conjugate Gradients for Solving Linear
Systems”. In: Journal of Research of the National Bureau of Standards 49.6 (1952),
pp. 409–436. DOI: 10.6028/jres.049.044.
[50] P. Hirsch. “Extended Jogs in Dislocations in Face-Centred Cubic Metals”. In:
Philosophical Magazine 7.73 (1962), pp. 67–93. DOI: 10.1080/14786436208201859.
[51] P. Hirsch. “The Interpretation of the Slip Pattern in Terms of Dislocation Movements”.
In: Journal of the Institute of Metals 86 (1957), p. 13.
[52] J. D. Honeycutt and H. C. Andersen. “Molecular Dynamics Study of Melting and
Freezing of Small Lennard-Jones Clusters”. In: Journal of Physical Chemistry 91 (1987),
pp. 4950–4963. DOI: 10.1021/j100303a014.
[53] S. I. Hong and C. Laird. “Mechanisms of Slip Mode Modification in F.C.C. Solid
Solutions”. In: Acta metallurgica et materialia 38.8 (1990), pp. 1581–1594. DOI:
10.1016/j.msea.2012.12.044.
[54] A. M. Hussein et al. “Microstructurally Based Cross-Slip Mechanisms and Their Effects
on Dislocation Microstructure Evolution in FCC Crystals”. In: Acta Materialia 85 (Feb.
2015), pp. 180–190. ISSN: 13596454. DOI: 10.1016/j.actamat.2014.10.067.
[55] P. J. Jackson. “The Role of Cross-Slip in the Plastic Deformation of Crystals”. In: Materials
Science and Engineering 57.1 (1983), pp. 39–47. DOI: 10.1016/0025-5416(83)90025-3.
[56] M. L. Jenkins. “Measurement of the Stacking-Fault Energy of Gold Using the
Weak-Beam Technique of Electron Microscopy”. In: Philosophical Magazine 26.3 (Sept.
1972), pp. 747–751. ISSN: 0031-8086. DOI: 10.1080/14786437208230118.
[57] C. Jin, Y. Xiang, and G. Lu. “Dislocation Cross-Slip Mechanisms in Aluminum”. In:
Philosophical Magazine 91.32 (Nov. 2011), pp. 4109–4125. ISSN: 1478-6435, 1478-6443.
DOI: 10.1080/14786435.2011.602030.
[58] E. Jones, T. Oliphant, P. Peterson, et al. SciPy: Open Source Scientific Tools for Python.
2001.
[59] H. Jónsson, G. Mills, and K. W. Jacobsen. “Nudged Elastic Band Method for Finding
Minimum Energy Paths of Transitions”. In: Classical and Quantum Dynamics in
Condensed Phase Simulations. Ed. by B. Berne, G. Cicotti, and D. Coker. World
Scientific, 1998, pp. 385–404.

104
 Bibliography

[60] K. Kang, J. Yin, and W. Cai. “Stress Dependence of Cross Slip Energy Barrier for
Face-Centered Cubic Nickel”. In: Journal of the Mechanics and Physics of Solids 62 (Jan.
2014), pp. 181–193. ISSN: 00225096. DOI: 10.1016/j.jmps.2013.09.023.
[61] K. Karhausen and F. Roters. “Development and Application of Constitutive Equations
for the Multiple-Stand Hot Rolling of Al-Alloys”. In: Journal of Materials Processing
Technology 123 (2002), pp. 155–166. DOI: 10.1016/S0924-0136(02)00081-X.
[62] U. F. Kocks. “Laws for Work-Hardening and Low-Temperature Creep”. In: Journal of
Engineering Materials and Technology 98.1 (1976), pp. 76–85. DOI: 10.1115/1.3443340.
[63] J. S. Koehler. “The Nature of Work-Hardening”. In: Physical Review 86.1 (1952),
pp. 52–59. DOI: 10.1103/PhysRev.86.52.
[64] A. N. Kolmogorov. “Sulla determinazione empirica di una legge di distribuzione”. In:
Giornale dell’Istituto Italiano degli Attuari 4 (1933), pp. 83–91.
[65] E. L. Kolsbjerg, M. N. Groves, and B. Hammer. “An Automated Nudged Elastic Band
Method”. In: The Journal of Chemical Physics 145.9 (Sept. 2016), p. 094107. ISSN:
0021-9606, 1089-7690. DOI: 10.1063/1.4961868.
[66] L. Kubin, T. Hoc, and B. Devincre. “Dynamic Recovery and Its Orientation Dependence
in Face-Centered Cubic Crystals”. In: Acta Materialia 57.8 (May 2009), pp. 2567–2575.
ISSN: 13596454. DOI: 10.1016/j.actamat.2009.02.013.
[67] L. P. Kubin et al. “Dislocation Microstructures and Plastic Flow: A 3D Simulation”.
In: Solid State Phenomena 23-24 (1992), pp. 455–472. ISSN: 1662-9779. DOI:
10.4028/www.scientific.net/SSP.23-24.455.
[68] T. Leffers and O. B. Pedersen. “The Activation Energy for the FCC Rolling-Texture
Transition as Related to the Activation Energy for Cross Slip”. In: Scripta materialia
46.10 (2002), pp. 741–746. DOI: 10.1016/S1359-6462(02)00065-9.
[69] T. Leffers and R. Ray. “The Brass-Type Texture and Its Deviation from the Copper-Type
Texture”. In: Progress in Materials Science 54.3 (May 2009), pp. 351–396. ISSN: 00796425.
DOI: 10.1016/j.pmatsci.2008.09.002.
[70] P. Lévy. “Sur certains processus stochastiques homogènes”. In: Compositio
Mathematica 7 (1940), pp. 283–339.
[71] G. Leyson, L. Hector, and W. Curtin. “Solute Strengthening from First Principles and
Application to Aluminum Alloys”. In: Acta Materialia 60.9 (May 2012), pp. 3873–3884.
ISSN: 13596454. DOI: 10.1016/j.actamat.2012.03.037.
[72] G. P. M. Leyson et al. “Quantitative Prediction of Solute Strengthening in Aluminium
Alloys”. In: Nature Materials 9.9 (Sept. 2010), pp. 750–755. ISSN: 1476-1122, 1476-4660.
DOI: 10.1038/nmat2813.
[73] J. Liu et al. “Deformation Twinning Behaviors of the Low Stacking Fault Energy
High-Entropy Alloy: An in-Situ TEM Study”. In: Scripta Materialia 137 (Aug. 2017),
pp. 9–12. ISSN: 13596462. DOI: 10.1016/j.scriptamat.2017.05.001.
[74] X.-Y. Liu et al. “EAM Potential for Magnesium from Quantum Mechanical Forces”. In:
Modelling and Simulation in Materials Science and Engineering 4 (1996), pp. 293–303.
DOI: 10.1088/0965-0393/4/3/004.

105
Bibliography

[75] Z. S. Liu and V. Mohles. “Solution Strengthening of Various Elements in Aluminium

Alloys”. In: Materials Science Forum 794-796 (June 2014), pp. 473–478. ISSN: 1662-9752.
DOI: 10.4028/www.scientific.net/MSF.794-796.473.
[76] J. Lytton, L. Shepard, and J. Dorn. “The Activation Energies for Creep of Single
Aluminum Crystals Favorably Oriented for (III)[T0I] Slip”. In: Transactions of the
American Institute of Mining, Metallurgical and Petroleum Engineers 212 (1958), p. 220.
[77] R. Madec, B. Devincre, and L. P. Kubin. “Simulation of Dislocation Patterns in Multislip”.
In: Scripta materialia 47.10 (2002), pp. 689–695. DOI: 10.1016/S1359-6462(02)00185-9.
[78] M. Marcinkowski, K. Sadananda, and N. J. Olson. “Cross Slip of Extended Dislocations
in Disordered Face-Centred-Cubic Alloys”. In: Crystal Lattice Defects 5 (1974),
pp. 187–198.
[79] J. Marian, J. Knap, and M. Ortiz. “Nanovoid Deformation in Aluminum under Simple
Shear”. In: Acta Materialia 53.10 (June 2005), pp. 2893–2900. ISSN: 13596454. DOI:
10.1016/j.actamat.2005.02.046.
[80] W. Meyer and H. Neldel. “Über die Beziehungen zwischen der Energiekonstanten E
und der Mengenkonstanten a in der Leitwerts-Temperaturformel bei oxydischen
Halbleitern”. In: Zeitschrift für technische Physik 18 (1937), pp. 588–593.
[81] G. Mills and H. Jónsson. “Quantum and Thermal Effects in H2 Dissociative Adsorption:
Evaluation of Free Energy Barriers in Multidimensional Quantum Systems”. In: Physical
review letters 72.7 (1994), p. 1124. DOI: 10.1103/PhysRevLett.72.1124.
[82] M. J. Mills and P. Stadelmann. “A Study of the Structure of Lomer and 60◦
Dislocations in Aluminium Using High-Resolution Transmission Electron Microscopy”.
In: Philosophical Magazine A 60.3 (Sept. 1989), pp. 355–384. ISSN: 0141-8610. DOI:
10.1080/01418618908213867.
[83] D. Mordehai, I. Kelson, and G. Makov. “Cross-Slip and Annihilation of
Screw Dislocations in Cu: A Molecular Dynamics Study”. In: Materials Science
and Engineering: A 400-401 (July 2005), pp. 37–39. ISSN: 09215093. DOI:
10.1016/j.msea.2005.03.077.
[84] N. F. Mott. “A Theory of the Origin of Fatigue Cracks”. In: Acta Metallurgica 6.3 (1958),
pp. 195–197. DOI: 10.1016/0001-6160(58)90007-5.
[85] N. F. Mott. “Mechanical Properties of Metals”. In: Physica 15.1-2 (1949), pp. 119–134.
DOI: 10.1016/0031-8914(49)90034-8.
[86] F. R. N. Nabarro. “Dislocations in a Simple Cubic Lattice”. In: Proceedings of the Physical
Society 59.2 (1947), p. 256. DOI: 10.1088/0959-5309/59/2/309.
[87] W. D. Nix, J. C. Gibeling, and D. A. Hughes. “Time-Dependent Deformation of Metals”.
In: Metallurgical Transactions A 16.12 (1985), pp. 2215–2226. DOI: 10.1007/BF02670420.
[88] W. G. Nöhring and W. A. Curtin. “Dislocation Cross-Slip in FCC Solid Solution
Alloys”. In: Acta Materialia 128 (Apr. 2017), pp. 135–148. ISSN: 13596454. DOI:
10.1016/j.actamat.2017.02.027.
[89] W. G. Nöhring and W. A. Curtin. “Thermodynamic Properties of Average-Atom
Interatomic Potentials for Alloys”. In: Modelling and Simulation in Materials Science
and Engineering 24.4 (May 2016), p. 045017. ISSN: 0965-0393, 1361-651X. DOI:
10.1088/0965-0393/24/4/045017.

106
 Bibliography

[90] W. G. Nöhring et al. “Wedge-Shaped Twins and Pseudoelasticity in FCC Metallic

Nanowires under Bending”. In: Extreme Mechanics Letters 8 (Sept. 2016), pp. 140–150.
ISSN: 23524316. DOI: 10.1016/j.eml.2016.03.001.
[91] B. Onat and S. Durukanolu. “An Optimized Interatomic Potential for CuNi Alloys with
the Embedded-Atom Method”. In: Journal of Physics: Condensed Matter 26.3 (Jan.
2014), p. 035404. ISSN: 0953-8984, 1361-648X. DOI: 10.1088/0953-8984/26/3/035404.
[92] E. Oren, E. Yahel, and G. Makov. “Kinetics of Dislocation Cross-Slip: A Molecular
Dynamics Study”. In: Computational Materials Science 138 (Oct. 2017), pp. 246–254.
ISSN: 09270256. DOI: 10.1016/j.commatsci.2017.06.039.
[93] E. Osswald. “Zugversuche an Kupfer-Nickelkristallen”. In: Zeitschrift für Physik 83.1
(1933), pp. 55–78. ISSN: 0044-3328. DOI: 10.1007/BF01331092.
[94] F. Otto et al. “The Influences of Temperature and Microstructure on the Tensile
Properties of a CoCrFeMnNi High-Entropy Alloy”. In: Acta Materialia 61.15 (Sept. 2013),
pp. 5743–5755. ISSN: 13596454. DOI: 10.1016/j.actamat.2013.06.018.
[95] R. Peierls. “The Size of a Dislocation”. In: Proceedings of the Physical Society 52.1 (1940),
p. 34.
[96] S. Plimpton. “Fast Parallel Algorithms for Short-Range Molecular Dynamics”. In:
Journal of Computational Physics 117 (1995), pp. 1–19. DOI: 10.1006/jcph.1995.1039.
[97] J. Poirier. “On the Symmetrical Role of Cross-Slip of Screw Dislocations and Climb
of Edge Dislocations as Recovery Processes Controlling High-Temperature Creep”.
In: Revue de Physique Appliquée 11.6 (1976), pp. 731–738. ISSN: 0035-1687. DOI:
10.1051/rphysap:01976001106073100.
[98] G. Purja Pun and Y. Mishin. “Development of an Interatomic Potential for the
Ni-Al System”. In: Philosophical Magazine 89.34-36 (Dec. 2009), pp. 3245–3267. ISSN:
1478-6435, 1478-6443. DOI: 10.1080/14786430903258184.
[99] W. Püschl and G. Schoeck. “Calculation of Cross-Slip Parameters in FCC
Crystals”. In: Materials Science and Engineering: A 164.1 (1993), pp. 286–289. DOI:
10.1016/0921-5093(93)90679-9.
[100] W. Püschl. “Models for Dislocation Cross-Slip in Close-Packed Crystal Structures:
A Critical Review”. In: Progress in materials science 47.4 (2002), pp. 415–461. DOI:
10.1016/S0079-6425(01)00003-2.
[101] B. R. Ramírez, N. Ghoniem, and G. Po. “Ab Initio Continuum Model for the Influence of
Local Stress on Cross-Slip of Screw Dislocations in FCC Metals”. In: Physical Review B
86.9 (Sept. 2012). ISSN: 1098-0121, 1550-235X. DOI: 10.1103/PhysRevB.86.094115.
[102] S. I. Rao, T. A. Parthasarathy, and C. Woodward. “Atomistic Simulation of Cross-Slip
Processes in Model FCC Structures”. In: Philosophical Magazine A 79.5 (May 1999),
pp. 1167–1192. ISSN: 0141-8610, 1460-6992. DOI: 10.1080/01418619908210354.
[103] S. Rao et al. “Activated States for Cross-Slip at Screw Dislocation Intersections in
Face-Centered Cubic Nickel and Copper via Atomistic Simulation”. In: Acta Materialia
58.17 (2010), pp. 5547–5557. ISSN: 13596454. DOI: 10.1016/j.actamat.2010.06.005.
[104] S. Rao et al. “Atomistic Simulations of Cross-Slip Nucleation at Screw Dislocation
Intersections in Face-Centered Cubic Nickel”. In: Philosophical Magazine 89.34-36 (Dec.
2009), pp. 3351–3369. ISSN: 1478-6435, 1478-6443. DOI: 10.1080/14786430903286201.

107
Bibliography

[105] S. Rao et al. “Atomistic Simulations of Dislocation Behavior in a Model FCC

Multicomponent Concentrated Solid Solution Alloy”. In: Acta Materialia 134 (Aug.
2017), pp. 188–194. ISSN: 13596454. DOI: 10.1016/j.actamat.2017.05.071.
[106] S. Rao et al. “Atomistic Simulations of Surface Cross-Slip Nucleation in Face-Centered
Cubic Nickel and Copper”. In: Acta Materialia 61.7 (Apr. 2013), pp. 2500–2508. ISSN:
13596454. DOI: 10.1016/j.actamat.2013.01.026.
[107] S. Rao et al. “Calculations of Intersection Cross-Slip Activation Energies in FCC
Metals Using Nudged Elastic Band Method”. In: Acta Materialia 59.19 (Nov. 2011),
pp. 7135–7144. ISSN: 13596454. DOI: 10.1016/j.actamat.2011.08.029.
[108] S. Rao et al. “Screw Dislocation Cross Slip at Cross-Slip Plane Jogs and
Screw Dipole Annihilation in FCC Cu and Ni Investigated via Atomistic
Simulations”. In: Acta Materialia 101 (Dec. 2015), pp. 10–15. ISSN: 13596454. DOI:
10.1016/j.actamat.2015.08.070.
[109] S. Rao et al. “Spontaneous Athermal Cross-Slip Nucleation at Screw Dislocation
Intersections in FCC Metals and L1 2 Intermetallics Investigated via Atomistic
Simulations”. In: Philosophical Magazine 93.22 (Aug. 2013), pp. 3012–3028. ISSN:
1478-6435, 1478-6443. DOI: 10.1080/14786435.2013.799788.
[110] T. Rasmussen et al. “Atomic Structure and Energetics of Constricted Screw Dislocations
in Copper”. In: Materials Science and Engineering: A 234-236 (1997), pp. 544–547. DOI:
10.1016/S0921-5093(97)00311-0.
[111] T. Rasmussen et al. “Simulation of Structure and Annihilation of Screw Dislocation
Dipoles”. In: Philosophical Magazine A 80.5 (May 2000), pp. 1273–1290. ISSN: 0141-8610,
1460-6992. DOI: 10.1080/01418610008212115.
[112] T. Rasmussen et al. “Atomistic Determination of Cross-Slip Pathway and Energetics”.
In: Physical review letters 79.19 (1997), p. 3676. DOI: 10.1103/PhysRevLett.79.3676.
[113] T. Rasmussen et al. “Simulations of the Atomic Structure, Energetics, and Cross Slip
of Screw Dislocations in Copper”. In: Physical Review B 56.6 (1997), p. 2977. DOI:
10.1103/PhysRevB.56.2977.
[114] W. Read. Dislocations in Crystals. International Series in Pure and Applied Physics. New
York, Toronto, London: McGraw-Hill Book Company, Inc., 1953.
[115] E. Rodary et al. “Dislocation Glide in Model Ni(Al) Solid Solutions by Molecular
Dynamics”. In: Physical Review B 70.5 (Aug. 2004). ISSN: 1098-0121, 1550-235X. DOI:
10.1103/PhysRevB.70.054111.
[116] S. Ryu, K. Kang, and W. Cai. “Entropic Effect on the Rate of Dislocation Nucleation”. In:
Proceedings of the National Academy of Sciences 108.13 (2011), pp. 5174–5178.
[117] G. Saada. “Cross-Slip and Work Hardening of FCC Crystals”. In: Materials Science and
Engineering: A 137 (1991), pp. 177–183. DOI: 10.1016/0921-5093(91)90333-I.
[118] S. Saroukhani and D. Warner. “Investigating Dislocation Motion through a Field of
Solutes with Atomistic Simulations and Reaction Rate Theory”. In: Acta Materialia 128
(2017), pp. 77–86. DOI: 10.1016/j.actamat.2017.02.001.
[119] R. O. Scattergood and D. J. Bacon. “The Orowan Mechanism in Anisotropic Crystals”.
In: Philosophical Magazine 31.1 (1975), pp. 179–198. DOI: 10.1080/14786437508229295.

108
 Bibliography

[120] G. Schoeck. “Beiträge zur Theorie der Versetzungen: Aufspaltung, Linienenergie und
Quergleitung von Versetzungen”. PhD thesis. Technische Hochschule Stuttgart, 1954.
[121] G. Schoeck. “The Cross-Slip Energy Unresolved”. In: Philosophical Magazine
Letters 89.8 (Aug. 2009), pp. 505–515. ISSN: 0950-0839, 1362-3036. DOI:
10.1080/09500830903092399.
[122] G. Schoeck. “Thermodynamics and Thermal Activation of Dislocations”. In:
Dislocations in Solids. Ed. by F. R. N. Nabarro. Vol. 3. Amsterdam, New York, Oxford:
North-Holland Publishing Company, 1980, pp. 65–163. ISBN: 0-444-85015-5.
[123] G. Schoeck and W. Püschl. “Prismatic Slip in HCP Metals”. In: Proceedings of the 8th
International Conference on the Strength of Metals and Alloys. Ed. by P. Kettunen,
T. Lepistö, and M. Lehtonen. Elsevier Ltd., 1988, pp. 239–244.
[124] G. Schoeck and A. Seeger. “Activation Energy Problems Associated with Extended
Dislocations”. In: Report of a Conference on Defects in Crystalline Solids. London:
Physical Society, 1955, pp. 340–346.
[125] A. Seeger. “The Mechanism of Glide and Work Hardening in Face-Centered Cubic
and Hexagonal Close-Packed Metals”. In: Dislocations and Mechanical Properties of
Crystals. Ed. by J. Fisher et al. John Wiley & Sons, Inc., 1957, pp. 243–329.
[126] A. Seeger, R. Berner, and H. Wolf. “Die experimentelle Bestimmung von
Stapelfehlerenergien kubisch-flächenzentrierter Metalle”. In: Zeitschrift für Physik
155.2 (1959), pp. 247–262.
[127] D. Sheppard, R. Terrell, and G. Henkelman. “Optimization Methods for Finding
Minimum Energy Paths”. In: The Journal of Chemical Physics 128.13 (2008), p. 134106.
ISSN: 00219606. DOI: 10.1063/1.2841941.
[128] C. Singh, A. Mateos, and D. Warner. “Atomistic Simulations of Dislocation-Precipitate
Interactions Emphasize Importance of Cross-Slip”. In: Scripta Materialia 64.5 (Mar.
2011), pp. 398–401. ISSN: 13596462. DOI: 10.1016/j.scriptamat.2010.10.041.
[129] R. E. Smallman and D. Green. “The Dependence of Rolling Texture on
Stacking Fault Energy”. In: Acta Metallurgica 12.2 (1964), pp. 145–154. DOI:
10.1016/0001-6160(64)90182-8.
[130] N. V. Smirnov. “On the Estimation of the Discrepancy Between Empirical Curves of
Distribution for Two Independent Samples”. In: Bulletin mathématiques de l’Université
de Moscou 2 (1939), pp. 3–14.
[131] R. W. Smith and G. S. Was. “Application of Molecular Dynamics to the Study of
Hydrogen Embrittlement in Ni-Cr-Fe Alloys”. In: Physical Review B 40.15 (1989),
pp. 10332–10336. DOI: 10.1103/PhysRevB.40.10322.
[132] J. W. Steeds and P. Hazzledine. “Dislocation Configurations in Deformed Copper and
Copper 10%(Atomic) Aluminium Alloy”. In: Discussions of the Faraday Society 38
(1964), pp. 103–110. DOI: 10.1039/DF9643800103.
[133] A. N. Stroh. “Constrictions and Jogs in Extended Dislocations”. In: Proceedings of the
Physical Society. Section B 67.5 (1954), p. 427. DOI: 10.1088/0370-1301/67/5/307.
[134] A. Stroh. “Dislocations and Cracks in Anisotropic Elasticity”. In: Philosophical
Magazine 3.30 (1958), pp. 625–646. DOI: 10.1080/14786435808565804.

109
Bibliography

[135] A. Stukowski. “Visualization and Analysis of Atomistic Simulation Data with

OVITO–the Open Visualization Tool”. In: Modelling and Simulation in Materials
Science and Engineering 18.1 (Jan. 2010), p. 015012. ISSN: 0965-0393, 1361-651X. DOI:
10.1088/0965-0393/18/1/015012.
[136] A. Stukowski, V. V. Bulatov, and A. Arsenlis. “Automated Identification and Indexing
of Dislocations in Crystal Interfaces”. In: Modelling and Simulation in Materials
Science and Engineering 20.8 (Dec. 2012), p. 085007. ISSN: 0965-0393, 1361-651X. DOI:
10.1088/0965-0393/20/8/085007.
[137] B. A. Szajewski, A. Hunter, and I. J. Beyerlein. “The Core Structure and Recombination
Energy of a Copper Screw Dislocation: A Peierls Study”. In: Philosophical
Magazine 97.25 (Sept. 2017), pp. 2143–2163. ISSN: 1478-6435, 1478-6443. DOI:
10.1080/14786435.2017.1328138.
[138] T. S. van Erp, D. Moroni, and P. G. Bolhuis. “A Novel Path Sampling Method for the
Calculation of Rate Constants”. In: The Journal of Chemical Physics 118.17 (May 2003),
pp. 7762–7774. ISSN: 0021-9606, 1089-7690. DOI: 10.1063/1.1562614.
[139] E. Vanden-Eijnden and M. Venturoli. “Revisiting the Finite Temperature String Method
for the Calculation of Reaction Tubes and Free Energies”. In: The Journal of Chemical
Physics 130.19 (2009), p. 194103. ISSN: 0021-9606, 1089-7690. DOI: 10.1063/1.3130083.
[140] C. Varvenne, A. Luque, and W. A. Curtin. “Theory of Strengthening in FCC High
Entropy Alloys”. In: Acta Materialia 118 (Oct. 2016), pp. 164–176. ISSN: 13596454. DOI:
10.1016/j.actamat.2016.07.040.
[141] C. Varvenne et al. “Average-Atom Interatomic Potential for Random Alloys”. In: Physical
Review B 93.10 (2016), p. 104201. DOI: 10.1103/PhysRevB.93.104201.
[142] T. Vegge et al. “Atomistic Simulations of Cross-Slip of Jogged Screw Dislocations
in Copper”. In: Philosophical Magazine Letters 81.3 (Mar. 2001), pp. 137–144. ISSN:
0950-0839, 1362-3036. DOI: 10.1080/09500830010019040.
[143] T. Vegge. “Atomistic Simulations of Screw Dislocation Cross Slip in Copper and
Nickel”. In: Materials Science and Engineering: A 309 (2001), pp. 113–116. DOI:
10.1016/S0921-5093(00)01624-5.
[144] T. Vegge and K. W. Jacobsen. “Atomistic Simulations of Dislocation Processes
in Copper”. In: Journal of Physics: Condensed Matter 14.11 (2002), p. 2929. DOI:
10.1088/0953-8984/14/11/309.
[145] T. Vegge et al. “Determination of the of Rate Cross Slip of Screw Dislocations”. In:
Physical review letters 85.18 (2000), p. 3866. DOI: 10.1103/PhysRevLett.85.3866.
[146] G. H. Vineyard. “Frequency Factors and Isotope Effects in Solid State Rate Processes”.
In: Journal of Physics and Chemistry of Solids 3.1 (1957), pp. 121–127. DOI:
10.1016/0022-3697(57)90059-8.
[147] W. Voigt. Lehrbuch der Kristallphysik. Teubner, Leipzig, 1910.
[148] A. F. Voter and S. P. Chen. “Accurate Interatomic Potentials for Ni, Al and Ni3Al”. In:
MRS Proceedings 82 (1987), p. 175. DOI: 10.1557/PROC-82-175.
[149] S. van der Walt, S. C. Colbert, and G. Varoquaux. “The NumPy Array: A Structure for
Efficient Numerical Computation”. In: Computing in Science & Engineering 13.2 (2011),
pp. 22–30. DOI: 10.1109/MCSE.2011.37.

110
 Bibliography

[150] J. Washburn. “Intersection Cross Slip”. In: Applied Physics Letters 7.7 (1965), p. 183. ISSN:
00036951. DOI: 10.1063/1.1754369.
[151] M. Wen, S. Fukuyama, and K. Yokogawa. “Cross-Slip Process in FCC Nickel with
Hydrogen in a Stacking Fault: An Atomistic Study Using the Embedded-Atom
Method”. In: Physical Review B 75.14 (Apr. 2007). ISSN: 1098-0121, 1550-235X. DOI:
10.1103/PhysRevB.75.144110.
[152] M. Wen, S. Fukuyama, and K. Yokogawa. “Hydrogen-Affected Cross-Slip Process in
FCC Nickel”. In: Physical Review B 69.17 (May 2004). ISSN: 1098-0121, 1550-235X. DOI:
10.1103/PhysRevB.69.174108.
[153] H. Wolf. “Die Aktivierungsenergie für die Quergleitung aufgespaltener
Schraubenversetzungen”. In: Zeitschrift für Naturforschung 15A (1960), pp. 180–193.
DOI: 10.1515/zna-1960-0302.
[154] H. Wolf. “Die Aktivierungsenergie für die Quergleitung aufgespaltener
Schraubenversetzungen”. PhD thesis. Technische Hochschule Stuttgart, 1959.
[155] Wolfram Research, Inc. Mathematica. Champaign, Illinois, 2016.
[156] Z. Wu and W. A. Curtin. “Mechanism and Energetics of <c+a> Dislocation Cross-Slip in
HCP Metals”. In: Proceedings of the National Academy of Sciences 113.40 (Oct. 2016),
pp. 11137–11142. ISSN: 0027-8424, 1091-6490. DOI: 10.1073/pnas.1603966113.
[157] S. X. Xia and A. El-Azab. “A Preliminary Investigation of Dislocation Cell Structure
Formation in Metals Using Continuum Dislocation Dynamics”. In: IOP Conference
Series: Materials Science and Engineering 89 (Aug. 2015), p. 012053. ISSN: 1757-8981,
1757-899X. DOI: 10.1088/1757-899X/89/1/012053.
[158] S. Xia and A. El-Azab. “Computational Modelling of Mesoscale Dislocation Patterning
and Plastic Deformation of Single Crystals”. In: Modelling and Simulation in Materials
Science and Engineering 23.5 (July 2015), p. 055009. ISSN: 0965-0393, 1361-651X. DOI:
10.1088/0965-0393/23/5/055009.
[159] S. Xia, J. Belak, and A. El-Azab. “The Discrete-Continuum Connection in Dislocation
Dynamics: I. Time Coarse Graining of Cross Slip”. In: Modelling and Simulation in
Materials Science and Engineering 24.7 (Oct. 2016), p. 075007. ISSN: 0965-0393,
1361-651X. DOI: 10.1088/0965-0393/24/7/075007.
[160] S. Xu et al. “Shear Stress- and Line Length-Dependent Screw Dislocation Cross-Slip
in FCC Ni”. In: Acta Materialia 122 (Jan. 2017), pp. 412–419. ISSN: 13596454. DOI:
10.1016/j.actamat.2016.10.005.
[161] J. A. Yasi, L. G. Hector, and D. R. Trinkle. “Prediction of Thermal Cross-Slip Stress in
Magnesium Alloys from Direct First-Principles Data”. In: Acta Materialia 59.14 (Aug.
2011), pp. 5652–5660. ISSN: 13596454. DOI: 10.1016/j.actamat.2011.05.040.
[162] J.-W. Yeh et al. “Nanostructured High-Entropy Alloys with Multiple Principal Elements:
Novel Alloy Design Concepts and Outcomes”. In: Advanced Engineering Materials 6.5
(May 2004), pp. 299–303. ISSN: 1438-1656, 1527-2648. DOI: 10.1002/adem.200300567.

111
Acronyms
F-E Friedel-Escaig. 3, 5, 8–12, 24, 28–30, 51

BEP Bell-Evans-Polanyi. 45

CNA Common Neighbor Analysis. 22

EAM Embedded Atom Method. 17–19

FCC Face-Centered Cubic. iii, 1–6, 8, 11, 13, 17–19, 22, 26, 29, 30, 42, 49, 52, 71, 73, 77, 80

HCP Hexagonal Close-Packed. 3, 6

HEA High Entropy Alloy. 19, 73

MEP Minimum Energy Path. 14, 22–24, 84

NEB Nudged Elastic Band. 20, 22–24, 81–83, 85

PES Potential Energy Surface. 14, 15, 23, 24

113
WOLFRAM GEORG NÖHRING e-mail: wolfram.nohring@epfl.ch
CURRICULUM VITAE

EDUCATION

2014–2018 PhD EPFL

2010–2013 M.Sc. (hons.), Advanced Materials and Processes,
Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany
Master thesis: Towards a Finite Element Multiscale Model of the
Accumulative Roll Bonding Process
2007–2010 B.Sc. (with distinction), Materials Science,
Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany
Bachelor thesis: Atomistic Simulations of Single Crystal Nanostructures
Subjected to Bending Loads
1997–2006 Allgemeine Hochschulreife (general qualification for university entrance)
Friedrich-Alexander-Gymnasium, Neustadt a. d. Aisch, Germany

PUBLICATIONS IN PEER-REVIEWED JOURNALS

1. W. G. Nöhring and W. A. Curtin (Apr. 2017). “Dislocation Cross-Slip in Fcc Solid Solution
Alloys”. In: Acta Materialia 128, pp. 135–148

2. W. G. Nöhring and W. A. Curtin (May 2016c). “Thermodynamic Properties of

Average-Atom Interatomic Potentials for Alloys”. In: Modelling and Simulation in
Materials Science and Engineering 24.4, p. 045017

3. W. G. Nöhring, J. J. Möller, Z. Xie, and E. Bitzek (Sept. 2016). “Wedge-Shaped Twins and
Pseudoelasticity in FCC Metallic Nanowires under Bending”. In: Extreme Mechanics
Letters 8, pp. 140–150

4. C. Varvenne, A. Luque, W. G. Nöhring, and W. A. Curtin (2016). “Average-Atom

Interatomic Potential for Random Alloys”. In: Physical Review B 93.10, p. 104201

5. A. Prakash, W. G. Nöhring, R. A. Lebensohn, H. W. Höppel, and E. Bitzek (Apr. 2015). “A

Multiscale Simulation Framework of the Accumulative Roll Bonding Process Accounting
for Texture Evolution”. In: Materials Science and Engineering: A 631, pp. 104–119

PRESENTATIONS AT CONFERENCES

1. W. G. Nöhring and W. A. Curtin (2016b). Cross-Slip in FCC Solid Solutions. Presentation.

MMM, Dijon

2. W. G. Nöhring and W. A. Curtin (2016a). Cross-Slip in FCC Solid Solutions. Presentation.

EMMC15, Brussels

3. W. G. Nöhring, E. Bitzek, and Y. Umeno (2013). Atomic Scale Analysis of Structural

Instability in Nanostructures. Poster. ICMM3, Warsaw

115
 4. W. G. Nöhring, A. Prakash, and E. Bitzek (2013). A Coupled Finite Element Polycrystal
Texture Model of the Accumulative Roll Bonding Process. Presentation. ICMM3, Warsaw

5. W. G. Nöhring, E. Bitzek, A. Kelling, and C. A. Volkert (2012). Deformation Mechanisms of

Twinned Au-Nanoparticles under Uniaxial Compression. Presentation. IUMRS-IUMRS
2012, Yokohama

6. W. G. Nöhring, E. Bitzek, and Y. Umeno (2012). Atomic Scale Analysis of Structural

Instability in Nanostructures. Poster. IUMRS-IUMRS 2012, Yokohama

7. A. Kelling, C. A. Volkert, W. G. Nöhring, and E. Bitzek (Mar. 2012). Deformation

Mechanisms of Twinned Au-Nanoparticles under Compression: Experiments and
Simulation. Presentation. DPG Spring Meeting, Berlin

ARTICLES IN CONFERENCE PROCEEDINGS

1. Y. Umeno, W. G. Nöhring, A. Iskandarov, and E. Bitzek (Nov. 2013). “Atomistic Model

Analysis of Local and Global Instabilities in Crystals at Finite Temperature”. In: Key
Engineering Materials. Vol. 592-593, pp. 39–42

REFEREE SERVICES

1. JOM – The Journal of The Minerals, Metals & Materials Society (TMS)

AWARDS

2017 Poster prize of the Gordon Research Conference on Physical Metallurgy

2012 Best student award of the master program Advanced Materials and Processes

TEACHING

2017 Supervision of a Bachelor student project

2015 & 2016 Tutor for “Introduction à la mécanique des structures” (lecturer:Matteo Galli)

116